14 Episode results for "Sumo Logic"

Continuous Intelligence with Kalyan Ramanathan

Software Engineering Daily

45:24 min | 1 year ago

Continuous Intelligence with Kalyan Ramanathan

"Logging provides raw data. That can be abstracted into higher level information logs are generated at every layer of infrastructure physical host virtual tool machine container pod and Kuban Andy's cluster logs are generated by network proxies servers. API requests and. There's far too much logging information to be read by humans log messages need to be refined into statistical metrics that can be put into charts a high volume. I'm of log. Messages can be used to detect anomalies across the system. If unusual behavior is present in a system the relevant log messages can be identified and sent to a human operator for that operator to triage and respond to Kellyanne Ramanathan works at sumo logic. A platform for log management and continuous intelligence sumo logic recently published the continuous intelligence report which is based on a study of over two thousand technology companies. It's a a useful data set for anyone who is looking to understand. Adoption of cloud products and Cooper netties. And it can be found at software engineering daily Dot Com slash sumo logic chick Kellyanne joins the show to discuss log management continuous intelligence and the data that sumo logic has gathered in the continuous intelligence report full disclosure. Sumo logic is a sponsor of software engineering daily Kellyanne Ramadan. Welcome to software engineering daily. Oh thank you very much today. We're talking about logging and log management and some higher level all discussions beyond that to start with the idea of logging. Logging is raw data. That raw data gets abstracted attracted into higher level information. Logging is often higher dimensional. It's high volume. Give some examples of common logging information and how that high volume high dimensional data it gets aggregated into something more abstract and usable. Yeah scrape glisten Jeffrey Great. Let me start by explaining waterlog For many of your listeners so a log is essentially a record of noteworthy events that are generated greater from software applications operating systems infrastructure. You name it right and log essentially keep track so of what is happening behind the scenes of the application and the system so that you know you have a detailed list off events that are happening so when you have a malfunction when you can go back you can copy law you can see what is working not working etcetera now every application stack component. Right right right from the application to the infrastructure to the you know the platforms may be on Prem or cloud for that matter that the application is deployed on are essentially emitting logs and what these logs do is as I mentioned before provides the ability to understand how these components of the application are working so applications are logs are typically written out by APP developers and they are tracking just about anything that is happening with an with an application. It could be a log in. It could be a transaction it could be errors or warnings that are happening in the application infrastructure. Logs logs are generally come prepackaged from infrastructure vendors and they generally give you visibility into the health and performance and any other interesting invents that are associated with infrastructure itself. My sense is that while logs have been used for as long as we've been doing software engineering the idea of aggregating all of this logging data and storing it and making sense of it is a newer phenomenon on. It has something to do with the cloud. But I'd like to hear your perspective. You've definitely been in the industry for longer than I have. You've seen how the world looked pre pre and post cloud was log management a thing before the cloud log management has been around for a long time right but it has existed existed in a very different form and perhaps what it exists today with the advent of cloud and You know high scale infrastructure you know this notion of infrastructure you know particularly network devices and operating systems writing logs and perhaps using assist log type protocol dropping logs into a centralized store that you could perhaps query and You you know get some visibility and perhaps do some troubleshooting has been around for a long time. This is a You Know Twenty Twenty ten or twenty twenty phenomenon This existed from the days of when Cisco Routers and switches and whatnot. Have you know have have been a common commonplace infrastructure within a data center. What has changed I would say in the last ten to fifteen years is the ability now to actually collect all these data's at very large it's kill and then more importantly the ability to to analyze the stayed at an even larger scale? So that now you can one identified needles in a haystack. Venue running into an application malfunction or a security issue but then your application or two. You can do very long-term mm trending off this data so that you can understand you know how you application is perhaps performing over. Let's say last one year or you know how many users I saw you get into your application in the last six months where are these users coming from. What are they doing with your application or perhaps look at security incidents that you may have had in your in your application going back months and months of Right so what the cloud provides you and what these high scales storage and analytics engine so providing you is it is the ability to indeed do this at scale which was not quite available once upon a time so once upon a time log management used to be the review view of maybe a team of five or six people now be of customers of products like similar logic we have hundreds or even thousands of people. Oh who are actively and concurrently using these solutions until the trends of the cloud have made modern modern log management solutions possible the cloud has also changed how infrastructure itself is laid out and how we're consuming software. You have the more layers of infrastructure. You have physical hosts you have. VM's you have coober netease you've pods and containers you have black box proprietary prior terry tools like Dynamo. DB The you don't really have a ton of introspection into with all this heterogeneity you. Can you describe some of the Best Practices for how to monitor all of these different layers all of these different systems. I mean it's an interesting question. The basic hypothesis right off monitoring starts out with visibility right that are multiple terms that people love apply to to visibility. Some people call it observability. Some people call it transparency. You name it but at the Harvard Right. You cannot manage what you don't see and what you don't understand so the essence of all of this starts out with. I need to bring back doc. All the signals may be logs and metrics and traces and you name it into a into a central depository so that I can understand stand the various components that makes up my application all the way from my application the platform as a service component own into the infrastructure as a service component. You know allow the host Sunday. VM's and the Containers that you just mentioned. So the essence of monitoring thank and the essence of understanding an application. Is that visibility. The the next thing that he had to do once you get control of visibility once you have a a good understanding the application is to is to stock monitoring the application and what that means is is to understand what is normal and what is abnormal in the behaved at the application itself. Now that are multiple techniques by which you know you can do this you can do it static methods you can do it dynamic methods and there are even machine learning or ai based approaches to understanding what is normal and what's not normal about an application and then finally when you have identified you no an abnormal condition. Obviously that's what you need to take some action. You need the right tools and the right analytics right troubleshooting systems in place so that you know you can very quickly understand the root causes of problems and then you can fix. This does does problems itself now. This problem has always existed. It has and the basic process has has always been the same whether it is on paramore the cloud and what changes in the cloud is exactly what you said. Jeffrey Edge. That ah you know the a lot of things change in the clock right starting from the cloud platform to the infrastructure that you deploy to the architecture of the applications that you're driving to even the teams that are responsible for keeping the application healthy secure. You name it and complexity of the cloud environment the newness missile declawed environment the number of components in this cloud environment. That's what makes this the act and the task of keeping cloud applications so much much harder and with like the addition of a system that allows me to have some observability over my infrastructure structure. How am I using that like? Who are the people that are using a log management system who are the constituents as is it just engineers nears or is it also product managers? Does everybody on the team wants to be using this kind of observability software. Yeah made made I think It's interesting here at some logic. Right we see many up. Personas who use the data that comes from applications Signals applications such as lock right. Be Obviously see in a devops teams and `sorry teams that are very interested in monitoring monitoring and ensuring performance and availability of applications. Logs are also at the heart of ensuring the security and compliance posture for navigation. So if you were to walk into a typical enterprise today and look up the security operations team. Chances are that they are looking being at logs and events coming from these applications to identify security threats respond to these threats to ensure the compliance date may be PCIE. Ah By you name it for the application itself but also starting to see logs being used in in in other interesting race you know a lot of the development tools Anything from you know get hub to CIC DI tools to deployment. Tools are right logs so if I am a VP of engineering. An I don. I want to get complete visibility into my devops process. You know the number of polls that I am making from guitar the number of times. I'm writing I am running my cic De workflows. The number of deployments that. I'm doing my production environment. Chances are that you know log management system that integrates many of these systems custom skin. Indeed provide you that complete into invisibility. Finally we're seeing logs You know being used in many newer applications for for example IOT applications. We have customers who are who right out iot logs and guess what if he can bring these IOT logs into a centralized I system you can learn a lot about your iot systems. You can see where these systems are being deployed. You can see how they are working or not working for that matter and it provides a lot of ability into not just your application but also the business that is being driven by your application you mentioned anomalies as little bit earlier you want your blog management system to detect anomalies because you cannot manually manually. Look at everything that is coming in to your log management system. I mean that's that's the whole point but the problem with out of the box anomaly. Early detection is that anomalies vary from infrastructure infrastructure. It might make sense for some companies needs to have a system with a Q.. PS of ten billion per minute and some other system if ten if it hits ten billion cube ps per a minute then that's dramatically arrogance q ps queries per second so permit. It doesn't even make any sense. But in any case anomaly is kind of a subjective phenomenon on an anomaly depends on what the infrastructure is. So how do you want to define anomalies for a log management enjoyment or a observability platform. Yeah I think you hitting on a problem. That has the speed around for a long time and Kamini observability systems and it doesn't have to be logs in particular. I mean you know. Even systems that collect signals and metrics from other from infrastructure and applications have have long been Stymied by these problems. There was some standards by which you know you can try address us right and missile will walk you through the different approaches. Here do this the way way off. You know setting up what I call Static thresholds right and Starting special orders. Where you know you may say that if my infrastructure produces more than X. number of warnings or editors per minute fight off in alarm or if my application has a latency of fled say more than twenty seconds? Fight off on an arm right So the there are many systems. That sort of work with these static anomaly based trust shorts. And that's been around on for a long time and the and the problem is exactly what you mentioned Ryan Bridges. How do you know what is that right threshold set? How do you know if ten seconds is the right pressure orchard? It'd it'd be it'd be thirty. I mean it's that's pretty much dependent on the on the infrastructure the application the particular use case. You Name Right Byu you've seen system sort of obviously improve from the static anomaly based detection mechanism and systems generally do that Through one of two please write one of two ways. One is applied dynamic. Tush old and dynamic thresholds is where the the system itself understands as to what what is the the right threshold that needs to be applied and now this can come through multiple mechanisms it can either add or multiple methods. I should say you can do do this based on historical basis so for example. You're an e commerce site and you know you want to try shored award for let's say ten. Am in the morning and let's say six PM in the are seven pm in the evening right. Typically Commerce sites generally will see more traffic Little little date later in the day given that people are going home. Now they're going to be transacting on. Does ECOMMERCE system you name it. The time or temporal system can look at Komo patterns of usage of the system over time and can apply in a threshold based on patterns of use it. So every day at six EX-PM I expect a certain treasured or a certain performance benchmarks for my application. And you know I can go back over the last one week. I can go back over the last one month and then I can arrive at what that threshold ought to be and I can said that as a benchmark so here's an example of an automatic automatic threat ashore or benchmark that is set by the system itself now another way of setting up. These dynamic anomalies is arriving at it through statistical. Means so what you can do as you can look at you know typical Data set and then you can arrive at. What is the normal for that data set and then set your treasure as a certain standard deviation from that data set so for example you know if you expected a certain performance at six? PM Am every day. Look at your last ten hours arrive at what does a normal for the last ten hours and then say anytime you have more than one two three configuration legal standard deviation from that from that point in time you set up a tra- shown and you set up an alert as a result of this so there are multiple ways as you can see by which you know you can set setup a threshold and thus identify an anomaly within the platform. Now the most recent way and I would say the the way that most people plus starting to experiment with Anomaly Management Right now is through these by using advanced technologies like like eme alumni where you can apply hi in machine learning technologies to now start to identify waters normal. And what's up normal. This is obviously brand new ground. You know still in experimentation it face some of these techniques work some of them. Don't work that well but you're starting to see more and more companies now start to work with. Mla Technologies to do identify in defy anomalies in their system. Can you give an example of how you would want your your log management system or your observability system to expose machine learning functionality. Yeah let me give you a good example of how sumo logic dusters right in. Sumo logic collects a lot of logs. Tens of pedal bites of logs from unique customers right. So it's impossible to sift through Penn.. peto bites of logs does this on a daily basis due to identify patterns of things that you may seen a lot right so what similar logic does is You know apply machine learning technologies technologies like clustering for that matter to two then Very quickly you know summarizes logs into pattern so imagine a log. UGH Data said that may have. Let's thirty four fifty thousand logs that maybe arriving into your system by applying technology a proprietary technology recall log reduced in our in our system. We can quickly now summarize these code forty thousand logs into in about ten to twenty patterns so rather within sifting through each one of these logs are trying to search through these logs and identifying this what might be happening in this lock said you better off by looking king at these. The ten log patterns Identifying that summarizing that you know eight of these ten lakhpat onto perfectly acceptable and normal for. Oh you system. But these things that were collected by some logic and that where reduced by similar logic into patterns are indeed unique lot doc patterns and perhaps those are the things that you should be paying attention to as you're trying to troubleshoot perhaps a problem in your system. So that's a perfect example of off how machine learning can be applied to large-scale problems in this case Log analytics to help you quickly identified fight. You know the needle in a haystack and then focus on just the right things other than get Drowned by the by the complexity and the scale of your log index problems. The Vision that your company's sumo logic has for how log management and an observability tools and metrics tools of today evolve. Is something you call continuous intelligence. Can you describe describe what continuous intelligence means. Yeah I mean it. Sort of goes back to the discussion. That have been having Jeffrey. What we are seeing is that menu for customers are making the move to modern applications and to cloud based systems and a venue run? Your applications is in cloud systems. You know what we are seeing that just about everything changes with your application Ryan. Your platform is different. Your your application. Infrastructure infrastructure is different your architecture's different. The teams that are involved in this management are different than some sense. You know. You're going from DAB SACK DOC and OPS team that were unique teams different teams silo teams. Do perhaps even Dev ops teams where there's one team that is responsible for or for managing this application one of the things that we see in this world. Is that your applications and also starting to be developed and released at a much much faster taste than you ever did before so what we are seeing is that an you know this is our hypothesis and this is the company I should say is that you need continuous genius intelligence. You know to run and secure these applications in a different way and in a better way. So what soon logics. Continuous intelligence platform enabled you to do is to essentially provide you that continues intelligence so that you can manage these applications that you can build run and secure these applications applications so the end better and meet the expectations of the business. Whenever there is kind of new term like this like continuous tenuous intelligence? It's often summarizing. Some lower level technologies. That have come before it. Can you identify what are the specific acidic technologies that are composed into continuous intelligence. What make makes that possible? Yeah I mean. At the heart of Katina's genus intelligence for sumo logic. Right is our service itself are so as the some logic service runs in the cloud and it's a cloud native service service and it's really the underpinning of what delivers continues. Intelligence to our customers are cloud NATO service leverages every aspect aspect of the cloud that we run on beer cloud native on Aws Amazon Web Services and what it does is to provide our customers marce incredible saleability Lasta city so our service collects data from you know your applications your applications and your infrastructure. Choose which can run anywhere right. These applications may sometimes run in the cloud may sometimes are on prime may sometimes be in hyperion environment and because we run in the a cloud Very much like like salesforce. All you have to do is send us your data be our service collects the stater and you know you can start to analyze this data You know in real time With the NAR service I mentioned that are so is runs in the cloud and therefore it provides incredible skill ability elasticity and You know let me sort of emphasis emphasis that a little bit. You know the the problems that are trying to address our big data problems the problems that are trying to address our problems where scaling Alan Elasticity is very important. You know we have customers who who have you know dynamic apps that some days may see you know let's say a ten thousand users and other days may see a million users when you see that level of dynamism in application and application usage the machine data that comes from these applications a signal. Does that come from these applications also dramatically and therefore if you need a management system that can collect all these signals and that can analyze all the signals. The management should system should also scale with the applications and with the use of the applications. So that's what sumo logic service provides because on in the cloud again you know leverage of power of aws cloud that on and and support our customers costumers and whatever dynamic Lord that that they may want to you know use within these melodic service. The one last thing I want to make a highlight here is is that you know the sumo logic service while it runs in the cloud is built. What security force principles in mind right so we started ten years ago building service service in? Aws we realized that our customers and their data is extremely sensitive and extremely critical customer. So three binge enjoy a Service with security force principles so our our services sock do typo compliant. It's got all right at the stations nations including GDP are be a federal already so so we make sure that we take care of our customers and their data uh-huh so there was a report that sumo logic put out fairly recently about continuous intelligence in it was just basically aggregating some survey data some conversations that you had had with customers. What's your sense of how how infrastructure usage is changing in twenty nineteen? What are the biggest changes that you're seeing a great question? And let me just give you a bit of a background on this report right. And then and then we can talk about some of the changes that we did. See and wear highlighted in this report right. Look I mean the fundamental reason we put out this report was to really provide wide. You know our customers and practitioners at large a roadmap off of how do build run and secure mission critical appplications in in cloud environments. And you know the way I always talk about. This report is that you know had we had the support ten years ago when we were starting to build our applications in the cloud commuted love to learn as to how do companies bid and run applications in the cloud. We didn't have that back then. What we have learned over the last ten years working working with over two thousand customers I would say almost seventy five percent of them running mission? Critical applications in the cloud. Is that the people who build applications in the cloud. Think about the infrastructure differently think about architecture there that applications differently and you know as a madman system that that manages these applications Sunday's infrastructure. Okay we have a unique vantage point unique visibility into how these customers are are doing what they do. And so. What we've done is collected this data anonymous minimizes data and then presented this in a forum that you know? Every practitioner can perhaps use learn and implement in their own environment. Now you know in terms of the report itself right. I think what we saw were were really five key observations. I him from us. What we see is that the whole notion of multi cloud environments? It's starting to happen. It's it's becoming more and more real now when we did the support you now at the first time at our in in two thousand sixteen osteen you know those. Were the days of multi cloud deployments You know there was one cloud to to speak off at that point that aws Azure was just getting out of the indicate the GCP was still very small. And what we can tell you is that things have changed quite dramatically right now. I mean almost almost thirteen percent of our customers who are running multi cloud deployment applications running on. Aws azure our GDP for that matter so so multicolored is not just a buzzword. I think There is there is definitely some some reality in multi cloud the other thing that we that we saw in this report is that many afar our customers are starting to adopt Kuban Netease. Then you know Jeffrey Dulas. I'm sure you know somewhat surprising right. I should say surprising. Not Surprising at the same time in our communities is still a fairly new environment and yet would be seen as at one and five customers of of some logic in aws or already using Carbonari ladies. What is also interesting is that you know when customers are thinking about multi cloud deployments? UCS sharp adoption and could Bonetti's so if you think about a customer running on just aws. As I mentioned before one in five is using Kuban nineties on the other hand if you think of a customer who is perhaps deploying on all three environments Environments Aws Azure and soupy be almost see eight and ten. That's like you know four and five you know however you want to slice it That that you scuba nineties you you know the takeaway from this is that cooper news is becoming the son of grand equalizer and a key enabler of multi cloud deployment so if you are an enterprise architect today and if you want to build an application that is no longer beholden to one cloud. AWS assure orgies feed. It would behoove you to think about using Kuban as as your underlying platform because that gives you the ability to port your application. You know rather seamlessly from from one to the other now. I got a few more observations in interest of time. Geoffrey let me know how you want to hear more. Tell me more because the thought was interesting about this report was there was a lot of information that that was aggregates of. I think what one hundred one hundred and fifty companies something like that. Oh No let me definitely correctly on that. I mean this is two thousand eight hundred. Seventy thousand accounts rather alarmed N. statistically accurate sample size. And you know we've gone to great lengths and pains to ensure that we only put out data for wait you know. We have a statistical significance in our data so the data that that'd be a presented in this report are are quite you know. I said the quite the norm for or for most enterprises. And you know something that you know. Every practitioner all day showed to definitely pay attention to But let me continue in the you know since since you wanted to hear more point point number three and this is something that anybody who goes to. Aws Reinvent eight abuses big show will definitely relate to so we all go to aws reinvent Vanden heroes aws announcing you know yet another fifteen services or twenty services for that matter are pushing the envelope of all the stuff that you can do in Aws Louis. Oh so there's always this question took aws as relentless engine that is that is pushing innovation every year. How many people do adopt these these these new services that are coming at such a fast clip from? Aws Right so looking at our data would be the alleges that when it abuse is indeed pushing yard a lot of innovative capabilities. The core services that are adopted I many of these. Aws customers are still very limited right or find out is that had on average and this is looking at you know over fifteen hundred customers afire who are using aws on average we see that only fifteen of these. He's one hundred and fifty eight of Louis. Oh bases are are really adopted by and large plethora of these customers right so customers go blew us. You know this is sort of the hypothesis coming out of this. Data does data point. Is that customers like the innovation pace at aws. Customers liked the the fact that aws is pushing the envelope when it comes to you know releasing new capabilities providing new services to enterprises but then the vast vast majority of them are still in the early phases still in the innings to three if So to speak of adopting these services beside a few customers are few enterprises as you say who have adopted many many many aws services but those are far and few in between a large majority of these of these customers end end up going to aws for their infrastructure as a service platform the EC to service the s three in the cloud formation. The the Ardi as and and the I am identity management services and are cutting their teeth at these basic services. You know the fact that. AWS offers all these. He's cutting services obviously is a is a good carrot and a good icing on the cake But the majority of the customers are still using just a quarter of the cake at this point point number four. Let's talk about so we're less and everybody is talking about sober. Less over less is being proclaimed as the as the next programming paradigm. That's going to change everything about the cloud and systems and how you used systems and would fear starting to see is that at least in the aws environment right So let's has has reached a tipping point would be aws as civilised implementation is called lambda. And what we see is that lambda adoption as indeed grownup dramatically. The last few years in twenty seventeen which was the second diarrhea. Were doing this report. We saw the put option of lambda right now. Twenty nine thousand nine. We we see almost thirty. Six percent. Adoption of lambda means one in three enterprises that are running now applications in aws today are using lambda lambda or sober less in some way shape or form so that is An interesting data point you know and I think that sort of tells you that people are really elite experimenting and deploying an implementing solar technologies that are at a much faster clip. Now I I do have one last point and and this is sort of Switching gears for more the architecture do. Let's say talking about security technologies within within the cloud so Jeffrey I'm sure you must have heard that security. The and cloud security is perhaps the number one obstacle or blocker or challenge as enterprises thing of adopting cloud and and there are some some do it. I mean you know enterprise sock teams are used to seeing the infrastructure. They used to coming their infrastructure. The US to seeing it in their in their data center so so that you know that that gives them the comfort of of their data being secure within the confines of that Data Centre her and now moving everything to the cloud may does bring up. These concerns that maybe my data is in a place that I don't control and you know what happened security then so what we are starting to see is that again. This is within the purview of AWS. A lot of the enterprises that are deploying their applications applications in the cloud and in aws are starting to use many of the services that aws offers in order to Specter Secure their their applications in the cloud so aws has these these interesting technologies rattler. A few names here. Scott Cloud trail cloud clouds drizzle. Think of it as providing the you know who walked when where how off axis of API's and services within AWS AWS asu offers repeatedly flow logs. which gives you visibility into? Who is accessing your systems? AWS has a new technology. Call Guard duty beauty. which provides you visibility into the the security state of your aws applications and what we are starting to see is that you know many knees similar logic customers or deployed in aws are using these services and are using these At a fairly significant rate so so that gives us confidence incidence that enterprises are indeed paying attention to security as they are moving to the cloud they are. They're building in the right capabilities. They are you know using the right services that are offered by cloud platform vendors themselves and all of this obviously a portends wealth for applications that are running in the cloud So sock names. You know if there's anything that you should take away from this podcast. Is that in security in the cloud is a little different but but it is. You know it's definitely doable. And there are all right capabilities available from vendors like aws auto from vendors third party vendors like some logic that can ensure the security of your apps and infrastructures in the cloud. I assume you've you've been to reinvent absolutely we'd go to arraignment. There's a big contingent engines of similar logic folks who got arraignment every year. So I'm going for the first time this year and just after seeing pictures and hearing stories it sounds completely overwhelming advice. I surviving reinvent man hydrate hydrate hydrate and and get out every every now and then From from the boots and from the from the session. So what can I tell you how about that. But then you know that Las Vegas has a whole you know. Reinvent has become the technology Mecca to be and to go to. You know. I'm dating myself here but I is it more more so than. CAS At this point. See yes is a different kind of show right. Jeffrey says he is a bit more consumer oriented. You know. I'm sure if you are selling. The latest style is not gadget or a phone. You know I see yes on 'em. WC The show in Barcelona is best. Perhaps a show to be at if you're doing B. Two B. Applications if you are targeting anybody who is in the devotee deb. OPS was a a site reliability engineer or perhaps Stephen a security person relentess is definitely the place to be the vibe in that place. The energy in that plays this off the charts heralds. What's the where technology is today and technology is headed for the next few years? You know rightly or wrongly. Aws is the eight thousand pound gorilla in the cloud. Space the building cool staff. They have some amazing customers. It's a it's sound like an advertisement for aws but the company does it really innovate well similar logic who was deployed in aws. We liked innovation that they bring to the it world and we like what our customers are able to do on on native Wegelius to ensure is amazing. Amazing I think is a sign of of what Amazon does right that it's like as a byproduct of the the accidental business that they created with aws they created like a billion dollar conference industry. Yeah I mean what is it I I hear that are on the run rate of twenty five billion. Whatever right now so reinvent I meant? Aws as business itself. I no I I don't know what the the actual revenue Magin. It's pretty profitable. Oh my mom was certain. It's it's profitable given the number of vendors like like ourselves who are there and the amount of money that we spend in demonstrating getting boots you know sending people to that event. I'm I'm I'm sure it's a profitable venture for the blues but look I mean at the end of the day right. I mean this is a place where where you meet the people who are building your connection APPs right by the fender guitar player who is building their APPs on there the next gen mobile APP on. Aws whether it's AIRBNB AIRBNB whether it's neck flakes or whether it's a tiny little company that's You know with like three guys in a garage today. Nobody would question you if you were to put theocratic card and build your first on us. So it'll be as hats and reinvent has become the show to to go to see what's going on in the in the the. It world for sure as we begin to wrap up a few higher level subjects. I WANNA discuss so the changing usage prototype of a log management system around observability system. We touched on this a little bit earlier but the fact that today not only do you have engineers and operations nations people working with your log management tool your metrics tool we have data scientists. You have customer service people bowl of product managers. You might have marketing people. As he's different constituencies want access to this logging data or maybe they want access to the post processed log metrics. How does that change the interface and has it changed the product design nine? That's a good question rear in early. I would say you know phases of understanding it ourselves but but I think what's similar Dick is starting to see is that you know in many organizations in our customers are not just the typical people right. I mean the SR is in the people who are on the security people who product people like exactly like you said or customer support people who may not have the sophisticated understanding of our curry. The language may not have the ability to create dashboards An alert and and other things that you know you would. You would take for granted if you're someone who lives in breeze. He's within the similar logic system. So what is starting to see more and more off as two modes of operation one is this notion of what I call is template iced iced offers right where I as an administrator of the product or as an expert user of the product. I met template ties What I may I want to do in the product and then expose a few very simple to configure things within my product and then offer it to a larger swath of if users so for example? Imagine if I have. Let's say three hundred customer support people and you know I need to support the customer support people and they need to look into sumo logic system. Every time somebody called with the problem they need to look up an ID and daddy need to see what's what's wrong with the product or the of the system at that point rather than expose the similar logic query language to do all of these people who I'm sure They come with in a different knowledge and different skills at what be condemn do is offered. A template is Interface or dashboard. But all you have to do is drop the product. Id and then you drop the product idea you know you are the customer and you get all the details about the the customer. that product used product the logs from the product the editors and the D. Bugs and the warning messages the product itself. So so this template is offering essentially in some sense simplifies. I don't want to use the word dumps because that has a pejorative meaning to it but at least simplifies the access and the analytics that these non-experts can also get from our product. The other approach that I will see you know and then from product. What is the ability for us to export Data from product into other systems so we have customers who explored our data into into Pablo into clack into Lucre. You know so that they can start to slice and dice data in somewhat different ways which may would be more in line with what these personas may be used to and mail to want to see it and end of the day. We own the business of getting insights from data and the extend that you know that are better than other systems that can perhaps do some of the stuff we have more than happy to export this data to integrate those Danine to other the systems. We also see some of data science customers. You know US notebooks and decimal logic as a simple interface to export or data into data data. Science notebooks there you can do other forms of analysis on this data which may not be core discipline logic which on the focus of some logic itself so look at the end of the day right. I mean you've been in the business of getting best insights from our data for customers and we'll do our utmost to make that happen. I want to thank you for coming in software engineer. Dale it's been great talking to you. Thank you very much. Jeffrey take

Aws sumo logic You Know Twenty Twenty Jeffrey Scott Cloud AWS Ryan Bridges Kuban Andy Jeffrey Great Kellyanne Ramanathan Cooper netties Harvard Kamini observability systems Cisco Routers salesforce
#095  Shahrokh Shahidzadeh: Education Is a Big Part of IT

Cyber Security Interviews

38:08 min | 6 months ago

#095 Shahrokh Shahidzadeh: Education Is a Big Part of IT

"Douglas Brush in. You're listening to cybersecurity interviews. cybersecurity interviews is the weekly podcast dedicated to digging into the minds of the influencers thought leaders in individuals who shape the cybersecurity industry. I discover what motivates them. Explore their journey cybersecurity and discuss what they think. The industry's going. The show, let's listeners learn from the experts stories and hear their opinions on what works and doesn't in cybersecurity. Hello and welcome to episode ninety five of cybersecurity interviews in this episode we're speaking with Chirac leads a day. Chirac is the CEO of acceptable. Chirac is a season technologist in a leader with over twenty nine years of contribution to modern computing architecture device entity, platform, trust, elevation, large, IOT initiatives in ambient intelligence research with more than twenty, five issued and pending patents. Before, except Oh, he was a senior principal technologist, contributing to Intel Corporation for twenty five years, variety of leadership positions where architect and lead multiple billion dollar product initiatives. In this episode. We discuss evolving authentication so an MFA. Challenges Anomalous Behavior Detection, enforcing least privilege his time with Intel Ai and machine learning multi cloud security, securing home users, and so much more I. Hope you enjoyed this episode as much as I did. Thanks for listening. Truck. How're are you today? Thank you for joining cybersecurity interviews. Thank you for having me. It's a pleasure house everything for you. Everything is going excellent, even though we're in this kind of crazy times. As I'm sure you're you're dealing with is is the constant adoption in? We stayed security. You have to kind of adapt and overcome. You've now we still military, but how? How have you and your team and organization kind of? Change adopted to this new way workings. Adapt and overcome is a is a fantastic overview of What what the what the situation is I? Think actually is. There is a silver lining behind all of this and all of us recognizing that. there is actually a possibility of being as effective at least in in our world and cyber and We're fortunate unlike some of the service organizations that they can't. Survive in such environment. It has been okay with us and we we have adapted, and we have overcome. And to my surprise them to the surprise of many customers, the they have been able to continue their business. And efficiency of the team members is good. Of course. Our heart goes to people who are in the service business. They're hurting both economically and also. Physically as well because there are more exposed to. To covert nineteen than the rest of us who can have the luxury of being able to work from? Australia it's it's interesting with a lot of the security teams. That I've had opportunity to work with over the people have. been able to adopt from the work from home pretty well. I certainly see some some areas that have had. Pretty ingrained socks where there's teams that are doing three shifts, and they've had, but it's interesting to see. The technology in place that really allows us to continue to monitor and maintain the safety, or at least some of the sanity has some of the networks. Tell us a little bit about the technology that you're working on now. I mean turn the kind of had this very interesting background which get into. Work that you did with Intel and your many many patents. But what? What? What are some of the projects that you're working on now? With in basically, what problems are you trying to solve? Well I suppose, thank you. I have I've been fortunate wrecking. Really cool technologies from. Onset of working for Intel all the way to two now so I've been surrounded by really good people. `specially partners in industry so today working one which is a continuation of what restarted several years ago. This concept of password, `less continuous authentication which fundamentally is. Based on the fact, that authentication is not a single event, but it is a continue. An what is cool about it? Is that basically get rid of binary authentications like passwords in? To a phase MFA's that are not. Effective enough and lose behavior analytics. Before at of indication and post authorization, which is the most important piece? To detect anomalies in basically ushered to the session to either log out tour or escalate and step up as needed. A very cool thing. And the reason is cool, is because We have really good set of customers that are frontiers in this area. Are early adopters recognizing that the old ways. Don't work. I was talking to. Purifying AC- so at. At a large organization over the weekend actually and she was telling me that she doesn't want the two thousand seven product anymore. She wants something that is twenty, twenty or twenty twenty five, and that's exactly what we're working on. We're we're trying to learn from our customer real time. That you use the word opening, adapt and overcome. That's exactly what we're trying to do on a daily basis. What keeps it very exciting for us? Yeah, it's interesting because I think one of the the traps that I think a lot of folks and technology and security feels. Okay, well, we solve the problem, so it's time to move on. You know we've had the issues with user names and passwords going back. You know forty fifty years. It's been been the same thing it's really hard to. Somebody's log Dan or Really authenticate this. Is this a good log in? Is this a bad log? In is known user they using multiple devices from two different locations. Certainly in a big push to single sign on and multi factor authentication, but what I've seen in the incident response, world is there's been a greater attempt now for attackers to do. Credential scraping of single sign on recently I saw one for. They were it was a very interesting fishing attempt where the organization received an email. Saying it was from a system administrator that. In probably doing some, DNS reconfigured out. Okay this this organization uses OCTA and they said you know we know through our zoom protocols is that you're not following proper dress code for remote meetings? Please sign into OCTA now to accept that you understand the new policies in you. GonNa, watch a training video, and so getting people you know, basically go into their OCTA portal and try to scrape those credentials, so it seems to be you know just just because we kinda fixed the problem, it doesn't go away. No absolutely I think this is a very very key, because it is a continuous battle that I mean I I mentioned that we do continue authentication, the battle against a the bad guys into threat actors is continuous, because if you think about it, it's basically a balloon. If I can draw the analogy to. Financial now, fraud fundamentally every time our financial institutions squeeze one side of the balloon. The other side POPs up some an issue of Roi for the threat actors. They're they're? There's fishing is such a low hanging fruit credential credential. Stealing credentials is such a low hanging fruit for or the threat actors I mean you have seen the verizon reports for the last five years. The. Credential. Exactly seventy seventy to eighty percent of always because of credentials. Five or five to eight percent are always miscommunication so right out of the gate. You have north of eighty five percent of your data breaches are the two basic things of configuration and poser station, and in the big ticket item is eighty percent of the credentials. Yeah, and it's you know. One of the challenges is certainly having done more incident. Response engagements than that can care to remember but. Once the attackers get in. It's it's very big on using credentials and a privilege escalation. It's their living off the land on using this, and then it becomes very difficult for sock team, threat hunters and I. Are folks to say Gosh again. They're blending in with the noise. Is this a good log in? So? How do we solve that problem once they're in and we know that we can. We know that they might be in there. They're not. They had to try to separate the signal to noise. The behavioral. If you, collected data than than the anomalies are easier to detect and the noise and the signals that and the controls that have lost their efficiency and. Efficiency they will basically. Stand Out because there. There is a norm of behavior. That's. Each individual accountable in their house, and all of a sudden when you come in, and you see the office than Doug Using downloading sixteen terabytes of data, need they never neither one of them ever do that one by the way for the last one week. They have been working very late at night. and that's very strange right so you can take a look at the behavior and the deltas and the behavior and identify anomalies and that way you know by doing a commonality analysis, you can identify the noise as well and and make sure that you don't. Insert too much unwanted friction to the system where the anomaly can be. Resolved by simple step up authentication or policy change or enforcing zero trust for a certain period of time till you get you gain your confidence again on the individual that is that is using credential. Does that make sense it does? It's almost seems like you know we're we're starting to. Go to the trust world we're starting to migrate to almost network like any network authentic. That -cation controls where instead of worried less about the device, but it really. Say Identity seems to be everything these days when we're. Managing the security networks always has to seems now more than ever with dispersed club computing. In apps that live in the cloud, and not on Prem seems to be the bigger the bigger concern. Well. Okay, so so it is a very good point, so it is, it is the device it is identity. It is the APPS. It is the type of data that you use. It is all fit together, so it's not just the fact that okay we can. We can let go of the endpoint security point, then point security becomes for for your identity right? We have a word that that we have coined called identity for life identity for life. Is Is. Something that allows you to create your devices, your network, your habits and everything else together to stab that normality about you and it, and then use a predictive model to say Doug is about to log into next five minutes using his Mac. Vice and his phone and his MAC are both of them are on the same ip address at the same Wifi, and the most likelihood that this is truly doug versus Shahrokh. Is Ninety four point three percent, and then when you happen to do this, depending on the risk I might not even ask you for an authentication. I might just give you an audit trail if I want. So you're absolutely right, it's it's it's. Is Not just a device it's it's whole bunch of things right? It's the APPs in the cloud themselves. They have a particular fingerprint of what Apps in the cloud you use. Do you use your? Giradi use your up right. Do you use your service now? Right? All of these APPS and they're using usage model becomes part of fingerprint and add as behavior year of you that that we classify that the first five minutes of your your log in Renault that you usually go to your g mail, and you go to your Jira and you go to your get up, and at itself is the graph of the events that are expected out of you. Funny. Having done quite a bit of dead box runs to. You you start to almost be able to see. Browser behavior just season as example on a daily basis. When I'm doing browser now second tell. People's Daily Habits, and then when they start deviating from that, you know whether it be two to three weeks before they might leave job you start. They start doing things they might not even be thinking of doing. That's detectable quite frankly. Absolutely this is this is. This is the mother of all recipes for. Detecting insider threats right when when somebody is on the on the rise in of leaving the company, and or or if they're you classify them that they might be disgruntled. You actually can see that. Have normalities The these are these are things like I mentioned earlier? Working late at night, excessive number of hours right. Now if they're working excessive amount of hours, but at the same time to his excessive hours are significant amount of data download. Or going places that they usually don't go. That is an early warning for you. How How? Sometimes there's a little bit of a concern. Let's say you when Oregon say cheesy and I feel like I'm I. Don't want to be spying on my employees, but to me. It's like watching a lot of what they do. This is just correlated. Do you ever run into issues where people say why I don't feel comfortable? Watching my my users this closely. I'll the enterprises that we serve. They usually have their financial institutions healthcare for mine. And even in cases of colleges, there is a fine line of privacy versus safety data secured. If you think about it. It is not an issue of snooping on the employees to see Hoy effective. It is an issue of and that's the job of the management. Right to make sure that your employees are effective, and you're fully utilized and everything else. It is about actually making sure that. The assets that are your Ip. Are Not exposed and you're not gonNA. Get a black eye because of a breach. Remember facts degree strike forget. So so hundred forty million, plus people's social, security and information in everything else breach, and then they lost almost thirty percent of their. Market Cap in this single day. When it comes to this, which is the livelihood of actually the employees of a company as well equally is not like the the big evil company is only carrying about themselves and Ip of a Pharma. Company is important to the employees because that's where they have. The assets are there or the data about an insurance company about the patients, and healthcare provider is very critical, so it's not that you're snooping on the on the employees, so Redo here. This and we don't differentiate. Privacy and security are tuna box by design. You need to to have a system that protects people on both on making them secure and prevent the Pi to be breached. Now if an employee is careless or policy is strongly has been there since two thousand fourteen and nobody has audited, and that is a root of. You No Doubt but this is going to cause a potential breach. Yes, you have to monitor if an employee that doesn't have access to financial data and all of a sudden trying to get all of the financial data, and their job is purely engineering wise at that looking at What is the projected earnings? And and do they have a business to do so? In that case I, most customers don't classify that as snoop ICK. It is it is about what data sets do you need to have access to? And as mention six percent of the data breaches are due to miss configuration. Do you really need to have access to this application? If your job does not require you to have access to it? And if not let me shut it down or at the minimum could step up authentication on it. Does that make sense does the old the old issues of? Concept at least privilege I mean we, we preach you. Really, do have to kind of. Force organizations to. Adopt that greater. Because in the end, yeah, why does this user need this much access? And it's not that you don't trust limits. Could be a problem that is our count gets compromise or they can even. A click on something bad things can happen. Accidentally and I've seen that happen Gosh more times than I can count to where. Maybe somebody just had access something they shouldn't have, so it's. It's putting in those controls and auditing them on a regular basis, Hec? Now let's let's step this back to a way that. Is More Practical Exactly I mean think about it and employees that has. That has. been lured to click on a link to go to OCTA for dress codes on zoom meetings i. mean think about the combination of that. In that case if you do such, click and you go there and you're configuration is not there and all of a sudden, I can detect that someone who has done an account takeover is going places that the good employees that unintentionally clicks on something by mistake. Now is looking at some data and downloading significant amount of data that itself is is a perfect example of why you need to have some sort of a continuous authentication and monitoring the behavior. The digital behavior of the individuals right? Deficit! What such that I want to get back to your time with Intel. You had a very. Very long hair that how did how did you transition from? You know from what I'm seeing is really working on the really early pentium designed to. Getting into the security realm, Transition I I think until as an amazing company and has always been a mega company that produces really good products that we. We had an Christianity. Intel I started as a chip designer at Intel and. While part of the chip design was provisioning in designing new security features, and then from the hardware security rent to firmware security to establish secure booth and for data. We needed to have some hardware features, and then from hardware security on firmer security into application layer security, and eventually in the last ten years I was focusing identity management and and behave around lyrics and and then. In between I worked on a project called Ambien intelligence research, which was about anomaly detection in a physical space. And my wife started acceptable at the same time, so she was working on cyber and I was working on physical space to detect anomalies, and then later on left Intel to join her to help out. Basically, we combine the two together and put our brains together with the city of the company to do this anomaly detection approach towards cyber, and and bringing a l. into into us for for anomaly detection and. Classification and Identification and early warning. For for our customers. Gotcha admit you you, you mentioned. What are the? What are the big buzzwords that we're using industry these days? I think deserves its place, but certainly as you machine, learning and. How do you differentiate those? And how do you see them? Helping? In ways that maybe a human can, but also you still need human a lot of what I've seen is. It's great at assisting people making decisions that they could normally make because there's just too much data a sifter. There is in am l., so first of all the difference between a NFL is is basically the they I every time that you write a code and if Dan condition that's basically you can classify that poor men's level as as as an ai m. l. is is the ability to Basically associate a behavior ear associated with that event and Learn from the past experiences, but it supervise or not. And react to Abnormalities in and whatnot, so it is the difference here. Is that one of the key things about machine? Learning is ability to. Go through significant amount of data. So that's significant amount of data is Allows us to do classification and do commonality analysis so then when I see if I was GONNA go after A. Company that has hundred people and a hundred thousand people versus hundred people, and do commonality analysis on their use cases, or were there. Their account has been taken over or whether there is a commonality of a particular endpoint, a device that is coming in for twenty different employees, trying to making an attempt to law again through the repeal. That commonality analysis does is allowed by the machine learning piece, and and the ability to sniff through all the data that we have an identify some sort of commonality as well as some level of abnormality. Yeah, it's funny because I think humans overestimate their ability to. Make those correlations, but it just does not scale. Does Not scale well. It's really where where machines can. Can benefit us where we just can't. There's just too much for you. Consciously process at any given time. It is I mean the sheer ability to to produce behavioral models and and associating some level of risk with those. Those behaviors, it really delivers an amazing transformative technology To that allows us to mitigate identity impersonation. and. Minute comes to. Fraud Right. We would what is key here is. Analyzing verifying the identity ended behavior by inferring contextual data, and also not only contextual, is contextual and behavioral, because contextual data are again are binary, and they can be duplicated, but behavioral is different. That's how you detect. Whether your spouse or your child is happy or unhappy, because you detect that you have the ability to infer their behavior based on the past history that you have, so you definitely know if they're happy or sad or Tired or not? That's exactly what we need to do. In order to deliver a higher level of platform trust. Not only at authentication, but actually pre authentication and post authorization. Because posts authorization is very all the will happens right. So. What are? Some of the upcoming technologies, yeah, we certainly see the more cloud adoption. Not. How do those fit into this model I mean? Are there areas that? You have either concern with or that you're actually already building towards. Knowing that this is maybe a new way that people are quickly that we're seeing now with covert. There's been a huge uptick in. Adoptive adaption of cloud technologies. Now when you look at the enterprise? How are paired? Do you really think people are for these new technologies kind of entering the overall enterprise design? Well okay, so let me let me divide their question into two pieces. Number one is what type of technologies that is. Coming and What are we working on? And the second thing is is about couvert, and and going to cloud, and what are risks there so? On the first one as far as what technologies different the from from everyone and their cousin you here today that they're the. They're going to use AIML and significant. The AML is a key technology that is going to be adopted by many of the vendors, and and for many of the customers to fight against the cyber cybercrime that exists. and. In order to do that again, you need to focus on the big ticket items. The seventy eighty percent that comes from the credentials huggy. US Am L. in identifying credential. You know, take account takeovers and whatnot. detect the behavior. And how do you rely on human behavior riches? Very Complex to to Mimic an. How do you armor against on the other side of? Our tax themselves The Bevan's ation of AI is is already a reality that is happening or as. So in that context what what we think that is going to happen in the next Basically from now have been working on it. Obviously, most of our competitors are GonNa go there as well is use of am l., and what's recalled the next generation authentication streaming and what I mean by. That is not on the using of the. How you stream data to your am L. engine. And that's actually what to be. The second part of the first question is what are we focused? We're focused on in ingesting data and feeding am Elaine Jin to detect anomalies in abnormalities and identify commonalities on the threat rectors that they hit an enterprise. How do we do that? We basically have. A smart data hub and that we ingest data from different sources that's tens and hundreds of millions of dollars have been spent already an enterprise to ingested data and associate that in just data to a session. I give you a specific example Let's say that it is case of your spunk in your your Simmons happens to be spunk or Q. Radar for whatever your favorite sumo logic, your favorite simple wider. The data that we ingest from the SIM will allow us to almost in real time, detect anomalies on individual accounts and on the corporate of. The users that they might have some sort of a Commonality s whereas the attack surface when somebody is attacking a particular organization in particular bank, or or whatnot right, and in that scenario by ingesting data from. Assets such as your your ACA is of the world and sponsor of the world and IBM. Curator of the world and sumo logic we. Unleash, the power of our AIML Injin to ingest the data on on the behavior of individual and a cohort that they prolonged. And that includes their devices, their browsers their texts that they have been using. The mobile device proximity the Wi fi the network that they're on the MAC addresses. All of those comes into the picture as individual binary That are integrated into one piece, and that that is the fingerprint of Doug versus stroke and the behavior I have. So. That's the first one the second question you asked was about. Cuvette pushed. covid nineteen in general transformation to clout. Organizations. Continue to have this hybrid model of. On Prem versus cloud for rethink that for the next decade. And the reason is because there are certain rules and regulations regarding privacy and security of the data that needs to be complied, and some people are going to continue putting some of their data only on their own data centers, which fundamentally make sense and for Daphnis in. You have to have a solution that is covering book. And that's exactly what the future is going to be is a hybrid mode the it is again the the balloons analogy you squeeze one side, and you all the way to to the cloud, and then you recognize that Oh, no, you need to bring some of the data back because the privacy regulations with GDP or the California one or whatever region that you're in requires that you protect some of this. Data, and the only way to do that is making sure that you have full control behind your own firewalls right because at the end of the day you have to assume. That everybody is reachable. And? What do you do if you're breach? And how'd you know? And how fast can you know? So postcode nineteen. We see that there's going to be a surge. That is driven by two two particular drivers. One is this transition to the cloud and number. Two is because we never trained our employees on how to work. A hundred percent of our employees to work from. Home. There's a surge of home devices. N. B. Y. O. D. devices. There is hygiene associated with the devices that needs to be in place. All of those post-code with nineteen before see that is going to. Result in a surge of attacks and birth of new types of attacks right. Obviously you can't. Monitor. The behavior of the people in inside their home, and what cliques they do, if they have their own before and we need to be super careful and educate more than anything else are employees Jeff I mean it's such a good point of seen many posts the various communities that I'm part of. Where people say, how are they? How are they monitoring VPN traffic for people at home. Do they have access to those devices? Should they be scanning inside those? There's a ongoing pen tests you know, and these things have they. Don't you know depending how the VPN up can have access to that? Brings in a whole bunch of things that I don't think people really thought about. When we just all of a sudden, everybody's working from home in the security landscape suddenly changes. Know go back to our beginning. You have to adapt over common. It's. It's interesting new challenges that we're going to have as all of a sudden we have more and more out in the field, and where everybody's on a different ISP connecting into various cloud services and maybe VPN at the same time. The the silver lining is actually this is forcing us to educate our employees delivered more and getting them to to recognize that. Their hygiene win at home. Might actually trickling to the corporate. Corporate disaster for enterprise that they are working for and they actually. Enjoy the job that they have and of course in this economic. Disaster situation disastrous situation that we're in. The last thing that you WanNa do is do harm on your on your company. That has kept you on board and your continue working remotely. So you WanNa make sure that you're well educated on on the topics. The education is a big part of it. really big part of the soul equation of covid nineteen the example that you started. The conversation is is exactly that. Don't don't click on anything that looks like a suspicious really dress code on zoom right And and kind of stab wish a protocol on if it is a corporate e mail that is coming to you is going to have this type of pay. Attention to these type of the signature is on the email and and whatnot right. It's got. It does at present I think that's that's the fear. A lot of organizations have we don't want to talk about how what attack vectors might look like. But if now this is the time more than ever to get out there and say hey, look, we're seeing uptake in these type of Wacky zoom dress code. Things make sure everybody is educated on this. These are great learning opportunities. Exactly. That's. Truck. Where where can people find you online? We, basically the company name is called a little dot com, except over to tease that that's very important except to do something it's it's basically an assertion of your intent, so accepted dot com, and if you search for behavior, indication or password, authentication or continuous authentication you should be able to to find this. Particularly if you're interested in getting rid of your binary authentication, we have some. Really Good Demos in capabilities and violets that you debts people who are interested in this area we can. We can provide them. We also provide some sort of a consulting a manifesto for the next generation authentication if you'RE A. Front runner in an early adopter of technology. loved to have a chat with you and see how we can be part of it as a technology partner for for people who are listening to podcasts. Very cool. I'll be sure. Put all that in the show. Notes and I greatly appreciate you taking the time today. Thank you very much. Chatting with you. Thank you so much for joining us today on cybersecurity interviews I hope that you enjoy this interview as much as I did. Please go to cybersecurity. Interviews Dot com where you can find every episode including show notes and links for each guest's there. You can also find social media. Links had to sign up for new episode notifications. Thanks, we'll talk soon.

Intel Corporation OCTA Chirac Doug fraud sumo logic Australia Douglas Brush principal verizon CEO OCTA Elaine Jin IBM NFL
Replicated Software Delivery with Grant Miller and Marc Campbell

Software Engineering Daily

1:08:51 hr | 1 year ago

Replicated Software Delivery with Grant Miller and Marc Campbell

"Distributed systems are required to run most modern enterprise software application services need multiple instances for scale ability and fail over over large databases are sharded onto multiple nodes logging services streaming frameworks and continuous integration tools. All require the orchestration of more than one server. Deploying a distributed system has historically been difficult because the nodes of the system must be managed by the underlying infrastructure. If I have a distributed system that I want to deploy the complexity of that deployment is going to be different depending on whether I'm running on. Aws or VM. Ware or my own bare bare metal server infrastructure heterogeneous server infrastructure makes it hard to sell distributed applications that get deployed to that infrastructure. A vendor that is selling a distributed database would need to figure out how to make their database work on the infrastructure of any given customer. Kubo Netease has simplified the process of deploying a distributed application Kuban. Andy's is a container orchestration system. That has steadily grown in popularity. The point where the ecosystem is mature and the software stable and now that the software industry has a reliable portable means of deploying a distributed application the enterprise software market is becoming easier to enter for the companies that want to sell a distributed. Application replicated is a company that builds products addicts for software delivery replicated allows for the distribution and updating of applications that would have been hard to deploy in the past grant Miller and Mark Campbell symbol or the CEO and CTO of replicated and the joined the show to talk about the modern enterprise software market and the process for delivering software to companies. That might otherwise have trouble. Consuming it full. Disclosure replicated is a sponsor of software engineering daily over the last few months. I've started hearing about retool. Every business needs internal tools. But if if we're being honest I don't know of many engineers who really enjoy building internal tools. It can be hard to get engineering resources to build. Hold back office applications and it's definitely hard to get engineers excited about maintaining those back office. Applications companies like Adore Dash and Brex an Amazon use retool to build custom internal tools. Faster the idea is that internal tools. Mostly look the same game. They're made out of tables drop. Downs and buttons and text inputs. Retool gives you a drag and drop interface so engineers can build these internal. You is in hours not days. And they can spend more time building features that customers will see retool connects to any database. Api Pi for example. If you want to pull in data from post grass you just read a sequel query. You drag a table onto the canvas if you WANNA try out retool you you can go to retool dot com slash s daily. That's R. E. T. O. O. L. DOT COM SLASH S daily. And you can. Even host retool retool on premise. If you WANNA keep it ultra secure. I've heard a lot of good things about retool from engineers who I respect so check it out at retooled dot com slash s daily Scranton mark. Welcome to software engineering daily. Thank you so much. For having us you guys have been building replicated for about five years and your company requires you to have have a detailed understanding of how enterprises are buying software described to me the modern large enterprise and how it buys a software. Sure yeah that's a great question. I think there's a lot of context that we can unpack there. So I think that specifically for replicated we think about software really needing to be enterprise ready and so we actually built this guide like three or four years ago called Enterprise Ready Dot. Io and that guide is sort of underlines how how SAS and software companies should be building software that enterprises want to buy it turns out. There's sort of the standard set of common features features that every enterprise software application really needs to have and I think for a long time like people didn't really like they kind the thought about those features as one off features. Maybe sometimes there were special requests but ultimately over the last few years it's really consolidated down to these like core features this is things like single sign on audit logging role based access control change management product security you not management integrations reporting right and so those whose features need to be present in the software that enterprise is going to buy and from there. Obviously they need to have like core business value. But that's the thing that really differentiates how enterprise is going to evaluate a product versus how a SMB your mid market company might. It'd be evaluating. There's a shift from the top down purchasing process that might be led by a CFO or CIO. Hi Oh that's the historical form of purchasing. But now there's more of a bottoms up purchasing process that can often involve engineers. Maybe the engineer buys is a cloud service or an API and start using it in their product or they start using the free version and then eventually many people are using the free version within the company. Any and then. Eventually there's an enterprise deal that's more of the bottoms up approach at an enterprise. How does the breakdown between different software products distributed distribute between bottoms up and top down? There more bottoms up or top down software distribution and purchasing. Yeah so you know I. I think you're right. This is definitely changing in. The world is shifting more and more towards this sort of like bottom up model and the tap for handful of different reasons. The first part you mentioned right like just just the availability of cloud services and software and tools. That folks can just start adopting. I'd also say that the open source has really been a huge factor in that as well right. It's just not that hard for someone to find an open source project included in a product that they're building and then for that to become a core part of the infrastructure. So you have these sort of like you know. Empowered developers and empowered sort of frontline folks who are finding tools and projects. That really help them do their work. And so as they start to use those oftentimes these are premium or low cost or open source in so oh they lack a lot of these enterprise ready features that we talked about so as those applications and projects become more widely adopted adopted in an enterprise you start to see demand for more controls and more functionality and more features in so so as that happens you know you start to get this like you know start with one or two teams then maybe it spreads out ten teams. And so there's this acknowledgment within the enterprise that okay. We need to take this new tool that starting to become really important in our in our workflows and our processes in make sure that it's enterprise ready and so that's often the time win that sort of bottom up motion starts to really become sort of a a combination of top down so they'll engage a sales team and they'll start to talk about you know the sort of good better best pricing plans. The move who from that sort of easy to adopt premium model to the next highest plan. Or maybe a plan after that. Maybe they'll talk about spreading it throughout the entire organization right so an enterprise could look at a piece of software and say while there's a lot of success at these. Ten teams are getting from the software. If we got our entire organization they use these and we created an enterprise license agreement. We could really be at an advantage for us. And so they'll start to negotiate what that looks like and then again the asking for these features around security and scale ability. How do they integrate with all the other software? And so that's when a sales person or like Aquino had a product someone's going to step in and kind of help lay the road map for how all that's GonNa work together. I think many of the people that listen to this show are either working in enterprise or there were gonNA software vendor that sells into an enterprise. So maybe they work on a database ace or an authentication solution or security product or a log management product and these software vendors. Some some of them are going to be selling purely cloud products. They'RE GONNA be selling purely through the cloud marketplaces but many of them will eventually get to the point where they're going to need to sell to a wider array of customers including large urge enterprises with a variety of servers. And I'd like to know from your perspective water the biggest struggles that I'm going going to encounter if I'm a software vendor as I am going to market if I want to actually sell a software product into enterprise. What are the biggest struggles that I'm going to encounter? Yeah so I mean first of all. There's like all of the standard go to market challenges. Right how do you get awareness. How do you get people to you? Know want to buy agency but then on the actual development side you know. I think that's where one you're GONNA have to develop these features like these enterprise ready features and then one of those core features. I think you're kind of alluding to is this idea that you know. A multi tenant SAS application is often in times like not very tenable inside of these large organizations right the data that AI application or a developer tool might require could be seen very sensitive information or sensitive data in so that large enterprise might start to ask for. We call like an on on prem version of that product right. So that's kind of goes by many names self managed self hosted but a version of that software that isn't hosted and managed on servers that the software vendor controls but rather as deployed to servers that the enterprise controls. Now those services could be you know in our racked and MM stacked in a data center somewhere or they could actually be innate of us or in Google or and Azure but just in the enterprises account right so these are controlled rolled from like a logical level by the enterprise and so this is a really complicated problem to solve. You know it's always been sort of in software over the last twenty years. There's been this like it's like almost this religious belief that's like you can't do both. You're either are a SAS company. Come a multi tenant SAS company or you are an on prem software company and. That's because like there's just a lot of challenges about doing either right. So if you're doing on prem software off where you have to think about releases and disconnected supportability and how do you make sure that you know keeping your customers up to date rather than having them all in disparate versions versions and so there's all of these really interesting challenges that come from that and plus just the fact that I mean you think about ten years ago building. Reliable software was just so hard right and there was all these different sort of tools in in ways that people are going about it. And ultimately I just don't think that we had the right patterns and primitives for building truly reliable software. But that's that's all changed right so what we've seen the last five years. Is that almost. All software development is moving to the sort of like cloud native architecture right and really in the last two years we've seen just domination of COOBER. Netease in so cooper netease in our opinion is truly the canonical patterns and primitives building reliable distributed software. And with that it allows a software vendor to use the exact same deployment model right that they're using to deploy multi tenant Saas application to reuse that and deliver that manifest plus those images that sort of create that communities application and deliver it that privately to enterprises to run their own version. I mean there's a whole bunch of other interesting challenges. Once you start doing that but generally that sort of model we think creates what we call sort of modern on prem right and this idea that you as a software vendor you don't need to four code base you don't need to think about it as a separate product you need. To just leverage these new patterns turns in Primitives and deliver a Kuban Eddie's based application to your customers so they can run it privately and that's not forever customer right that's going to before the largest enterprises who are really concerned about the security of the data that they want your application to process. Can you describe in more detail. Why and how currencies has changed the relationship between the enterprise buyer and the software vendor? Yeah I think there's a couple of different things that it's it's changed. Their one is instead of every time that enterprise receives software where they have to understand how to deploy. What are the requirements like? You know what what resources do I need to provision for this in. What steps do I need? What one off process inside my organization in order to get the software up and running and monitor and operate it now it becomes? Oh this is a coup bonetti piece of software so I can apply all of the practices all the knowledge all the skills that we've been building inside our enterprise so I can deploy this thing. I can monitor it using the tools that already have monitoring mccoubrey netease infrastructure. Then on top of that is that the application Shen will conform to those Cuban Eddie's primitives around these common interfaces that Kuban Eddie's provides networking storage container. Run Times interfaces like this so if if the application developer has written you know traditionally an application developer will write code and say it needs this. Port expects this volume like this directory. He is hard coded inside the code. Whatever it is these assumptions they make about the run? Time are now configured through that interface. So the enterprise when they're operating it and when they're installing it they can choose Tuesday relocate that they can choose to create a service abstraction on top of it they can apply service Mesh to it. They can run the storage where they want to so it allows them more flexibility for that last mile. Oh configuration of the application that historically was just tightly coupled into the application. Yeah I think it really becomes the common substrate for both vendors And enterprises to build and deploy software and you know this has tons of benefits beyond just like portability right. We're also talking about the the fact that you know developers. Their skills can now translate from company to company. Because everyone starts to use this canonical stack. You know you're not entering into facebook. And trying to learn their you know system of container deployment which they called Tupperware or going to twitter to do Meza's doing like everything his moving to Cooper Netease and that means that developers are skill sets can really transfer and we can walk into a new company. Pick up their stack. No sort of the patterns that they're using because we've seen this many times before today's sponsor is data dog a scalable monitoring and analytics platform that unifies metrics logs traces and more use data dogs advanced features to monitor and Manage S. L. O. performance in real time. Visualize is all your S. L. lows in one place easily search filter and sort ask Ellos and share key information with detailed intuitive dashboards. Plus data dog automatically calculates and displays your era budget so that you can see your progress at a glance go to software engineering daily Dot Com slash data dock and sign up for a free fourteen day trial. And you'll also get a complimentary t-shirt from data dog. Just go to software engineering daily DOT COM slash data. Dog Sign up and get that free t shirt. The first idea that you're talking about there. Is that the fact that Cooper Netease has become the canonical Michael Way of managing infrastructure. Means that as you always employees from enterprise to enterprise you're going to have a consistent experience of of how you're managing your infrastructure as opposed to the status quo today or perhaps you know certainly two three sorry for years ago where you had to learn. Whatever at Hawk infrastructure solution the companies built to manage their different servers? Whether they're doing some kind of chef situation or they've just got a bunch of servers that they manually. You know run scripts on some kind of distributed in in home in house bash. Solution that they've built today. Many many. Many companies are moving towards Kuban Eddie's as the unified way that they're going to be managing editing infrastructure and then downstream impact of that. Is that if you want to sell software to these companies you now can have some expectation speculation of the kind of medium that your software is going to be deployed onto and that simplifies the sales process for distributed applications because most of the important software that is getting sold to an enterprise. Today is some kind of distributed system some kind of database or logging solution or something like that that is sold that has to be distributed has to be managing multiple servers multiple containers. There's and in the past. That was very hard solution to sell. Because it was going to be the deployment medium was GONNA vary from company to company and the management intruments. Different servers was going to vary from company to company. So if you have a steady infrastructure medium than you as a software vendor can have an expectation of how you want to deploy employ a distributed system and then I think one other benefit is if you have an expectation there it becomes much easier for you as a vendor to work with the enterprise that you have sold to on an ongoing basis because usually when you sell software to them. There's going to be some kind of maintenance agreement there's going to be some kind of ongoing solution and you're going to have to do a little bit of work with the company so if you are a vendor and you've sold software to an enterprise. How is that consistency of coober? netties making your life doing maintenance easier. Yeah it's a great question so oh I mean. Part of it is because the underlying platform of coober netease provides so many important reliability primitives that you sort of know that your application is going to be able to scale horizontally and really have a lot of the other sort of like core things that you want to have an expectation expectation for higher. Application can run successfully. Right so you're getting all of these consistent patterns and primitives know that it's going to work there if they can run Kunis horrendous applications you can you know one of the things that we try to provide are like we call preflight Jackson. We'll talk about this a little more later but you can validate that. The conformance of their cluster meets the requirements of your application in. So you're basically able to say. Hey look here's this common pattern. We both follow to deploy scalable applications. Here's our you know. Manifest that describes how this application it should be run and we are so. It's it's almost almost like taking away this manual operations concept where I used to you know. SSH into a server and make configuration changes and now instead we're just in writing these manifests and then we're having that Manifesto Cooper Netease and we're letting cooper netease manage the application and the underlying infrastructure to conform to the desired state that we described described in that manifest in so if I'm handing that to my servers that are in aws. Why can't my enterprise customer just hand that to their servers? That are needed to be wes and have that same desired state applied run the application just as Scaly just as successfully. Tell me more about the typical operations that a the vendor might want to help an enterprise with so five sold a piece of software to the enterprise. What does my ongoing life look like? I know it's not just a one time. Self Service sales. So what kinds of war come I going to have to do with my enterprise. Customer as time goes on so I mean there's a handful of things right number one you have to think beyond just day one right like deliver the software and get it running and you really have to think about about day to in sort of this long term operations of the software and so that's your number one. How do you make sure that you're delivering updates in a way? Hey that the enterprise can process those and deploy them through the same processes they use to the ploy their internally developed software right. So that's a core concept that we believe in the other part would just be you know things `red like we don't think that the world is perfect by any means yet. I don't think that like Cooper Netease. Applications nations like this is an air freeway. There's probably going to be issues. You're going to need some type of troubleshooting that you can do long term and do that in a disconnected way where you're not taking control of the machine remotely. You're trying to do something where you're still getting access to all the data so you know you're gonNA. We've we've built frameworks around this. You know we call replicated troubleshoot which is a framework for identifying you know different log files and commands might WANNA run in your customer's environment to identify potential issues and then run it through redaction tools and analyzers that will sort of supply next actions actions the customer can take to troubleshoot installation or some issue that might be happening. And there's a handful of other things. I think you're going to need to be doing you know. Part of that is is making sure that your updates are continuously successful like as you add new components as you add new parts of your software. You need to make sure that the environment is is able to to manage and handle those so finding ways to validate that every time you deliver as also really important. So the reason we're discussing this is because replicated you have been focusing on software delivery the software delivery process process that improves things for the vendors and approves things for the enterprises that are consuming it from the vendors. Describe the stack stack of software. That you've built to enable that software delivery process. Yes here's Marco into a little more detail but we basically have developed upped a suite of open source projects. And he's a really purpose built Kuban Eddie's projects that are designed to sort of help facilitate tate the common challenges around delivering and Managing Coupe. Randy's off the shelf software and so we've open sourced all of these into that was a really a big part of what we do. And so you know. We call it the committee's off the shelf. Software referred to as COTS with a K.. And this is kind of a play on traditional commercial off the shelf software which is called taught. That's kind of how the industry would talk about. You know shrink wrapped or packaged software. Fifteen years ago. So we're kind of revitalizing that term Cooper nineties era but we have these different cots sort of components that are all open source and really tackle different challenges. I'll let mark dive into those so at the core of that. We have this. The main cots products seal I. It's in as a coupe control plug in so it executes on the client side and we made that decision just so that we could work with existing existing clusters really well without having to ask for higher security requirements or you know cluster admin level permissions in with that coop. Control Plug in and an enterprise can just receive the upstream mural of an application in a licensed file and then install it and when it installs it actually installs the next component that we have which is called we call it caught stadium. And that's the cluster admin console that's presented as a Ui that served inside the cluster in. It has a few. You know core functionalities like a dashboard. So that you can add some prometheus metrics in the health of the application in general view the licensing the license if you've requested quested changes or the expiration has changed on it and then like view the downstream sorry view the version history that are available in all the downstream clusters that you have deployed too so you you can see the difference that are applying. You can check for updates. You can make those last mile. customisation changes one of the things that we built core into cots and Casa Diem from the beginning as deep integration gracious with a project and the Kubis ecosystem called customize with a K.. And that allows the enterprise to take the YAMBOL and make last mile changes just to it that are relevant to their configuration but probably not relevant to other configurations and other clusters that need to run it and then we also have built in in support with preflight checks troubleshooting. That grant mentioned earlier so if the application stops working you can click a button and it'll collect a bunch of support bundle handle type data collected into an archive redacted and then even perform automated analysis against that to show you what might be wrong based on what. The application vendor has has packaged into the application. A lot of the stuff that we've done and then the last part to mention is that it enables you to kind of move away from this click to deploy like my application received achieved an update. So you click a button and allows you to integrate this into a get ops workflow so you can connect your repository. Whether that's key hub. Enterprise bit bucket get lab wherever that is and then updates to the application whether their license updates configuration changes customization changes you've made or the vendor itself has updated the application. They'll just be He created as commits into that git repository. which then you can run that through any kind of workflow you want for security scanning or any type of like post or pre eh deployment validation of those manifest before they get to play to the cluster? Yeah I think that part's really important rate. So you know I think last year on the show. We talked about a project that we called replicated ship encodes. Adm is really the spiritual successor to replicate it ship. We sort of took what we learned with buildings replicated classic plus replicated ship and we created this cots suite of tools right and one of the core concepts of ship was both like you know getting customized and that sort of last mile configuration. It was also we really believe in this concept of gaps. Right and in the folks that we've works kind of have pioneered this idea but get ops is really about version mm controlling all of the sort of configuration as code so all the manifest before you actually deploy those to a cluster right you you should check those into version control. In order to use the change management processes that were also familiar with version control in from our perspective. The really cool thing at this does is if you think about like how you manage third party software version thing like for a long time you might be doing like you know in the world you might have done. Unlike a Hallam upgrade or helm install these sort of different like commands that are just gonNA operate directly in the cluster. And now with what we've done with COTS COTS cots. Adm You can actually manage third party software verging through version control right and we think that just makes sense. It's like it's automated automated that you don't have to do anything extra anytime the upstream application ships new update. We're going to automatically download those manifests. Take those images. Push those into internal registries that the enterprise might have rewrite the images on top of the manifests and then commit that directly into the version control system at the enterprise has set up from there can flow through their internal processes around policy enforcement image. Scanning whatever else they want to do before it gets deployed employed. But those deployments you know then have this really clear. Record of all the versions that have ever been delivered to the cluster. Yeah I think one of our goals that the main the reason we wanna do that was to be able to receive third party software using the same pipelines at. You're already using to deploy your first party software so if your team's writing API's into playing them as part of of your product and you're also receiving third party software you need to deploy to that cluster. There shouldn't be two different processes involved with getting those released. This should all work through the same pipeline. Because you've you've already invested in the tooling visibility and everything into that. So this set of tools that you built is about five and a half years in the making king and I'd like to understanding of how you arrived at this particular stack and maybe some of the errors or blind alleys that you found along the way. Why did you arrive on this set of open source tools caught caught? Adm preflight eight and so on as the set of solutions for being a software vendor that wants to deploy software to an enterprise prize or an enterprise that wants to consume software from a traditionally cloud vendor. Yeah so you know. All too many innovations in the making right so I I think that's one you spend time working with software vendors and so our customers are folks like Hashish Corp Circle. CI AND TRAVIS CI folks you've had on the show like sneak. In bugs nagging ubt be all use replicated as well and they've been using replicated classic for years so we work with you know fifty percent of the fortune one hundred hundred to deliver applications. There's I mentioned into these enterprise environments in totally secure ways writes some of these clusters air gapped inside of you know a truck in the back of the Middle East right so like you have to be able to do this in a very disconnected secure way. In order to support report. The the most demanding enterprises are tools of really been designed from the ground up to do that. And then the other part right is. We've just seen Jeanne Cooper netease become like. This project is so widely adopted and it just has such velocity and so for a long time and we we actually wrote our own orchestration Asian scheduling. We first started the company but events in our goal there was not like to be an orchestration scheduling company was just like fill some of the gaps who felt like we're missing to deliver software for this way but then maybe three years ago we delivered a version of replicated. That was Cuban is compatible and this version that we've delivered is the first truly mccoubrey metis native way to deliver third party software and I say that because it really meets the entire spectrum of end customer needs right online or air gap we can solve both it also does something we talk about as existing cluster installations or embedded cluster installations so differentiate. Those is really quick. SOM- enterprises have existing Cuban Eddie's cluster. And they need to install your application in it. Maybe we'll create a namespace for you that they wanted. Deploy your software to but you need Cooper Netease Native tooling to make that happen. And so that's where you know. Mark Kinda mentioned this idea of Cuccia plug ins but these coupes detail plug ins are used by the cluster operator and QC T. l.. Anyone that has Cuban and he's like has CTL on their workstation or on some machine they used to control their cluster and so extending coops et l. with functionality right like preflight or troubleshoot or cots really gives a native experience to the end user to manage cluster and then the other side is. There's some organizations that you know that enterprises don't have really any idea what cooper nineties is yet so that's starting to become less and less but they're still probably fifty percent of the market. It doesn't have any kind of Kuban. Initiative or cluster that. They're running yet. And so when you deliver them your software. You need to package up or embed Kuban Eddie's with with the application and so we open sourced are embedded Kuban. Eddie's installer that we call curl with a K. so this is your Kuban Eddie's url and a choice cly Kuba district district creator. That can package up some of the core components of a cooper netease cluster. Right because key. Role Cooper Netease incubate. Em doesn't provide all all of the necessary components or add ons that you would need to create like a usable cluster. So curls is built on top of Kuwaiti M it adds in these different add ons and it creates like a very usable cluster that your application can be deployed onto and that can just be installed either from an air got package that we generate or from a URL H url. That has like a little hash at the end that that is an installer for your application and again is is totally open source so like we open source that we open source of all these tools because we think number one that anything that the cluster operator is running. That's it's going to power the installation in the management of one of these third party software applications. It should be open source right. It just makes it more transparent it's easier to adopt and and ultimately we have figured out a business model where you know our customers. This offer vendors so we have a commercial product that helps them operationalize and scale the distribution of all this right and we talk about that in towards the end but like we don't need to monetize or to keep these cluster operator tools proprietary so we open source them and they're really focused on this idea of of sort of optimizing the experience automating as much as you can and reducing amount of human interaction in terms of supporting and updating in managing these applications. So we spent a lot of time with the existing tool set and and you know ultimately we decided that these are the core features that we need to have. And these are the core features at any like Cooper Netease application. That's it's being delivered to an enterprise should have any third party software application delivered. The committee's need some way to be configured and managed and then a way to troubleshoot it and ensure that it's going to run correctly on every update and so creating these frameworks and these tools is our way of sort of helping the industry move towards this idea of modern on prem software. So let's go through some of these open source building blocks walks that you have one of the projects you have is called cots which is Cooper Nettie off the shelf. He explained what that is. And why you needed to build a Kuban. Eddie's distribution sure so cots is not a commodities distribution by itself. It's a tool. That's designed to deploy a commercial off the shelf software and help you operate commercial office shelf software. Inside any cooper nine cluster and that can be a cluster that was embedded in with the application. It can be a cluster running on Amazon E. K.. Asser G K or as your or can be open shift it can be whatever you want. Air Gap non air gap the challenges that like application vendors worshiping their application in various ways you know helm charts are a very popular way to distribute applications other application vendors. Would just write application the animal and deliver or that to you and then you'd have to configure that more so there was still a little bit of disparity in how you would configure and manage that and then enterprise needed to inevitably make last mile changes that were only applicable to their enterprise to their cluster to their installation. So cuts is really a tool that tries to help normalize all that through do a coupe control plug in so you can operate this. From your workstation pointed to whatever cluster that you're coop configure coop context is already set to so so you can say coop cuddle cots install whatever application it is and that'll install the admin console. Get ready for a licensed in the application ready to go. From which point you can just take run with in built it into any workflow that you might already have got it and surf the confusion there and then so who cut. ADM is a sidecar tool for managing the applications that you create with costs you describe what cuts adm does Chur yeah. It's exactly the way we like to talk about it right. It's a SIDECAR application. Think of it like an admin console for the application it's optional but when it's their provides a lot of value around a dashboard and visibility into your license the ability to troubleshoot it it creates a Web. Ui Around preflight light checks so that you can see if your cluster meets the minimum requirements or has suggested changes that you need to make our upgrades you should make to it. And then the the ability to visually see the diff- Kuban Eddie's manifests are becoming that common substrate that we talked about so it allows you to visually see the dip in the cluster. Operator can click in when there's an update available and even see the lines of the Kuban Eddie's application that have changed with that update and then they can decide whether or not they want to deploy that or they can then as grant mentioned earlier integrated into their existing pipeline if they have that and be able to configure to just push all those updates to their get repo but Caccia is is a Sidecar Admin console that will deploy alongside and application. It's a pretty small not very resource intensive but it provides a lot of admin functionality that every application developer would probably have to create through like. CLI's or their own little web. Okay so I think the cotton caught. Adm of the six different open source projects that you have if we want to talk through the process of using your suite tweet of technologies as a vendor. The main idea is that if I have some kind of software that I want to deploy to an enterprise's is coober netties cluster. I'm vendor. I'm going to make my software compliant with Kuban ities off the shelf golf and I'm going to be able to deploy that to the enterprise. COOBER netease cluster. Like if I've got my database Abass like if I'm Yuga by DP create my database. I make it possible to deploy to the coronet off the shelf version and caught cots will let me deploy it to that kind of Kuban Eddie's and caught adm will let me remotely manage aspects of of my vendor distribution and cots adm kind of provides the medium of creating updates and doing kinds of operations other administrative duties and creates a medium of communication between the vendor and the enterprise. It's actually consuming the vendor's software. Yeah I think it's pretty. Close the key piece there is that Yukos's adm lives inside of the enterprises cluster and it becomes this like tool. That automates some of the toil. That the enterprise is it admin might be doing in order to manage updates or to troubleshoot the application so it's being deployed along with the application. It lives in the cluster. There's no like remote management as much as it can make an outbound request to find updates that use the software vendor published and then it can also extract logs and other information. redact those bundles up. Either for in cluster analysis or to be delivered to the vendor for remote disconnected troubleshooting and so we should think about costs. Adm as this like really powerful in cluster sidecar component for any software vendor. That's delivering committees application. And then I think probably the most important part right so like out of the box. The end customer is going to do this sort of they're going to put in their configuration configuration information around like. Oh what's my host name where my T- alerts like. How do I can figure this APP? And that's very similar to like a helm values but then the important part is that once they've set up the application have it running they can actually set up day to operational automation. And that's the stuff that's going to push images into the enterprise's internal version control system and their internal image registry and so that automated process means now now instead of like having task on my list every two weeks ago check for an update and download the software and run it through some scanning and unlike deployed out I can basically just use my version control system to merge in these commits to my production deployment and that is sort of will run through all of my standard processes. And I'm using to deploy other Cooper Nettie services that my internal software teams are deploying. And so so it's piggybacking on that same process that you use to deploy internal software and saying look now third. Party software CAN BE Q.. Managed to the same way in very automated process. Take me inside the engineering at replicated. So you've got these open source projects you also have the the previous version of replica ship. And you kind of have the the evolution of the of the replicate platform to support. I just just like to know how engineering at the company is organized. And what your process of communicating with customers and iterating on the product is. I mean it's a great question. We have three different versions of replicated that we have out there. That were supporting right now. In they weren't ever a complete rewrite eight from the ground up. We would continue to build on on all the tooling that we had built in the previous iteration. So even our Communities Distribution Lists Curl Project Act that allows you to create a one line Cubans distribution is really just an open source version of the Kuban. Eddie's install scripts that we've been running for the past three plus two years for somebody who had a coupon compatible version of replicated classic so we've worked really hard on the engineering side to make sure that we don't have three totally different sets of code bases in three different product lines that we're actually offering. They're all built on the same thing. And we're actually working hard to take the replicated classic Doc Code Right now in that instead of having it have a fork of the open source stuff that we've created and released under the cat's name dish to be able to use that so we we don't Have like they're all using the same common code base internally and that's because we're a go shop right and so a lot of time vendor these in yeah exactly I mean we. Almost all of our code on the back end is written. All of the open source stuff is Ingo and so they re use each other a lot in a lot of the things that we do our around Kuban Eddie's operators incriminates custom resource definitions also. And so we have these custom kinds that we've created inside are different projects and so when we went to use them we grant mentioned we've vendor that men and we just just create custom resources that are able to use each other. What's on the roadmap? What is important to implement in order to improve the process of software delivery from your point of view? Yeah I think the interesting pieces like we've been working towards the replicated cots what's offering. I mean really for five years right like you know. It's sort of over several iterations. Maybe three years ago when we started to really make Kuwaitis a first-class last part of what we're doing. We had some of these ideas. And you know it's been a long road to sort of find the right way to offer this. I mean partially because is the ecosystem is moving very fast right if you think about this idea of coop. CTO PLUG INS like. The whole framework was rewritten. Maybe like you know nine months ago or something and and we think that's a really important part of how cluster operator should be. You know managing these third party components and so as everything has been changing. You know we've been trying to make sure we're taking advantage of the latest and greatest different ways to offer this functionality and so this project in this suite of tools is really exactly what we wanted to have for a long time and I think that there. It's an incredibly strong foundation and for solving some of the most important problems around configuration troubleshooting updating like automating away this toil and so when I look at the road map right. It's continue to integrate these projects with the broader committees ecosystem. Right so are curl project. That's the open source sort of coober. Netease Destroy Creator. You know. We're going to integrate that more closely with the add-ons projectors actors or Sig that's happening in the community because system integrated more closely with the cluster. API and so. There's some really interesting things we can do there. But ultimately our goal with that project object is to continue to commodities the installation and updating management of just a raw cooper news cluster right. We don't think that should be some special at hard thing to do. So you know we just WanNa see that. So that everyone can have a cluster. No matter where they are then these other projects I think about you know the ability to you do preflight checks into trouble. Shoot these applications. You WanNa make sure that that project is well adopted in in the standard and becomes a standard for application unlevel troubleshooting in the Kereta Chico system. So we're trying to work with his some of the folks at measures fear on the CU project and we just like to Z.. Like we would like to collaborate with more focus on these tools and really build community and consistency so that you know that these are not just one off tools that we created but they become and we migrate them towards standards that we all sort of agree will create a better ecosystem for delivering delivering these types of applications. I'm sure Marcus mother thoughts on on Road Map as well. I mean I think our goal with with cots is you know to make it so that there's compelling enough reason and it's it's integrated enough into the enterprise's existing systems and process that they have the whether or not the application is being delivered through a replicated vendor. It's just a good way to run third party software. Because it's you have guaranteed success that is going to install is going to update it's going to be compatible with all the workflows and everything that you want so we just really what's to become an open source good tool that you can use to run any third party software inside Kuban et cluster. So that's you know could be home. Charts could be raw commodities manifests could be operators. We think all of these can benefit from some of the core principles around get ops and automating workflows does and you know having a consolidated dashboard for all your third party applications. So that caught theam product actually has a really cool feature which you know if if you use as a software vendor to distribute your application your customer can use it to manage that one application. It's totally white labeled and it looks like your admin console but then over time is that operator really loves these patterns around you know having their third party application updates automated version control in their into registries. They're not doing the work on that. They can actually start to add additional applications to that adleman console and it becomes multi APP just like kind of how slack can go from like one and slack works based two to three to five like this tool used a similar experience to allow you to manage multiple applications as a cluster. The operator in this very automated way as businesses become more integrated with their software than ever before it has become possible to understand the business more clearly through monitoring turing logging and advanced data visibility. Sumo logic is a continuous intelligence platform that builds tools for operations security charity and cloud native infrastructure. The company has studied thousands of businesses to get an understanding of modern continuous intelligence and then compiled wild that information into the continuous intelligence report which is available at software engineering daily Dot Com slash sumo logic. These SUMO MO- logic continuous intelligence report contains statistics about the modern world of infrastructure. Here are some statistics. I found particularly useful. All sixty four percent of the businesses in the survey were entirely on Amazon web services which was vastly more than any the other cloud provider or multi cloud or on prem deployment. That's a lot of infrastructure on aws. Another factoid I found was that a typical enterprise uses fifteen. Aws Services and one in three enterprises uses aws. Lambda appears service. Service is catching on. There are lots of other fascinating. Statistics in the continuous intelligence report including information on database adoption Coober Brunetti and web server popularity go to software engineering daily dot com slash sumo logic and download the continuous intelligence report today. Hey thank you to sumo logic for being a sponsor of software engineering daily. Could you tell me more about division for these open source tools actually develop into an ecosystem. Because you're hinting. There are at the reason that you had to go from replicated ship or you chose to go from replicated ship to the suite of open source tools tools. The idea there being that accompany like may so sphere and perhaps other companies in the future would want to use some of these different tools to help facilitate the software delivery process or things that are components of the software delivery process. Yes and under trying to understand your vision for the future of how software is delivered. How softwares purchased the process Assis of deploying it onto coober netease clusters? Clearly you are envisioning a very large potential market because you're indicating that the market is so big that it would not only support multiple commercial players. That are going to want to do stuff like this. But so big that these different commercial players are going want to be contributing to shared open source resources. So I'm just curious about your your perspective for how this market is going to develop over time time. Yeah I mean ultimately we really think that the world of enterprise software like it's always been this this sort sort of duality of of on prem versus SAS and our vision of that those two worlds should really blend into one and we think that coober netease is the primary driver ever for that blending ultimately the idea that like every company is going to run a thousand different SAS applications locations in send data off to all these different you know. Multi Tenant APPS. We think that's insanity right. We just don't think that the you know large organizations are gonNA want to expose their the data to that many different vendors. We don't think that that's a secure way to to think about you. Know your data and you know you as an enterprise this is really a steward of so much data and you have consumers at trust you you have other businesses that trust you and you start to just like sort of Willy Nilly Tossed Austin data into any vendor. You let anyone sort of bottom up use these applications. Without the considerations around their compliance and security. You know even if you're trying to do you know vendor security questionnaires for every piece of software that you use. I mean trying to manage the security purity in surface area of thousand different software vendors is just. It's really hard right. You're just have so many different points of attack. And so we see a world where you have. Fewer trusted embedded software like infrastructure providers. And those could be the big three hyper clouds like Amazon and Google and Microsoft where you're using their as a service but it's really just infrastructures of service and and some of you know maybe you'll have a handful of other SAS applications that you also use. Maybe it's cloudflare. Maybe it salesforce. Whoever else but the number is definitely zero vendors that you trust with your data and the number is definitely not a thousand in terms of vendors you trust your data so that number has to be somewhere in the middle but we think that the the world of software like you just you have to be able to use a lot of software and so the only truly secure way to use software in our opinion is to deploy floyd where you have control over the data that you put into it right? You're not exposing any of this digital data and so if you can use lots of third party applications deployed into servers that you control. Well that's sort of like the panacea perfection in our opinion so if you can automate the process around around how those applications are actually like operationalized and how they're updated and how they're managed in they really become self healing. Well then you've just you you know like I said okay now. I don't have to think about the data security around these applications nearly as much because I you know it's the same data security that I use for internally developed and deployed applications nations. I don't need to think about the operational overhead because the same operational patterns that I use to deploy internally written applications and so we can start to expose more and more software applications that your team can run in your infrastructure and they can start to take advantage of lots of different tools and automate more processes and really achieve that like concept of overused term of digital transformation and become a software company. And you know that vision I think is just it's so enormous because you're talking about taking a market of multi tenant SAS applications. It's about one hundred billion a year adding it to a market of on Prem software deployment which is still about four hundred billion a year right blending those this together in a way that accelerates the adoption of software so that this model and when you look at all the alternatives right you look at like enterprise key management which is like a solution around. How do you encrypt data? But keep the keys with the actual enterprise like these data still require trust in these models to require trust around the data and you know maybe someday we get to a foo leaf and super fast version of hobble Morphou conscription corruption. And it's not like a science project but it's actually something that we are operationalizing and makes sense to us for terabytes and terabytes data. We know everything we I know about. That doesn't seem to make that really reality. At least not in the next twenty years. He's not with our current computing technologies. So we think that this idea of modern on Pram Cram is really part of the future and its core driver for how enterprises will make that digital transformation with cooper. Nettie is at the center. You know we just see it as this hugely like there's a huge amount of momentum behind it already and we think that it's clear that running third party software in Cobra's applications is like the best way to run third party software like no one should be delivering. Va's or jars or wars and we see a lot of these traditional software companies that had built their applications and have tens of thousands of different enterprises using that sort of like legacy on from software. They're all making the same transition as the SAS companies to create Kubis versions of these applications. Because that way they can can take advantage of the patterns primitives of reliability. That Cooper News provides and their customers can run their software much more reliably and reduce that operational overhead. You're seeing these markets really converge into one way of delivering. What we call coburn off the shelf software and so for us? We we want to provide the set of tools that help. Spur that forward and we are thrilled to work with anyone who sees that same vision and wants to create tools to make that more possible. And we think you know the success of Kuban overall like when you step back and you compare it to docker or something else. We think it's because it was so open. It's because it was community owned and community driven in. So you know that's why we're worth thrilled to work with anyone. That's making these like that. Wants to see this reality come true. Because ultimately standards words are really important and they create the longevity and the opportunity behind a house something becomes not just a five five year opportunity but at twenty or fifty opportunity. I agree with with what you said. It's it's hard to exactly how the market will shake Kowt but it certainly conceivable to me that it is. It is big and it's going to be changing as much in the direction of coober netease as you're indicating reading but just to give a little bit more color on the particular area of software that you are solving that is the the delivery process of a vendor delivering to an enterprise that process actually has a number number of rolls involved. And it's not just like I'm a vendor and I'm going to deliver the software to my enterprise is and it's GonNa all be handled by a solutions. Architect does actually multiple different people. They're gonNA ball. There's going to be customer success. That's going to be sales. Els is going to be engineering solutions architecture. And all these different roles within the software vendor within the company that is maybe the the database so the logging solution that selling to the enterprise and my understanding of one of the things you're trying to solve is the communication channel where they're all of these different roles within the software vendor are talking to the Enterprise D. Do I understand correctly that there is just like frustrations or difficulties around the communication channel between that vendor and the particular customer and you're trying to consolidate eight seven that communication. Yeah this is really part of our commercial offering right which is the vendor tooling to operationalize and scale the distribution. You should Kouba Netease off-the-shelf applications right in. So you know. This is a product that our customers from using for years. And it's really just focused on things like workflows and processes cusses for Release Channels Licensing entitlement management is like how do you create a customer license. How do you create different values and enabled different features per customer costumer? How do you make sure you're delivering the right version? Teach customer in assigning them to different channels and then. How do you manage the the process around you know support in making ensure that you have a clear process to troubleshoot those customer environmental initiatives rise? And to your point we've made all these features be very cross functional and so the idea behind the commercial tooling is. Hey you're going to deliver a KARANIS applications images to a customer. If you're going to do that like once or twice ace like maybe some are open source stop. It's going to help you out. You can write these little manifest that will invoke the preflight troubleshoot or you can make the Koch. ADM valuable but if you're GONNA do this and try to deliver your software to tens hundreds thousands of different enterprises will that's when you need defined workflows in processes and that's what our commercial product really offers. So you know that's when you would find that you're getting involved more than just an engineer to send some Cuban Eddie's manifest manifests but you're pulling in your customer success folks who want to know what version of they on. How are they using the software you know? What's the latest you know like? When did they last check? Check to see if there's a new version your support folks. You're GONNA WANNA manage the support bundles that are generated by to these customers and then sent up once they're redacted you're gonNA WANNA have you've you know folks that are in pre sales can create a trial license and maybe that try license is generated from a salesforce integration. And any time you put a new lead and salesforce we generate a new license to them and you give your pre sales team the ability to walk in to a customer give them a great demo set up an environment so they can try out your product right. So there's all of these different front folks that have to be involved in the process of going to market successfully with a coupon off the shelf application and that's what the commercial replicated tools are really designed for to close off what has been the hardest part of building the business over for the last five and a half years. What was the most difficult point? I'll let mark to crack it at first and then I'll give my my perspective. Okay from engineering a product perspective. I think the most difficult part is definitely been around. Just how fast moving in the ecosystem is five years ago. When we started the company docker containers were around but Kuban? Eddie's didn't even have the first Alpha release yet and so we needed to build a container schedule orchestration system to solve this and then swarm arm came out and if you think five years is you know it's it's not a long time to have gone from a proprietary scheduling orchestration system. Rewrite it to we could have support for dockers swarm warm and support for Cooper Netease and then finally to where we are now. Where we like? It's Cooper Netease native and it supports Cooper needs so in general. You know I'd say one huge technical challenge in product challenge. Just really ben to staying ahead of in where we need to be to be compatible with. Both the vendor who often is as you know bleeding edge in wants has a Kube Annuities. Application has the latest and greatest and being able to still support enterprise customers who may be run disconnected did offline older Santos in rural environments that they need to be able to make that compatible in from my perspective and there's so many different challenges. Yeah I mean this is the second business that mark and I have run together and when we first started that we felt like we had learned so much from the first business and now I look back five years ago and realize like how how much we've learned now and you know one of the core challenges that I've faced personally is just like. How do you describe really complex six and technical solutions that you're building that are super powerful but like you know when the market doesn't necessarily like not everyone's up to date on all the technologies easy you're talking about but their foundational in what you're doing? How do you describe that to folks no way they really quickly and see the value and want to learn and more right and dive deeper and reach? Docs like that's a really big challenge. Particularly kind of couples on on top Marcus talking about about how fast the market moves right so the Karinna Chico system is just been exploding and so if you want to describe it you have to reference you know the things that are neighboring your technology but you Kinda need folks to the up to date so they have to be. Maybe on hacker news every hour in order to know the latest and greatest or you know be following ever release and reading every every common blog post. That's out there but it's it's those are those are real challenges around building this kind of business guys. Thanks for coming in software engineering daily. It's been great talking to you. Thank you so much when I'm building a new product G two. I is the company that I call on to help me find a developer who can build the first version of my product G two. I is a hiring platform. Run by engineers that matches you with react. React Native Graph Q.. L. And mobile engineers who you can trust whether you are a new company building. Your I product like me or an established company that wants additional engineering help G two. I has the talent that you need to accomplish. Your goals us go to software engineering daily DOT COM SLASH G to I to learn more about what g two I has to offer. We've also done several shows with with the people who run you too. I gave Greenberg and the rest of his team. These are engineers who know about the react ecosystem system about the mobile ecosystem about graph. Curiel react native. They know their stuff and they run a great organization in my personal experience. G Two I has is linked up with experienced engineers that can fit my budget and the G.. Two I staff are friendly and easy to work with. They know how product-development works they can help you find the perfect engineer for your stack and you can go to software engineering daily DOT COM SLASH G to I to learn more about G to I thank thank you to. Gti for being a great supporter of Software Engineering Daily both as listeners and also as people who have contributed did code. That have helped me out in my projects. So if you want to get some additional help for your engineering projects go to software engineering daily Dot Com slash g two. I

Kuban Eddie Cooper Netease Kuban developer netease Kuban Eddie Amazon engineer ADM Kubo Netease Sumo logic Road Map
The Full Array of Big Data Applied to IoT

The Internet of Things (IoT) Show with Bruce Sinclai

58:44 min | 1 year ago

The Full Array of Big Data Applied to IoT

"Uh-huh. Coming out on they say beside the is e ink business show. Big data projects. They need to start small experiment. Innovate fields. Phil Austin feel false. Learn from it. Really all the lean startup methodology comes in page when you're working with begetter IT projects. Because you wanna you wanna hit rate fast. You when you wanna you wanna change false when something doesn't work and you can't do that with be. Big data's big it's another entire industry. That subsumed by the internet of things. There's a lot to consider, but it all starts with your business information requirements. What you get by transforming the raw sensor, data you collect into business value. You can use in this episode of the business show, I speak with big did expert Mark van romina to conclude are seven part miniseries on analytics by covering the entire array of what's available. Oh, this animal on this episode of the ink business show. Thep people, the business and its hat -nology of the next generation, and senate's this is the C Inc business. And now he's your heist, Bruce sin CARA. Hello and welcome to the business show. This show is made possible by sales of my book not Inc. Published by McGraw Hill, and I o t inks certified IOT professional or I c I p online training and certification program become a certified IOT professional by completing the programs three courses by technology business, and I say peace strategy and digital transformation. Details of which can be found at WWW dot IO IT dash, Inc dot com. That's WWW dot IO t dash. I n c dot com. With me. Today is Mark van remnant, Marcus founder data flock a one stop shop for big data. He's an entrepreneur public speaker and strategist as well as the author of the book think bigger developing a successful big data strategy for your business. I thought Mark would be a great guest to wrap things up on our mini series on analytics, Mark, welcome to the show. Thank you for having me. So it's Friday, Friday night for you, and you big plans for the weekends. Yeah. I'm gonna ask some good to parties out of me. So it's gonna be a great week and again parties uh. So what type of parties birthday parties that's horrible things. These little birthday parties are big birthday. We'll find out tomorrow. Well, you tell me about yourself and your background in big data. And specifically in IOT sure while I started working in big data about three years ago when I when I first founded a big startups, which is the predecessor also data. Flog. Since then I've been in, in, in the in this area, which I very, very much enjoyed because it's this is the area where those things happening, and big data and data is the future, of course. And of course, also the internet of things where everything when we have a connected world, where massive amounts of data will be generated as always fest. Fascinated me. So, yeah, I, I started a company, you really say the idea to help organizations. And at the moment to the idea, the emissions of data flock, is really to empower organizations to create value from data, but really from a business perspective. So with data flow, we don't have any, any technology at the moment, but it's more really from a strategic point of view advise. How do you start with big data where different vendors that you want to approach events jobs etcetera? Well, I mean for listeners to this show, they'll know your man after my heart, you said the word strategic. He said the word business. He said the word value, and I'm all over that. So no, that's great. And so what drew you to big data? Did you have this in your in your passer is just something that you're interested in? It's just something that I'm interested in started working. Well, I just had a few almost three years ago, and I've always had a bit of a good overview of the different trends in the world. And I will big data is, is, is one thing that said it's going to be big. I want to be part of it. What's your background? Are you technical guy? Business guy, this guy. I've most in business. All right. Well, good. Good. Because we've had way too many techies on this show. Okay. Beside being a business guy besides being a party animal. What what's another interesting fact? Okay. Well, you said, well, I don't know other party weekend, but well, tell us something else, something interesting that most people don't know about you. I enjoy traveling. I enjoy some good adventure. So if you use ago, I assume navigated Australia on the push bike for genetic chills, fantastic. Experience than something I recommend everyone. So it's a yeah. I like I like doing things off the beaten track. I like that push bike. So what is that exactly push? It's normal bicycle but okay. So just. Warren speed kind of biker, specifically four for long distance. Okay. Yeah. No, I do I do a lot of cycling myself. I do some I do some long distance cycling myself. But I haven't heard it called push by a push pushback before, but that sounds good that you have Pani as and you decamp and all that. Hundred days to, to cycle over fourteen thousand kilometers. I think that's about nine thousand miles. Nice nice. We can't around the roads. And we, we managed to get almost twenty five thousand euros for, for charity during the lists also. You're also were raising raising money than now. You did. That's right. I don't know if either of us mentioned it, but you are the Netherlands aware in the Netherlands, are you I am in the Hake, which is our political capital of the Netherlands. Okay. And, and I've been to before, but remind me with respect to Amsterdam where, where resides about fifty kilometers south of. Yeah, that's right. I remember it was it wasn't a suburb, but not a suburb, but it was very close to term the net United States. It is a several events then. Very small country, a very small country, but very fun country. And you can almost say a party country too, but anyway, it's been a long time since I've been in Amsterdam, but I think, I'll be going there, actually in about a month and a half on, on a different trip that'll be in Europe. But anyway, so as I said, sort of, in the intro, we've been going through a lot of the details on analytics. So how it works. You know, getting into the mechanics of it getting into different types of the different types of analytics. So I thought you'd be a great guest one to put a business perspective on it, and to then also give us an idea of the packages. So what I wanna do for this interviews. I wanna talk we'll, we'll use the classification. We'll start with you know we'll start with real time analytics, and sort of take it from there. But if you can start with real time analytics maybe discuss it from, from a business point of view things that need to be considered maybe best practices. And then what are the commercial products of? Able because as we know platforms are not platforms often will have an analytics component. But sometimes you may have very specific needs or they may not be complete enough for what you need to do. So it also like you to maybe just give us some examples of commercial products party products that are available in each of the categories. So let's with real time analytics. Give us give us a bit of an overview on that. Well Ruth Olympics. This is very, very interesting area, actually, I wrote an article about it a couple of months about, okay, why real time. Analytics? And the internet of things are a perfect match. I got interesting. So it's a link to it in the show notes for sure them. Yeah. It's, it's it wasn't it wasn't an info. Graphic diet had found enough from company and a wife, you look at, at two minutes for into things, you, of course, you know, with the internet of things, we have massive amounts of data, streaming outweigh and realtime. Another enables you to, to add. Is that data to, to collect it to story to analyze to visualize it within milliseconds, and of course, very interesting from our business perspective, because all of a sudden you, you get inside about you company about your equipment about your customers on in mealtime instead of every week or every month. Of course, when that happens. Does require a different approach from your from from a from a business perspective goes, all of a sudden issue, if you, you get these insights, streaming to you all the time you need to take action as well. Right. Right. So, so how do you do that? Like, do you. How do you connect the dots, I guess, so to speak? Well, I guess it really depends what time organization, you are, of course, because if you are a temple a manufacturer where you have your folks only industry incidental, which is basically the also the things, but then from an industrial point of view, you can start working with, with predictive analytics or predictive maintenance for your for your equipment if if if something comes in, in real time that something needs to be done right now. It requires an action, and that's, of course, different from the policy, where in vocations reports reports would come in once recruiter, if they were lucky and only had to take action every week. So this is just one of the examples that, that, that one of the items requires a different approach within your organization. Now. Are you seeing real time? Analytics hooked up to business business systems such as lamb or CRM or anything like that or again. I'm just thinking, you know, you're getting this data's you said I mean, there's a lot of data coming in. You're doing the real time. Analytics, perhaps you're looking for some anomaly to trigger to trigger something, or perhaps you just looking for, you know, going in and out or of a certain segment of above or below a certain certain value segment. But how do you what about on the business side? What are you how you how you taking advantage of that from mechanics point of views his again? Is it hooked up to two business intelligence system, or or is it done by hand? How what have you seen? Oh, if while I talk about the initial manufacturer you have systems that, that, of course, trigger an alert. That's what's all about a predict air, predictive, maintenance, something something happened in the system is being created that's being sent to, to, to the person that needs to take action. That's something that from a business perspective than that person needs. To do something in order to ensure that the continuation of all of the processes that wasn't involved in. If you look at from different expect perspective. For example, more into the IT operation. It becomes really a part of the vital infrastructure of a lot of companies, especially in the tech industry. You need to have these real time. Insides in what's happening with your with your tag with the IT systems to ensure that you have continuously operational system, for example, in the banks that the needs to be oepration all the time. These real time analytics can help you to, to stay online all the time. Yeah. No, we're going to talk about predictive analytics visualization in a few moments. But, you know. Again, taking action on this date is at that you're getting a text or getting an Email or is it actually you're hooked up to another another system with an IT in what have you seen like 'cause I understand the concept. And I think most of the listeners understand the concept of real time. Analytics and the result being some notification, basically, it's gonna be some it's going to be some sort of notification. And I'm just again wondering how is that notification manifested as it? I mean, I've heard tax heard emails, connecting up the system, you know, maybe connecting up to another system. What's your experience? I've, I've, I've heard the same as you as you did. Okay. These are the manual trigger say, regime menu somewhat notification is triggered and someone has to take action also have the medic reactions where notification is triggered within the system and the system takes action based more on logic. And if this happens than that, of course, so you can have the manual reaction where someone in the company needs to do something, or you have to system do something. And, and how the notification is being funded by text or Email, or whatever that, that, that doesn't really matter. Yeah. No one. I think that's I think that's right. I mean like you said, there's the manual there's the automatic and the automatic, you know, from the way that we've been looking at the technology, there's the there's the app been looking at it from a product point of view, so does the app that's being developed, and in that app, it will be looking for these triggers, and then it may do something to the system, if there's a closed loop, for example, a may actually go back and modify. Something or may, you know, set some visualization. But yeah, I think, yeah, I think that confirms kind of the intuitive thinking of manual automatic when I like the way that you kind of broke it in the two. Now what about some? What about some examples of some commercial products are available for real time analytics? Well, one company that came to mind was a sumo logic. Let's okay. Right. Nobody company being backed by investors will and they have created is cloud based service that collects manages analyzes data in time. So it really allows companies to analyze old data from the different apps network systems regardless of typhoid immoral caption, and that moves information about yeah, what to do in strict. Yes or no? Based on shorten a various you've set. That's very interesting company to doing. Very well. And there's another company that came to mind, which is was jar- te'o. And, and they re aim to help companies make better decisions with with. Yeah, we intuitive in real time. Analytics dashboards that use data from different sources and chill in real time. What's happening in a really from a business perspective. Usually take action. Yes or no? Now. I mean you brought up an interesting point cloud, and real time and I suppose real time is really just really depends on your definition, right? Because you know what we've also discussed the show is where you're doing. The analytics you could be doing it. You know distributed a closer to the source, you could do it in the cloud. And with the first company mentioned, by the way, we'll be listening all these companies in the show notes, so you don't have to take notes as you're listening. We'll have him there just than in your click on the link. But yeah, so if you're if you're analyzing in the cloud, and obviously, there's going to be the latency of getting up to the cloud in the computing. And then coming back down. So real time is really, I guess it really depends right? I mean on on what your definition of real time is? Yeah. Yeah. If you if you if you take into account seconds, it takes to, to go to cloud in back of core for phone companies that way too long. Think of investment companies on the Wall Street they can money in those milliseconds true for them, realtime is more real time than for for, for different for another company show. Yet, you're right about that. It really depends. But that's basically, I think with everything within big data. And the things is that everything depends on what kind of company, you are, what kind of industry, you're in, there are different possibilities in different industries, and different use cases in different industries with different companies. And that's also what it makes difficult for generations to really cross big data because it means something different for everyone. Yeah. And that's a good point, you know because of the amount of data also I mean doing the real time. Analytics, you're mentioning predictive as well. Which will will discuss in a little bit further. In that case you're gonna be doing some, some preprocessing, and then you're gonna be comparing to that pre process model in real time. But, you know, depends on the amount of data to right? I mean, if you're if you're sending WADA date ups at a cloud in office, that's gonna take a bit more time. And, and like you said, I mean it's, it's really difficult to get into specifics. We have to talk in generalities, because there's so many different use cases, that may come difficult. Yeah. Yeah. And so if you're advising, for example of someone came to you and said, Mark, you know, you know, we need a real time. Analytics solution we are making a product for example. And we wanna know when this product is, let's say running out of battery, or something like that. How do you guide them? You know, like where do they where do they start? How do you how do you sort of take him through this, this path? What is there? Is there kind of a logical sequence to go through there is there is a logical sequence, but it's? If if they are that far already today that they will they develop the product. They know monetary. It's time to see. Well, we in which. Well, of course, geographical location, also helps. Sure. But also which what are the requirements? What are essays what's really required for, for for the system. And based on on on these different criteria. The price do you want open tour or not point, only different criterias are to be to be a taken into account for, for users, if you if you if you if I'm going to give an advice to, to the company? Actually interesting to know, for, for listeners well, is that this particular thing is very difficult for innovation to find the right to a bit of promotion on the flock. There's a search engine that just does that really at it's going to launch the end of August and it to help really patients fines derides data solution, whether it will be, for instance, things or any other data area based on different criteria. Nice nice. No. I'd like to talk to, you know, maybe off line further on this. But that's a good resource in so I suppose by time this episode airs will not be available right? I mean, but, but maybe there's a page of scrubbing I can put that in the show notes. Oh, we'll something in shots. Okay. We'll put something in there. Yeah. And I think like you said, what's most important, and again, as listeners of this show. No. What I what I promote is very much building your information requirements in advance and the information requirements, of course, are going to be defining the model that you need to use the application that you need to use also the type of analytics, and I think the other things that you mentioned price obviously being an issue and somewhat related or directly related to open source versus versus a commercial platform for the first two, I mean, are all things that, that you need to have sorted out beforehand. What I've interested is just an has just talking to you again, off-line and Mabel do some sort of a follow up. But the information that you're looking for very interested to know you know, the, the different the criteria that you need for your search engine just to see if it lines up, you know, just how you're looking at it in the type of information, but we can do that at a later point to civically on the two coming. That you mentioned are these are these are these open source. Am are these open source companies using open source database technology or they, I guess closer I should say, no just regular vendor regular vendor base. I'm not sure. But I believe they are regular under racia-. Okay. Yeah. And but as we've both said the that's an important part to consider as well. All right. Let's shift gears. So real time analytics. Let's go into descriptive. Analytics and descriptive. Analytics? There's probably different definitions, one of the visualization I think that that's considered in their taking -lica- did, and making a model a mathematical representation of that. So you can describe it. So again, let, let's do the same thing for descriptive. Analytics, discuss discuss it from business point of view in what needs to be considered best practices, and then and then I'll prompt you on the on the products as well. Okay. Yeah, it sounds good. Well, what I always if I need to explain to people did. The difference between descriptive. Analytics, predictive analytics and prescriptive. Analytics, yes, I'm jumping bit of too much into the show. I think it's interesting to, to, to give the explanation because when you look at descriptive Dixie it's always as as looking behind looking into history. Now you driving a car and you're looking in the mirror. You see what, what has happened? That's descriptive that undertakes everything from from a second ago to ten years ago. Yes. Predictable really looking in out of your front. We're window to see what's happening and mind to predict what's, what's, what's coming out of you. Use all the data from the descriptive under these, you have as well to, to, to build a model to predict what's going to happen and bring a prescriptive analytics, which basically the future all big data as as I see it. Which is still in its infancy. Is really building in a an infinite amount of scenarios of what's going to happen. And based on those infants among scenarios, you make a decision and systems going to recommend or even take action. What needs to be done, and that requires dramatic dramatic amount of data and good example of, of product does that is, is to go self driving car. Right. Because they use prescriptive founded the stool, and to understand where where should go to car go to and which which routes should a car take. So these are the three differences between the three different types of energies, which are out there, which, of course, scripted out of the most fascinating one. Yeah. Known. I like I like that because we haven't really touched that much on prescriptive analytics in the show, you know, we've broken it down. Maybe incorrectly. But real time analytics scripted ventilation and predictive analytics. But very much so prescriptive analytics, like you say is similar to predictive analytics except for your closing the loop, right? I mean you're you're, you're looking at a bunch of different predictions, and then you're actually using algorithm, ical you're making a decision based on those different predictions to take certain action, and I liked that. So you see this as being the future of big data big data in general. Yeah. Because he did yet the of course, the objective is what we see right now is you to predict what's going to happen anyone which customers you need to, to treasure when predict direction into need to need to states when predict winsome plan something breaks down. This takes it one step further. And it kind of actually incorporate algorithmic lead to do those predictions in and then make an action that you don't even have to be involved in or that the that the user doesn't even have to be involved. And that's fascinating, I think, and the possibilities in that area are just endless. I think they are they are. All right. Well, I wanna get to that. But let's let's, let's go back to the script of analytics, I think the descriptive and at least I think that's you had a good perspective on it looking in the past, even though the past could be you said a second kid, even be half a second or less. It's still in the past, you know, because sometimes there's real time descriptive analytics, but it's not as we were saying it's not really real time but descriptive analytics okay, from the business point of view. What are you looking at? Is that depend on the problem that you're solving, I suppose, is there any general guidelines that you can give our listeners to win their? They do need a script of analytic solution. What, what should they be house, should, they start on this journey for me? A descript fantasies, basically just Bisson. So reports on what's happening with the. Within your company reports on the sales of your products across across the globe reports on new customers holiday behaving on your website of, to me that is a basis intelligence presented. The exist is business reports that help you to, to make better decisions based on, on the data that you have. And of course, the mortgage you have you can combine you can combine them with, with, with each other. And the more and better insight you get out of it. And where does visualization come in the play in? Visualize is extremely important in this area because it really helps you to play around with what's what, what out there. And and to, to better understand, how different data source are connected with each other. Jeff, you companies, of course, probably familiar with them that to blow software. One of them do does a great job and clear story data as will these are two visualization companies that really are able to combine a lot of different data sources and visualize that. And give the user the ability to play with and to understand what's going on. And to dive, deeper into the into the into the data, but visually and dating of descriptive undertakes detailed shoe to, to dive deep into the data to see what had happened, but then visually. Yeah. No. I think you know, I kind of look at it from the perspective in, in correct me if I'm wrong. But, but when you using like you said, you can have these reports they could be number based, you can actually develop models and they can be mathematical bay mathematically bays. And these are discrete representations, however, when you do a visualization instead of using the mathematics to kind of come to some conclusion, using the brain, and sometimes the brain has a lot higher capacity to kinda make correlations between what it sees obviously, you need a good package that you can go in and out of all the different dimension to build to do so. But is that a good way to kind of look at it visualization is kind of using the brain the process it, and then or your mathematically representing it, whether it's charts or where mathematical equations? Is that a good way? Is it is it helps you making better decisions lie giving you the tools to, to discover what's going on? Also people call a discovery entities. You discover what's what's going on within the data to have a better understanding about your company, but your business about your processes about your whatever. And then by visualizing that all that data, you really can grasp it and better understand. And, and by that you are able to make better decisions. Okay. Well, you've mentioned Tableau and clear story data and others are more in the visualization size. Are there any are they all visualization base, or, or they're kind of any on the other side that we talked about, like the like the modeling you know, putting a model on something that happened in the past are there any packages you can recommend or that you can, you can mention there, or is it more? This visualization kind of more covering it for me visualization is more covering it. But just my perspective. I know a field, our visualization companies of it for me. It is. Yeah. The most important thing. I'm not of my mind. No mulling complete no, no, no problem. I think you're right. I think all in all when people talk about the script of analytics, generally talk about visualization, but it should just everyone should just have in mind that there's also different ways of describing things not just the visualization like we're like we're just talking, mathematically. You know, whether you're. Offing it or the you're displaying as an equation. But somehow you have to take action on that. So I suppose mathematically than you're really going to be feeding that into maybe what we're gonna talk about next predictive analytics, and using using using mathematical model for predictive analytics, but before move on the predictive analytics anything else, you wanna talk about on descriptive. No, I think we've covered whatever to say about it. Good. Good. Okay. So last talk about predictive analytics, we've, we've talked about predictive analytics and show, you just mentioned a couple use cases, the big one, of course, is pretty of maintenance, but, but let's talk about it a bit more from a business perspective. What when someone wants to do predictive analytics? And the reason they know they won't do predictive analytics thought through the information they require, you know, in the case of IOT all incremental value of IT products really comes from the intelligence that you can gather by transforming the sensor date. That's what it comes down to. So if you look at it from that perspective, again, we're going to start with the business intelligence requirements, and we're gonna go down from there. They're going to know that they wanna they're gonna wanna perhaps predict when something's going to fail. They're one of maybe predict the time of something, I guess, a time is that is that is that basically the, the major dimension or is there other are there other things that, that we should be considering predicting, I think our Milton dimensions time location. Right. You know, they're, they're really it's not only about when, when something happens, but also where from thing happened in housing have been good point is. It's really a lot more Milton dimensional. Yeah. Sensors in all the data that streams from from it can be really mill two dimensional to see what's going on and, and make prediction from there. Yeah. Okay. So I kind of cut you off earlier. But from a business perspective, what are we looking at what's important? What should be considering were when we're thinking about incorporating, a third party predictive analytics package will, I think what you should be aware of this. This is typically these, these project are, are not that easy to implement and okay, I've heard you say this a few times. But anyway, keep going, and they, they can can get out of hand can't get out of hand at quite false. I know interesting use case from UPS. I'm not sure if you've discussing journey previous note, no have an audience. But, but UPS did they for them the their big data IT project is a one billion dollar project while fell? That's what I mean with that they can get out of hand, it can be really big Dave done is, she's very interesting is they have installs all kinds of senators within Karsh. Demont cars performing and and, and to see how the drive is driving into see how they can improve that to see if they improve the reduced the fuel usage. But when it's more interesting today start combining that censor data from the from the tricks with all kinds of data such as whether they'd as as. If you'll prize data such as traffic, gems, it shows, batches wherever they have to go cetera. And based on that the system calculates the most optimized rooting for for the tricks. In addition, if dissenters note that something about to break within the truck, that's where you have the predictive maintenance, the system, incorporates in the routing that at the driver needs to stop at the maintenance ship for, for fixing the problem there by just gusting as little time as necessary. So right fascinating example of the extreme possibilities of, of big data and communication with IOT. What's capable when you start combining the data from your sensors in using multiple senders, and starting started start when you start to combine it with different data sources you? And I think that is that's a great point that you're may. Making and something that the listeners have to consider as it's not just the data come from the sensors, you know, we call it. I call it external systems you know, attached it in fabric. External systems are in my in my mind. They are third party analytic systems, but they're also services from the weather company as you remembering or they could be pricing where you hook up to an information service. It could be again, we were talking about business business systems that could be your P L M your CRM, and it could be other IT products, for example, other cars in their case of UPS. It could be other other vehicles. So that's, that's real important. I think number one point is that consider when you're looking at when you're looking at all the analytics really, it's not just a sensor data since data's probably a big part of it. But it's not just that the second point is the hand very fast point. And I don't know they're eighty best practices here. I would think logically, you just want to start small expand from there. But. But, but how do you is that is that the extent of the advisors or anything more that you can add to that did the main thing is, if you want to start with IT's start small start, experimenting and start started getting shot to get going? I think that that's that's very important. Because if he is, there are so many different factors that are involved with with mighty product as we just mentioned. That if you if you want to do and large scale, you sure you're gonna you're going to lose it. So, right. Right. Really start. Small think big. Yeah. Yeah. And that's, but that's every big data project, I need to start mole experiment, innovates sales. Ill often feel false. Learn from it. Really all the lean startup methodology comes in patriot working with big data IOT projects. Because you wanna you wanna hit rate files when the wanna you wanna change false when something goes on work, and you can't do that with me, big project. So. Big data and IT also requires a different way of working at are already discussed the beginning different approach different culture. So if you that also needs to grow when you start working with, with your big data are not project. So starting small starting to get involved all the different people within your organization and and slowly expanding from their own. That's, that's the best way to go on my specter. No, I agree. And I think I think you hit on the two different points. Yes. Start. Small technically think big start small, because I think if there's only one known is that things are going to be different than what you thought things are gonna change. Right. So do that. And then you bring up another point which is also important that, that we've talked about, as well is within the company, you know, if you talk about implementing this within the company start small again, do not involve a whole business unit. That's not gonna work. You cannot do it by committee. You're going to start with one guy. He's going to be the founder. He's a technical guy. He's going to bring on a business guy. He's a business guy is gonna bring on a technical guy and interest to of you. And then you're gonna hit rate on this, and you're gonna rate on it, and you're going to build a team internally, it's a good point. And it's it's important to keep in mind. Innovations what they do is they put this outside of that company. So, yes, African wool mart. They have Woolmer laps him. That's what they take it out of the company. Give them all the freedom and really to start a methodology. Go go play and then, then they come up with these fantastic features. So that's, that's the way to do it. No, I agree in there can't be the pressure, the Pia. No, there can't be the pressure the quarter the quarterly. And so, yeah, whether you go, whether you go to that extreme where you start absolutely like incorporated, different company. Or whether you just have a startup within your company, same thing, right? You need to insulate that team start at small than an ingrown from their packages third party can give some examples of some third party products that do the predictive analytics into speaking about. Yeah. Sure, one of them that I came up with. We talked about UPS company transmit Trix is interesting. I don't even from Europe and what they are doing is trying to predict the freight within transportation companies to optimize it. So they is specific very, very specific industry industry, specific example. But I like it very much because the predictive analytic social, they are developing. And which is valuable as a as product really helps transportation companies to predict the shop shipping volumes. Sulfur learned I believe it learned from from pass shipping histories and use external four costs to make these predictions, reading the things that would just explained and it really gives them the. Yeah. It helps dress tradition companies to, to save money and to them by predicting the right freight transport very interesting example of how you, you know, in a very specific industry. Can take example can take a benefit from from predicted. Analytics? No, that's, that's a good one. Any others that you can that you can let us know the added company in for if you're familiar with that no, I'm not either. They deliver data our business relations win. More customers and also cloud base -lution really focusing on, on, on different signals from from the web to predict where you're next customer will be up to find new customers. Forget for, for ecommerce websites for, for fall short for, for ecommerce companies, it's gonna help them to, to, to, to, to find a new segments in different areas that didn't think think of four. So with predictive annotate it can't react to customers. I'm doing a little predictive analytics on what you're saying, and I'm, I'm thinking back our, our last guests Neil from Kuala sent analytics, they did predictive analytics, but very specific on manufacturing and finding defects on the manufacturing line last two that you mentioned inform and trans metrics again. They're very industry specific is that is that the trend. I mean, is it that when you're looking for a predictive analytics solution, you need to look for one that's within your industry or or they're more? Or can you go with the more generalized package as well? I would go with Moore civic goes, okay, if there isn't ones, authorise added a generic one can eat to tailor to wash wash industry because every, as I mentioned every industry has as a different approach requires there are different requirements for diff. Industry define intially industry needs different data sets than the retail industry and energy industry needs different from the carbon picturing mystery. So it really helps if you have. A more specific tailored company or package that knows industry that has they designed to understand the specifics of assorted industry. I think that really helps. No, I think you're right. I think what it's telling our listeners is too. I look in your industry may be used your search engine that might be one way to do it. But look, within your industry, I see what's there and then also considered generalized packages. But knowing but knowing that you're gonna need to customize them in some way or form. And you may need a little bit more data science. I guess I pee in your company to build a us them. But that's yeah. And that's that's a good perspective on, on how to look at it. Prescriptive analytics line, a we've, we've sort of touched on them, and it seems just to go back again, real time analytics, and descriptive analytics that almost seems independent of industry because of the functionality that, that it's, it's performing predictive. Analytics and prescriptive analytics will seem like you were saying very more specific. I mean, certainly if you can find a more specific package than it sort of gets you a little a little further. Is that a good? Is that a fair characterize? Absolutely. If you are few companies at work with prescriptive Dixit have developed prescriptive analytics, algorithms, and manages and one of them is yet. And you see is at a focus predominantly owned the oil and gas industry. A lot of data is involved in gas industry. Finding new oil wells are maintenance predictions on the only drilling drills so that requires a lot of different data sources to, to make those predictions, and to really understand what's going on so out, focuses as built these prescriptive packages predominantly for the oil and gas industry. Yeah. And, and any others any other prescriptive analytics package that you can use his examples. Came in. It's pretty cutting edge, and it's pretty in. It's pretty, you know, it's, it's, it's pretty new stuff. But I think the point remains the same in the same point that we're talking about is that. Yeah, look in your industry, but, but don't discount generalize. Analytics as well because, you know, sometimes a custom solution might be better for you just may be a little bit more work, you know, within the company, however, you know, one way or the other my advice to, you know, to, to companies that are looking at analytics, and looking at the modeling and looking at the apps is this is stuff that you have to do in house. This is not stuff that you should be farming out. This is my perspective. This is the intellectual property. You know the other stuff the plumbing the network fabric, the sensors all that. Sure, I mean, a lot of that you, can you can farm out. But when we're talking about Pacific -ly when we start getting into value, and specifically, how we're gonna get that values to information than that's a stuffy need to do in house. So. So, yeah, I mean, you're going to need some form of data scientists in house now, whether that is to use a customized prescriptive or predictive. Analytics package or two. Maybe even go a little bit further customize a general, a general one. I think that's an important under. If you have a different if you have a different or anything else to add to that, but that's important consideration as well. You're going to be doing this in house. If you're developing an IOT product or you're bringing out into your into your company, the help help, make it more efficient or drive some different sources of revenue. This is something you need to bring in houses. Is that your view or do you have a different view? Absolutely agree with it. Okay. All right. So we're on the same page with that. All right. Well, there's at least two of us that agree on that. I think I think I think just logically speaking again, when you're talking about value, and when you're talking about, you know, I always look at things from a product company perspective billing and not product because you can look at it for many different ways. But again, if you're building a product, there has to be some intellectual property there. Differentiate yourself and. An analytics being one of the three horsemen I've mentioned before analytics in, in my vernacular modeling, which, again is related and then and then apt development, these are three that you need to you need to bring in house. All right. So I think the last thing I wanna talk about real quickly and maybe maybe maybe we, it's really independent. But we, we did touch on this. But distributing the analytics, we've talked about fog computing, cloud computing, and obviously everything in between even sensor computing. There's some platforms now that have a little bit more from the embedded system point of view. The little bit more horsepower memory and so forth. What's your perspective, there from a big data perspective, is that again, just based on the use case? Everything's open. All options are on the table. Or do you have any is there any guidance? You can help us with when we're looking at, you know, making these decisions, whether we're where we're going to put this where we're going to put the analytics, I guess the question very important. Because if you have little sensors and the generate a lot of data which which some industries that happened. Do you get the question are you going to transmit all the data? And then the question is, do you want to transmit old role data back to your to your cloud, whether it's private cloud hybrid cloud, or public cloud or do you want to do some on this injure using using computing, to do more locally, and then transmitted, right? Right. Very important. Question that you need to ask yourself, how much data will my Senator generate and ketone can I deal with the head a sending transmitting and receiving the raw data, or do I need some, some sort of analytic, some sort of screening to be done locally because otherwise it becomes too much. Yeah. I've seen exams. I don't have got the names of these, these examples. But where they did just that on the local system. Only Senator do so Mattis's into to remove ninety percent of the data and only send on the ten percent really reversible. No. And I think that, that, that is an issue that has to be considered. However, I think when considering that issue need to look at it who listed because there's also there's also the potential future loss of information if you don't keep it. All right. You know. So there's a balance you have to make in the sense of. There's one there's a transmission cost, you know, then there's a storage cost, and then there's going to be the computational cost of, of moving that data. However, if you do you know if you do some local analytics, and by definition that kind of means that you're gonna, you're gonna be filtering the data, you know, whether it's gonna be some sort of there's gonna be some filter. We'll look at we'll talk about it from general point of view, that means you're not going to be storing all the raw data. If you don't store all the raw data than some point in the future, you're not Bill to use it. And so it's not an easy question. But there is that balance right of, of keeping that original data. 'cause once it's gone. It's gone, you know, and you're not gonna build the monetize it, you're not going to use it for any other purpose. However, that being said, I think I think general statistic is. At least today only five percent of data's being used as being acted upon so it's not easy, right? I mean, I don't know if you've if you have an experience in this with your with your clients, but any advice other than being very general, what we just said. I'm just consider consider all these different things. Yeah, yeah, yeah. There's not much that I can add to that goes. It's going to get boring. But it depends on the type of company type of project ministry. So, yeah, you really have to define for your cell. What is important? And again, as as you mentioned, all these issues are for, for your company, you have to determine what is important. For your new raw data that's gonna cost, you do don't want, do you don't want that, that's fine. Daniel gonna lose something in the future potential opportunity cost, and that's something that you have to look, into and these decisions, I might look very small now but they can have massive consequences in the future when you built the wrong systems. Yeah, yeah. And interestingly enough. However, they're, they're a gut decision because you don't have you don't know you don't know what's going to happen to the future. And so it's really the session, and I guess, maybe it's a budget decision you, you save as much as you can budget for. Yeah. I don't know. But, but yeah. It's just it's just important for him to keep that in mind. All right. Mark world. That was really good. I really appreciate your perspective on it. I like I like that. We're aligned on the value, the business and so forth. A working people. Find out more about you and your company where where should they go? If they go to data flow, come and in the queue that'll be in the show notes. They can they can find all kinds of information on big data on, on Edison, Janu will be there events are there, if mission about me is there, also Twitter Arlington. We can connect a lot of I'm quite visible online. Okay. Well, we will get all your coordinates, and we'll put them in the show notes. In case people wanna follow. You also notice that you have a course, which seems a really interesting to me on your on your webpage. So, yeah, visit data flock dot com will put in the show notes and with that I'm going to thank you and have a good weekend. Thank you, Bruce. Thanks for having me here. Good job, man. Good job lead you like it. Yeah. Yeah. How'd you think you think that we cover the stuff? Okay. Yeah. Definitely. I think we recovered everything that we wanted to cover. So yeah, I'm I'm happy about it super. All right. Well, this is going to it just depends either gonna come out next week or the week after that, but I'll definitely send you all the information and we're going to be on social media. We're going to be promoting it. Into it. Well, they'll be they'll be almost ten thousand people. So it's yeah, it's a big number minute. Well, big number, you know, it's a big number for for, you know, very specific technology. But yeah, I mean and it just Innis along tail right? It just keeps on over time. I have I have episodes from six months ago. I mean, the every day, they're getting like thirty forty lessons. So, so, yeah. No, I think I think this'll be good and. Yeah. Interested Aleka saying online. I am interested in finding out more about your search engine. I think it's an awesome idea. And in particular, I'm really interested in how you, you know, the data that you're that people would be typing in, like what information you're looking for to make these recommendations. So, you know, I don't know if some point I wouldn't mind if you could share some of that with me so I can just take a look at it and maybe share it with a with a with my audience as well. It's going to it's going live next week. It'll be okay. Well, then it will be actually live, then when by the time this comes out. I will announce in a few days later. Okay. But yet it on August twenty fourth it's out there so August twenty fourth so ten ten days now. So it'll be the Monday of the phone, we based, right? Well, we can also meant the show notes even if it this is either going to come out most likely the nineteenth or twentieth or the twenty six or twenty seven again, I'm not sure which one, but even if we don't put in the show notes at that time we can add, it later, it just means that we won't get the initial the initial people because we have there's two types of people where there's people are subscribed. They're automatically downloaded to their phone and enters the people that haven't subscribed yet that will see when they hear for the first time it'll be there, so. The best be on the twenty six th. Yeah, yeah. Yeah. No, I understand. Yeah, I'm gonna know I'm going to no more later today. I'll be able to tell you kind of the, the schedule as to as to the timing. But, but yeah, I mean I, I also some also some other considerations because I have a sponsor now for the show and. Yeah. So that's a good thing, my first one. And so they want to, you know, they have, they also have a consideration of when this comes out and so forth. But all know are one when podcast comes out so it depends. So I guess what it means is if I'm doing another podcast and this'll be on the twenty six. If I'm not doing another one. This will be on the nineteenth. So it just depends on know that later on though, the in one hour and fifteen minutes, more or less. Sation good. Let me know when it's out there, I will share it as well. Okay. I certainly will provide you a little graphic, and you can use that. And, and yeah, yeah. And I'll be looking out, then for the for the search engine, I'm interested in how you're going to do it. Because that is a problem. It really is one. And I think, you know, the, the point that you brought up, which is a good one that we haven't really talked about to date was this whole vertical. You know having a specific predictive. Prescriptive analytics package for for a vertical industry, and I don't know. Are you going to take that in consideration as well for your search your searching or? Based on categories based industries based on a short just annot basal offending receive based on size of defenders based on your location, based. Anything? Yeah. Yeah, good. So you yeah, so, yeah, you're you've done what I'm what I'm doing right now. I'm producing database of different IOT technology companies and I sit him. That's what you did. And then once you have database you can queue it in, in a variety of different ways. Companies in the database, so you serious. Wow. So did you did you is this all by hand that you that you kind of found them? Or is there like did you find a repository or how thousands of big number whole nation of, of all of? Automatically and artificial artificial intelligence artificial intelligence to artificial artificial intelligence artificial artificial. It's basically not not really artificial. All right. Well, I'm gonna let you go. I know it's later in your day. Have weakened have some fun birthday parties. And when I said, big and small, what I meant was it is a children's birthday party or party of mine. Friend of mine, then it's a big, big birthday party. All right. We'll, we'll be in touch. Thanks. Thank you. Bye. Okay. That was a good talk with Mark van rim. This podcast goes vertical deep diving into different topics each week if you prefer a more horizontal and structured approach to learning IOT business, and it's orbiting technologies. Check out my book, I t ink published by McGraw Hill, or become a certified IOT professional, like completing the icy IP training and certification program, where details, just go to WWW dot IOT dash Inc dot com. Also go to WWW dot bio, t- dash INC dot com for analysis of this episode links to things that were mentioned during the episode and very importantly, the episodes PDF transcript. Just search for the name of the episode or the guest if you knew this podcast. Subscribe that way you'll get every week, Sep Assad deliver straight to your device. Or if you've been listening for a while, there are three ways you can support the show. He can leave a rating or view on, I tunes, just go to is not dash, Inc dot com slash tunes. Takes one click lever rating, a little bit longer to leave review. You can share it on social linked in to lesser extent on Twitter. And of course, you can sport the show by buying my book, I t ink or the ice p training and certification program. So I pay the bills. Next week's episode is producing the digital twin for the industrial internet of things with the Mitri Volkmann. I hope you can join me, then I'm your host, Bruce Sinclair. Thank you for listening till next week, may your Pattaya business, a big one. You have been listening to the is e ink business show.

IOT Mark Europe Bruce Phil Austin founder senate Mark van romina Amsterdam United States Senator sumo logic typhoid C Inc
Equity Monday on Tuesday 09/08

Equity

08:43 min | 4 months ago

Equity Monday on Tuesday 09/08

"Does your start up to get a sock to report to close big deals or do you already have a sock to report and want to make it easier to maintain year after year Vantaa has built suffered. The makes it easier to both get and renew your sock to with anti you avoid anxious auditor interviews and you don't have to capture hundreds of screen shots proving that you are sought to client to your auditor companies like Laddis user testing and hundreds of others have successfully gotten. There sought to reports with Banta appleby listeners can redeem thousand dollars off advantage description by visiting fans dot com slash equity that's vanda dot com slash equity. Hello, this is Alex and you're listening to equity Tuesday for September eighth twenty twenty. Now, have you wondered how it feels to be part of the company that owns talk as these social service becomes an international political kickball. Well, it's so much fun to work for by Dan's the famous china-based Unicorn has declared a rare half months salary bonus for employees according to Bloomberg. This article cites the desire to calm the company's workforce presuming everything over there is lovely. And in turn against the narrative Netflix's Reed Hastings came out over the weekend against working from home that he sees no positives from it according to Hastings. The future of work is probably four in-office days and one work from home. Perhaps a bit more flexible smaller companies. We think maybe well, most are heralding a changed the world that fixes boss doesn't seem to quite agree with it his perspective here good point to a more traditional world post Kobe, something that many software startups and both private and public investors are betting against. And the continuing saga of China versus the world continue with more fights breaking out between Democratic Australia and autocratic China this time a few Australian journalists in China appear to have barely gotten out of the country while another Australian citizen. Chiang, who worked for the China? Global. Television network has been put under residential surveillance according to the country the tension between China and countries that have representative governments continues to get higher over time. This could be draw world boosts the importance of India to global tech companies and start ups alike and could draw the future faultlines A. Shattered Internet, and of course, whenever we talk about China in a business context, it is worth repeating that human rights matter and one party governments are bad. That is what is top of mind for me this morning. Welcome back to the work week. It's Tuesday here in the United States and we are doing equity today because well, it was a holiday yesterday. So we are catching up we are getting to the email maybe but before we start equity is back on Youtube, you can check that out over at YouTube dot com slash tech crunch clips from the show, and of course, follow equity on twitter at twitter dot com slash. pod And with that, let's begin. As always, WE'RE GONNA kick off with some news from the weekend or from very very early this morning and beg us out according to me is that there are new IPO filings from both Jay, frog and sumo logic to companies that filed in that great IPO wave of a few weeks back we have been tracking both companies impending debuts because we care a lot about them, their software companies, it's been venture capital money at play that puts them right in the very middle of what we care about here at tech rudge. So the news. To s a filings when it comes to developer focus J. Franck, it is going to target thirty seven dollars per share at the top end according to this initial range, of course, and could raise as much as four, hundred, ninety, two point, two, million dollars, and we did some very. Rough math this morning. So you know have the jobs. So in fact, check us but we believe could be worth as much as three point three, billion dollars at three seven dollars per share it not counting select shares reserved for its underwriting banks. Turned into sumo logic it will target twenty one dollars per share for an IPO that could be worth three, hundred, Fifty, seven, point, four, million. That's how much could raise in the debut and if you do all the math. It could be worth two point, one billion dollars again, not counting shares reserved for underwriters more of both of those to come on the site of course, but it's great to see those ideas moving forward making stores markets. It means we're not going to be bored for weeks. There's GonNa, be lots to talk about. So get excited in the stock market struggling and we'll talk about that in just a second. And two big pieces of funding news. Before we talked about earlier stage rounds, I wanted to make sure we mentioned these because they matter and they are kind of thematically lined with stuff. We've talked about on the PODCASTS A lot. So here we got Indian online learning giant buys you raised five, hundred, million dollars at ten pointed billion dollars valuation according to reporting from. Tech richest manish Singh. Now. By you confirmed the investors Silverlake Tiger General Atlantic Alabama's all that they did not confirm the dollar amount evaluation. We got ourselves Ed Tech. Is, of course hot in two thousand twenty as the world learns how do work from home and learn from home thanks to covid nineteen. So not surprised to see more ED tech rounds but of course, a five, hundred, million dollar investment is a large. And Molly offers API based payments tech for other companies has raised one hundred and six, million or ninety, million euro and around led by. That pushes its valuation north of the one billion dollar mark, which makes a Dutch company a Unicorn according to techcrunch dot com. You may have heard of them the company's quote on track to process more than ten billion euros in transactions. This year representing growth of one hundred percent on a year ago. Of course, this is the space that Phoenix Stripe, and other players play in. Now onto funding rounds, there were a few funding round note from the last couple of days I'm going to give you some nails on three more links in the show notes. Of course, run calm I will have all the other stuff that you need for these US goods are staying hot as fashion file a quote preowned online shopping firm according to payments dot com as thirty eight point five, million dollars in a series B. from new spring growth ability by Jews. Round wasn't a shock ecommerce hot anti-tank is so. Seen fashion file raise a lot of money not a surprise. What was fun though and this guy was fashion file was founded back in one thousand, nine, hundred, ninety, nine when I was ten that's that's that's an old company fund. SEATTLE. Based cloud density has raised thirteen million according to thin SMS DOT COM forged point capital led that round, and as you can kind, of guess, by the name Cloud entity cloud entity does cloud identity and security stuff. Unsurprisingly and finally suffocation think a French company has raised twenty seven million dollars according to techcrunch. The Mar- tech startup is also old having been founded back in two thousand one. So today we really saw a couple of companies that have some years on them raising good says giants, which is always a delight. You know we talk a lot about Y combinator startups and the Hugo founded last week and all this. Sometimes, it takes awhile sometimes you guys cookie building that company and then you raise money and that's without today. So obviously, the busy market is still nice active. We are nearly at length today, but we can't close without taking a look at the stock market. Frankly after a truly epic run, it turns out that was partially fueled by Mariachi, son making huge complex bats in the value of tech. The stock market is melting a bit tech shares have fallen sharply from recent all-time highs in the United States and are set to open once again, sharply lower as I talked to you about nine. AM here on the East Coast now, a few days of bad trading does not a trend make but with NASDAQ, set to open three point four percent lower than Friday and Saas and cloud shares already in a technical correction, it's a warning sign that the president good times do not have to persist. They're still here probably but. We're. Not Do them. Owed us they could go way whenever sure lots of folks are still chanting that stocks only go up and everyone has a theory why that will keep being the case but stocks do do go down they do and they sometimes do it very sharply as we've recently seen. So as we keep working for or building or investing in startups, keep at least a little bit. Of your personal vision on the macro climate as it will have an impact on your life and business, the slide continues naturally I don't know what's going to happen but certainly doesn't look particularly good this morning and I was equity Tuesday September eighth twenty twenty follow show on twitter at equity pod or myself on twitter I M Alex Fear Back Friday morning unless big breaks.

China twitter United States sumo logic auditor Banta appleby Youtube Vantaa Ed Tech Reed Hastings Alex Dan Netflix Bloomberg Laddis India giants Kobe Chiang
Drones with Chris Anderson

Software Engineering Daily

47:36 min | 1 year ago

Drones with Chris Anderson

"Drone applications are easy to imagine drones will deliver food to us. Drones will be able to extinguish fires. Drones will be used to relay relate Internet signal and make the world more connected. These all sound like great ideas. So why aren't there more drones in the sky. Today there are many answers to that question. Some of which relate to engineering and some of which are about regulatory barriers Chris. Anderson is the CEO of Three D. Robotics a drone company which he started seven years ago before three Dr. Chris worked for many years as a journalist writing about technology and science. He was the editor in chief at wired for eleven years a writer for the Economist for seven years and spent three years at both of the leading scientific journals nature nature and science. Chris is highly eloquent and has lots of interesting ideas he also wrote the long tail. which is an influential essential? Two thousand four book which described set of emergent Internet trends. I read that book back in two thousand nine and it was enlightening. Chris joins the show for a discussion about drones journalism and his perspective on modern technology. This podcast is brought to you by Pager. Duty you've probably heard of pager. Duty Teams Trust Pager duty to help them deliver high quality digital experiences to their customers with pager. Duty teams spend less time reacting adding two incidents and more time building software over twelve thousand businesses rely on pager duty to identify issues and opportunities in real time and bring together the right people to fix problems faster and prevent those problems from happening again. Pager Duty Helps Your company's digital title operations run more Smoothly Pager Duty Helps you intelligently pinpoint issues like outages as well as capitalize on opportunities empowering teams to take the right real time action to see how companies like Ge Vodafone box and American Eagle Rely on pager duty to continuously improve if they're digital operations visit pager duty dot com. I'm really happy to have pager duty as a sponsor I I heard about them on a podcast probably more than five years ago and so it's quite satisfying to have them on software engineering daily as a sponsor I've been hearing about their product for many years and I hope you check it out at Pager. Duty Dot Com Samson himself. Jimmy Daley thank you I. I saw your name when I read the long tail in College. Excellent Book I've followed volunteer work since then did spend many years as a professional editor writer. But your background is as a scientist in two thousand nine and you started three. Dr And no longer had as much time for writing or editing if you were full time writer or editor today what. What area would you be covering? Gosh that's a good question. I backed into writing and editing simply because I inside done physics and Physics was kind of a dying Profession for reasons that had to do with like the cost of accelerators and things like that my parents had been journalists and I swore I would never do journalism for exactly that reason but what you know I I tried to find a middle ground which was okay. I'm not going to be a journalist but I you know rather than doing science all I work for the science journals and you know and and and write about or read it you know science so I still felt like I was in science and academia but it was technically media nature and science with the two journals but again I didn't think of writing as being my thing that ended up writing. You know in in that in that context doing journalism although science journalism. And then I went from there to the economist where I was. You know. Starting Science and technology and at the economist. If you've read the Economist got very distinctive voice and what I learned there was I picked up that voice and you can think of the Oxford's debating society voice Prime Minister's Question Time Voice. But it's it's it's it's a voice that has confident. That's a sure that's quite opinionated. That sort of you know assertion evidence assertion evidence and when you get to the economist. The reason I'm telling you this story is because you know I think that once you get that voice it becomes very portable and that voice carries you onto books and beyond so I'll I'll explain explain. Why that voice carries on but when you get to the Communist? That was a poster on the wall and the poster on. The Wall had the canonical economists sentence and the canonical. Oh economists sentences wrong period. A matter of fact is economical economist paragraph. Now when you think about that what does it take to be able to get away with wrong wrong period as a paragraph and what that means is that you basically. Have you set up an assertion. Somebody else's assertion presumably. You know the the Prime Minister of Indonesia has said that palm oil subsidies or the route to the country's prosperity wrong period. How do you get away with that? And the answer is that you're not just not some neutral observer you're coming from a an intellectual foundation. Basically the anglo-saxon Philosophical Foundation of free markets and free people and Democracy. And all this kind of stuff. And you're summoning all this kind of western philosophy to take down an argument using the kind of the power of the brand the T. One hundred fifty years and becomes traditions. The you know the the world worldview of the liberal British British perspective. And then you sort of you know and and then the rest of it is just you know once you've internalized that voice and once you have you know even turn allies. The you know the Anglo Saxon you know economic philosophy then you can go around see the world through that Lens and say you know wrong long period and explain why and at the end of the day. People come away informed perhaps enlightened by that perspective so that confidence and that voice. which did you learn and You know the editors sitting on editors floor every Monday morning as the debates on what you're GonNa say come across that voice then you know once you come out of there you realize well. There's a lot a lot of the world that you can apply that that voice to you. Could you could talk about technology. You could talk about the environment you can talk about science. It's not necessarily be need to be obnoxious contrarian but it gives you the confidence to to move beyond the the the the the failure of American journalism failure American journalists the neutrality. You know what people call the view from nowhere just like Whoa on one hand. People say this on the other hand people say that and American journalists aren't allowed to have opinions whereas whereas elsewhere in the world especially in the UK journalists are allowed have opinions especially informed opinions and so that power to have an opinions or carries over to writing. And then when you write a book like the long tail free or or makers etc.. That book becomes a thesis. It's basically just a a long informed. Opinion argued out. And now you have now you have the ability to have opinions. You have the ability to to inform those opinions at the bill and the ad the skills to communicate that opinions you know. Now the world's your oyster so if I was writing today I would not I listen to you basically go and you said what is the most interesting thing that's the most poorly poorly understood. Those are the those are the best the best ones so I think. Ai Probably Right now is very interesting and poorly understood. There's lots of people writing code but good books about it so I wouldn't. I don't need to the ad that think autonomy in general is interesting poorly understood. I'm tormented by headlines about AI ethics. Which I think is a complete head scratcher veteran? I don't even know what that means. So that would be the reason not to write about it because I don't want to get caught up in a in a straw man arguments but no general right now. Technology has accelerated faster faster than society's ability to to deal with it and that strikes me as a good opportunity to bring clarity and persuasive argument to the the case there are so many applications of consumer drones that could improve our world. Why don't we see drones in the sky every day? Why why is the sky? Not Dark with drones. I ask myself every morning so you are with me here. In Berkeley. At the headquarters of three are also. Berkeley was one of the birthplaces of the modern Auden. Drone drone industry. This is a military drones started the nineteen fifties. But this is the this is the consumer that you're describing it. Was You know here. With the University of California's a Stanford Myself in about ten years ago and at the time. It was obvious that you know that Adrian's could be cheap and ubiquitous Quddus so I started something called. Diy Drones on the the notion that you could stick like literally the letters. Do It yourself in front of military industrial thing and it could work. It was clear. Clear that drones were going to be the future we're going to descend from smartphones. Not from triple seven's so that Sorta told you something about if you know. smartphones are ubiquitous. Presume that drones could be as well so economic reason couldn't and then the next question was only GonNa be you know what are they good for. You know we've answered the question. What are they good for sufficiently because there as a regulatory barrier between inventing a drone being able to use a drone the regulatory barrier by the totally appropriate? It's true for autonomous cars as well. You know anytime you're it's silicon valley is all about asking forgiveness not permission and I can tell you the original drones. We did not look for permission. We just I did it and San Francisco out there is got floor. That Bay is scattered with crash drones. Nobody got hurt. I don't think they leaked bad things into the foam etc but lots. Lots of drones crashed in the course of doing this as we went from the DIY phase to the commercial phase and then and then and the consumer phase as we started to have to comply with the regulations and the FAA regulations initially didn't allow commercial use at all. It was only recreational. Use those really like not. You couldn't make money from drones until two thousand sixteen so there's one answer why this guy's not dark with these things. The next answer is that even once commercial use was allowed it had certain restrictions you had to stay within visual on a site you say below four hundred feet. You couldn't fly over people. You couldn't fly at night. And also there had to be one pilot per drone so although their autonomous they don't need piloting they're actually has to be a person standing there and even today are fully. Autonomous Drones cannot be launched into a human being touches night bad. That's so I think they just have to touch an IPAD and the actor touching the IPAD. Validates there is a human being present watching in case something happens. Emergency Services Headquarters helicopter or something like that. So we actually haven't achieved any real efficiencies yet because these restrictions now we're very very close to being able to to to break through and there's an FAA process this is called type certification and once vehicles certified as safe then it will be allowed to fly beyond visual Site over people maybe more than one to one pilot to human piloted drone but could be one to twenty one to fifty one to one hundred. And then we're going to start to see the efficiencies that automation and robotics brings Beck that's going to I mean our our type certification's going to be the first and that's going to happen before the end of the year. Starting next year we may actually start to see them flying. Initially we're gonna be doing things like infrastructure the structure inspection pipeline powerline inspection but dams your state delivery things like that and so. I think that I think the answer. Is You know we thought it. There's a technical problem and we saw the technical problem and five years it was actually a technical and then a regulatory problem. The regulatory problems can take ten years. And we're about five years into it so give us give us another few years and and we'll get there. What would they be doing anything? Satellites were doing in the seventy s and airplanes. Were doing the nineties in drones can do can do you better not not. They don't have higher resolution both spatial temporal resolution. So if you want to see something. Millimeter resolution every hour. Only only a drone can do that if you WanNa see the globe once a week that whatever. That's that's a satellite but the long answer to your question but I believe that we in silicon looking valley have a holy mission. which is that we were? We were gifted the Internet. We're born with you know the most important technology are ages electricity of our time and our mission in his to extend this gift to the world. Extend the Internet to our homes to our arms to our cars to the air to space etcetera. And why. Why would we do that? Well if you extend the Internet into the world the world becomes smarter all these devices connected to the Internet are smarter Ford and the Internet becomes smarter from being able to kind of measure the the planet and are fundamental principles. That you can only manage what you can measure. And we've been doing kind of kind of bad job of managing the planet from environmental and economic perspectives. And so better. We can measure the planet with drones and satellites and everything else and feed it into the Internet. Will the better. We will be managing right. Now we're just picking the ones that where the where the value is the highest We're here in. Wildfire season in California are drones are being used to fight wildfires. How do they do that? The thermal cameras you can see through the smoke and see what the fires are before the fire. The drones can spot the fuel the needles and the believes that will create those fires and then after the fire we can spot the hotspots. They're still there so they don't turn into fires again. Climate change our drones right now. All the water infrastructure the dams and levees of America or built assuming one amount of rainfall level rise and they're all being challenged by climate change. So those you know now. We're seeing floods we're seeing you know the sea level rise etcetera all that infrastructure needs to be re re skinned impossibly. Expensive we know with on on the ground and trivial easy for drones and so you're seeing the US Army Corps of Engineers using in your Jones. Look at all those levees there figure out what's what needs to be reinforced. And what doesn't and so on looking for a job is painful. And if you're in software and you have the skill set needed to get a job robin technology. It can sometimes seem very strange that it takes so long to find a job. That's a good fit. friel veteran is an online hiring marketplace. That connects highly qualified workers with top companies. vetera- keeps the quality of workers and companies on the platform high because because veterans vets both workers and companies access is exclusive and you can apply to find a job through victory by going to veteran dot COM com slash s. e. daily that's V. E. T. T. E. R. Y. dot com slash S. e. daily. Once you're accepted to veteran you have access to to a modern hiring process you can set preferences for location experience level salary requirements and other parameters so that you only get job opportunities opportunities that appeal to you know more of those recruiters sending you blind messages. That say they are looking for a Java Rockstar with thirty five years of of experience. WHO's willing to relocate to Antarctica? We all know that there is a better way to find a job. So check out veteran dot com slash. Sec Daily and get a three hundred dollar sign up bonus if you accept a job. Through veteran veteran is changing the way people get hired and the way that the people hire so check out veteran dot com slash. Save daily and get a three hundred dollar sign up bonus if you accept a job through battery that's V. E. T. t. e. r. are Y DOT COM slash. Save daily thank you to veterans for being a sponsor of software engineering daily over the remaining five years of that regulatory time line and then in the successive years after that. How will your business strategy unfold? We at three. Dr Started by building the components for drones autopilots such than we became than we built. The drones themselves awesome became America's largest manufacturer drones then went to the Chinese. Got Really good at it. We got out of that and we moved over to the software so essentially software company. We just look at the data from from Jones. It's not the drums itself. That's not entirely true. We do actually have one of our cells for for people who can't earn on allowed to buy Chinese drones which is like the US government by and large where we're on the data side and so we used to be that it was quite hard to use a drone gather data. Now it's trivial easy to touch a button. Magic happens and the question is where's the return on investment on that day the highest so we start with construction then we went to the geospatial industries. Like the you know the birth work inspection that we talked about before your sting Pu- public public safety fire. Police picking up his well. We have just scratched the surface of. What's possible here? I think that you know what the regulation allows us to do is to go beyond that sort of visual site perspective so we're looking at like like bigger areas so although we started with autodesk and construction we're now actually working even more with with a geospatial stuff with like with as raise geospatial giant and. That's our main partner in this. Whatever they were doing with satellites years ago? They're increasingly doing with with drones. And the Nice thing about these regulations is that now we can. Now we can fly tens of kilometers you know so beyond visualized site so linear infrastructure power lines gas pipelines roads roads bridges tunnels airports. All that kind of stuff now is. That's now within the reach of what we can scan with drones. So I think you're GONNA see us. You can see more of the world fall within the scope of drones. We've already the battery. Life's already there. You know the flight times already there. The soon the regulations will be there and now you know think of it. We've been looking at pixels Oregon start looking at screens. What are the remaining technical barriers that feel the most acute right now? Ah almost none. I sort of feel like all the technical problems were solved years ago. I'm not that's not entirely true. But you know fully autonomous in including Centene Voids Company skied. Yo Right now doing amazing amazing work. But that's computer vision. Basically you know navigating through forests and through through leaves etcetera just using just using cameras. That's flying low when you fly higher you want to be able to avoid other aircraft and that's a harder computer vision problem. But there's this company called Iris Automation. That's doing that again with with cameras you know. The vehicles can fly almost any distance drones afloat across the mid-atlantic just use different fuel gas engines or hybrids or whatever. The radio links can go tens or hundreds of miles. The computer vision is amazing. GPS get better and better the software's kind of done on On all this this this almost nothing I can think of the drone delivery precision landing all that stuff. We basically benefit from the advantages in a and computer vision out there already so I literally cannot think of any technical problems right now. They haven't at least been solved at the university level. What about security well into fine define security? I mean the military has secure drones and has and has for many years. It's just you know. Do you want to do you want to have it like going. Through satellites hates encrypted sure can be done right now at the security is kind of whatever people want among the commercial space. It's standard Wifi to fifty. Six bit. Encryption encryption you can go you can pay more for for other security on the cloud side. We use fed ramp. There's really it's kind of what he want to pay it all exists. Let's imagine a construction site. I want to map that construction site or ensure the the safety I want to have an understanding of. What's going on in that construction site and so I'm going to use a drone to do that? Walk me through. What the drone is doing as it's flying over or through the construction site? How is the data getting recorded? How's IT getting sent to the cloud or is it like sitting on the drone and then the drone lands and you have to upload it take me through the technical process so I'll I'll describes her the optimal process and so far larger customers? They've kind of got it all all all very efficient. The optimal process is that this is being done every day. Maybe in the morning and the evening so you know the temporal resolution you know not just not just the spatial but because we objective here is to create what's called digital twin back in the day you know. Construction was started on screens with cad file but the moment they started digging it was analog paper. Blueprints need notepads and things like that we want that digital file that the digital plan to reflect reality so you know as they say you know no no no plan survives the first shot you know. Construction projects survives the for spader. Yes something changed. And so if you don't update the digital filed and that digital file you know there's interpreted it it sort of loses its relevance as it becomes less and less reflective of reality. So there's something called reality. Capture and the objective is to have the digital file software. You wanted people commit their software back to you. Know the the you know the master. So the masters nautical so in this case. The it's not it's not aww repository. It's a it's a cad file and that cadfael should be updated every day to reflect reality. How do you reflect reality? Well back in the old days people have to type in. Here's what I did too laborious now you want it. You want the scan to automatically capture reality and then update the file to show what happened to win so that that's what we're doing at at the headquarters they've said okay. This is this is the site and we want it to. Let's say capture the whole scan both horizontal and the vertical structures WanNa capture at in seven o'clock in the morning and at five o'clock at night and so the plan is the sitting some spot in the box. What's going to happen is that somebody's going to walk in the the site in the morning on lock? The gates locked the trailer. Turn on the generator and open the box and take the drone out maybe stick battery and at that point somebody will. Construction worker will touch a button on an IPAD and a plan that has already been loaded to that. IPAD from headquarters is going to be uploaded to Joan. Joan will take off. It'll do a lawn mower pattern hatter nor Peru or circular pattern or spiral pattern depending on what the site is at the point. It'll take about some between nine and ten minutes to do the whole site. It'll it'll take probably about two hundred images flying about about two hundred and fifty feet and then a land on its own at that point is someone who will put the back in the box touch another another button on the IPAD and the injury from that tyrone will go into the IPAD and then automatically buffalo to to the cloud to cloud at that point all those photos then get through a process to photograph maitree all those photos get get sort of analyzed and basically the way photograph works. It's called called structure from motion. But when and you see the same object from different perspectives using the paradox effects. You can actually see the depth you can have that so although the the photos to D- when you combine a bunch of two D. photos you end up with a three three d three D model and that'll be a point cloud or mesh or something like that. So that's automatically generated. Then that is automatically syncs up with the the cad file and when you have these things called ground control points so in the course of flying over its certain features are known had known position they have like an ex or some sort of fiduciary optical fiduciary automatically maquis recognize. And so that aligns this three d model to the same locations and so it snaps into location and now this becomes a layer in the cad file and and you can basically scroll forward and backwards through time and see how and see how things change and because these are meshes they're actually geometry's which can be snapped on. Snap and Khanin in alignment with the underlying catfight and you can say oh. That post was supposed to be here. But it's actually two meters over. There's probably a reason why they moved. That post two meters overdoes rock or something like that. Okay okay well. That's good to know now that now the digital twin says okay guys going forward. No that that post is now. It was supposed to be now over here. So when you put the trench now now put Trenton. I have to move the trenches well and you know when you're when you're going to be cutting the That you know the steel beam to go in that post note. The steel beam can have to be changed as well so now all all that information goes into the supply chain in the scheduling going forward and they make better choices because it reflected reality. Your software is open source or some of it It's it's actually not so the software and the drone so the drone might be using our the the software that we originally you know developed or or the software that we're now are working on as part of the drone code project is part of the next foundation says suffer on the drone is probably open source might be a degi vehicle which is closed source or might be one of one of the open source wants based Tantrum Code. That's just you know to operate the drone and mission and all this stuff. The data on the other hand goes into the cloud. And that's all closers. What's the reasoning behind? And so the open source project is you just join or CO founded the foundation projects credit. Yeah Okay what have been the ramifications of the open source. I mean just to give some context. I You told quite a great story at at the Open Core Summit Guy I you know. We can't go through that in the entirety but maybe you can give a condensed highlights version of that story. Perhaps the evolution of your code being used by constituencies of various ethical. Flavors got it. Got It okay. Yeah so when I started I started as a as a hobby. CBS editor of wired. It was doing. My kids became a community. It took off and everything I do as a community has always been open source. You know whether it's you you know creative Commons or or actual code so it was just a default open because it was a hobby you know and and then as it got as it got bigger it became better organized and you know proper code development processes and and maintainers and things like that as it got bigger yet it became clear that we had the opportunity to create attention the android of ABC's Amand Aerial Vehicles. And I was like okay. Well you know this is starting to look like smartphones. Dj was doing really well and DJ. I was very much modeled modeled after apple so TGI came out of the Shenzhen Pro River Delta and was one of those companies that was formed by people who had been you know building. iphones you know in the factories around there. And for the last ten years and taking notes the whole time so very much modeled after after apple close source vertically integrated but with an apple layer. And we're like like well okay. We think there's probably room for Android as well opens good and you know it. The Nice thing about android is allows proliferation of form factors and with drones. You also do you want a proliferation of form factors you want airplanes and vertical takeoff Many want big big helicopters little helicopters. And you know we're too soon too. I mean with phones. You can argue. There's sort of maybe a limited number of form factor small medium large etc but with drones. It's like everything deliveries military terry the work so open platforms make sense when you want to separate the software and hardware that said Dj doing such a good job. There really wasn't sufficient demand. Everyone was like you know. Gosh we're just getting crushed by DJ. Nobody's big enough to be the Google of this operating system if you will and so. It was kind of academic project. I think for a long time what happened then. Is that like who wouldn't want to use. Dj who'd want open source most hobbyists for sure. It was academics. But then it was also a guy mm-hmm and we found that as you know because we've made drones really limiting make something. Diy it becomes very cheap and open and easy and the vast majority of the uses. Were good but there. We're clearly some people who were there. were terrorists there was isis. who were we're going to be using it for ill and we thought about this and we said you know? What should we be doing about that? And so we talked to our our friends at the CIA and the NSA FBI et Cetera. And we said look you know we want to be super transparent about this. We know that people are using the software for ill and thanks for telling us and we're like we don't know what to do differently we can't close it off because what's the point. We can't put in back doors because people won't use them. We can't track these things because full. Just turn off the trackers it's open source etc and they're like yeah. We can't think of anything you can do either. If you see anyone doing anything bad just let us know and so we told our community. Look if you if you tell us that you want to deliver fifty five pounds two thousand kilometers. We don't think that's a good idea and you shouldn't do that. We're probably going to tell our friends at the FBI to this is happening now. We don't really know who these people are you. Yeah so we just say hey. FBI check check this thing so we were pretty transparent. That was we're going to do. And then you know we said well you know look. We really can't stop people from using the you know the code but we can encourage people to use the code as well so we spent a lot of time in forming the US government about the advantages of the open source drones. And we said look you know right. Now you've got a symmetrical warfare you've got. You've got isis using the carbon together drones with hand grenades and then meanwhile you've got these multimillion dollar anti aircraft missiles missiles etc its disproportionate. You should probably also have cheap open drones and by the way it's free and here you go at it and to their credit credits hearts of the government including the White House during the Obama gene totally embraced openness and the Office of Science and Technology Policy actually created whitepaper encouraging. This in subsequently you know it's been a long long process of of educating the government including the military and the police etc about the advantages. So these kind of consumer grade drones. And there's we're coming around to it and you know today you're starting out There's come to call. Andrew has a drone killer drone one throw in the kind of chases and kills another drone groaned that drone happens to be based on our on our software. You know you're starting to see the small drones based on our on our software our software being the PX for software. So I think I think the message is just getting through and I think all we can do as a community sort of bend over backwards to educate those who are trusted to protect us and otherwise just remained consistent to the to the spirit of openness and not try to kind of you know sneakily put in back doors try to try to gain the system. Just just educate. It's no surprise. Surprised that you have to bend over backwards to educate the regulatory bodies. What about customers? If you're trying to sell to construction companies or oil refineries or these likely candidates as early customers of consumer drone technology. How ready are they to to buy? Not Very ready is the answer. You know this too hard things about selling to enterprise business to business and stuff you know one hard thing. It's just that sales cycle for large companies. Have to kind of you know you're asking them to build a return on investment investment decision if it changed the processes that's a long process and then when you have a brand new technology that's highly regulated. It's even harder yet. So unlike mature technologies. Where it's like I'm GONNA use this? HR software versus that HR software when you're saying like flying robots on your construction site. There's a lot of people who need to be convinced the general council the the CFO. You know the site manager you know maybe local constituencies like you know the city government etc construction workers etc so it's been a long a long process and typically to be what happens it starts with a proof of concept we won the one sort of works. Okay then you sort of extend the proof of concept and you say okay well. It's always flew the site we got the data and now you now the data's available to people who aren't on the site you know the client or the you know the construction manager or the CFO or whatever and they're like Oh okay. Now I can see the day that I can actually I we use that. Then they win a lawsuit because they had evidence that they were not at fault and then like Oh okay now I can see the the virtue recording this and then they're we're like okay. We're going to standardize on this. Now we're going to do more sites. We're GONNA do all of our sites this way and then like one. Construction Company sees another construction company standardizing. Oh Okay we're talking ten years years. It can take you know to do that. So that's been a challenge. They actually don't care whether it's open source or not. They just care about the data and we want them to be totally agnostic about the capture side. Push a button button. Magic happens. The open source stuff is more relevant has become relevant for two big reasons one. Is that the paranoia about about China. The hallway stuff it's CETERA has led the US government to discourage the use of degi vehicles over critical infrastructure ports transportation transportation energy etcetera and also military infrastructure. So there's been this vacuum in the market. Basically a lot of fleets government grounded because of the stage I ban and so there's everyone's I was like well. There was something else out there. That's trusted and that this has been a big motive force for the adoption of the open source stuff. The second is that the FAA created in this new new certification process for drones. And we're going to I going through and the FAA also wanted to embrace a kind of an industry standard rather than a single companies one and so these open source wants are going. I'd be the first to be approved certified and so they are going to be able to one the first to be able to fly beyond visual Linux side and all those other things so in a sense the government is driving the adoption Donovan source by a banning closed the close source alternative and be standardizing on the open source. Development process is being the one that will be that they trust to be certified as businesses is become more integrated with their software than ever before it has become possible to understand the business more clearly through monitoring logging and advanced data visibility. Sumo logic is a continuous intelligence platform that builds tools for operations security and cloud native infrastructure. The company has studied thousands of businesses to get an understanding of modern continuous intelligence and then compile that information into the continuous intelligence intelligence report which is available at software engineering daily Dot Com slash sumo logic these sumo logic continuous intelligence report. It contains statistics about the modern world of infrastructure. Here are some statistics I found particularly useful. Sixty four percent of if the businesses in the survey were entirely on Amazon web services which was vastly more than any other cloud provider or multi cloud on prem deployment that's a lot of infrastructure on aws another factoid. I found was that a typical enterprise uses fifteen. Aws Services and one in three enterprises uses aws. Lambda appears service is catching on. There are lots of other fascinating. Statistics in the continuous intelligence report including information on database adoption Cooper Netease and web server popularity ability go to software engineering daily dot com slash sumo logic and download the continuous intelligence report today. Thank you to sumo logic for being a sponsor of software engineer. Daily in your talk you described competing with China on a piece of drone hardware as I think you said your first encounter with superior species. Yeah Yeah can you give me a nuanced perspective and prediction on our relationship with China as a business and technology community. Sure I some disclosures. Biggest disclosure is lifted China for four years so my children were born there. I'm a huge sign of file. I think we're going to lose in many ways to China and I'm completely okay with that. So just putting that aside you know okay you can take away my. I buy mega hat and look. I am not nationalistic. I'm British by birth living America. My children raised right. I mean I kind of like ah pro technology and just like want the best technology wherever it happens and if one country does better than other piece Gaba with you that said I'm running a business business and I want to do well and win so when I lived in China from ninety seven through two thousand and it was the kind of the dawn of the modern Chinese so I was like camping out on the floor if a hallway and it was pretty clear to me then that everything I've been told about China was wrong I was told. Oh you know. The can't innovate they only copy. Oh don't worry. They can only do hardware. They can't do software. Don't worry there the global marketing. They can't go global reach. You Know Oh you know Oh. Don't worry that understand user interfaces pretty clear all that was wrong it was like you know. Don't worry China can't ex- I could not find an ex and I was like okay so when I started the company through the are and we actually started twenty thirteen although the the community of the company started earlier but the company that you know I run today starting twenty thirteen. We're like okay. We know there's no ex- China can do everything. So how can we possibly possibly start a company that would compete with China and and I thought I think rightly or wrongly that I had thought I found the ex and the experts open source. I said you know the one thing that we have going for us is that we really understand open source. We build communities motivate communities and and for whatever reason China has used open source but they haven't really contributed to open source they haven't built a kind of internal culture of them source and so we think that open sources our secret sauce and the other thing cheaper than Chinese engineers for engineers. Okay so sadly. I think think I'm right about that. China has still failed to really embrace open source in the contributing side. That said where I was wrong is the notion that open source engineers or free engineers. They're by no means. Free engineer isn't if you're right you're basically paying people to contribute to an open source project so we didn't have the economic advantage that we thought we did. In addition at this image sure stage of the business. You can't just sort of open source the software and assume that somebody else is GonNa make great hardware. You actually had to vertically creative hardware software combination and there there we were at a massive disadvantage. TGI Is I think one of the best companies in the world certainly one of the best companies in China. When I talked about a superior species did is is a kind of a twenty-first-century Chinese company now like you know the old ones that kind of migrated this is one who is kind of you know born in the born in the clarifying fires at the apple supply chain? You know they did everything right. And they kicked our ass fair and square raise more money. That had more engineers they were faster. They innovated et CETERA. And you know on the hardware side. Although we were manufacturing in then we weren't native and so we were always going to be slower more expensive. who were less funding wingless engineers etc.? So once I realized that I was like I don't think that American hardware companies are thing. Carter should be done in China and so the next question. The American software companies. A thing and the answer is yes they still are. Why are American companies? Still thing and I think number one. I think we really do. Do I think open source really is not just sort of you. Know technically opening stuff it is the community building and the sharing and the poor quests are much more important than the downloads. If if you will and pull requests for a really unusual thing in China why would you. Why would you submit your secret bug fix to your competitors etcetera? So I think that's. That's a bit of an imaginative. We still have here and the other is that the great firewall China works both ways as software becomes more cloud and more data and less running on devices. You know you're starting to see people. People don't want to put their data in a Chinese cloud in Chinese cloud is considered subject to Beijing's influence. which is I think? It's increasing the case and so that firewall. Although there's no firewall WOM hardware tariffs aside. The really is a firewall and cloud and data. And so I think we're starting to see that you're actually to Internet's you've got the you know the bats the by do Alibaba tencent Internet on one side. Then you have the fangs. The facebook Amazon Microsoft and Google. One thing I don't know how to pronounce that exactly CETERA. And those two worlds really. I do see them continuing for a long time. So I'm bearish on non-chinese hardware but bullish on non the Chinese software and communities. There's a whole lot market ask you. `Bout there but we're nearing the end of our time. So I just want to ask you a few things that I can also get a distinct opinion on you are deeply familiar with the scientific journals nature and science now. What's the future of those periodicals and their influence that it's a great question? So Science and nature of the two premier scientific journals journals and they're a career maker you published in science and nature and a few others tenure citations all that stuff so academian signs of particular is built on repre reputation economy. You know your your grants your positioning diversities. Your ability to recruit students etcetera is all based on your reputation as a reputation station economy to carriers the reputation of the journals and their citation authority. If you will. And that's just the way. Science works also peer review. And all that stuff. But it's but the reputation economy and reputation is formed by the journals nature and science will be fine forevermore but there are also about you know hundred thousand other journals below them of diminishing mincing reputation and they're the process of going into the one of those journals involving a year appear review publishing something. It costs a lot of money. Not being able to give free access to other things becomes actually a hindrance to science and so there's this open journal method. It's been taking off for the last few years especially especially in physics computer science and but now now also in biological sciences where people are saying. Well why don't we just open source signs you know and create a community appear review. You community that will edit share and references CETERA and take a monetary take all the disadvantages of the commercial journals out of that process. And that's working. Well it's slower than you would expect and it's and sewer you expect because again the reputation because the reputation economy is so entrenched in traditional academic incentives. There's really hard for someone to say. Hey you know what my great paper. I'm not going to publish nature. I'm just GONNA publish it in employs to which the Public Library of sciences an open open when access journal. I'M GONNA publish employees do now. I think that's awesome and I'm super glad they did. They may feel that it gives them slightly less credibility and slightly less chance of getting tenure and is so important important that they may choose to kind of support the old system because it's a little thing for the system but it's a lot for them so I think that long term what you're gonNA see a winnowing. It's a little bit like media so media. Newspapers are show. Unless you're the Wall Street Journal The New York Times maybe the Washington Post you know so. You're seeing three entities that remain have the reputation or the or the backing of Bazo or whatever to maintain. And then the San Francisco Chronicle's Michaels of the world or just host. I think you can see the same thing. In scientific publishing the natures in the sciences in the cells and a few others are going to be fine. And then there's a bazillion you you know. Smaller journals especially in fast moving industries like computer science. They're just GonNa go away and that will all go. You know the long tail of publishing is going to go open access us and the heads going to remain the premier commercial journals and in the limit in both those domains. Do we want this kind of esteemed and voice. This esteemed trusted voice. Is that useful or do we want everything to be crowd sourced and open. Well I mean. The reputation of the science and nature's comes not just from their title but because the rigorous process that go through to decide what they publish them what they don't Peer Review But but editors etc scientists compliment science wrong. That's cool that happens. That's just the way scientific works figuring out what sciences wrong and not wrong is very hard. You often need peers. Here's those peers are busy. Sometimes they're conflicted etc so. I think that we really do. Need you know editing process for science review process a way to figure out you know what sides sides is better than other signs and the question how do we pay for it. The journals pay for it commercially. The open access ones. Do It with non-monetary motivations you know just the same way. How Open Source Works Essentially non-monetary motivations but they too have a business to run and editors to pay and so? I think that the experiments can be played up. I in computer science and computer science is starting to move away from the commercial journals. Physics is also moving away from the physics journals. And so I think we're GONNA see this play out in computer science because it's fast moving. There's there's lots of other ways to discern what's real and what's not and you know. Watch that Space Watch. How the academic computer scientists start building the reputation in a more open access way and? That's going to be the path for the rest of the disciplines Chris Anderson. Thank you very much. Thank you as a programmer you think in objects with Mongo. DB Sodas your database Mongo DB is the most popular document based database built for modern application developers and the cloud era millions of developers use Mongo

China FAA US America Economist Chris Anderson Dj apple editor Jimmy Daley Dr. Chris Jones Sumo logic Google writer Prime Minister of Indonesia
Roblox Engineering with Claus Moberg

Software Engineering Daily

1:00:28 hr | 1 year ago

Roblox Engineering with Claus Moberg

"Roablocks is a gaming platform with a large ecosystem of players. Creators game designers and entrepreneurs the world of roablocks is a three dimensional dimensional environment where characters and objects interact through a physics engine. Roadblocks is multiplayer and users can interact with each other over the Internet. roablocks is not one single game. It's a system where anyone can design and monetize their own games within roadblocks over the last fourteen years ears. Roadblocks has grown to be massively popular as the product has grown. The software has evolved to meet changes in consumer demands and engineering constraints client devices. Include mobile phones desktop computers in virtual reality. All of these clients must have a consistent experience in graphics and functionality. Thirty the back end platform has to support a high volume of concurrent players who are accessing a high volume of content. The networking needs to support multiple players operating rating in an environment. That Demands High Bandwidth Klaus. Moberg is a vice president of engineering at roadblocks and he joins the show to discuss the engineering of roadblocks and the future of gaming we are hiring a writer and a operations lead. The writer is somebody who would be writing about software engineering and and computer science and a part time capacity and the operations lead will be helping us run our business more effectively and that's also a part time role. If you're interested in either of these roles send me an email. Jeff at Software Engineering daily Dot Com. We don't exactly know who were looking for. We don't exactly know what skillset that we're looking for so don't be shy. I'd love to hear from you. Cox Automotive is the Technology Company behind Kelley Blue Book Auto Trader Dot Com and many other car sales and information platforms Ormes Cox Automotive transforms the way that the world buys sells and owns cars. They have the data and the user base to to understand the future of car. Purchasing and ownership Cox Automotive is looking for software engineers data engineers scrum masters and a variety of other positions to help push technology forward if you want to innovate in the world of car buying selling an ownership checkout checkout Cox Auto Tech Dot Com. That's C. O. X. A U. T. O. T. E. C. H. dot com to find out more about career opportunities. And what it's like working at Cox Automotive Cox Automotive isn't a car company. They're a technology company. That's transforming the automotive automotive industry. Thanks to Cox Automotive and if you want support the show and check out the job opportunities at Cox Automotive Gooda Cox Wchs Auto Tech Dot Com Claus Moberg. Eric Welcome Software Engineering Daily. Thanks super excited to be here. You work at. Roadblocks roadblocks has existed for fourteen years. It's a large large gaming. Ecosystem scribe the roadblocks platform in its current form. Absolutely so perhaps the easiest thirty thousand thousand foot description is that what. Youtube is to video. Content roadblocks is to three D multiplayer video game content. Ed that is that we produce a set of tools and a platform that lets anybody in the world create a three three D multiplayer online experience or game and we also build the platform that lets random people from the Internet. Come MM and play that game in a superfund social context so we have about a hundred million players that common experience these these games that they themselves created every month. And it's just amazing destination where you can go play Subaru birther original creative gaming experiences with your friends online and if you want you can actually make them yourself. Give a few examples samples of the kinds of games that people build and play. Yeah it it really looks like a microcosm of the greater gaming aiming industry at large right. Now there's everything from social online hangout games to first person shooters to mobile as as to you know you name it. Whatever your favorite game genre is? There's probably a version of that that's live. On roadblocks today traditionally traditionally our audience has skewed sort of towards the nine to thirteen year old age demographic and so a lot of the games toward a sort of skewed towards is that demographic is well though especially as of late. That's really starting to change. We see some more competitive content. That is skewing towards older age demographics as well being becoming very popular on the platform describe the process that someone goes goes through to make a game for roadblocks so roadblocks provides a software a game development environment that we call roadblocks block studio. It is a full featured game development environment if you're if you're familiar with the unity. Heidi ear the unreal idea. Looked pretty familiar to you with some specific differences by within that you have control over. You're basically everything in your game. You can Script game play in our scripting sandbox the the language that we provide to users Lua which is a very popular game scripting language in the industry. You top roadblocks experiences will include about yep upwards of fifty thousand lines of code but they also have full custom three d meshes that are uploaded to roadblocks asset services include clued full maps and and three D. environments. That you bill. We have terrain editor that. Lets you build beautiful three dimensional multi material terrain. At Yeah Yeah basically. It's it looks relatively similar to what building a game in any other major popular three D game. Id looks like today Eh. The the one big difference is that sort of central to the roadblocks. Ito's is trying to make that game development mid-process dramatically more efficient for the developer. So most of the other sort of AAA development environments that are out there air you start with a complete blank slate. All your defaults are empty when you start a project and you have to build everything from the ground up. Roadblocks blocks really starts with the idea that we are going to provide our developer with starting point with rational defaults. Thanks for everything from the strength of gravity to a base plate. That's the basis for the three D world that they're building to a set have players scripts that describe how our three D Avatar would interact with its environment in the game. The developers always able to modify all of these defaults defaults in build their own player scripts or are their own gameplay elements obviously radically changed the base plate. Man All that kind of stuff but we think the by providing a reasonable defaults. They can actually move dramatically faster. And I think that's what we've seen with our development with our content thus far are is that because we provide this really strong set of default sort of mimic even like real world behavior of stuff. You know real physics real character. Actions animations that mimics sort of anthropomorphic expectations and stuff like that that they're able to interrelate much more quickly Liam spend a lot less time building experience on our platform than it would take them to build that same experience on most of the other game. Id's that are available in the industry. History there are many facets of the engineering that will get into because roadblocks is fourteen fourteen years old and over fourteen years. There's lots of ornate architecture all decisions that I'm sure have been made and lots of interesting pieces of innovation that we can touch on. I want to continue with a bit of a top down exploration for what this product is. Because I I know that there are a lot of people that are on. I was unfamiliar with it when I started digging into it. But I understand now that this is one of those ecosystems that maybe under the radar for a Lotta people but has a gigantic following. I mean there's a lot of the Internet enables these kinds of amazing gigantic GEICO systems. Can you give me a brief history. For how the product has developed over time. Yeah absolutely so I. If any of your who listens listeners. Haven't heard of roadblocks. I would encourage them to either. Go Talk to the closest twelve year old that they have in their in their life or or go talk to any parents they know of kids net like nine to thirteen year old age demographic and ask them about roadblocks you will get a outpouring pouring of emotion from anybody Either of those demographics if you're in that nine to twelve year old age demographic the odds are in the United States as today. It's better odds that you do play robots than that you. Don't we have over. Fifty percent market penetration that age demographic and it is a huge phenomenon. Yeah so more than fifty percent of the twelve year. Old population plays roadblocks in the United States and other English speaking countries today. Yes Okay Nice staff so going back to the history yeah so our our CEO founder and CEO. Dave Basuki. We started the company with one of his close friends. Eric Castle like you said fourteen years ago so two thousand six day first released our development development environment roadblocks studio two years later it was sort of a almost like a demo at that time but the concept up from day one has always been really consistent that the idea is that we are building ways for people people to do online three co experience with their friends. The initial idea to found the company Really came out of Dave's previous company which was a two D physics simulation tool for the educational space basically allowing like kids to do their physics his experiments not in a real life lab but on a computer this is like back in like the the Macintosh one kind of days so they built this platform that let kids you know. Sort of attach blocks that had masses smash them together in to do these like physics experiments. To see you know sort of what would happen from Newtonian perspective. A big block in a little box Fox block with momentum that kind of stuff and they created this curriculum and and the company got acquired for a nice little exit but what they noticed while building that is it wants. The curriculum was over. Once the kids were done with the official experiment class. They would sit there like build cars and then crashed them together. Or they make these little like short little obstacle courses that you had to navigate your car around. There are basically using this physics simulation software as a gaming environment and they thought That's really interesting. And they decided after exiting that company to start a company. Those explicitly focused on building building a user generated content gaming platform. So like I said. They founded the company in two thousand four. Two thousand six. They released robuck studio studio which was the first time people could really build three D multiplayer experiences that were online and hosted for them. The company from that point for the next like eight to nine years grew grew relatively slowly until appoint in in late twenty fifteen when something something changed and we have a lot of conversations internally about what that thing was in the answer is it wasn't just a single bill. Who is a confluence of different factors? But all of a sudden the companies went from growing relatively slowly. You know twenty five percent year over year or something like that to growing at over one hundred percent year over year and just completely blew up and we've really continued that growth trajectory ever since late. Twenty fifteen early twenty sixteen sixteen to the point now where we have over a hundred million monthly active users and it. It's really been a wonderful sort of Silicon Valley Ali Growth Story where that hockey stick actually happened and seems to be continuing to happen even to this day. Okay so I know that a product act like this basically if you look at any angle of the software is going to be some interesting stuff we could discuss discuss. I've done a few shows on gaming one thing that seems to be characteristically. Interesting in in these shows about gaming is networking and what I mean by. That is when you've got a massively multiplayer. Online game like like a roadblocks. You've got these people that are independently exploring a three D. environment as a character. That's running running around so you've got you've got like a camera that's positioned kind of behind the character the characters like a three D. model running through space and and so that's that's your perception and in reality. The perception is like okay. You've got the whole world on your phone or you know on your your your PC. But of course this world is like so gigantic that there's no way it makes sense to load the entire world onto your a computer and continuously sync it with the server so in reality. You've in many of these memos these massively multiplayer online games you've got got a gigantic world that is hosted on a server and it's periodically sinking with each client plant and giving the client enough world to explore on their client device and this creates a very interesting networking problem. Because you have to ask. What are we passing in those packets between the back end in the front end? What are we giving to the client? What does the client? What is the minimum amount of information that the client needs to render on their device and is their path to graceful degradation if the client is on a mobile device on a spotty cellular network? Can we give them a good experience. Can we pass them enough. Data from the back can so I guess I'd like to just explore the networking challenges of building an M. O.. That is available across us all the platforms whether you're on an xbox or an iphone on a T.. One connection or cellular connection tell me about the networking. Yeah so first of all. That's a great tea up to the problem though I would take it. One step further and say that most memo's his benefit from being billed by US sort of vertically integrated studio where the same company that's building the network infrastructure to support support. The game is also building the actual content of the game. So you know you can have the level designer sitting down with the networking engineering sort of making decisions. On trade offs between whether the networking infrastructure can support the crazy number of triangles. That he wants in the volcano Hainault. Or whatever right robots is user generated content platform we have zero control over the content that our community is building. And so we have to actually actually do the Much harder thing which is to solve the general case and build a system that basically opens up as much capability capability for the game developer as possible across all of the devices that we support so the first question is what front end client devices. Do we support As of today we support IOS devices all the way down to the iphone four s which is at least by modern standards ancient in its CPU view GPS capabilities etc on obviously to support the xbox one. We support P. C. N. android down to. I don't remember the the minimum supported version of android right now. But it's essentially the equivalent of four us on the west side so the big differentiator for roadblocks. When it comes is to this type of networking is our game engine? We have a fully proprietary custom in house game engine that we have built from the ground up explicitly to do multiplayer online. Quote Unquote Streamed Three D. experiences. variances where you have some geometry and rendering done locally on the device but you have the exact same version of the game engine running on the game server and supporting all of the simultaneously connected clients. So there's a source of truth on the Game Gaiman on the server running our seedless game engine and then we have obviously a robust networking layer that arbitrates. What aspects of of the game world are going to be passed off to the client in terms of ownership for physics simulation? Obviously the vast majority of rendering happens on the client it and that's all sort of the optimization that are game engine was built with from the ground up for from day one and it's it's really from that perspective the the only three D game engine on the market that was built with this as its primary or sole use case even when you compare it to unreal or unity both of those were even in their current conception built to have the vast majority of the triangles they're rendering downloaded a large. DLC or whatever when you first download the game. Roadblocks is very different in that all of our geometry. The tree is streamed from the game engine from the moment that you actually join a game server. None of that is local to your device prior to joining any of our individual experiences and as a result the game engine is actually optimized for not just passing up and down ownership but also passing down the specific geometry of the world that the developer has is create. Being on call is hard but having the right tools for the job can make it easier when you wake up in the middle of the night to troubleshoot the database. You should be able to have the database monitoring information right in front of you when you're out to dinner and your phone buzzes. Because your entire application is down you should be able to easily find out who pushed code most was to recently so that you can contact them and find out how to troubleshoot the issue. Victor ops is a collaborative incident response tool victor drops brings your monitoring data and your collaboration tools into one place. So you can fix issues. More quickly and reduce the pain of on-call go-to Victor Dot com slash. Save daily and get a free t shirt when you try out. Victor ops and it's not just any any tee-shirt it's an on call shirt when you're on. Call your tools should make the experience as good as possible and these tools include comfortable comfortable t shirt if you visit victor ops dot com slash se daily and. Try Out Victor. You can get that comfortable. T shirt Viktorov drops integrates with all of Your Services Slack Spunk Cloud Watch data dog new relic and over time. Victor OPS improves and delivers. There's more value to you through machine learning. If you want to hear about how Victor ops works. You can listen to our episode with Chris. Riley Victor ops is a collaborative incident incident response tool. And you learn more about it as well as get a free t shirt when you check it out at victor ops dot com slash se daily. Thanks for listening. And thanks to victor ops for being sponsor a game like roablocks. This has something called a physics engine and a physics engine essentially outlines the rules of the physics of the game. If if my character jumps how does gravity interact with that character. Is You know. Is it like the real world. Where it's you know? Nine point nate meters per second squared. Or whatever or is it is it like you and you can double jump for something you know exactly. Excellent is the physics engine. Something that is simple enough that you can keep entirely on a client device. Or is that something where it's like. You have to make a round trip up to make physics decisions so the physics engine itself is definitely something that could be kept on the device but you actually run into the networking in ownership problems. That we were just talking about right. If if you imagine a game like grand theft auto where you can be driving a car are and I can be driving a car and we crash them together right if you assume that there will be some round trip time from my client to the server and from your client to the server from my perspective rendering in real time on my device if you make a last minute adjustment right before our two cars crash rush. I'm GonNa not see that because that last minute adjustment didn't make it to my device by the time my last minute input push got sent up to the server at a new sink of of the real time. Geometry locations came back down from the server to my device. Answer you might see a different different car crash that I would see on my client so really one of the core competencies of a game engine like what we've built it. Roadblocks is how you arbitrate ownership of physics. And frankly how you do some predictive modeling of even things down to user input to try. Try to come up with the best. Guess of how these multi player. Physical interactions are actually going to work in the game space. Do you WanNa make it something. Where all physics is owned by the game server and so you know you have a completely level playing field but a lot of the input and controls than are gonNA feel super laggy because in order to see the result of an input you have to take the input on the device? S- ended up to the to the game server which is obviously some hundreds of miles away from you and in almost the best case scenario and then send that you have that determine what happened from a physics standpoint and send the results back down for rendering locally on the device this or do you want to have the no lag option where you're actually have direct control and rendering of your car or whatever it is locally on your device vice but where the reality in your game client can as a result diverge from the reality on the game server and consequently from the reality on another players local client device because all of its having to be arbitrated in sinked back out of real time from what you're seeing being rendered on your screen in in robotics. We have a really complex set of code. That basically says well if there's no other players near you obviously we can have physics owned locally ugly on your Client N. as other players in other dynamic interactions come closer and closer to you this essentially physics ownership code base kicks in and tries to determine the optimal placement of stuff either in the cloud or on your local client and which version of the world takes precedence. That's crazy I mean. This is the kind of thing I was. Why like talking to people about the gaming the gaming business do you have these guys that crazy problems the a fourteen year trajectory? What's interesting about a fourteen year? Trajectory is that takes you back to pre cloud days. There was a show we did within two it. It fairly recently and into it made this crossing the chasm on Prem to the cloud and this is pretty interesting area of debate or architectural decision making because there are many companies that have investments in Colo's or their own on prime infrastructure. And you know they get to a point where we've got the infrastructure it's useful but there's there's these cool cloud services and maybe it would be cheaper if we didn't own any infrastructure and so they have a set of very interesting decisions about should they get rid of their on MM permit for structure. Should they move to the cloud with roadblocks. I assume it's a little bit simpler because the real traction with the product didn't start until like the post cloud era but I am nonetheless. Curious are there uses of of on Prim infrastructure everything entirely cloud. No We so there's actually two aspects to that one is. You're absolutely right that when they started roadblocks the word cloud hadn't really even been coined yet in in fact if you ask the engineers who are at the company at the time we actually have our game server infrastructure internally is referred to as RC. See Roadblocks Fox compute cloud and they swear up and down that the industry had not used the word cloud before they decided to call it that internally. I don't don't know if I believe them. I wasn't here at the time. But if you believe them they think they coined the name cloud which is hilarious in terms of our current infrastructure. There's actually another aspect of it as well. Which is the operating company at our scale? We are now having to make decisions about not just whether weather or not as effective to move stuff out of our legacy on prem data center in Chicago into the cloud but also at this point our scale justifies building out our own global essentially private cloud infrastructure. So one of the big investments that we've made at the company over the last two years is is literally acquiring acquiring our own global fiber network standing up a a series of. I think we're at sixteen or seventeen edge termination pops around the world and running all of data between our own fully owns data centers in Chicago and Ashburn Virginia over our own private fiber infrastructure infrastructure to those edge termination nodes so from our perspective. Our the way we think about this is if you're a sort of full stack application engineer building service service. You shouldn't actually care what is what infrastructure that service is being hosted on there should be an abstraction layer that says hey I just need to host the service that needs to be able to serve you know X. amount of volume and and and to be able to scale at why why rate up to future volumes and then beneath that abstraction layer there's Microsoft's platform team. That doesn't actually care explicitly. Okay what your service does. They don't care if it's serving the social graph or if it's serving the in purchase functionality or whatever they agency a service that needs hosting with a specific geographic distribution of consumption and specific growth rate with individual certain signals around volume and stuff like that and their job is to place that service on the best possible infrastructure for structure whether that's public cloud infrastructure our own bare metal for cost reasons or our legacy data center. Because it's it's part of one of our like you know sort of legacy model lists that haven't been upgraded to the latest stack yet so from our perspective basically try to have a team whose sole mission in is making optimal decisions around what infrastructure is best to host individual services into abstract that away from the the work that in actual application engineer would need to do to optimize that the resultant user experience does that make sense. Yes so speaking of services. I assume that for many years roablocks had a monolithic code base. Because you know I think uh over the course of fourteen years there has been this increased promotion of the of the breaking up of monolith into domain specific services services or service oriented architecture or micro services architecture or whatever the term does your is so that probably happened. You know that. Kind of breaking up of the monolith or addendums to the monolith that probably start happening before you joined the company three three years ago. Can you just tell me the the rough picture of the architecture today. There's some central monolith and then some other services or is it. Is it like a lot of just a flat services architecture so in an interesting way this kind of goes back to a core roadblocks engineering principle which is sort of set your north star in terms of with the the way you want to solve a certain class of engineering problems and then sort of make sure that any near-term firm engineering technology choice that you make moves your stack towards that long-term Northstar so essentially is another way of saying make make sure you're doing the engineering investment necessary to avoid any sort of unnecessary technical debt. This is a core roadblocks engineering principle. Something we think about. Every day as it relates to architecture vis-a-vis like model. It said micro services we still do have a core four of Monolithic code base that powers. Some percentage of the roadblocks dub dub dub website and associated services the The reason that still exists is that it has proven capable more or less of serving the the fifty to one hundred x traffic that increase that is experienced over the last three to four years. As we've we've gone through this crazy exponential exponential growth curve varies an internal mandate today and has been for for years. Now where if you're doing significant modification to existing existing services. You should simultaneously do the work that's necessary to break them out of the monolith and stand them up. As a containerized micro service independently scalable manage dependencies et Cetera. And so that we have like internal tracking sheets that show the migration of core services out of the monolithic into these independently hosted micro services that now are significant majority of our backend infrastructure after. But there's not a abstract internal mandate that says. Hey stop all product work until the monolith is dead and basically draws back to the idea idea that when that Monolithic code base was created. Yes it was a monolith but it was also created with a ton of attention paid to scale ability. The end of future magnitude of demand concerns in. It's actually even to this day operating relatively wealth from a scale ability standpoint. The only part of it. That's actually a really major negative drain on internal resources that at our current scale of headcount in the engineering department having everybody iterating on the same model at the code base can be kind of painful at the release schedule becomes a little bit harder to manage and stuff like that and so we think. Moving microscopes actually unblocks individual teams allows them to sort of control their own destiny in terms of shipping their own services and not have to wait for the daily monolith release to get their code out into user's hands. Give me some general perspective for where you're at in in terms of assessing technical debt and resolving technical debt versus building out new features. Yeah so oh you know what I often say to engineers. That I'm interviewing. Is that for fourteen year old company the amount of technical debt. We have is astonishingly low. While uh-huh I also say that with a grain of salt because I've never seen the code base of any other fourteen year old company in my career so it's easy for that. By the roadblocks again the robots operating principle from an engineering perspective is look were a platform. Were very unique in the game industry in that nothing we are building today. A HASTA ship tomorrow. If you're at a traditional vertically integrated game studio you know you have your content line in your game has to be done by by October first so it can be cut onto optical disc so it can be on store shelves by November. I you can make the holiday season or whatever and so the entire development exercises exercise is is basically like nine months of trying to figure out what you can actually get into this release killing yourself through time for eighty hour weeks or whatever whatever for those nine months to get every last line of code you possibly can in by that deadline and then you're done and people go and take a vacation or if the game's not successful they get laid off or whatever until they sign up again in two or three months later to do it all for the for the next year's release rohbock totally different. Everyone on my team could go on vacation nation for six months and some seventeen year old in Dubuque Iowa would still release a new title tomorrow. That is marginally better at retaining users or marginally better at monetize a user's and as a company we would still hit every single one of our user acquisition short-term user acquisition and modernization goals. Even if the big thing that my group is working on doesn't ship this quarter the right so dramatically more important for us to build the thing writing than it is for us to build it on any arbitrary timelines time-line and internally it's it's frankly celebrated if somebody's working on a project and they come up with a an implementation that's twice as good headbutt takes twice as long. That's always something we want to see. Is People taking the time to build things the right way in a way that avoids technic unnecessary technical debt and gets the product out. We still want to do that in iterative culture and we don't want to you know decide to make a big bet. Spent three years has has down no product feedback just building the thing just flip a switch and see if it works. That's not the way we work. But it does mean what we normally do is set some major aspirational guiding principle. A North Star. THAT WE WANNA work to over the next. Say three to five years and then every near-term sort of iterative rid of choice that we make we evaluate whether the near term implementation or the near term sprint or the near-term test is moving us. Generally towards that North Star guiding principle or away from it. If it's moving us away from it we don't do it. We don't ship it and we won't take that debt but if is moving us toward it we will. What do the most incremental solution we can that still validates? Our long-term assumption does that make sense as a result is is actually again like you could consider instead of our existing monolith as technical debt. I frankly don't I say it's actually a marvel that a significant amount of code base that was written five to ten years ago is serving a hundred to two hundred x the traffic that we had at the time that code was written. I don't think that's technical data. It doesn't mean it can't be significantly improved and we're doing that every day but it is an example of the amount of forethought and disciplined. That was used when that code base was was originally created. And it's something that we try to carry it through to everything that we're doing today. What are the programming languages that use at roadblocks? And do you have rules around what programming languages developers can use as a great question especially for my group so at a really high level will that the proprietary game engine that I mentioned is is predominantly C. plus plus. I don't think any surprise there are web stack. The monolith is a dot net. Had you know sort of Microsoft stack though. We now have a Microsoft's architecture that gives developers a pretty broad mandate to choose the best language that they think think is applicable to their individual problem that they're trying to solve so we have code hosted in in go or python or a whole bunch of other sort of languages. which is your on? The application side are tech stack is extremely interesting in non traditional so you know for most mobile applications especially cross platform applications. You have this sort of devils choice to make. Do you want to build your the application logic you I in the native language for each platform which generally means you get best in class results us at the cost of having to re implement each feature at least once for every platform that you support so for us that's at least IOS android a pc Mac and xbox and VR. If not different implementations for both IPHONES IPADS or do you want to use some sort of have cross platform code base historically embedded web views more recently react native or something like that where you get the benefit of only having to write each feature a single taivon having it shipped everywhere but the result. Kinda sucks roablocks chose Third Door. That's really unique to us. It struck us that the very first thing we do when we port roadblocks to a new platform is integrate our proprietary in House C. Thus whilst game engine at the lowest level graphics. API's that are available to us so yet metal on IOS Vulcan on android open jail wherever awesome Cetera in our game engine itself has to D- Ui rendering capabilities. But also our application infrastructure the stuff that users navigate navigate to play the Games before they actually join a game server and stuff where they choose a game to play or curate their social networker chat with friends or by robots are digital currency Consi. All of that is both Tutti but with a huge smattering of Three D. Elements your your actual avatar itself. The way you decorate your Avatar within the the Avatar editor in equipment Three items and all that kind of stuff and so we came up with this idea of saying what if we actually used our game. Engine engine to render our player facing APP. The thing that a normal game soon he would call the APP Shell and we've taken the last two and a half years to basically migrate from our previous mobile infrastructure which is a mix of native and embedded web views to now when you download roadblocks wchs and open it from your home's green on your iphone everything you interact with after the splash screen is actually a AH roadblocks a game scripted in Lua. which is the coding The sandbox that resurface to roadblocks developers and rendered heard by our proprietary game at the benefits of this have been huge. Because any time in doing this that we found the performance of our game engine wasn't up to our standards standards for making a first class social mobile experience we were able to go into the underlying C.. Plus plus game engine in fixed. It's Tutti Eh rendering capabilities so we made our scrolling frames dramatically more performance. We fixed a whole bunch of stuff in terms of how we manage our data models and all that kind of stuffed dog food aspect of this was huge but at the same time we get to build a feature a single time and have it shipped to literally every single platform warm that we support and the result in product is in the vast majority of the time indistinguishable from something that was built using that devices native. Ui Libraries it's it's been pretty phenomenally successful for us and I think is really unique in the industry. I think when we didn't have you with Google Google Earth. They had done something kind of of that. Yeah so that's that's exactly right. We've heard of a couple of other companies that have been in a similar space where they had some sort of proprietary in house client rendering capabilities and decided to sort of dog food for their own purposes is again for us. The results have been fantastic. It's just you have to go through the hard work of building those clients capabilities which it roadblocks is like a five year effort for it to create our proprietary in house game edge as businesses become more integrated with their software than ever before it has become possible to understand the business more clearly. Through through monitoring logging end advanced data visibility. Sumo logic is a continuous intelligence platform that builds tools for operations security and cloud native infrastructure. The company has studied thousands of businesses to get an understanding of modern continuous intelligence and then and compile that information into the continuous intelligence report which is available at software engineering daily Dot Com slash sumo logic. The sumo logic continuous intelligence report contains statistics about the modern world of infrastructure. Here are some statistics. I found particularly a useful. Sixty four percent of the businesses in the survey were entirely on Amazon web services which was vastly more than any other cloud provider or multi cloud or on prem deployment. That's a lot of infrastructure. Aws another factoid. I found and was that a typical enterprise uses fifteen. Aws Services and one in three enterprises uses aws. Lambda appears here's server. Lewis is catching on. There are lots of other fascinating. Statistics in the continuous intelligence report including information on database adoption in Coober Netease and web server popularity go to software engineering daily dot com slash sumo logic and download the continuous intelligence report today. Thank you to sumo logic for being a sponsor of software engineering daily in your vp of engineering right correct. So your Howard. The engineering teams within roablocks arranged. And how do they interact with each other. So roadblocks is a very flat organization. We have just from a size perspective. We have about just just less than six hundred total employees about eighty percent of that product in engineering. So we're incredibly heavy on on engineering especially and essentially there are. We are arranged as collection of over thirty relatively autonomous autonomous full stack teams that have some full stack product ownership over a certain feature set within the roadblocks offering so I currently oversees about five of these teams in. My team's purview is is on the robots application side. The stuff we we're just talking about basically you divide all of roadblocks software into the bucket of stuff that helps seventeen year olds. Build awesome three three D multiplayer online experiences teams. Do any of that if you the other half is stuff that let's random people from the Internet come in experience. Those awesome awesome three D multiplayer online games with their friends. My team does a lot of that. And basically owns the entire user experience from downloading a a binary onto their device to actually joining a specific game servers selecting inexperienced actually joining a game server for that game in getting dropped into survey. UGC environment so what that means. Is that the these you know. Five teams that reports me. There's a social team. A game discovery team sort of owns owns the roadblocks your equivalent of the steam game store or whatever. There's a what we call a universal APP team that owns the infrastructure and architecture behind signed this Lua application where we're dogged shooting our own game rendering capabilities to build the APP itself and all of those teams have you you know. At least a director level product leader director level engineering leader their own roadmap of our backlog of features features that they own and that they're maintaining looking to build in the future and they're basically looked at as a sort of micro startup or a small autonomous company within the larger roadblocks construct. So they'll have their own status reports with our CEO who also is the ultimate sort of product visionary. Mary for the company and they're in complete autonomous control over what they do on a daily basis how they hire people into that team what their prioritization nation is they have to justify their decisions to their stakeholders but apart from that it sort of showed Iran and the way we think about scaling the company moving into the future is we like these autonomous teams of ten to thirty engineers and associated products designers data scientists etc.. So we'd we'd like to see the number of these teams proliferate as they themselves in their size. Stay about the same. I mean that's sort of waving operating throughout the last eighteen months in his working. Really well for us. I'd like to give the listeners. A little bit of a picture for how the the roablocks game economy economy functions. I realized that's not the you know the area that you spend the most time in but can you just give me an outline outline of how the roadblocks economy works and what sorts of engineering problems that economic engine creates eight. Yeah it's it's actually fascinating. So the Roadblocks as we've talked this user generated content platform for and the fundamental idea. Is that people around the world. Come to roadblocks use our tools to build these three D multiplayer online experiences as they host those experiences on our infrastructure they distribute them to our user base and we give them all of the tools rules hosting in infrastructure. They need to do that for free. What they're able to do is sell digital assets within those experiences experiences and those digital assets the whether it's a skin for a character or whether it's a access to VIP area area within a game you know you get sort of the entire ecosystem of sort of premium monetization mechanisms that exist in the gaming industry at large? And you'll see some version regime that unroll blocks today when they sell. Something on our platform Our players purchase it with our digital currency which we call robots robots is obviously purchased with real currency hard currency from wherever the player is coming from around the world of the United States. They're buying robots with US dollars and developers Who Sell these digital assets within their game can actually cash out there robots and earn? US dollars back so so the top game developers on our platform are earning over a million dollars a year from the game experiences that they're building on the the platform itself in. It's it's a fulltime job Easily a fulltime job very lucrative fulltime job for them to build and maintain these top quality experiences the exchange exchange rate that they take that robots in cash it out for is called the Called the process of exchanging robots for US dollars that basically exchange rate basically is how we pay our bills and become profitable so there's a different exchange rate for buying robots from cashing it out but inherent in that is essentially they're quote unquote APP store fee so instead of paying thirty percent to apple for just the privilege of listing listing your mobile game in their act store and having users download it. They pay a slightly higher percentage but have literally zero other costs other than their time time to develop their game for publishing on roadblocks and as playable across basically every gaming platform in existence again. IOS ANDROID PC MAC xbox CETERA. You're given your front row. Seat to the development of the roadblocks economy. And how how the software is advancing and. I'm sure you have a lens. Into how other aspects of the game economy or developing twitch and minecraft and. What not all these other verticals? You just see up into the right growth. I think I saw some stat about like take the proportion of people who watch video games versus the p the proportion of people who watch sports and it's something like ten acts of five ex-soviet so he absurd that most people wouldn't expect but gaming is not going to stay the same as it is today and it's definitely not going to shrink so at that certainly leaves only one alternative outcome. How is gaming going to evolve over? The next decade is a great question. First of all if I knew the exact answer to that I'd be doing something else for a living other than just building software but if I look at the sort of trends that exist in the industry right now there a few things. That seem really clear. If you look at what's happening across the industry it going back to your network and questions about where the actual computation for the experiences happening. Twenty years ago it was one hundred percent on your computer. Senator or your console or your device and the trend has clearly bad over the last fifteen twenty years pushing those computations means to somebody else's computer up in the cloud When you see things announced like Google Stadia or any of the other competing platforms words that are being announced by other large companies in the gaming space it seems clear that some semblance of game streaming that enables is players to access triple a quality content regardless of the local processing power of their device is is a a big part of that future? This thing. That's been part of roadblocks vision for again literally twelve to fourteen years at this point. The idea that we need an architecture that basically produces a best possible experience for every player regardless of the actual will client side hardware. That's running that experience Our approach is pretty significantly different. From what the pure streaming entranced prince recent entrance into the market. do where Google is basically taking your user input sending all of that up to the cloud doing all of the rendering on the cloud and then streaming basically video back down to your device. It's nice because it makes all of those really hard things around physics ownership in rendering all that kind of this stuff really easy you just do it all on the game server. It's hard because it places big constraints on the size of the pipe that you have to send stuff up in San on stuff down and you had inherent lag on your user input because it's not being processed locally at your fingertips is being processed hundreds to thousands thousands of miles away on somebody else's computer so that's sort of a trend that I think will continue. But I think the the implementations that we see in that space will become significantly really more complex than what we're seeing from from a bunch of the new entrance in the space today when we take a step even further back and look at the trends in the industry. The big big question is how far in. How soon do we get to full immersion right? Four years ago everyone was like oh my gosh it's finally time for VR and four years later it seems pretty clear that VR is still an A. R. as well it is a wonderful tech demo. That for a bunch of different reasons has not had its mainstream moment yet when I was hired at roadblocks three years ago so it was to lead a small. VR team we don't have a VR team today men largely because the player based just hasn't materialized serialised yet. There's not enough people with compatible devices to justify US having full time engineering staff dedicated to support of that as is a platform even though our game engine is capable of doing it and the you know we're we're actually actively available on both Oculus Jalousie and five platforms. So I think when roadblocks thinks about the future of immersive experience were pretty confident that at some point in the future it could be three to twenty years away there will be this idea of fully immerses online digital co experience where you are in using a bunch of different digital tools to consume a alternate reality with your friends. It's the sort of ready player one multi-diverse or whatever you WANNA call it. That's definitely what we feel like. We are building and is really a question. One of when will consumer adoption of the hardware catch up to the digital platform that were creating. The great news is unlike a lot of mother startups in the space we don't have any need for Vr to take off to be successful where we're growing incredibly fast on to consumption of devices. His today were already sort of very very profitable and Doing very well with the current technology infrastructure this distributed around the globe. But we also think we're fully ready for a future where three D. consumption of these. Same experiences predominates user experience clouds. Thank you so much commotion. I'm not sure it's been really fun talking absolutely I really enjoyed it. Thanks for having me. The better dot com is a software startup with the goal of reinventing the mortgage industry mortgages his are a thirteen trillion dollar industry. That still operates. As if the Internet doesn't exist and better dot com is looking for engineers to join the team and build a better mortgage experience. The engineers at better dot com are attacking this industry by bringing a startup approach into an industry filled with legacy incumbents better dot com automates the very complex process of getting a mortgage by bringing it online and removing the traditional commission structure sure which means that consumers can get a mortgage faster easier and end up paying substantially less better dot com has a modern and software stack consisting of no J. 's python react type script coober netease and aws the interim quickly and ship code to production auction fifty to one hundred times every day. BETTER DOT COM is one of the fastest growing startups in New York and has just announced a series see that brings the total funding to two hundred and fifty four million dollars. If you're interested in joining a growing team checkout better dot com slash. Save daily better. Dot Com is a fast growing startup in gigantic industry. They're looking for full stack. Engineers FRONTON engineers data scientists. They're looking for great engineers engineers to join their quickly growing team. And for more information you can visit better dot com slash. Save daily

United States developer Victor OPS Eric Castle Sumo logic Microsoft Google Cox Automotive Cox Automotive Youtube Subaru engineer Cox Automotive roablocks Software Engineering editor hockey Cox Automotive Gooda
JS Party with Kevin Ball

Software Engineering Daily

1:04:44 hr | 1 year ago

JS Party with Kevin Ball

"The Java script ecosystem stretches across front end back end and middleware. There are newer tool such as graph Q.. L. Gatsby and web assembly and there are frameworks like react view an angular. There's complex data. Handling with streams caches and tensor flow J S Java. descript is unlike any other ecosystem because a single language can be used to construct every part of an application because Java scrip- is used for such a broad spectrum actress use cases the amount of tooling available can be intimidating to someone new to the ecosystem. Kevin Ball is a host of J S Party. A podcast on the change log network. Kevin joins the show to give his perspective on the Java script ecosystem in this episode we discussed. Es modules the Jam Stack and the growing number of tools libraries and workflows used by Java scrip- developers. We are hiring a software engineer. Who can work across both mobile and web? This role would work on software daily Dot Com. which is a huge application our APP and and our ANDROID application? We're looking for somebody who is very flexible. And who learns very quickly and can produce high quality code at a fast pace if you're interested in working with us send me an email. Jeff at Software Engineering daily Dot Com. We're looking for somebody WHO's hungry and somewhat entrepreneurial so I would love to work with you if you are well-versed and a fast learner just send me an email jeff. At Software Engineering daily DOT COM UH as businesses become more integrated with their software than ever before it has become possible to understand the business more clearly through monitoring logging and advanced data visibility. Sumo logic is Z.. Continuous intelligence platform that builds tools for operations security and cloud native infrastructure. The company has studied. Thousands of businesses is to get an understanding of modern continuous intelligence and then compiled that information into the continuous intelligence report which is available at software or engineering daily Dot Com slash sumo logic. These sumo logic continuous intelligence report contains statistics about the modern world of infrastructure structure. Here are some statistics. I found particularly useful. Sixty four percent of the businesses in the survey were entirely the on Amazon web services which was vastly more than any other cloud provider or multi cloud or on Prem deployment that's a lot of infrastructure on aws. Another factoid. I found was that a typical enterprise uses fifteen. Aws Services and one in three enterprises uses aws. Lambda appears server. List is catching on. There are lots of other fascinating. Statistics in the continuous tenuous intelligence report including information on database adoption Cooper Netease and web server popularity go to software engineering daily Dot Com sumo logic and download the continuous intelligence report today. Thank you to sumo logic for being a sponsor of software engineering daily Kevin Ball welcome software engineering daily. Hey Jeff good to be here. I'd like us to take a tour through the modern world of Java script and the place I'd like to start is es modules explained es module as well. Let's start by going back a little bit to talk about how modules in javascript sort of evolved over time so unlike some languages that start with this concept concept of code isolation modules and things like that javascript when it originated everything was in global scope. There was no concept of a module in separating things in pulling these out because it started as is a play language on the web and then as people started to do more serious software engineering with javascript. They wanted to use good practices like code isolation and things like like that and so initially there were a bunch of what you might call user space solutions to that folks who basically built up using the language as existed ways raced to create modules. That's where you get things like. AMD which was one of the the first sort of specifications you could call. It was essentially like once again user space space specification. If you write your code in this way it will work with this tooling and we can loaded dynamically you get nice isolation things like that over time the language which became more mature and folks started saying. Hey we should actually have a first class solution to this. The first thing we got to that was when no J. US came around and they said. Hey we're reading Service side packages. We need a way to do this. We're not going to be shipping stuff up to the browser. Were not an isolated. In the same way we don't have as as limited scope in the same way we're we're just GonNa make something happen and that was based more or less on. AMD and common jazz which is another sort of specification that came on that and that became the the defacto standard as node and require which is common approach and then it went up one more level and it said okay node while very very large and very popular and able to drive this defacto standard. That's not actually the language we need a language level solution for Howie encapsulated slate code and allow it to load in different ways and that was when. Es Modules kind of evolved so this is something coming down from the TC thirty nine specification community the script specification nation where this is how modules function in the language of Javascript as defined at the SPEC level. So it's no longer user space. This is actually baked into the language language and accessible. Now because it's the web and it's messy and all these other things there's been all these back and forth about okay. What about file extensions? How does this change different things? Because many of those defacto standards around node had built up lots of tooling and magic around the Senate in lots of ways that turned out to not actually work when you try to do them in cross environment situations like. How does this work browsers compared on a server? So there's a lot of stuff there that has really been getting hammered out and still coming in. But that that's the high level lizzy s modules are the language level. Now we're bringing it down into the actual specification solution to a problem. That's been solved in user space. For years there's a classic example of a module that we needed is J. Query right like J. Query is this big blob of things that we need out of our Java script infrastructure and historically we would just import it on a global basis and we would have it available to our entire Java script application and that was not perfect but it did the trick. What is wrong with that architecture? Why is it problematic to have a global variable? Why is it problematic architecturally global variable? I mean this is actually an interesting question. Because I think that's a lesson that gets over applied and it gets fought about a lot in the web world when you start talking about CSS which is still in many ways global in different ways the fundamental challenge with having having things that are global is. It's really easy to break them. If you're trying to pull in code from lots of different places that have perceptions options about what this thing is in. Is it going to be there or not going to be there and can I manipulate if there's just one than those things can mess with each other and break with each other and it especially if you look at the javascript ecosystem today. The trend is towards lots and lots of small packages the tooling rolling around packages and package installation dependency. Management is so good that people said why have large packages when it's just just as easy to do ten small packages than each one of those as a tiny surface area. I can test and do things. That ecosystem doesn't work if everything's global because if I'm installing a thousand packages it is and each one is depending on something global that they can actually mess around with and mess up very quickly. I'm going to end up being a territory where one of those is expecting one thing. The other one has already manipulated in some way and so it's not quite matching expectations. If you're going to be integrating different pieces of code where you don't control everything at which is fundamental to modern software development general right like old days you look back browser applications. They're relatively simple. They're small the entire code base can be owned owned by one person or one team whereas nowadays if you look at a modern web application look at your g mail or facebook or one of. He's really advanced products. They've got maybe half a million lines of Java script if each team that's working on that that's probably spread across five six ten twenty teams like if each one is able to muck with these variables. You'RE GONNA shoot each other in the foot real quick just to revisit what problem problem do. Es Modules solve. Yes Ma tools solve. How do I isolate code into its own package? Essentially whether it's within within a single application or package or an external one and reliably pull that into my own application or into another package to us. It's the same problem problem. That essentially like gems. It's actually. It's a little complicated because it's also how you do code imports so like you could think of a package. Any yes module is the equivalent of like even just importing code it built into the language of python or go or something like that where you can import code from one file Allie to another file that literally did not exist as part of the language before it was packed together with user space tools. That would sort of put that together for you right right. So it's this fundamental problem that most languages had built in from the start like you wouldn't imagine writing a python application where you couldn't import code owed from one file into another file but that was all user space solutions that wasn't part of the language prior to guess module why have. Es Modules been controversial. Well partly because they're in the web and the web. It's probably the I. I don't think it's out there to say that the web is sort of the widest and most diverse set of software. Stuff that exists right. Gets you have things that are running a distributed environment across every device known into man. You have no control over how this thing is running where this thing is running other than it in the browser and so people did all sorts of crazy stuff on top of you've got server side solutions javascript with node where you have that more traditional environment there were already solutions that existed there that had been built up that are subtly incompatible with yes modules. So you have the javascript. Language being specified by folks were mostly concerned about the web. They're mostly people coming from rob Browser companies thinking about that use case. And they're trying to build something that is also going to be utilized on the server where there are already subtly the different approaches being used. I think they're also just missteps along the way there were you'll waste of people trying to make them kind of different aren't than just being javascript. So there is an introduction of okay. We're GonNa tell what the module and what's not by having different file extension and that was problematic because in the node world where where people were mostly thinking about packages like everything is just file extension. It deals with do that for you. Don't even have to include it file extension because note. Has this magic about looking up. Is it here is it there. And what have you so it you created these situations where you had had could have discrepancies. From the way people were used to thinking about the world one other aspect that I think is worth bringing up here is pretty interesting in also relates to why EH. US modules have been very controversial is in the Java script world. We've gotten used to using features before they are fully specified because there is incredible tooling that essentially allows you to transport within the language so you can extend javascript at new functionality and transparent trans pilot back to older functionality. The tooling for this is called babble. And it's a transporter and the original use case was okay browsers. Thor's are slow to update and in fact looking back five years many of them did not automatically update so a user had to go and actually do something to update their browser which meant that even though Oh javascript was moving forward in adopting features that were valuable turn bringing it from being a toy language into a first world are high productivity extremely powerful language. You couldn't write code in that way and run it on those little browsers unless you had a way to translate that new tax in this new code back discipline. Those browsers could understand so John Has Gotten used to using features and compiling them back to older browsers now. This meant that when people started talking about Yes yes modules and they said Hey. This is a really cool syntax. We're GONNA do this. They could use it before it ever got specified or built into any platform. It was not built in any browser. It was not built in to not at all and people started using it by using babble to Trans file it and using web pack to and similar bundlers to package things together however however they were doing that based on the assumption that hey we can just call it a DOT J s file the same way. We've called every other javascript file we've ever worked with end it'll just work. And then the specification morph initially said. Oh no you're going to have to have different file extension for these modules. Because they're really they're different from the old stuff. So I I like that. was fundamentally a misstep but why was so controversial was because we as a community had already started using these things as a way that we assumed we would be able to and then it turned out that that wasn't going to be quite right. Well one way that applications in the job script ecosystem get condensed and and presented to the end consumers through a bundler key. Explain what a Java script Bundler is yes so this this comes from a couple interesting things so one thing to remember. Whenever you're talking about Java script is the number one target for Javascript is the web that means that any code that someone is going to run has to get loaded by their browser over Net connection probably the majority of web access. Now it's probably through a mobile phone connection and many of those connections are pretty slow and so you need to your. There's there's long been in work on saying how do we make that amount of code that were shipping out to the browser as minimal as possible especially when there was no equivalent of assembly language out there. I couldn't ship binary it had to ship actual Java scripts in so butlers are the next step of going back a number of ways we had this approach years and years ago where we would concatenation nate all of our files at them. So what that would get you is. Hey I'm going to have a single file that has to be loaded so don't have to issue a bunch of HDP requests West which is less relevant now that we have a two and and things like that so it's not as expensive to issue more requests but it's still a thing so I'm GonNa concatenation so you only have to into a single requests which is another one of these transportation things where any type of variable name or codename or whatever is GonNa get swath down to very small letters single letter function names Etcetera Etcetera and transformed into the as essentially as tightly as I possibly can compress this file and then I'm GonNa ship that single bundle of Java script up to the web so that's an old tradition? That's been around essentially certainly as long as I've been doing web development. which is I don't want to say how long that's been with us more or less since the beginning of javascript is? We're going to put these things altogether ship them out now in the old days when everything was global. You just had to make sure that you were putting those files in the right order so that anything that depended on one thing happened after that thing was defined defined and kind of put them on a file and go but in the new days that dependency is more complicated. It's more complex in a good way. You have much finer grain control of it because you you are importing modules whether it's es modules or you're using amd or using common Jason An old school known modules. You're pulling in code from all sorts of different ways in a complex independence a tree and we still want to take all of that together smash into maybe one file maybe a set of files but a small number of files and have it all worked together so bundlers taking charge of that core would a Bundler is doing. Is it's crawling that dependency graph figuring out. What is the set of code that you're importing from all these these different places that is needed to make this thing run smashing it together into a file which Butler's at a more advanced level may split out into multiple files for different types of optimization but smashing smashing together at potentially potentially doing other things on that so that you have that Blob of Java script that you can ship to the browser and how does a bundler handler fit into my workflow as a Java script developer like you've just give an overview of what purpose it is solving? How does it actually fit fit into what I am doing on a day-to-day basis? Hopefully you can ignore it because somebody else's set it up because they are to this day nightmare to configure though there's been progress on that but conceptually definitely. It's the you if you're thinking about this from old school software development standpoint if you're may file or it's a piece of how you're you're making your project so you're writing your code hopefully. Hopefully you don't have to worry about your butler. Because it's already set up and then when you're ready to run it the bundler packages and ships to the browser and typically a setup we'll have a development mode which is doing hot reload so anytime a changes made the Bundler rebe bundle and automatically refresh your page for you. And then when it comes time time for deployment is just GonNa make it all up into your final package files and ship it out one concept that may be useful to talk about. There is the concept of entry file. So we're talking about this dependency tree. A tree starts at a single So what you'll do is you'll tell your bundler. Hey this is the entry point to my application action or this library or whatever it is and it will take that as it starting place called on that tree and then bundle things up into a single file that is named predictably in you can customize or configure that how we want but started at that single entry point so a common set up in a relatively simple you probably have just one entry point. That is your top will file. That's your ABC Jay. Or whatever it is in the Bundler will crawl down. That package it up you end up with a single APP compiled jess that is then ready to run. And and considering we started with a conversation about modules how do modules in the emergence and the rise in popularity of modules rules. How does that affect the ideal workflow that we would have with the Bundler? Chances are doesn't affect them at all for you because if you're using a bundler it's because you're already using some form of modules in fact. Chances are if you're reading javascript. Right now you're probably even using something that looks like. Es Modules us. And if you're compiling your code to be a single blob the send out your bundler. CONFIGURES GONNA stay. Essentially the same the fact that we have. Es Es modules shipping natively to the browser does enable us to do some things down the road as more and more of those modules exist in our supported you. Can you could see a world where we don't have to do that. Same Level of bundling because the browser handles all of those imports for us. But did a day right now yes. Modules is a great way to structure your code. The bundle will keep working the way it's been working and I don't think there's any advantage to China move away from that when you say the browser houser could potentially handle the importing. What do you mean by that? What would that look like? Yes so right now. In a typical bill bundled setup. A javascript. File is not loading other javascript files unless you're explicitly writing javascript that says hey go and fetch this file added to my html or something like that like. It's not doing that dependency tree crawl if you had an import statement it wouldn't know what to do with it when you add yes modules. It now knows what to do with that so as long as that. Imports statement is pointing to fully qualified path. Basically a you are the browser can pull that going get it come back and put it in place and run it as you would need it that in Bundler world. They're doing that for you. They're crawling it. They're packing up so those import statements get compiled away to. We're putting this code in this place in Lincoln things properly so from a developers standpoint. Chances are it's not gonNa make that much of a difference to you you one way or another which way you do it right now except it's more of a pain to do it with the browser because you have to do fully qualified paths in all of that with a Bundler it can be smart and you can set up. Aliases and you. You can do all sorts of other smart things. So I'm far from the biggest expert on yes modules in particular but on Jazz Party just didn't episode of that Ruler Picking The brain of one of the folks who is an expert in this area in overwhelming message. I came back from is. Don't worry about it yet. If you're writing modules if you're wanting to explore something new go go out and you take your modules and make sure yes module compatible. They're they're shipping that if you want to explore writing tooling in that but if you're line developer writing javascript. Just keep using a bummer. For now. It's it's not GonNa hurt you and it's GonNa work better right now. Being on call is hard but having the right tools for the job can make it easier when you wake up in the middle of the night to troubleshoot the database. You should to be able to have the database monitoring information right in front of you when you're out to dinner and your phone buzzes. Because your entire application is down you should be able to easily find out who pushed code most recently so that you can contact them and find out how to troubleshoot the issue. Victor ops is a collaborative incident. Response Tool Victor ops brings your monitoring data and your collaboration tools into one place so that you can fix issues more quickly and and reduce the pain of on call go to victor ops dot com slash. Save daily and get a free t shirt when you try out. Victor ops ups and it's not just any tee-shirt it's an on call shirt when you're on. Call your tool. Should make the experience as good as possible muscle and these tools include a comfortable T.. Shirt if you visit victor ops dot com slash sl daily and. Try Out Victor ops. You can get that that comfortable t-shirt victor ops integrates with all of your services slack. Spunk Cloud Watch data dog new relic and over time. Victor ops improves and delivers more value. To you through machine learning. If you want to hear about how Victor Works. You can listen to our episode with Chris. Riley Victor ops is a collaborative incident response tool and you can learn more about it as well as get a free t shirt when you check it out at victor ops dot com slash se daily. Thanks for listening. And thanks to victor ops for being a sponsor so jazz party is definitely a better podcast to dive into for people who are very serious about their Java script in there and if you want to hear like kind of a talk show experience for going into the Minutia of Java script and more introductory conversations as well. It's definitely a better podcast test for that but just sort of channeling my own. Inner J. 's Party for this episode do want to continue down kind of a Just a list of things that I've been thinking about our or exploring in in other episodes or things that have come up in other episodes that we've done about Java script or our Front End Development. There is a term Java script fatigue or cooling fatigue. What does that mean what other people fatigued about who was fatigued is new developers developers every developer? What are we talking about here? Great question this is coming from the fact that javascript ecosystem is massive and moves faster than probably any other ecosystem am aware of. There are if you look doc or number of packages in different ecosystems and you look at how many pipe packages consult with python. How many ruby gems are there? Whatever javascript is essentially at an order of magnitude above everyone else and growing wave faster? So they're like five hundred new packages do javascript. Packages added to the N. P. M. Registry everyday we day is ridiculous and the language itself is evolving evolving relatively rapidly the language. Has I think since two thousand thousand fifteen. There's a new SPEC published every year continues to make advances in progress. And so what this means is that there's even more more than in any software engineering job. There's a tremendous kind of hamster wheel effect of trying to keep up trying to keep up. What's new what's different what's new? What's different the approaches? That were modern and correct and the right way to do it. Two years ago are perceived now to be old school in out of date and not there are and some of this is just perception. You know. I just didn't interview with somebody from SEM. She said their their mantra is. We like boring tech in. There's a a lot to be said for that right like you don't have to be using the latest and greatest fanciest javascript framework and all of those different pieces to be writing good software. Not Not in any way shape or form but there is this perception of constantly having to adopt new things constantly having to adopt new changes. And that's exhausting. And and there is some amount of truth to it in that the types of things that you can create an right now using modern javascript. Framework are like reactor reactor view or angular or even one of the more sort of newcomers. Felt like the type of application you can write that level of Dynamic Amac interactivity that you can create in a browser is simply worlds beyond what you could do five years ago with a most J. Corey based application. It's incredible and you can write. I much more productively because you have all this tooling in place. So there's there is some real need to to learn to grow into adopts. The new frameworks worsened technologies. That are coming available but so much of it is just perception. Is this feeling of keeping up. You've got to keep up and there's a treadmill and you can't catch up and that creates fatigue. There's definitely a sense that well I mean. I think this is one of the reasons why there is this really really rapid pace is there's a sense there's almost a palpable future. Where Front and development is as easy as dragging and and dropping or as easy as a low code tool where you're just you're building user interface with the whizzy wig and you're easily putting together the he's Ui components? You're you're putting together basically a front end application. That does everything you need to do. And if you wanted to dig deeper into it and optimize the performance of a particular react component for example. You could do that. You would have no problem doing that. We're not there Yet we are still in a time where the front end developer has to do a lot of debugging in tweaking and typing code. It's not a drag and drop. Ui Experience yet. We still need separate roles for the designer minor and the front end developer although there does seem to be like some palpable future. Where perhaps those two roles will will intertwine fine and hard to know where that ends up but I think one thing we can discuss that seems to be hinting at that? future is these component libraries. Where seems to be? There's component systems that facilitate workflows between designers and front end end developers. These places where the designer in the front end developer can mind Meld. Can you tell me the modern workflow between a designer and a front end developer ver. I can but I I WANNA push back a little bit on the thesis here because I think there is a sense of others. There's all this new stuff we can do. That is making things easier but no front end developer that I have talked to feel like their job has gotten easier. In fact there's a sense that is getting more complex ex- because more and more things are moving to the front end so it used to be that almost all of the front end work that you're doing with presentational and you're doing most of your stuff with each Tim Allen. CSS assess Maybe a little bit of Javascript but all the heavy lifting in the logic of your application was living on a serger that world is long past so more and more of the complex software engineering. That's going on moving to the front end than you have logic. That's moving there. You have even data management and things like that. So you've got these complex estate management systems. You've got redux. You've got Ma box you've got all these different things going on to manage more and more of what used to live in the back end in the front end so I think it is a little bit like there is a sense of. Oh my Gosh Dragon. Dropping no code is getting so powerful it is but the result has been that we've continuously wanted to do more and in fact we wanted to do more at a more rapid rate than the tooling has gotten better so the amount of complexity laxity that has happened. is happening on. The Front has simply skyrocket relative to finance a while ago. So I think it's good that we're having more tooling. The state of the art of what you can do without having to dive into code is going to continue to go forward. We've had no code web development systems for forever. We've got wordpress. We've got squarespace. People have been able to build websites forever and even some amount of interactivity but whenever you want to go beyond that do something new in different. You need to get into the Code A.. And the amount that we've been wanting to do interesting things there on the front end has far outpaced the ability of that tooling to catch up so the front and teams at most companies. I've talked talk to are growing and expanding and more people are feeling the pain of we don't have enough skill on the front end than I see on the back end as well coming back to the question about design and development and how those two things interact it varies a lot by company but what we seem to be. Moving towards is this concept of Designed Systems linked component libraries so having within design a set of concepts a set of specifications standards. This this is the type face that we use here. The font sizes that we use. Here's the spacing hero components. Here's how we're thinking about all these things so that when designers working on something they have a fixed library of tools to use and this has been a thing in software engineering for a long time. We love our libraries. We love building out reusable. Bits that we can use over. And we're gonNA recombine in different ways. We'll that's coming to the design world and then the translation of that into the front end is often a component library plus some additional styling stuff around And things like that. So when you've got that set up were when you. You're moving towards that. A lot of that interaction comes back to this discussion. Russian of okay. What are the things that we have enabled in our design system right now? How do we fit the things that we want to do that? And if we can't what's the way that we can extend that system to do what we want to do now and how does that play into component library. And if you think about this future development workflows workflows in. I mean who knows what it'll look like but if you think about just some high level things that we have today we we have. These thriving frameworks six. Actually let's let's go deep on the frameworks right now and then. We'll get to two more two more futuristic things but I'd like to to level set in the present and talk about the present day frameworks works so to my mind. The main prominent frameworks talk about our view and react. I think you still have angular stove. Dove a lot of people in angular thriving angular ecosystem. And you do have a spelt. People are telling you felt episode. I don't know anything about Svelte yet. But view in reactor really the two elephants in the room. As far as I can as I can tell and they're in their thriving they're growing really quickly he could you contrast view in the react ecosystems for me. Sure I let's. Let's talk a little bit about one thing that sometimes this gets lost here is a lot of these frameworks have Latin comment. We on the front end have more or less universally gone to a model where our development is organized by components and we we think about the world in terms of components which except properties from your parents you can think about that. His arguments passed into a function. They may have some internal state they may be don't and then they have have child components that we're building these trees of components. The Component is the fundamental organizing block. If you look back to the Jae crowder days or things like that. That was not obvious as the development philosophy that might have been true in the U. I but then the way you organize your code is different but that has been essentially universally adopted as the organizing framework for how we think about. Front end frameworks in front of Java script and react is doing that view. Is doing that angular suing that celts doing that embers doing that doges doing that. Like whatever framework. You talk about the pretty much. That's the approach there doing. So if the positive thing about that is no matter where you start on a framework if you really dig in and start to understand and how did you component oriented development how to think about your code base in that way you're going to be able to take those skills in that understanding from you as you go from framework to framework now now to your question particularly about reacting view. There's a number of differences both in the communities and in the way that the code is is organized and thought about Uh. I'm GonNa Start from just how the projects are. Run react is a project from facebook. They have a heavily staffed internal team that are paid by facebook up to work on react and that heavily influences the direction of their development. And they're very open about that. You know Dan Abramov is one of the most visible members of that court statement. He says you know what y'All should know we are making decisions based on what facebook needs. The things that we're doing may not be optimal for you. If you are not at facebook scale or complexity city and that I think shows up in lots of little ways but it is something to be aware of and I think there's a lot of can hype around. Oh this is the react way to do it. We're GONNA do it this way. That actually does not scale down very well. It works very well at scale but then when new folks are trying to do it. Those practices actually don't work very well always for small applications and there's a lot of I almost want to say embrace of complexity eighty in that ecosystem where to a lesser degree than angular. I think angular actually goes even more in the direction of they have. The learning curve on. Anglers is really we slow and long in. It's very complex. Framework with lots and lots of different interlocking pieces because the mindset is enterprise and the people working on the framework and the primary users the framework have these massive enterprise applications where they are going to have complexity anyway angular scales down very poorly to small applications you the individual components that you might embed in regular Nila website or things like that react has a little bit of that problem as well and a lot of their time and energy is focused on the needs. The facebook has which is fine. They're open about it. The one of my wishlist items for the future is that they might put react out into a separate foundation the way they did with graph Q.. L. So so that it can be a little bit more responsive to the actual community using view on the other hand from a development stance perspective started as a Bedia fell style L. projects so benevolent dictator for life. Avenue created it. They are in the process of moving to being completely community run organization us been and essentially delegating more and more pieces than giving ownership. There's a well-established courtroom now with different areas of specialization Avenue is still sort of acting as a bit a bit of a dictator there but not nearly as much if adopted a very solid community are FC approach to adding new functionality and features in that has very much influenced -fluenced the development of view. Three which is supposed to launch any day now. So that's kind of the approach and development. I think that actually ends up playing out a little bit in what you see. In the communities the view community feels much more bottom up. It feels much more however say like can of global and grassroots driven. The reacts community. There is massive. There's lots of people. A lot of the very loud voices tend to be sort of your more traditional tech voices. There's a lot of loud white men in that community to not too too bad whereas I feel like I've gone onto view related conferences. They tend to be much more diverse along gender directions but also racial dimensions also H. Dimensions. I found a lot more. You both young and old ends of the spectrum that view communities than I have the react which just seems to be very traditional audience centered in terms of approach. Using the frameworks frameworks. There are some ways that they're pretty different so react has very much embraced a functional programming. Approach they try to do all sorts of things with Having immutable state in. Everything's returning you're trying to get to sort of very functional ways of thinking about the world even sometimes when that's very confusing to users to new folks there's kind of a philosophical approach. Their view has not approach that much as much and they've they've kind of centered around this concept of reactivity where you are changing objects rather than having immutable objects that than get processed in return new things so you're changing objects. It's and observing those changes having things react to those changes so interestingly enough react to that in the word does not use this concept of reactivity they use state changes as you're you're doing some sort of functional transformation than replacing the state whereas if you you're changing an existing blob of state and things react into that totally changing the subject rapidly. Because I I want to move through several different other topics in the gets perspective on on where you think the future is headed. How would you define the term jam stack and how his the rise of the JAM Stack affected fronton development trends great question so I just simple? What does the acronym means so jam stack concerns for Jam stands for Java script? API's in markup and the concept of Jam Stack is essentially going further and further towards this concept of separated front end and back in so many applications have started to be architecture where you will have a javascript application as your friend and then some sort of server application that is your back end but all of the communication in between those things is via. API's contrast this to I don't know ten years ago when most web applications your server was rendering your html and then when you were also putting out javascript that might manipulate that the client so the trend is towards this concept of Single page application. Where you just you have javascript APP? That is doing things and maybe it's just it's loaded all at once or maybe precompiling some things that you can load a page. That's extreme but you have the separated front end for your back. It Jam Stack is going way in that direction. It saying okay. Let's forget that idea kind of having a server application at all. Oh yes it'll be still be there but we'll call it. API are just API is in fact a lot of times we'll try to use third party. API So we don't have to build those pieces ourselves. We're going to focus entirely Harley on building this front end of the application. What does that look like? Can we generated all using javascript and markup and then anything dynamic we talked to the. API's big big picture. It's part of this trend towards pre compilation. We've sort of over time observed that one. It's it's often faster especially it's more responsive as you go to ship a bunch of javascript out and then only once you can cash in a CD and then only ship data out from from your core servers. So that's that's one step but oh there's some slowness there because that jumps has to render a page. That's not responsive right away. Well what if cupid ship just you can pre compile your html and ship your javascript. So all of that's going out and then all the data's only loaded from precompiling more and more things. Maybe we could pre compile all of our pages or almost all of our pages and have those out put them out on a CD and so it's very close to whoever's loading it have it load very fast and then only go back as much as possible or as when you need dynamic data only go back to Napi then everything else stay on what they call the edge. So that's that's the the concept that javascript urge jam stack is pushing towards saying. Let's push things further out unless pre compile more so a one of the Big Frameworks in the tech world. Is this framework called Gatsby which is a react based framework for generating websites and applications but the concept there is. They're trying to pre compile as much as possible. They WANNA precompiling every page even though you might have your data loaded from a database traditionally. Maybe you've views a wordpress background or maybe you have traditional database API got a bunch of content. There that's going to generate your pages maybe product pages or blog pages. Or what have you gatsby. Gatsby going to go and fetch that data not when somebody requests the page but at compile time when you're shipping a new version of your application it's going to go out crawl through everything it needs to crawl through to get the data to render your pages pre render all of those pages and then you stick them up on a CDN so that when somebody tries to load it. It's all there right away. It doesn't have to touch any sort of database or processing processing to get there. Got It when I'm building a new product G two. I is the company that I call on to help me find a developer who can build the first version of my product G two. I is a hiring platform. Run by engineers that matches you with react. React Native Graph Curiel and mobile engineers who you can trust whether you are a new company building. Your first product like me or an established company that wants additional engineering. Help G two. I has is the talent that you need to accomplish. Your goals go to software engineering daily DOT COM SLASH G to I to learn more about what g two I has to offer. We've also done several shows with the people who run G to I gave Greenberg and the rest of his team. These are engineers seniors who know about the react ecosystem about the mobile ecosystem about graph Q.. L. React native. They know their stuff. And they run a great organization in my personal experience g two I has linked me up with experienced engineers that can fit my budget and the G.. Two staff are friendly and easy to work with. They know how product-development works they can help you find the perfect engineer for your stack and you can go to software engineering daily DOT COM SLASH G to I to learn more about. GTI thank you to GTI for being a great supporter of Software Engineering Daily both as listeners and and also as people who've contributed code that have helped me out in my projects. So if you want to get some additional help for your engineering projects go to software engineering. Daily Dot Com slash gitai continuing. We are trend of this joint topics. We'll we'll bring some of these together is web. Assembly Impacting Front End Development. Today good question when you say impacting the thing there's a couple of ways I could interpret that so one as your typical javascript or front end developer. No you're not thinking about it at all however it's already being included in ways that you're using you're not aware of so for example Bell birds using different types of tools and looking At when you're doing that MRT and compiling of things you need to willing to be able to go back to. What your original source code Aziz things called source maps that basically map from compiled code to your pre compiled code the tooling that is working? On the source. Maps is written in shifting Wembley and so are a bunch of other libraries and and so I think your average javascript developer. That's going to continue to be there or your average front end developer. That's GONNA continue to be. Their interaction with web. Assembly is under the covers. They're gonNA use a library that happens to be implemented in Web Assembly. What Web Assembly does for you though is it allows folks who are not traditional front end developers offers to bring or not traditional web developers to bring their applications to the front so for example? I talked with a guy last year. Who he he has a gaming engine that he had built with unity and it was it written in c plus plus was functioning Download this game engine. Whatever and just use it as isn't installed application but pilots web assembly? He built a react based front end and suddenly his entire gaming engine in your ability to write games was available in the browser. So that's where I think. New website is going to impact the front end. It's not for how it's going to change current front end developers there for them it will most likely be hidden in libraries unutilized. It's another tool in the tool chest. Import it the same way you would javascript. But what it does do is it allows you to bring things that are traditionally not web applications or people with skill sets that are not traditional web development skill sets and move them into the front indeed so impacting in small ways today almost inevitably impacting us in big ways in the indeterminate. Time Horizon Future. One Way to think about the web browsers is the web is the most widely distributed in most those popularly used delivery system in the world and it used to be that that was only available to documents. It was a document delivery network then we added applications nations. But only if you're willing to use the particular programming paradigms and languages particularly Java script of the Web Web Assembly does is say. Hey you know what screw that. Whatever programming language you want to use you now have access to the largest APP platform in the world way bigger than you know any sort of mobile APP platform way bigger than any sort of desktop APP platform? Everybody can use this and you can now access that from whatever language you are right defined the term micro front end for me micro fron is being. Put it in the context taste of twitter. But I'll put it in. The context of is something that I think your audience may be more familiar with. which is this idea of? Micro services so micro services on the back end have been A popular topic for a while with lots of opinions on whether they're good or not but essentially what they let you do is organize your code base as a set of independent services that can evolve independently that don't have links to each other other than via an API and whatever guarantees you on that API that has the downside of dramatically increased operational complexity but the upside of each one of these particular things is isolated. Probably easier to test and more importantly can be owned any devolved Its own team using their own technology choices without impacting the other services very valuable particularly in large companies where you have lots of teams and synchronization costs. Between those teams is very high microphones is saying. Hey we do that same thing on the front end. Why not we can Dan? Have this team owned this part of the Front Abo. If I'm Amazon or something like that like I might have a whole team focused on the navigation and and I have another whole team. That's focused on the product page. And I have another whole team. That's focused on the shopping cart. And all these different things and some folks said okay. Well let's let each of those teams make their decisions and ship. Each one of those atavism microphones and then they can only interact the the challenge there is substantially more operational complexity. There there isn't micro services because one you have to be really rigorous about what are the key is how do they interact with each other but to because because we've got all of these different frameworks and since everything's running. In the browser the framers themselves have to get shipped out and they have to go over the wire to the browser to run. So you if you're writing microphones in reacting view in all these other things like anything. That's running those microphones needs those libraries out there in a a micro services I don't have to worry about your servers doesn't matter what's running on my server because all that's coming back to you is the data from the in a micro front and you can quickly. We ended up in our environment. You can quickly end up in a situation where. Your friend is loading. Hundreds of kilobytes Maybe even megabytes of Javascript to running to these frontlines now there are other solutions. There are folks who say well we're not going to restrict you entirely but we are going to restrict you to one framework so they don't end up with that or they do some sort of integration layer where essentially have a proxy inbetween weaves together the front ends rather than loading them all from directly to the browser. We've lived together rendered output in a proxy server of some sort and then ships that final version out. There's there's lots of different implementation things. But the high level there is. It's tried to apply The concept of micro services to the front end but it turns out there's even more operational complexity getting it right to say that this is something that people don't actually use. This is just. It's like a twitter talking point. No I think there are folks actually using it so by the way it basically being the idea like. Let's say I'm a sizably large company. Let's let's say I'm like AIRBNB. I don't think AIRBNB does this. But let's AIRBNB. We want to give the developers lots of freedom and we want to give them the ability to for airbnb experiences. They can develop their system in view and AIRBNB. The home home sharing platform is is developed in in spelt to end. These different teams can work with their different front end systems. And that's all Hunky Dory and that's the idea of the micro front end it's really the bring bring your own front end kind of yes though. Let me clarify a little bit more. It is not that uncommon to have different sections of your application owned owned by different teams potentially using different framers so for example an admin dashboard. That's written in view but a customer facing experience it's written in reactor or even in an old customer I was chatting with somebody from Oetzi and there. I believe it was their internals. have been rewritten with react for a long time but the customer facing stuff is all still being rendered by Jake work. That does stuff right like. That's not that uncommon divided by product area. And it's it's often not that bad because you don't have often the same people aren't using the same all the things at once. Maybe you. It's divided by a user type in essays case might be customers versus this shopowners or even just like when I go to a page. It's going to be like a brand new page refresh. But that's okay because it has to load this new library where microphone and and I think really start to talk as when you're talking about Subdivision at the component level so I have a single page with a dozen different components on it at this component is owned by this team and that component is owned by that team and those things can evolve independently. That's where it gets really complicated in terms of allowing fully different frameworks. And you you have to be really thoughtful or you'll quickly this situation where you have can have catastrophic levels of Java script going to the browser. I think there are people trying to solve all of this because there are there are some sort of benefits to being able to isolate things in this way I think most of the people actually using this still put a restriction. So they say okay you. You can own a lot of things independently and we're going to stitch them together and we're not gonNA have them as part of the same code base may be but you still have to use react. And so they at least eliminate the multi the front end loading lots and lots of different front. End Frameworks on the same page okay. There's a whole bucket of other things that I didn't get to explore with not that I wanted to things like machine learning and tailwind CSS in Graph Q.. L. And so on. And maybe those we'll have to wait for some future conversation or something but one thing I I really wanted to get your perspective on just because it's come up in a lot of recent episodes is how we do get to this dragon. Drop world assuming you believe this is a reality like I mean I see it. I see it happen I. It is happening happening to some extent in in the low code environment and and to me. This seems like a very important trend. I I mean I think some people who listen to the show think have latched onto like a brain virus of the low code. No code stuff but to me. It seems like this vision of whizzy wig. Software development is finally coming to fruition in some sense and it's hard to know how mature it is. It's hard to know where I stand exactly how widespread the uses and the biggest thing. I'm curious about that. I feel like we can actually discuss. Today is what bridges the gap between these two ecosystems. The low code dragon drop perhaps in many cases proprietary based interface builder ecosystem versus the We build your own from the ground up Java script based Classic Framework World. How how do we bridge the divide between those two ecosystems and what is the nature of the divide as it stands today? Yeah that's a really interesting question so I think think one way to think about this is rather than thinking about it as a. There's a no code code. World think about it as east increasingly powerful powerful abstractions and increasingly powerful tooling. So the amount of stuff that we can get out of the box. Today whether it's front end or back end for doing development is astounding and both as a developer looking at tools like reactor view on the front end or looking at managing servers on the back end with Cooper netties or not having to manage servers on the back end. Because you use a platform as a service Roku or things like that like the amount of stuff that I used to have to worry about that. I don't have to worry about it. All is shocking and that is playing out on the UI builder side. As well you I- builders have gotten more and more powerful awful more and more integrated able to do more and more things and I think those trends continue to rise so the question is where is the line where the majority of situations that we're wanting to accomplish are below that line of it's already been solved as automated away. I just have to wire things together and I think actually jam stack is is an interesting driver of that. It's pushing things in that direction. Because it's pushing it's creating an ecosystem system where for example ZAP here is a viable company where I can build a business that is only about creating dragon drop relationships between third party services and I can do that because there's enough demand from people building you is to do that or there's enough demand from people who are building building things on wordpress or who who are essentially doing no code right. Now that want to do things that traditionally you'd have to write code to do you'd have to run a little server somewhere will instead instead of just going to wire them together with Zappeion. It'll just go so I don't think it's a line in the sand before this point we're going to be coding and after this point we're going to be no coding. I think it's just saying like what is the level of work that we need to be doing. What is it that we're trying to accomplish? And how much of that can be done without reading custom code and I don't know where the line is today because I am actually. I don't honestly care that much about no code because I like coding. That code has gone. Come very far. I think were going along those lines when we talk about building a bespoke application from the ground up. There were zero per people doing that today. Folks finally equivalent zero because everyone's building on top of a framework like react or they're building on top of even if they're building on top of J. Query right right. There's this massively developed piece of tooling that has had thousands of hours of time put into it. That makes your life easy now now and the typical javascript application that you ship may touch us either in the actual code or as part of the bill process over a thousand independent open source packages. I like you install the application template essentially from create racked. React APP which is like. Here's a standard template for building a react application and it installs a thousand packages. Most of those are things that get built are part of the bill system compile compiled away. But it's touching all those things that's all software you didn't have to build and so it conceptually. It's no coat right like if I if Conoco just needs code. I don't have to worry about So yeah I think we're already getting there. In many ways and people who are doing application development right now are on sitting on top of the same foundation of the people people who are using dragon drop platforms to create applications. It's the same stuff it's open source packages that are and available. API frameworks that interact with each other. That anyone can access all right. Well just a discussion at the end here about software software podcasting so podcasting as a way to explore and disseminate information about software engineering. You and I both do this thing. How does this fit in to the educational no path of a software developer? I mean we all know that you can only succeed is a software developer. If you're continually learning I mean I guess it it it does depend on. You're on your definition of success. Because I mean you can just learn kind of one paradigm of software and just kind of maintain code basis in in that paradigm for the rest of your life and have a very good living so you don't necessarily have to continually do the reinvention sort of thing. But certainly most people make some habit out of it. But what's your perspective on the software podcasting medium. And how does it look going forward. Is it a durable medium. Oh absolutely absolutely and I'm not just tooting my own horn there because I obviously do have a little bit invested in podcasting doing well but for me podcasting casting gives you as a listener scalable access to the type of content or information that you would otherwise have to go to a conference to get we actually Ajay Pretty. We didn't episode amusingly labeled the wonderful thing about figures. But we did this episode on learning and how How different how we learn about different things and one of the things that came up there that I thought was fascinating was different mediums? Help you learn at different different levels. So going to conference is great for learning what you should learn about. Its great for getting excited. It's great for inspiring you. It's great for getting the big picture. It's really bad for digging into the nitty gritty details on building out a tactical skill and I would put podcasting in that same bucket. PODCASTING is a wonderful way to discover what you should be learning about degraded it. A sense of how people are thinking about things. It's a really bad way to learn a particular learn tactical skill but I think that piece especially if we look in a world that is so incredibly filled with abundance. I mean look at what I mentioned in terms of Javascript. Five hundred new open source packages ages every day figuring out. What you should be learning about is a huge part of the challenge of learning? What should I learn? And so podcasting odd casting is a scalable way to disseminate that same type of curation that same type of inspiration and that type of access to what are the brightest people people in the field learning about thinking about speaking about in a way that you can listen to in your pocket as you go for a run. You don't have to travel halfway across the world to go to a conference on it. You can listen to from anywhere anywhere Kevin Ball. Thank you for software engineer. It's been great talking absolutely as a programmer you think in objects with Mongo DB sodas your database Mongo DB is the most popular document based database built for modern application developers and the cloud era millions of developers use Mongo DB into power. The world's most innovative products and services from crypto currency to online gaming Iot and more try Mongo. DB Today today with Atlas the Global Cloud Database Service that runs on aws azure and Google Cloud configure deploy and connect to your database in just a few minutes check it out at Mongo. DB Dot com slash atlas that's Mongo DB dot com slash atlas thank Hugh to Mongo DB for being a sponsor of software engineering daily the

developer front end developer facebook AMD Sumo logic L. Gatsby Kevin Ball Amazon Software Engineering Jazz Party Jeff good Dan Abramov Victor ops software engineer Aws Services
Okta Engineering with Hector Aguilar

Software Engineering Daily

54:22 min | 1 year ago

Okta Engineering with Hector Aguilar

"A new employee at a software company needs access to a variety of tools in order to get started working the employee might need slack e Mail Google Docs and Amazon web services and all of these require an account with a username and password setting up all of these accounts can be time consuming because the company needs to go into their admin portals and create the accounts and then the accounts need to have the right security policies in configuration settings and when the employee leaves the company. All of these accounts need to be shut down. OCTA is a company that builds identity and access management software such as a single single sign on tool that allows users to log into all of these different types of accounts using only an OCTA. Lakhan Octopus started in two thousand nine and has grown steadily since then going public in 2017. Hector Aguilar is the president of technology at OCTA and he joins the show to talk about these software stack back of OCTA and how the company has evolved over time as it has become a core infrastructure provider and hire large engineering team and scaled and built additional national products. It's an interesting story about a company that has grown and built new products and skilled engineering workforce out. Enjoy it the log. DNA allows you to collect logs folks from your entire. Cooper netease cluster in a minute with two QB C. T. L. commands whether you're running one hundred or one hundred thousand containers you can effortlessly aggregate and Parse and search and monitor your logs across all nodes and pods in a centralized log management tool. Each log is tagged with pod name and a container name a container. Id and a namespace and a node log. DNA is logging that helps with your cougar nineties clusters there are dozens of other integrations with major language libraries in aws and Harajuku and fluent D. and more logging on Kuban Andy's can be difficult but log. DNA simplifies the logging process of Kuwaitis clusters. Give it a try today today with a fourteen day trial. There's no commitment. There's no credit card required. You can go to software engineering daily. Dot Com slash log. DNA to give but a shot and get a free t shirt. That's software engineering daily. Dot Com slash log. DNA thank you to log DNA for being a sponsor of Software Engineering Engineering daily actor Aguilar welcome to Software Engineering daily thank you thanks for having me. It's great to have you you joined OCTA in two thousand twelve and we're GonNa talk about your journey and the OCTA product right and how often has evolved describe what the product did when you joined. What did OCTA do well. I think initially OCTA was focused on basically single sign on the idea that you would need to have credentials to multiple cloud services services that you need to get your work done and OCTA started a product that would allow you to have a single credential for everything that that you would use to to get your work done and so if we go back eight years ago there were a lot of cloud services certainly less than now but there were still alive glossaries already. You have yourselves. Force your box your dropbox and anything that that you would use cloud service in the idea would be to have a single sign on for for everything that you had. That's that was a long time ago. Why is that kind of identity management useful. Why do we need a product for that that well. I think there are several issues we'd identity. The first one is really security ride you what was happening back. Then is is that organizations were provisioning all of these accounts to these cloud services and then you would have to set up a password for each of those cloud services all of those cloud services would have different password restrictions different security settings and then at the end of the day the user would have to remember all those passwords and so what ends up happening is that did you end up writing all of your passwords in like little posted node which is not great in addition to that if one of those services gets compromised and you're using the same password for all of those services ervices obviously all of your services get compromised as well and so the idea of having gone a single sign on protocols with strong indication so that you can sign in in once get kind of strong authentication multi factor and then you basically get access to everything that you need was and continues to be very appealing to our customers so that's kind of one thing from the security perspective from the second aspect is the provisioning of all of these accounts to all of those services used to be. MANOHLA swell ride you have to go to box and imprecision a user and then go to salesforce impression user and OCTA also allows you to actually provisions and provisions those account that means that when an employee joins one of our customers they automatically get probation to all of the services all of them all of the services that they're going to need to their their work kind of day one and that's basically one of the one of the Nice Nice things about October that your day one experience becomes really great in addition Asian to that when somebody leaves the company or when you have to the provisional employees will go on the prohibition they accounts or suspend the accounts on on all of those services which means that because that's done automatically as well you don't have the concern of users having access to your your company data after they leave the company which was also again one of the problems that we had as an industry longtime ago so there's a few types of products that I'd like to contrast with Octa just to help listeners who are unfamiliar with this this kind of product so I run a small company for example. We're based around Google APPS and Google title APPs is my notion of identity and it works for us pretty well because we only have a couple of employees. We don't have a large scale all employees based. We don't have a bunch of new accounts that need to be provisioned on a regular basis. How does OCTA differ from that kind of experience experience the experience of the small business owner who just uses google APPs for their identity well. I mean everything that I have described. So far is the perk that we have eight years years ago. The broadcast has evolved quite a bit so the problems that I was talking about is basically the way the company started. That'd really looking at the exact problem that you're talking about. When you have a small kind of business that has let's say let's say goal as as their email and maybe be there on a document repository. That's basically how they start using OCTA but then I'm using it hub and I'm sure you're probably going to start juicing gear up for your backtracking definitely get hub right and maybe they'll use confluence. Maybe you'll use some sort of wiki pages or maybe you'll use. Amazon web services is due to provision your infrastructure. Maybe you'll have some monitoring systems like maybe maybe sumo logic us wipe front or Zombie so oh at the end of the day. It's not only what you use for collaboration. It's also anything that you use to really run your run your product and I'm talking about engineering engineering but eventually you're going to be using quick books. You're GONNA be using nets with. You're going to be using salesforce as well and so. All of these services is exactly what OCTA can can do for small businesses and again. That's that's how we started and so now you have this concept of an identity for your company in any employee can get provisioned into all of those services and if they are engineering they will get basically access to get help and maybe confidence and maybe Jira and maybe whatever you use for backtracking whatever you use for your roadmap planning and then anybody that joins finance will also in addition to go account. They'll get their salesforce account. Maybe the nets with account and so at the end of the day you really really create a central view for that entity in your company in how that I did use for all of the services that you use and that's basically when when companies are a small right ask. The companies continued to grow. You're going to continue to need access to old more services and that's basically how continued to evolve into really becoming a platform former identity for everything that you use because it's not only the cloud services that use maybe you'll have some internal systems. Maybe you develop your own your own provisioning learning system for for accounts in in your own service for example well. OCTA can connect to it or in the more can of newer scenarios. Maybe you're a small company that doesn't even want to build at entity on product use the platform product which is basically the second part of our business that we also have where we can actually manage the windy for your product as well so that our customers that have gone beyond what we call the workforce which is people are using cloud services to actually embedding OCTA into their service. Because of course we are entities so we have. Api so you can actually directly call our API's to actually power your entity for whatever product or service you're developing so we have the developer side and the it side of you will so you have everything that I have that have been talking about this kind of what the CIO would be interested interested in or the the director of it would be interested in but we also have kind of what the city are would be interested in which is really using OCTOPI is to build this platform do power identity for their service and then they don't have to worry about not only username and password but also they don't have to worry about multi factor indication supporting multiple standards words. I Dunno phase of ide- or thumb authentication all of that can be done via. Just OCTOPI is so the first engineering earing challenge that stands out for a product like OCTA is that you essentially have the security properties properties of a password manager and that means that you could in a worst case scenario you could have this this honeypot of password password information that if it was exposed it would really created a significant security problem for a customer so security is something it's been paramount to the product from day one can you describe how security management factors into the engineering nearing process at. OCTA YEP absolutely so securities is embedded in our DNA in our culture engine concession of course and I've been insecurity for for a number of years to know that security is not a digital thing is not secure not is really a continuum of things that you need to do and I think part of it is just embedding it really in the culture in constantly evolving it so from the way we train our engineers in kind of common security. Gotcha in code or concept to the way we have our security team constantly pen testing constantly doing source code reviews constantly hiring of another other firms to constantly look at the security of the now again continuum newnham things that spans across several civil aspects of of security now if you look at the protection of material if you will that also has evolved over time how we use came services we actually use it. Ebola's came as service to all of the all of the private material and we also have have can of segment of Keesa customers for for obvious reasons so there is a number of things that we can do. One thing that I think is important to notice you talked about kind on of user names and passwords octaves much much much more than that because we also do multi factor authentication so it's not only can your your cinema passwords also all the prevalent material that you're gonna be using to authenticate to actor so for example when you log into to actor our customers can also configure multifactors indication which means that for example you could get a push notification to your watch and to your phone after you type in your credentials to refi who you are or you can use web authen for example interviewer now. Microsoft surface you can indicate with with your face and so there are a number of things that go into strongly authenticate a user and all of that at private material we have to protect and they're kind of different ways that we do that in addition to that we have a ton of data that we can use to do some analytics just because of kind of the tapestries are we are encounter scale that that we have gotten to we have lots of data and we can also entity signals to protect basically look at what people are doing and because of their intimate patterns and we cannot identify signals that can help us also strongly authenticate users. OCTA was started in two thousand nine. This was the earlier days of Amazon Web. Services gaining traction was OCTA built out in the cloud from day one yeah so I mean our founders from salesforce are used to be the used to run engineering for for salesforce in the very early days and so from the first line of the code October designed to be a cloud service We've never had an on premise server till this day irritating. Mrs Cloud and we decided to go on on idol. US but also you need to do to think about the fact that this was ten years ago so a lot of the of the infrastructure and services that has they didn't exist back then but they had enough that we could really build our infrastructure. Nedeli so everything is cloud how from the beginning multi tenant from the Beginning Beginnings Scalable enact cloud way in from the beginning basically to be to be a service and describe the software architecture as it stands today. Obviously the infrastructure options on aws have improved a lot. You've also got a whole variety Bridie of other cloud providers. You could be using you've got a variety of virtualization tech and containerization technologies you could be using what does the software stack doc look like today. Well are so forth. STUCCO has evolved right. If you think about ten years ago I'm not even sure that contains existed. Certainly docker was not what what it is right now and so are architecture has evolved but if you look at its traditional enter breast doc where we have basically multiple tiers of of St Louis servers bag by a database and a bunch of caching engines and and back in jobs so we're building Java I would say ninety five percent of the product is built in Java and we're back by my sequel databases many of them and then we use radius for cashing in also pc so and a high level we asked we we started learning about new services of aws for example would really look into what other services that we could leverage but also very careful about not just chasing the latest it is shiny thing because a big thing for us is reliability from the very early we we understood that we were building infrastructure service and a low of our customers kind of think about US power Internet you know if you have a business cannot continue cannot work especially the customers that are using us to back their their service identity in relevant is incredibly parliament for them and so because of that we cannot put our architecture at risk so what we do is that anytime there is a new series that we wanna use we typically create an obstruction before even started to trying it out so that we could always kind of have ensure that there's no single point of failure in the architecture and so I would say that the way they architecture has evolved has been mostly around reliability and skull ability in the sense that had we need to understand the failure mode for all of our components and at different levels of stack and make sure that the stem is able to work around those failure modes and react very very quickly and hopefully at computer speed and not not human speed and so as you know I mean this is infrastructure services that are out there they they don't provide redundancy by themselves so or did they provide redundancy. We cannot even rely on that. Redundancy is basically building an architecture where we can build redundancy on on top of redundancy and we can make sure that at any point in the architecture there's no single point of failure that has been there the relive Lucien of of the architecture and everything that we have been adding is built that way so for example if if we need to send an SMS if we want to send an SMS as part of your factor so that you type in your user name your the password and then we also send you an SMS message right well we cannot rely on the SMS provider because Sims providers could also be having a problem so we have two of them and so and we automatically are doing dynamic routing communes that if one of the providers he's telling problems or we our customers have not we have basically some curious is that we have built into the system and the system will automatically switch to other providers and we have to at any at any any of the architecture so it's not only the services that we use Broza how how we use them to ensure nursing failure email has been around for longer than I've been alive but there's been surprisingly little innovation in inbox management. Sane box is a new way of looking at your inbox that puts features like snoozing and one click unsubscribe and follow up reminders as first class citizens. It's if you are overwhelmed by your inbox and you're almost ready to declare email bankruptcy tryout saint box and the on boarding process sandbox inbox analyzes your emails and helps you sort them into categories. You can get a free fourteen day trial and a twenty five dollar credit by going to sane box dot com slash. Save D that's S. A. N. E. B. O. X. Dot Com slash. Sad these days. I spend more time in my inbox than I do in front of my coating environment and back when I was programming I would spend hours configuring my coating environment because I wanted to maximize productivity if you spend as much time managing email as I do. It's crazy not to set yourself up for success with your inbox so stop the craziness. Get Sane with saint box go to sane box dot com slash S. E. D. and get a free four day trial as well as a twenty a five dollar credit. Thank you sane box for being sponsor of software engineering daily. Let's take containers in coober netease as an example. This is a piece of software that has become much more accessible to the average engineering organization ten years ago I yeah I don't know if I think containers were a thing in the sense that Google had internal container technologies. You know there there were see groups in name spaces in Lennox and those could be turned into things that people called containers but today a year you have the the product is Asian of containers in terms of docker or the usability improvements in terms of docker and then you have the container -tainer cluster management system in terms of coober netease now. That doesn't necessarily give you anything that you didn't have with just e c two oh based virtual machines on aws but it does give you some greater flexibility some better economies of scale gail in terms of managing your infrastructure and so with that kind of architectural decision. I'm I'm curious. As to how quickly you would move in that direction or or if you would even move in that direction because maybe it doesn't even matter that much. Maybe that's not something you even care about that much. Because every you know the thing that's most important is reliability ability. Maybe this is just such an important server component that you just don't even want to mess with it so as kind of a case study in OCTA OCTO infrastructure. How have you approached the newer containerization options yeah so the reality is that as the OCTA architecture has evolved. Also products have evolved as well right. We have new offerings. We have kind of new things that we're offering to our customers. So I'll give you one example a long time ago. We didn't have an elder interface we basically build our own directory so we can be the master for for users but our API's were not held up compatible and so we started basically having customer that actually wanted to connect systems that we're using the API directly to OCTA and we didn't have an interface for that and so we had to build a new interface a new translation layer of you will that would take the elder API's and and translate them into something Dr Understand that when they were two options one option could be you know. Put it in in the indexing system. Just tell you even more code to the Latte translation layer and then you're done and the other option was basically to say hey logistic followed the technologies that are out there right now in how we can leverage them and see see we can create our surveys that it's kind of look more modern and gives us all of the kind of new things that that we could get from containers from anything that uses right and so we decided to actually go with the micro service and so they held up interphase that customers can use it's actually backed by a micro service and then micro service statelet service that basically those that translation and that's build with containers and build kind of all of kind of new new new things that are out there now so it's a matter of like anything that these new we tried to leverage kind of the new technologies and the new stuff that it's out there that is also changing very very quickly and then for the stuff that is kind of more legacy gacy would try to find when is the right time to basically extract maybe a portion of an and make it an tried to use the new technologies but there's stuff that is going to continue to work and it's going to be avert reliable so we don't really want to touch it and when we touch it we wanna make sure that we converted in a way that we always have a fallback mechanism. That's why does any any change that where we actually touch model that module that has been working for like now five ten years it really takes months if not years for us to make ktar transition so we were very very good for about it but we of course we. WanNa try to leverage the latest things that are out there so we do have an initiatives that are kind of taking some of those modules in evaluating when we need to use can newark new technologies but everything new and so being kind of in whatever kind of new architectures are out there and so being being based in does something at the heart of autism engineering is integrations so as you said in your initial description of the product if I get an OCTA account when I sign up for the company if I joined the company as an is an engineer the OCTA account when it gets spun up is going to spin up a user account for the company's box and the companies get hub hub and maybe the company's Stripe Account or Twi- Leo and all these different SAS tools that I might need a log in for four OCTA becomes the sham over that logging experience and that means that you have to build integrations with all of these different. SAS providers and integrations can be a really overwhelming challenge for you know for a company that does have to build a lot of integrations because as I see it that the number of SAS tools that you have to integrate with is just it's growing and growing and growing and you have to maintain tain those integrations over time so how do you systematize the testing and the expansion and the the demands that come from a growing wing surface area of different products that you have to integrate with yep it stuff. This is a good way to think about the problem so the reality is that when we actually developed OCTA. We probably need to focus on. Maybe like twenty. Maybe fifty products ready had the boxes of the world sales force of the world world in everything was basically build in a monolithic way. If if you will right so we would be developing those applications they ended up being kind of like a fixed hard headcorn classes in China that will connect to those services eventually. It was very obvious to us that we needed to really take more of like a platform approach and so over the years we basically created a fool on Meta data based architecture so that we can develop these applications much much faster. They're kind of new services calm along in some cases. We don't even have to write code because at the end of the day you start looking at kind of things that are common across multiple services and then and you just need to configure its own becomes marriages configuring a template of you will template icing this this this conversation so over the years we have evolved that piece he's but he's definitely one of the biggest things that that we are. I think I would say that October three things so we are an infrastructure company we are security company and we are of course an integration company company and not necessarily in that order and integrations has been really big for us as I said we haven't made a data system to do all of that we have application analysts that can and tweak or fixing application very very quickly but also we have a lot of risk based monitoring because it's it's almost trying to live in a world where eh a lot of it is unknown and also you don't really control it because for example tomorrow one of the thousands and thousands of providers that we support could change their API without telling us because of course there they don't have to tell us and so we need to be able to take that and fix it very quickly now of course we have really good partnerships with a lot of ISP's but there are some smaller vs that are not even going to know that they're breaking integration with other other singles protocol or the provisioning protocol so a lot of what we do and this is very big for us is being able to monitor identify patterns identifier can of kind of regression so that we can react very quickly. Try to find that's many microaggressions we have and also being able to evolve. These applications as new services are added in addition to that. We also create framework so that customers can build their own applications so if you have an internal build system Hawk that doesn't have to build that application anymore that used to be a time when we did now. It's basically more more of like a platform approach where customers can actually build their own integrations and that basically exploded as well in terms of the numbers at one point. I remember several real years ago. We were we had about three or four thousand applications in there what we call the integration network and when I did a quick search to see how many customers awesome had build their own applications. The number was about thirteen thousand and so it's definitely more than a order of magnitude because once you let customers is built on your platform and they have they may have like five or ten internally build applications that they want to integrate that in the numbers becomes really exponential and just fast forward two four three three four five years and then that that continues to grow last not least which is also very important integration is that we we also extended our reach because before we only used to do cloud applications now we can do on premise and now we can even do locations behind the firewall so applications that even our our system whereas the developers do not even have access to you can still build them with on premise provisioning or on premise single sign on that we can that we can support as as well so the number of applications has grown dramatically over the years and we also do and because of that we also have a lot of data and we produce a business at work report every every year where we actually show all of the Trans implications in what people are doing with with application network so yeah big big thing for us one of the trends in the last ten years has been the growth of API company so it's obviously box has an API but the boxes also a rich platform where you're managing files the more recent things that people think of as API products products are things like Twi- Leo or like Stripe how does octa integrate with these different kinds of API's is there anything notably notably different about octopus integration with the API companies relative to the more classic SASS product types. We'll we are an API company in the sense that there is a lot of customers that are using OCTA for the product and service so the traditional workforce workforce is one of our areas the other side is basically using octave API's to power your system so there are some even consumer websites that are using OCTOPI is that you're not even be able to tell that octaves behind the scenes and in authenticating the user but customers can actually use her API's and when we we started having that kind of line of business we basically started getting really good at really indicating API's so for example we can be an an open been ID token provider for example where people can actually secure those is API access management with OCTA in so that's basically more on on the a CTO said that I was talking about where we actually become really authentication for those services and we can give you can of tokens where you could even kind of indicate your own services so now in a world of micro services for example where you even want to authenticate your own API is you could also use Oxen Open Eddie provider in have client ideas and in create kind of authentic gators all built in October. Now that is a very very different thing than what we have been in describing us us basically the workforce or connecting to cloud services. This is more of like what a developer would do you go to for example developer DOT COM and and you're trying to build an indication for API's. It's going to be a very very different experience that you go to to our website in and go to our workforce side which is more of off again this year versus the CTO so yeah I mean the short answer is very very very different world but actors an identity platform we also so think about not only indicating people but also getting services nothing getting APS. I'd like to walk through a you log in with octa in some technical detail. Can you just explain what happens when a user logs into OCTA OCTA give me some of the breakdown of the security key management and the workflow of different services behind the scenes while that is one of the things that I think is part of our value develop W. bring to our customers that we make the complex. Look very very simple so let me there are a number of things that that could be involved in in an actor anticipation. I'M GONNA try to do is one of the most common ones would would may or may not be one of the simple so let's say you want to authenticate to you click. Let's let's say you you get an email and you get a link from sells for record that you wanna that you wanna see so the way it works. Is that UC a salesforce link on email email then you click on it in your browser. It's going to take you to that. SALESFORCE page. salesforce knows that you're trying to access a specific attendant. Then salesforce is configured to use OCTA or cellphones now dad four so we'll know that occupies your entity provider and so then salesforce is basically going to redirect you to Octa with some magic redirection and then your browser is actually going to show an OCTA page and it's going to be your sign on page. There's going to be customized to you so you're we already know the tenant. We're GONNA show you whatever page it's customized to meet whatever your tenant anemic and then he's probably going to be using Password Ryan the most simple simple scenario so you'd have been your username tapping your password dad since those grand French else to Octa they're not there are some policies that are evaluated and then at that point. We evaluate where you're coming from. If you're you're even allowed to the authenticate if you have any GEO location restrictions for example if you have any specific policies around your advice about your around your Ip address anything thing that gets gets evaluated in the policy so let's say you're policy requires a multi factor authentication so then at that point. We're going to return different pages to you you so that you can use one of the multi factor that you have configured and let's say that that factories push notification to your watch and so at that point the page that we send you it just going to be like hey you're going to be receiving push notification accepted and then at that point the backend is GonNa via the APN if you're using or whatever they apple push notification service or the notification service. You're going to get it to watch in your watch. You're going to see the location of their request. You're going to get some information you get in your watch. You say except that except go through different low. Obviously whatever you're watching using for wildfire lt and it's that's that's GonNa talk them out. There is going to make that connection is going to know that okay. I just had a partial session and the verification from the second factor that is going to create create a strongest indication now okay now. We know who you are now. We know who you are and then we're going to send that wasn't gonNA basically create a similar assertion and then we're gonNA that is really a token of you. Will that that we provide for your browser so that you can actually present that to salesforce so that can it gets to your browser. In the form of direct. You're basically to to salesforce salesforce will basically verified that signed by OCTA and they were so usually previous trust that was created when sales were initially configured and then finally cells. I will say okay no who you are. The salvation will have everything that they need to authorize. Is You and then you're. GonNa see the page that you had access to. That's a simple one. I can give you a little bit more complex. One wonderful now this. L. Associates one facet of OCTA. Which is that okay? This is core infrastructure and it needs to be quite fast and there's a lot of integration points. There's a lot of network hops so I'm wondering kind of instrumentation do you you have to be able to understand the trace of all of these different network hops and to be able to to diagnos latency issues that might be occurring along this complicated request path which by the way is just one of. I don't know how many integrations you have but you need. You must have some kind of standard instrumentation infrastructure for being able to detect. What's what's going on in this kind of trace. Can you tell me about that yeah. In this particular scenario I choose not to use multiple multiple identity providers or delegates indication for example averse an observer or some active directors indication there was involved in this flow which happens also for certain certain authentication paths right so you're absolutely right aware monarchic about monitoring and I guess the simplest answer is there is a request I I did that is basically past left and right but there is also a lot of systems that are involved in this and so we also are redone in the way we monitor our service and so we have several approach for ABM sold for log management and so we can actually track all of this and all of the of everything that is happening and we can tell of these requests Chris together but we cannot monitor that at human speed right that would be impossible at this scale that we are and so we also have created a lot awed of alerts and dashboards in curious sticks in analytics around all of these requests that are happening in our system and one thing that is I don't know how how unique it is but it was definitely unique ten years ago. Maybe I don't know if it's still unique now but our testing what's traditional like. Qa team team are testing is basically engineers that are not only kind of like doing the traditional black box testing. They're basically writing tests. In reading synthetic transactions sounds that are constantly having a doctor but in the just monitor what's going on but in addition to that they also create dashboard also create automated alerts they also create these humoristic to identify trends are are changing in that basically how we can monitor and react very quickly hopefully hopefully before customers nor or is there is there is a problem and then once kind of our automated systems alert. We have basically all the traceability of everything that is going on and we have threats halls across everything that I talked about for for example in delegated authentication when we have to negate to another system that is constantly being monitored when we train where even like a few customers are seeing light delays that basically triggers what we call tribal alert and those travel alerts go to allure of architects and engineers that are basically looking to see if there's a problem and if we we are anti that there is something that could potentially cause problems then they engineer calls what we called a flare and then when you call a flare is basically pager duty eh calling a number of again architects engineers and managers just looking to see if there could be any any customer impact after that we escalate to other phases of yellow. Oh and read just to make sure that we can try to take things that are not doing great in the service even before customers start noticing but again. This is something thing that we have evolved over ten years and over. Obviously I'm over simplifying it but yeah you hit the nail on the head. I mean we have to have traceability for everything that is happening because the complexity it is incredible and also this has to happen within seconds ready. Milliseconds looking for a job is painful and if you are in software and you you have the skill set needed to get a job in technology. It can sometimes seem very strange that it takes so long to find a job. That's a good fit for you. Veterinary is an online hiring marketplace that connects highly qualified workers with top companies vetera- keeps the quality of workers and companies unease on the platform high because veterans vets both workers and companies access is exclusive and you can apply to find a job through Vettori by going to a veteran dot com slash S. e. daily. That's V. E. T. T. E. R. Y. Dot Com slash. Save daily once you're accepted except to veterinary you have access to a modern hiring process you can set preferences for location experience level salary requirements and other parameters so that you only get job opportunities that appeal to you know more of those recruiters sending you blind messages that say they are looking for a job of a Rockstar with thirty five years of experience who's willing to relocate to Antarctica. We all know that there is a better way to find a job so checkout veteran dot com slash s daily and get a three hundred dollars sign up bonus if you accept a job through veteran veteran is changing the way people get hired and the way that people hire so check out veteran dot com slash s daily and get a three hundred dollars sign up bonus. If you accept a job through veteran that's e e R Y dot com slash. Save daily thank you to veterans for being a sponsor of software engineering daily. You've been with OCTA for seven years and the ORG has grown a lot in that time. Can you describe how the Engineering Org structure has evolved since you joined the company and how it looks today yeah absolutely I mean when when I joined we're very small. I think we've probably had like maybe ten people in Engineering Service was very very small. We have hundreds of people now in engineering but I think the core thing was that I think that they were getting station on structure has evolved as much as the company in the business has has evolved. We're very lucky and knocking on wood here but we do have a lot of people that have stayed in the company for many many years as well. We have people that have been here three four five six seven years so we have a lot of the same people that were in the company when the company was a startup continues to be in the company because again they're excited yeah about the mission they believe the work that they do is impactful and they basically have evolved with the company as well and so they have created basically teams around them and so we're constantly obviously improving the talent the talent pool as well but also the company has evolved to the point that I think one one big thing that I that I think I could give you as that's an example of a structure that has changed is that before we used to have a team that would just develop letting the product any features in the same same team would basically look at performance reliability so it's basically years one team and at one point in evolution of fog that we decided that it was sawed such a critical thing to make sure that we thought like an infrastructure service that we actually divided the team into the team that only deals we'd good infrastructure scale ability reliability technical operations and a team that deals with the product features a huge transition for us because at that time the plot management team didn't appreciate that we're basically really like carving out a big piece of the engine organization with with really strong architects were we're only going to be looking at really bitten bites and not any features and then also big transition for the company but it proved to be something there was very very powerful very powerful message organization until this point. I mean we have our ratio of people that are only focused on that and these are not just the traditional system operators. I mean these are engineers engineers architects that have been in the valley for many many years that are developing products that are only focused focused on on API performance on reliability on his Kabila on on like self healing and they're not basically they cannot be distracted for example by by customer feature that we need to release very quickly they they are only focused on the infrastructure of the service on the core kind of bones and beaten bites of the service that basically one example but I mean it. The transition has been quite big of you think if you think about kind of how the company has has grown over the past few years what's the hardest engineering problem You're solving today at OCTA. I think the hardest problem there's a lot of problems in part of my job is always being worried about kind of the next thing and all the being paranoid but I think the hardest problem is just to make sure that as an engineer station we can get ahead of the business growth and we've been very fortunate. The business continues to grow our customer getting bigger and the types of challenges that we saw are getting are getting bigger and more complex. I do remember that a long time. AGO would be scared when when there was a customer there was going to go live with like five thousand users ride and we were like my God I hope that system can can handle that and now we have basically legalize with with literally millions of users and so it's very very completely different different game and I think our big challenges and engineer organization is to to make sure that we can anticipate the business grow growth in we can build technology. There is going to be ahead of the earth so for us. It's all about business growth for us. It's about business growth and so if we were kind of the business was at the stage where we were on boarding thousands of users. We need to make sure that we the system can do ten thousand users right and so we basically need to be always one order of magnitude ahead of business growth. That's a huge challenge for a company like us where literally rolling in the span of this interview hundreds of thousands of people indicator doctor right and so we always need to make sure that the technology can scale much much faster than on the business so a long octaves growth you have expanded into this customer identity segment mint and from what I can tell this is. This is the largest product adjacency that you have moved into the largest brand new product adjacency. I mean obviously it's related to the core workforce identity product but it's something different customer identity. If I'm trying to manage the identity of my customers of my e commerce company or my media company that's a very different product than managing managing the identity of people who are working for me. So what have you learned in the process of building out that product adjacency yeah absolutely I think one of the things that that I just mentioned as well was you know in the workforce or the house there there is a lot of companies that have millions of employees right but on the customer side of the House that is a lot of companies that have millions of customers so it's definitely a different rent game as as as you mentioned and I think the biggest thing that that we have found is that the actual kind of infrastructure structure in the API monitoring becomes much more complex when you're talking about people that are going to be authenticating literally millions of the customers that are going to have people are going to be indicating really millions of times. A month versus an employee that would offend kate. Maybe once a week in maybe once an because then they're they're going to whatever they use for for their their business right. Also this despite there are some very interesting interesting specs that we see when people go live with new products or services but I I think the biggest learning has been or us the way we have to tune the monitoring to understand the use cases of our customers I in in what's important for them and so I think that we are we have been getting very good at understanding what our customers are doing with our product in there are kind of different different different ways to think about it and depending on what they're doing and what their businesses so for example of you are grocery store that has potentially intially thousands of ovarian millions of customers that want to get coupons every specific day then we know kind of how the usage patterns Ktar and so we have to build into our migrants so that we know that everything's fine for that particular use case whereas we have other customers that have kind of pro lives in things that are sending that that happened all the time then we also need to get ready for that and also the depth of customers that are like a set schedule of things that happened in every month so basically what what I'm trying to say is that we have to tuning and monitoring that the Taylor's very specific use cases in the consumer space. It's much much more than the workforce I identity in in in the variety of those cases. It's exponential as well okay last couple of questions. Tell him that you believe about modern software engineering hearing that I would be unlikely to hear from anyone else. I'll give you a two and these are two things that are really really embedded. In the culture of OCTA The the first one is that continuous integration lobes are not enough. I listen to that specifically when when we talked to small companies that have never really seen large scale and they'll tell you own. Just make sure that your test cases are well done and then you'll never have problems in production function in a in a product like OCTA. That would be very hard because we're always looking in designing for their known. We don't know how our customers are going to be. You can of integrating with our products and services we know kind of integrations are our customers are building and so our test cases have to be really good of course when they have to prevent everything that we know but we also need to get ready for the unknown so the pre-testing of you will becomes or the posting of you will become as important as predestined testing some monitoring synthetic transactions and really understanding how the product is being used become sometimes even more important than when you do pre-testing so that's one thing really follows that that that are continuous integration continuous deployment. Lopez always an offer quality. I think that just the beginning there is much more things you have to do to ensure a service with with high quality and the second one is we have hashtag the doctor called Hashtag no mysteries this and that has been something that has been very evident dr where because we're dealing with so much complexity sometimes engineers will kind of default to saying. Oh you know what there was an Edward Blair or maybe a customer's doing something that we are expecting they've stopped and we don't have to look into it and so because of that we have these Hashtag no mysteries because for every single thing that we are identifying the product we need to find the root cause for it and sometimes we spend literally months and looking into exactly what the root cause was for a specific thing even if it didn't have exact customer impact and because we're maniacal about that we we have a lot of very interesting things in our software and the progress that we use or even if an inner providers that have helped us build much more robust in high quality service so for us the no mysteries in the post testing of you will is a fundamental part of our culture that. I don't think it's needed in in all organizations. I think it would help all organizations but I don't think it's as critical as it is for Octa to have those two. Things really nailed okay hector. That sounds to close off. Thank you for coming on Software J. D. It's been great talking to you. Thank you Monday. Dot Com is a team management platform. That brings is all of your work external tools and communications into one place making cross team collaboration easy. You could try Monday dot Com and get a fourteen day trial by going to Monday dot com slash S. e. daily and if you decide to become a customer you will get ten percent off by using coupon code S. E. Daily what I love most about Monday. Dot Com is how fast it is. Many project management tools are hard to us because they takes so long to respond and when you're engaging with project management and communication software you need it to be fast you needed to be responsive and you need the Ui to be intuitive Monday. Dot Com has a modern interface. That's beautiful to look at. There are lots of ways to use Monday but it doesn't feel overly really opinionated. It's flexible can adapt to whatever application you need dashboards Communication Kanban boards issue tracking if you're ready to change the way that you work online give Monday dot com a try by going to Monday dot com slash s daily and get a free fourteen day trial. Gal and you'll also get ten percent off if you use the Discount Code S. E. daily Monday. Dot Com received a Webby Award for productivity tippety APP of the year and that's because many teams have used Monday dot com to become productive companies we work and Phillips and WICKS DOT COM try out Monday dot com today by going to Monday dot com slash. Se Daily thank you to Monday dot Com for being a sponsor of software engineering daily

OCTA salesforce OCTA OCTA engineer Software Engineering Amazon aws Google Software Engineering Engineeri OCTA Lakhan Octopus developer Cooper netease sumo logic director
Amazon EC2 with Dave Brown

Software Engineering Daily

33:24 min | 1 year ago

Amazon EC2 with Dave Brown

"Amazon EC two or elastic compute cloud is a virtual is server product that provides the user with scalable compute infrastructure. EC Easy to was created in two thousand six as one of the first three AWS services along with S. Three and simple queuing service. Since then EC two has provided provided the core server infrastructure for many of the companies that have been built on the cloud a large-scale virtualization product like e C Two requires its engineers just to have a deep understanding of scheduling in multi tenancy in previous shows. We've touched on subjects. Such as hyper visors. The noisy neighbor problem the cold start problem and and other aspects of multi tenant infrastructure to make EC two successful. These issues must be continuously revisited and resolved at different areas of the stack. Dave Brown Brown joined the EC two team in two thousand seven and now leads the EC two compute networking and load balancing teams as vice president. Dave joins the show to discuss the the history of EC two and economical problems of virtual is server infrastructure. We are hiring a software engineer. Who can work across both mobile and and web applications? This role will include work on software. Daily Dot Com are IRS APP and our android application. We're looking for someone. One Who learns very quickly and can produce high quality code at a fast pace. We're looking to move beyond the world of just being a software podcast into more of a platform arm of information about software if you're just didn't working with US send an email to jeff at software engineering daily Dot Com. We're looking for somebody who is hungry angry and wants to learn quickly and wants to build lots of software if you are that person and you're hungry it doesn't matter what your experience level is as long as as you've built and shipped meaningful applications Simeon email jeff at software engineering daily Dot Com. If you are a SASS or software vendor looking to modernize your application distribution to gain more more enterprise adoption checkout replicated dot com replicated provides tools to deliver your Kuban. Eddie's based application to enterprise customers I as a modern on Prem private instance that means your customers will be able to install an update your application just about anywhere bare metal servers in a cloud. Vp See gove cloud in their own Kuban as cluster vs fear. This is a secure way. Ah Your customers can use your application without ever having to send data outside of their control instead of your customers sending their data to you you send your application to your customer now. This might sound difficult. And maybe you're not used to it. Because you're assassin vendor you're a software vendor but replicated promises that recent advancements from tools like Kubrick make it far easier than before four and the replicated tools can help vendors operationalize and scale this process the replicated tools are already trusted by noteworthy customers like Hashi Corp Circle. Ci Sneak and many others and as a result over forty five live of the fortune. One hundred already have an application deployed via replicated in their infrastructure. That's a strong sign of adoption option GOTO REPLICATED DOT COM for a thirty day trial of the full replicated platform. You can also listen to an interview. With with Grant Miller the CEO of replicated that we did a while ago thank you to replicated for being a sponsor of software engineering daily and you can check it out for yourself self at replicated dot Com and get a free thirty day trial. The Dave Brown. Welcome to software engineering daily grind man. Thanks for having me on your one of the original engineers working on e. easy to EC two was originally built on the Zen hyper visor. What kinds of modifications did you have to make to then to fit the specification for what you wanted out of e e C Two? Yeah so when I started building. EC Two one of the most visited was really look at modifications. Who Hyper visor? There was a lot to do. How the distributed system I can actually run virtual machines scale take requests from customers? Remember back then there was nothing like it really on the Internet that allowed you watch machines and run services and so. We kind of wanted to simplify simplify things initially and so we just took pretty much standards then and use that and that was back in August of two thousand and six as we over the next few years started to make a few modifications mortification since there are a number of things we did across the in one of them was around the scheduler scheduling. That was the logic changes we did. There was for our T.. One instance type when that cannot not much has an insight that allows you to burst most of Isis types. Today give you kept the best and you get exactly what you would you lost four without t-series you actually able to get the apple to burst and so we did a lot of changes to the schedule there. The other area we've invested massively in in Zen has been around. How do we do security updates and be able able to perform updates to the hyper visor without actually impacting customer workloads? And so for the most part when you updates in you would normally have to affect the customer workload where we get the machine and we've actually been able to to improve in it and make changes that allows us to do those updates with you call live update we'd never impacts the customer workload at all and some of those things have now actually made their way upstream into the current version of then so I do still. Obviously we still have a large fleet. NITRO ZANU hyper visor. Still have a lot of visit instances and so some of those things that made anyway upstream now as well. I mean look very closely with his community so initially not much but over the years of who've changed quite a bit how do you choose between making a change in Zen itself Elf versus maybe building. Some kind of new system or module or scheduler on top of Zen as opposed Israeli comes down to what needs to be done to actually get the job. Okay so what do you need to go and do. And what's the most efficient thing to go to make way to make the change and you know with the scheduler. There's really no other place to do that. The security updates. They used to do that ultimately what what happened. Over time as we realized that we needed to move a lot of our components to hardware because we knew that dude get much better performance there then continued to run with inside a software. Hyper VISOR IS A. And that's where we ultimately went and both let it components outside of Zan and Boulton which ultimately became the nitrous system. What was the hardest engineering engineering problem? Starting to build hardware based advisors. We kind of good. We've got into sort of relatively slowly. I mean it's it's enormous project and so when you think about taking the hype of EC two and replacing that that's a multiyear project. And they literally chain almost every single component but within the data the plane and across the East you control plan and where we started wasn't twenty twelve. The first thing we wanted to address was to get better network performance and back then we were struggling with breath Taylor agencies on network and so the network latency itself wasn't wasn't bad for what you get from a software hyper exam the tailing sees were challenged with some customers. And so what I meant but as you see jitter and you'd have sort of a baseline of latency but then you spikes every now and then of latency that you just couldn't find a way to solve it just comes down onto the fact that our hyper visor was sharing the same physical core the same server as the customer workload and so anything. That's pretty core is going to affect other components can give you jitter. And so that's what we decided back then was to offload the networking to a physical card. And so essentially we could run links on and we wrote on networking stack to run on that card. On that became part of the Knicks. The network packets would come into the nick it with the entire network in the process in that. We need to do which includes encapsulation of those packets to make sure that it can work within Z.. And we went out and software defined network there and there's a lot of stuff that has to happen before it gets to you instance. That all happened on hardware and we launched that with our seat. Three instance set reinvent in two two thousand twelve so I reinvent actually and that had massive improvement in network latency and speaking about just completed and so that was the first sign that these hardware the right the right Pasta and from twenty twelve through to the end of twenty eighteen or twenty twenty seventeenth. Sorry which is when we launched our C. Five vincit Israeli this ongoing process of afloat we started with networking eds so offload ubs storage and. Then we did envy drives and so control for for India drives with I three instance and then eventually we said you know what this just move everything coming into a place with the actual coal that the customer's going to use the processor that we make zero percent of their core if you go back to head on sin about twenty percent of their core was dedicated to easy to processing and eighty percent into the customer where we are today is. His zero percent is dedicated to to to one hundred percent of their core is given to the customer. And that's why we're able to do things like bare metal. Bare metal is essentially running an easy. For instance without a hyper visor. And so be able to run the entire heightened nitro system and then remove the hyper visor completely and so you literally have mental access to landline align hardware. And we'll just being quite a big step forward and so it was a journey. I think one of the biggest was just honesty committing to their journey. And the amount of time you think about as a business knowing you that you want to rewrite almost every single component is quite daunting. I'm very happy we did it. I think it's given us an incredible advantage and our customers just amazing performance. Obviously two instances today the engineering earing challenges. That you've seen evolve with the popularization of containers. How closely have those containerization engineering challenges mirrored? What you you saw prior to that with the maturity of virtualization so I'm not sure drew many parallels there I think you virtualization back in the day? I think everybody's just accepted that it wasn't going to be as performance as running on actual hardware and so there was a big challenge there. I don't really think that's translated into containers. I think we we struggled more to think a little differently about containers and all this you the civilised spaces well is the rate of change right if you think about where we started when we first started to virtual as you would maybe launch an instance on where some other virtualization system and you'd run that instance for some period of time when easy to came around you. Launch an instance sense and unprepared at time you would shut it down. Instances became a lot more mutable and I think in the container space and you get the solar space things are just becoming more and more immutable. He wants them. You're on them. I E seconds because you need to have a task and you shut them down as well. And so. That's where we've had a lot in terms of the hyper visor. In how much time does take us to boot instance for example if that takes too long you can't run these sort of very ephemeral workloads that you'd want to be running with containers and civilize transactions. And so we've occupies changed a lot to become a lot foster to be able to support container workloads but then also things like firecracker which is a new hyper visor is that we bought specifically for civilizing container workloads sort of built on top of Nitra Hyper visor but allows you to launch the machine in A few hundred milliseconds. That's a lot of even in support more of these workloads and I think he's going to continue to see that. Workloads are going to be more ephemeral. Where you don't have a machine running until you absolutely need to start it up? Make his shut John Again. And that's really where the container space has been going as well looking for a job is painful and if you are in software and you have the skill set needed to get a job in technology it can sometimes seem very strange that it takes so long to find a job that's a good fit for you. Veterinary is an online hiring marketplace. That connects highly qualified. Fight workers with top companies. vetera- keeps the quality of workers and companies on the platform high because veterans vets both workers and companies access is exclusive and you can apply to find a job through veterinary by going to a veteran dot com slash S. e. daily. That's that's V. E. T. T. E. R. Y. dot com slash se daily once. You're accepted to veterinary. You have access to a modern hiring process. You can set preferences for location experience level salary requirements and other parameters so that you only get job opportunities that appeal to you no more of those recruiters sending you blind messages that say they are looking for a Java Rockstar with thirty five years of experience. WHO's willing to relocate a to Antarctica? We all know that there is a better way to find a job. So check out veteran dot com slash s daily. And get a three hundred dollars. Sign up bonus if you accept a job through veteran veteran is changing the way people get hired and the way that people hire so checkout veteran dot com slash. Save daily and get a three hundred dollars. Sign up bonus if you accept a job through veteran that's V E R Y dot com slash. Save daily Louis. Thank you to veterans for being sponsor of Software Engineering daily. The cold start. Or you're basically saying that the cold start problem was not as much of a cold start when people were more focused on virtual server infrastructure structure because those vm's were longer lived than the containers. Seems absolutely I think back in the day. Beams that run for a longer period of time people tended to launch a machine and make use of it. They wouldn't thinking about this idea of washing machines and shut them down. We didn't need them anymore. I remember the first time I saw a customer. Really do that at scale. I I believe was back in two thousand and eight remembered was Thursday and there was there. Was this company called Moto. Do you remember them. Down the AMI demote any motor and they had this thing basically I think it was you point them at your facebook stream and they would generate like a video of your your music with crazy graphics ethics as you've there's this cool little gimmick and it was a Thursday afternoon because I remember this sort of doing this thing where they would launch a new instance every movie that somebody subscribed to and they went viral and and suddenly saw this massive spike of launches easy to and we. We kept it up. But I wouldn't say there was a number of behind the scenes making sure that the early easy to it wasn't I'm good a full of increase. They were spinning up. A new VM. For each video for each way which I think we later found out it was actually a bug in their code but that was the first title we actually saw this idea of. He'd literally was very little capacity to thousands and thousands of machines suddenly starting up for the first time we saw it. Obviously we've seen begins for short periods of time. And then you get the Solis and the cold start saying you want those lambda functions to start up instantaneously. And so we've been working a lot as EC two do with the Landau team the secret. We shouldn't tell anyone but service actually has a server behind the scenes and that sort of lambda is easy too so we've been working with them to solve the problem a ministry to things we had to do was give them a hyper visor. That has a very very fast start time. And that was firecracker which actually ended up open source seen as well so you can go look at Clark record out there. And it's really designed to be incredibly lightweight giving the same security boundaries that we have with the NITRO VM. which is something we've never compromised on? We'll never compromised on the boundary between customers. Never want to take any chances there but can start up a VM. And a couple hundred milliseconds hundred two hundred milliseconds of time so that that's all one part of the problem of lander. The other part of the problem was for customers using BBC with Manda. How quickly can Einstein state the networking resources? How does it take to attention John? I for example and so we've done some work with the team to get that down to a couple of milliseconds as well and so I believe landers. They've announced and they've all done to most of the fleet. So the the solution to cold spot and slammed two functions ousting foster and supporting those vary ephemeral closed the game. which is where the wills? So how does the firecracker occur based vision for service infrastructure compared to and. I don't know if you're familiar with this stuff. But the K.. Native suite of projects from coober netease community so. I don't know too much about Canada but we have been working a lot with a number of those vendors so what happened with with firecrackers. Is We both firecracker and we did use a number of those tools and k native the name of the other one. I'm looking at right now. Remember shortly but we ended up launching and it was largely reinvent last year it was is actually the the the the number one guitar project for two days and so it's pretty pretty insane. We couldn't believe that end. It just showed that the desire and the community was something in that space right and so we've been very excited to contribute their from an open source point of view and some firecrackers team which is actually based in Bucharest Romania. An amazing team up here that both firecracker they've been very engaged in the community and so a lot of those teams are working together on K.. Natives involved but then we've also got another project called recipe on 'em uh-huh which is sort of building a vm. Management system unrest. Which you've already? Let's rush at the moment and so there's a lot of engagement there and again it's a you know what these officers probably just being. Really great is sort of the whole. Competitiveness is kind of goes out the window and it's all community coming together and saying. Hey what can we actually. Will you know whether it's ourselves without providing any of the other the clock provided so any of the other chip manufacturers. And that's what's coming together and working on that project and so it's early days or that stuff from excited to see where it goes does firecracker hard dependencies on specific hardware. So I don't believe so I was here at the moment. They believe. We've just finished one running on so that's an extra six processor. And it needs to be a hyper visor to run on metal on so you can run it on a lot of customers that went on. EC Two mental today. There's customers that run it themselves on any other way that they'd like to use you know we've had people even look at us in like small devices embedded devices and things like that it is very very lightweight and save you need in sort of environment something different okay. So the workflow for using a firecracker instance is or for using firecracker ker you spin it up on an EC two instance and then you spin up your own server infrastructure on top of it or are you spending up aws lamm does on top of it. Can you help me understand what the workflow is so the way that it works with. Lambda is lambda uses firecracker. But you don't actually see director. Lambda us uh-huh interfaces talking to lambda starting up a creative function and then executing the function wetlands behind the scenes is they taking a mental easy to instance and they're installing firecracker onto that instance. And didn't they manage firecracker is if it was a VM. And so much the same way you would have done with then they creating the necessary. Three COM VM. Instances containers. Whatever you WANNA call them votes within firecracker and so they could put thousands of pipe? Machines Service instances on that physical machine and each one of those tied to a customer function. And so when the function executes attorney inside they'd Firecracker Vm. And so that's how you would manage it if you ran for yourself. You downloaded from get hub and then install it on the machine. As a as a hyper visor. Much the same way as in does interact with the cricket. Instead of the is the lady create images and then instances and use it as okay. So firecracker is a hyper visor. Yeah okay got it. What went into the design that allowed allowed you to spin up images faster and reduce the cold start problem? What were you able to strip away? Or what kind of performance areas were you able to improve Gregor. It's it's making it incredibly light weight right and in reducing the number of devices that you actually one of the charges. USB You probably know USB printer what happens a lot of these hyper vices they emulate pretty much device. That's been out there in the podcast right to look at what was happening Zan Mangku. It's just a math. A Lotta time in the emulation a number of other optimizations as well. I don't know the details on all of them but thirty stripping that away and getting it to boot and then obviously making sure that the image building can also be started up very very quickly and so thinking about the operating system. And what were you bring into memory there. The other side of it was just making sure that the network performs a lot faster and at the state of BBC's being pushed out a loss to and that got us down into the sub second time range will coleslaw times on them so that seems like a really good approach approach to the cold start. There's also an approach. I've heard some people talk about where you pre load or pre warm a bunch of containers with like no Oh J. S. or with python on it so as soon as the workload comes in that requires python on a container. Then you can just schedule that workload onto the container that's pre filled with python and boom. You know you run it really quickly. Do you think that's also a viable approach to reducing the cold. Start suddenly approach. We've we've looked at and we've we've definitely don't pre warming in other parts of Easy to in. Aws Right so they're number services that cooper services so there's an use them It depends a lot on a couple things I think about their one is the cost and so does your premium pool need to be all right. Because you're essentially keeping capacity around that you might use at some point in the future you want to have enough capacity because when you don't have capacity in Fremont Pool you end up with the slowest time with these in that model but if you have too much Cabestan the people in Pool. You're spending money that you shouldn't be spending. That's one of the things. The other one is from a security point of view. Where's the security boundary when you when you have a pretty warm pool scooby? Nothing that premium pool that you wouldn't want to give any custom and so when you spend that machine blake is ready for that customer or is it a pre one pool and then allocated to a customer. You also think about whether you account boundary so you can't move a machine between accounts since a you know if you do have a service where each uses different. Aws accounts does it work in that and so there are a number number of things that many services are views pre one of the teams that's in my organization as they let go balancing team as well and they had US pre warming we've used premium elastic. Load balancing some of our other tosses where it runs. EC Two and when we need to have a new machine on your note for the advanced be able to pull it from our preamble and then you really avoid the boot time to able to get another machine into service look cluster. So there's definitely places we've used that very effective in shock. Customs duty as well as businesses become more integrated with their software than ever before it has become possible to understand the business business more clearly. Through monitoring logging and advanced data visibility. Sumo logic is a continuous intelligence platform that it builds tools for operations security and cloud native infrastructure the company has studied thousands of businesses to get an understanding of modern continuous. I intelligence and then compile that information into the continuous intelligence report which is available at software engineering daily Dot Com slash sumo logic. These sumo logic continuous intelligence report contains statistics about the modern world of infrastructure. Here are some statistics six. I found particularly useful. Sixty four percent of the businesses in the survey were entirely on Amazon web services which was vastly more than any other cloud provider or multi cloud or on prem deployment that's a lot of infrastructure on aws another factoid. I found was that a typical enterprise uses fifteen. Aws Services and one in three enterprises uses aws the US Lambda appears. Service is catching on. There are lots of other fascinating. Statistics in the continuous intelligence report including information on database adoption Cooper Netease and web server popularity go to software engineering daily dot com slash sumo logic and download the continuous continuous intelligence report today. Thank you to sumo logic for being sponsor of software engineering daily. It inside of Amazon you some extent get a preview of what the rest of the world is going to be seeing in a decade. So you see this in internal marketplace things that people are building as well as aws services that are in Beta or services that are an Alpha or services that are in pre Alpha Alpha. And I know you can't reveal too much secret sauce. Can you give me some broad predictions or areas or things that I can keep my eye on on that in ten years are going to seem as inevitable as service or edge computing. Ten ten years is a long time five years. Whatever your prediction in horizon is sure three months? Okay are we can look at it. Well one of the things that think about. It's just how how much of things changed in the last ten years right and easy to start at fourteen years ago and when I joined these team we had no idea what this thing would be sure and where where it is today with clouds today. I think things chance to let things change an enormous amount in ten years. I think some of the interesting things we already spoke about the very workloads things. You're going to be spending up machines and using the shorter periods periods of time. Civilised solvency would be a mass of movement in that direction. I think some of the other things we were excited about is We don't know at reinvent this year so last do we put out there. That was what we called out on the eight one instance and was really just to test in the market and putting something out there and saying. Hey here's there's an arm processor interested to see what you do with it arms being very big in the mobile space but hasn't really had a big play in the space and so we were very very happy with what happened with on because it provides provides you with the initial one provided you with forty five percent cheaper price performance for your workloads and so a workload that ran on excessive process. If you could put it to arm you could theoretically he say forty five percent which is a massive saving. Would we announced this. We could reinvent is the gravitons to process. And we've been working on this for a number of years is now ready and we're very excited about the performance where seeing and so. It's very first. Six generation instance type and giving us significantly better performance than what as you can get on current exit six processes. And so. It's just a leap forward in processor technology and so if we think about where we're going to be going over the next couple of years it's really up to the community now up to talk customers and engineers and developers software engineers to think about how. How are they going to be? Are they going to be put into on his arm. Really going to become you know a very viable server go rhino. I think we've we've put one out there now that is bringing a leading in many ways. So it'll be interesting to see what engineers do that. I think more and more stuff. He's going to continue to move to the cloud. I think one of the big challenges. We're we're we're having now that I think I'm thinking Lord over the next couple of years and be very interesting to see is latency seeing networking and what's really happened is if you think about how far we've come over the last ten years. Everything basically improved to levels. That ever moved a lot of blocker for our workloads so you are now going to point where they very seldom the blocker for any sort of workload anymore. Moore's law died about four or five years ago and so we really at a amazing place. Seaport viewpoint. Few memories made it credible progress envy drives and is just made an outstanding performance. And the one thing that I think is limiting somewhere closer when I'm starting to see is this needful needful lower latency. The world is mobile. A lot of the stuff. We're consuming from funds of LT networks and think about autonomous vehicles. You think about Iot going. Do you think about Robotics and what the data centers machine learning vision. You know you've got things happening at very high speed. We need to make decisions and do infringement. Latency is becoming a problem and what it comes down to is the speed of light and we're still working on solving that problem. Can we make go any faster job at. I think anybody can. And so. That's what you've seen us do things it's like well. What do we have to do to solve this problem? Latency we've got to move workload closer to us and closely with our customers uses and so we. We spoke about a number of projects this this week. It reinvent his well one of them being outpost which allows us to bring our hardware into your data into local zones which is idea of being able to put an aws availabilities close to the base. I'm excited to be I when they're in Los Angeles which is actually be supporting the movie industry and so they doing. There is forming forming movies. Komo TV shows on a daily basis those the content which is much higher resolution than four K. that's uploaded into aws and the next year designers designers and editors and animation artists working on it on easy to for instance through the night to make sure that it's ready for the next day and then the next that's crazy higher high-velocity production so we're putting an latency. That's very important. You can't do it without you. Know sub sub. Ten ten millisecond latency and so our region in Los Angeles is giving that that pardon industry experts to very low latency compute and I think one of the things can have the biggest impact is five g and five g you you know. I think it's not just a the next version of four G.. That has a number of different networking features. A two of them is obviously much much much higher through good so when you can get a couple of tens of megabytes from your current four G. Connection you could theoretically get up to gigabits one two to five gigabits from a five g connection from your mobile device which is just crazy and in the agencies is going to go down to single digits milliseconds and so that completely changes what we can do from an engineering point of view. If I'm able to have a mobile device they can get. That's what latency. Suddenly the poverty ready do games that hosted suicide and have a really sort of client side experience with that right. Autonomous Vehicles can upload data potentially even make decisions of not not whether I should break or not but decisions that are remote to their vehicle. Maybe maybe pause for routing and things like that driving factories for example. Could people get get rid of WIFI completely completely and really. Just move to five G.. And so it just becomes this ubiquitous connectivity. It's going to be everywhere and I don't think we know what low latency is gonNA give us tens of software applications nations. I don't think we knew what was going to give the iphone and android phones are going to give us now much. They were going to change of Los Angeles but I think that's the thing that we're really watching is getting latency down. And it's going to drive a whole new workloads and I think we were going to be very pleasantly surprised about how that changes our lives and what you thought it was going to be out there Dave Brown. Thanks for coming on the show. Fantastic thank you very much in this. PODCAST is brought to you by Pager. Duty you've probably heard a pager. Duty Teams Trust Pager duty to help them deliver a high quality digital experiences to their customers with pager duty teams spend less time reacting to incidents and more time building software over twelve thousand businesses rely on pager duty to identify issues and opportunities in real time and bring together the right people to fix problems problems faster and prevent those problems from happening again pager. Duty Helps Your company's digital operations run more smoothly pager duty he helps you intelligently pinpoint issues like outages as well as capitalize on opportunities empowering teams to take the right realtime action to see how companies like Ge Vodafone box and American Eagle Rely on pager duty to continuously improve their digital operations visit pager duty dot com Tom. I'm really happy to have your duty as a sponsor I I heard about them on. A podcast. Probably more than five years ago and so it's quite satisfying to have them on software engineering daily as a sponsor. I've been hearing about their product for many years. And I hope you check it out at Pager. Your Duty Dot Com

Aws Software Engineering cloud Dave Brown Brown Sumo logic Amazon EC Easy Los Angeles BBC Zan Mangku John Duty Dot Com Kuban software engineer Hashi Corp Circle US
Equity Shot: Everyone filed to go public Monday

Equity

28:48 min | 5 months ago

Equity Shot: Everyone filed to go public Monday

"Hello and welcome back to equity. Terence's venture capital Focus podcast where we unpacked the numbers behind the headlines. My name is Alex I'm joined by Danny CRICHTON Denisa. Hello. Hello. Natasha micheals say Hello Hi guys I am rushing because this is an equity shot and we have so much to get through record this Tuesday afternoon you're GonNa hear this Tuesday night. This is in response to Monday in which we were attacked y swarm of one file in. So this is the IPO themed episode guys. It was nuts. It was felt like they were just draw every time. You've got one figured out another one would land on US I do not recall a day in my life ever choir like that as a financial reporter Dante have you ever seen a swarm like that? Not In our industry. They're GONNA have seen any liquidity in general and our industries took. The the Unicorn of of IPO's has now turned into a fleet of horses. I thought it was kind of amazing that it was happening during Demo Day because it felt like this really like poetic moment of the pre seed and seed and series a and then always exit events and I was like I feel sequoia just like cackling in the background and it it just a moment that we had to me. It was like a college graduation next to a kindergartens first day you know it was weird weird mix talk for you and I were covering the Demo Day and then had to kind of pivot felt like to this other. World I put shocking anyways enough throat clearing their IPO filings will last Friday Mr Denny cried and broke much news about the pollen tear. One filing that hasn't dropped yet will start with that and second yesterday was unity Asana Jay, frog snowflake and sumo logic, and then also in the background was Corsair Gaming and ex. which is a Chinese company. So big Joe. A lot to get through. Danny started at the pound here just because it's important to get that out of the way we all expected the S. one on Friday. You've wrote a bunch of news about their IPO, filing than no IPO filing dropped are you to blame Sir for the? Public Pound here filing. No I don't think so we still have a draft copy of the Swan. What's what's nuts is it still has income though I mean we we you know we see this on Friday. It's now Tuesday night as of this second, that were recording pallares and filed, and perhaps part of the reason is is there's just not a lot of great news in the dock and maybe they're just sort of testing the waters. So blessed of the Numbers. So two thousand, Eighteen, revenues five, hundred, ninety, five, million. Okay. That seems okay. Remember this is a seventeen year old companies founded two thousand three to six, hundred, billion in top line revenue not bad. The the problem starts with the fact that the loss for the same period of time was roughly the same, which was a loss of five, hundred, eighty, million dollars, and so for every dollar in revenue the pound tear made last year it lost two bucks. The other direction spent two bucks, right spend two bucks, and then between nineteen two things get better was the perk up like a like a coffee or was it more like a like a DECAF coke and when flat it was more like a juicer Oh Machine Ju, Ju Sarah I believe that you Sarah Dairy go. The positive side is again, almost two decade company grabbed new growth twenty, five percent up to seven under forty, two, million they they managed to keep the loss in line so that the loss was only five hundred and eighty million boxes the the net loss percentage actually shrunk to negative seventy, eight percent and good news is that. We, we've been tracking the story for a long time about whether pound here's commercial company in two thousand eighteen was actually ahead. It actually had more revenues from the commerce side Dan from the government contracting side The downside is maybe because of crow virus or maybe because of other dynamics going on in the world, the government side of the business grew tremendously over the last year. The downside is the commerce business did not, and so government has actually retaken the lead for Pallares now representing fifty three point, five percent of all. Revenue compared to obviously the balance being commerce and what was interesting also is that it actually produced only forty percent of its revenue domestically in the United States with the other sixty percent international to I mean there's stuff to like their stuff obviously not to like chocking though it's a seventeen year old company right and keep bring this up because I think we're gonNA see snowflakes on these others, their companies that have you know in some cases a third to half of the period of of growth are much better fiscal position than pollen. Tear disappointment so far is my read of this and that's why I was Kinda joking did you ruin it because did you drop then everyone like, oh, it's not very good. Maybe we'll keep it to ourselves and I also think that with pollen tear they were they've been teasing out this one billion revenue number for a couple of news cycles and so seeing it not hit thought or not come super close to that is what I think made it a little bit more of a flutter response and reaction and then I also. Wanted to talk a little bit about one of the points that Danny made about. You know we knew that pounder had this two pronged business of both government and commercial and I think they want to distance themselves from the government narrative a bit but they landed a huge navy contract. The numbers show that it's still making up a bulk of their revenue. It was a cool moment to see like what a business wants to be framed as versus what it actually is making money from. It's like the best part of it s one. Unlike some of the companies we're GONNA talk about the the revenue concentration with this company is extreme. The company only has one, hundred, twenty, five customers and customers from the same say government department that are separate agencies count as separate accounts, right? So if it's the the example from the Draft copy that we leaked said that you know if you were in the CDC and the National Institutes of health is counted as separate agencies. Despite the fact, they're both part of the Department of Health and Human Services Butts nuts is that the two companies top three customers represent twenty, eight percent of the revenues last year and the top twenty customers represented two thirds of all revenues and so you know to me like a business model perspective i. mean this this honestly is why never went public I mean this is why it has remained a private company for so long I'm actually really curious why now I read this and I saw nothing in it other than they were able to downgrade kind. Of the expenses over the last six months the first six months of this year, they are able to massively cut back on expenses ear in in the first six months of two, thousand, twenty they massively able to cut back on on sales marketing costs on. RND. On s day, and so again, the margin improved dramatically as they were kind. Of. Going type Yo but this is not a strong business in my view. I was just GonNa ask like and I'm sure we've talked about it before. But truly what does pounder benefit from raising a ton of money when it goes public I don't I don't know what it would be used for, and this is ernest question but true question I have. The problem with. Shockingly profitable, we think we think a lot about growth companies in silicon. Valley is capital to fuel growth pounder isn't growing that much and it's still enormously profitable. I haven't seen the cash flow statistics I don't know offhand what it's operating cash burn wasn't say h one and two thousand twenty, but it can't have been fantastic you know and so that's expensive in. So public offerings are away to at least in theory raise this one more part to this I would have talked about besides the business model and this is unique to Pailin. So most of up heroes in terms of the voting and the. Shares Wall Street investors are going to buy get fairly consistent these days. There's a a ten vote share and a one vote share founders get ten vote chairs retain control, the company, etc. It's actually become very normal over the last decade in the tech industry was interesting with Palin tears. It's going to have three classes of shares. One share that has one vote per share classified as has ten votes per share, and then a special class F for founder chairs that will actually in perpetuity own, forty, nine, point nine, nine, nine, nine, nine, six, nine, seven, nine, include the nine in the earlier nine so. Six nights after the decimal very complicated. But basically, the the three founders Steven Cohen Alex Carp and Peter Thiel. In perpetuity have essentially control over the company regardless of their actual ownership stake in the company I mean even even for Silicon Valley, it's pretty played indoor. Colorado. Now 'cause they had mood but let's put a cork in that particular bottle because we talked about enough I want to get into the cool new fun. IPO's the filings that we actually got yesterday on, we should name it. Like awesome Monday, or like spectacular Monday when all the numbers came out the Tasha where are we going to start yet? Let's start with unity because I. think whether it regards POPs are not. It's going to be this huge metaphor again for twenty twenty because with gaming platforms and that brings apple that brings in antitrust brings in epic. So Alex stood the numbers of unity as one. Okay so I have to admit it's it's bigger and better than I expected is, might my I read of this company? So in two thousand, eighteen unity had revenues of three hundred and eighty, five, hundred, eighty, one, million give or take some more in there, and then in twenty, nine, thousand, nine, hundred up to five, hundred, forty, two, million rounding. A little bit, which is just absolutely crazy. That's so much money and in the first half of this year revenues were three hundred and fifty, one million. So this company is growing very quickly. It's very, very large and importantly its losses went up from two thousand, eighteen to twenty, one, thousand nine but they've kind of come down in the more recent periods. We're seeing an improvement profitability as the company scales, which is what you want to see, which is what pound tear kind of hasn't done as it's gone along with crazy about unity. Dan I don't know what happened in your brain Niessavis but they have like four hundred and fifty, three, million dollars in cash and like one, hundred, twenty, four, million debts of. Their net cash position just enormous and it's GonNa be a huge IPO. It's valid like six billion or something like that. So it's going to be just this huge cannonball of a splash all over the capital markets, which is a terrible analogy. But now it's on the show and it's I was impressed by the scale and also the the improving profitability. I think the other side with unity. turns Palette very complicated to see who's going to really make money that is unquestionable in unity's cates. So Sequoia, which actually did the series eight back all the way back in two thousand, nine eleven years ago owns twenty four point one percent of the of the company it was last valued in bid twenty, nine, hundred, six, billion dollars have you just do a little bit of back of the envelope math assuming the public markets are pretty favorable to the company I mean sequoias looking at a billion dollar return and what's most interesting about this is actually Sa- quiz actually bolstered their ownership stake in the company. A lot over the years. So they only own roughly eleven percent of their twenty, four percent in that early fund, and they actually have user growth funds three separate growth fund vehicles over the last decade to continue to buy up shares all the way through company's existence, and so they along with Silverlake, which owns eighteen percent really are gonna come out his huge winners with unity. Well I mean that's presuming the stock market reaction but given the crap that has had amazing. I days recently and certain offer insightful stay unnamed union you should be doing fine question then becomes before move on valuation Bingo does everyone's favorite pastime Tasha. Give us a number. What's Day one I'm going last time because last time I got screwed by going I? Eleven billion end of day one Danny. I. Think would be something similar play twelve billion. Tasha. I'll do nine in. The conservative one. Okay. It's important to note I would wrap up here. We obviously think of unity as a gay mentioned that's where it started actually started as a mobile. Game. Engine to help developers build on IRS android as well as other systems since expanded to and other devices. But the real magic to unity long-term is that it intends to expand engine to a lot of other different industries. So it's it's making headway into automotive simulations movies even movies have actually used it. It's competitor unreal. Engine is actually using the men, and so I think one thing to pay attention to is even though mobile gaming might be a little bit more of a mature market than it was a decade ago unity itself is trying to expand its town overtime I. I would throw in there talking about you don't see companies like Twi- Leo. When you're out there in the world unity is a lot like they'd be will walk down the street and talk about unity but the often play games on their phones on their desktops, whatever that are built by. It's definitely behind the scenes powering the world kind of company and those tend to be experts very profitable. My take on the company is that you know if you weren't taking gaming seriously before this is a really cool case in how big it is. There was a satistics on one of the stories that said developers start like one, hundred, fifty, thousand new projects each day in one hundred, ninety countries over unity, and that was just like crazy to see. and. So we think only thing about look where we think about to quit success. I see it really trickling down to two private startups down the road, which is the question I? Keep thinking about is like how do these like a million s ones that we saw on Monday impact the wiessee companies basically, we'll see radically liquidity trickle down to smaller firms. In time as the funds are recycled, we'll talk about gaming one of those popular game stores esteem when you free steam. You create no flakes and moving the log. Did you like that? I hope everyone enjoyed that was a lotta fun. Danny stole my segway segway planned out but his was much better supplied. Snowflake we're, GONNA talk about we're GONNA go from gaming. What what are the patterns see here? Is that basically everything other than, of course, they're gaming thing we're GonNa talk about today wasn't enterprise company right? So we were talking about six IPO's all in the enterprise gaming it's on and you know it doesn't target consumers. Unity targets game developers who is Enterprise B. TO B. C. snowflakes, snowflakes on the data warehouse space they help you go from the data lake into a process data format. It's only a couple of years old I to say it's like twenty thirteen to about seven years into the business snowflake is growing massively. So if you're looking at the fiscal year ending January thirty, first twenty, twenty snowflake. Had revenues of two hundred and sixty, five million, and that's almost two point seven, five x from the year before where the revenues ninety six, million at the same time the net losses sword. So went from one, hundred and seventy, eight, million, twenty, nine, thousand, nine to three, hundred, and forty, eight million. So that's a little bit of a Downer but. What's crazy is you look at the last quarter, the quarter ending July thirty, first twenty twenty. They actually managed to cut back a little bit on their net loss to seventy, seven, million bucks and so that net losses fifty eight percent of revenue as a comp compared to healthcare I think was that one hundred and fifty five percents ninety, seven I think. Many seven in the most recent. Yeah. Down from one, hundred fifty or so from the year before and that net loss person snowflake fifty eight percent is actually down from one hundred and thirty one percent in the previous quarter this year. So snowflake is obviously massively controlling expenses revenues are going up I mean who who's actually seen a three x as SAS companies going onto the public markets I mean we've never see this right. So this this is just hugely anticipated but I think a lot of investors question is, is it sast and I got into a debate the some VC folks on twitter about this. Because my life is awesome and super exciting and The general point was that snowflake is more pass or platform as a service versus SAS, and that sounds like a really small nuance of a difference. But what it boils down to his what kind of gross margin should we expect and if I recall snowflake has had improving gross margins overtime all the math kind of sit the last couple of fiscal years and the most recent quarters steady improvement. Exactly what he wants from a company as matures superagency solid just really good. But definitely lower than SAS in the mid sixties call not seventy, five, eighty, two percent. So small quibble on the economics but certainly, Danny's totally right company that's controlling expensive really really well scaling super fast and is broad net loss down, and now because of that is a path to profitability and it's time to go. Public and on top of that. So obviously, a huge numbers for for Snowflake as a company for its investors. Again huge win for Sutter Hill which incubated the company and put five million dollars in series back in two thousand twelve they are looking at least at the previous valuation at a roughly two and a half billion dollar stake in the company. The director from Sitter Hill is on the board actually owns ten percent, of Center hill steak directly. So He's very happy man on top of that. There are a bunch of major investors involved A. Sequeira Ed. So szekler growth came in just in the last two years. So they're not looking at a huge return about four x given his previous evaluation also read point altimeter and iconic all coming in with with huge returns red point around twelve to thirteen x given the last valuation and the other guys around eight to nine X. so I mean what an amazing story and I think this is going to be among this little cohort of call it the the Monday Edgar Explosion of IPO's this is going to be the real winner from this batch. And the last shares from Ron Miller story that kind of got reactions on people who've been tracking snowflake and one analyst made the point. snowflake relies a lot on the infrastructure of cloud giants like aws Microsoft and Google but if it goes public and does well, it could one day become a competitor to those two already kind of is, and so it's like it's growing. It's not done growing and that's going to be an exciting public journey to track. But let's move onto the next one in our carousel of Swans Assan. Alex. walk us through the numbers. All right. So Asana is a company that I've been pretty excited about the let's let's start with some clarifications. We jokingly call it an IP. It's not what it is is a direct listing so it's going to be a different sort of capital event it will go public but it will not raise money along the way the company took on I two hundred or two, hundred and ten million in debt a little bit ago, which is kind of how it's approaching this entire public market thing. So instead of pricing its shares raising money and then going public, which is traditional standard. It borrowed a bunch of money that would convert later onto shares, but anyways Asana is a I. Don't know guys team productivity and to do management application service for groups of people inside of company kind of like Monday dot Com and two other stuff famous in the world for having Dustin Moskovitz and Justin. Rosenstein early facebook irs with a lot of money power in this think there they were back in the game after. kind of not needing to do anything ever again, Colorado attention and four a company that is going to direct list I was expecting and I say affection because I've talked to somebody. So much of the years slightly better numbers than the profitability side, but there's some good growth here. So here is the math. So from the end of its fiscal nineteen ended in January of that year no I'm not GonNa, do it that way? What's a good way is fiscal years guys but that doesn't a thousand words long struggle to. Easy Way to do it. What is the fiscal year or January thirty one so Should I call that fiscal? Fiscal Twenty tried to say all the words and be boring. Okay You're going to say fiscal year twenty, twenty s mostly the calendar year twenty nineteen. That's gotta be that's going to be worse I think. It's it's off by a month. It's Christmas people. That's why they do it this way this is the the commerce is the wall calendar you and I both know that but I'm trying to figure what the best way to convince. All right I'll just. Masses. All right. So a SAUNAS numbers well, they have a fiscal calendar, which is a very annoying thing to talk about their year doesn't end December thirty first and ends January thirty first. So that means that the year ending January thirty one, twenty nineteen is mostly twenty eighteen bose actually ended in two, thousand, nine, hundred there's reasons for this. We're not GONNA get into what matters is in that year they add. Seventy six point eight million dollars in revenue, and they roughly doubled in the next twelve months to one hundred and forty two point six India the start at this year, the problem is they're losses over the same two time prayer timeframes. These two fiscal years I've actually doubled from fifty, point, nine, million, two, hundred, and eighteen point, six million, which is a much higher percentage of Revenue Demi Tennessee from SAS. Companies of a similar size in a similar growth rate niching down to the most recent quarter, which was the quarter ending April thirtieth twenty twenty Asana grew revenue from twenty eight million in the year ago period to forty, seven, point seven million, which is seventy percent growth, which is pretty good. It's definitely above average for companies company going public puts the firm at a run rate of round two, hundred million. Dollars a year but they also lost thirty five point eight million in the same three-month period more than twice what a loss in the year ago period. So good growth material scale big run rate surprisingly large losses with a tiny caveat, which is they now have some debt servicing expenses in there and their share based compensation costs also rose so that net laws numbers a little bit inflated, certainly a knockout. Danny Impressions Natasha Impressions. Would we think about these numbers through them? It was more impressive when you consider what their customer breakdown looks like. The factoid being they have three point, two million free collins, and then it makes all its money off of seventy five, thousand customers. That's pretty wild and so that to me maybe more optimistic about it. Once I colored and the fact that most of their accounts upset aren't giving them money. anyways. It was a free medium trap I mean it's hard to get people who are currently not paying to pay slack has been famously good at this I think but not every company has the same conversion rate but again, they still grew seventy percent year-over-year to a two hundred million dollars run rate so I don't WanNa diminish that cut them some slack. De. They don't have one hundred and twenty-five customers, right? I mean. Top talent here who sells to the the government three I it's three letter agencies three customers I think I think might take away from this is that it seems to me like a much more long term sustainable company I mean partly, I think Moskovitz and and risen steen's. Previous endeavors which obviously very profitable but Moskovitz actually comes out into the IPO thirty five percent of the company Rosenstein and sixteen percent, and I, think the largest I'm doing this memory I think the largest VCR benchmark owning above ten percent of the shares I think they did one of the very early rounds to not huge win given the previous valuation for benchmark although you know obviously nice multiple and invested capital but to me the the real magic here is it didn't raise all that much right. It raised about two hundred, million of equity race twitter million of debt. Probably. Some sort of SAS securitisation or some sort of a R line which we've been talking about a lot and so you know, yes, it's not the company is not growing the fastest. The Laws Aren't the best. But what's amazing is think I've ever seen a company recently in which the founders just outright own majority of the shares and I I. Don't know if they've invested personally we don't have all those pieces of information in front of us, but it is extremely rare for a company that's ten twelve years old to have founders who actually a majority outright of the shares. When they go through in there I read the headlines I figured this is true but I think dust may have been behind some of the capital borrowed, which would double down again and again into the business. Here's why Asana matters all. That was numbers kind of inside baseball for US financial. If you're listening to this, you want to know why you need to give. Here's why because they'd raise debt and then proceeded direct listing. This is a possible template for future all not feature public offerings and at this goes well, for Asana, we may see a lot of other companies follow suit because people are very peeved at the IPO pricing process that most companies go through today dry-cleans get around that. How do you raise capital? Well, maybe just raise convertible debt beforehand. So it's a test and that's the cool. Thing here I my first ever interviewed disrupt was just Rosenstein like there's a special place in my heart for this company, but that's the that's the real thing that from the market's perspective matters, I'll add to that Monday dot com, which is probably its closest competitor is also kind of gearing up for IPO's even though like who is in at this point so it'll be interesting to track Monday cotton to over a hundred and thirty million. Earlier this year when they came on the show actually. So yeah. I was GONNA. Say I think the when you look at Snowflake snowflake is definitely my winner of the set of like is GonNa be the most on the ideas that sauna is definitely my most likely to be acquired before the IPO actually hits or I should say direct listing actually hits award because I think the price point these numbers I can imagine a couple of different buyers in the text space that might WanNa buy out to compete with Microsoft Planner Arts Jinxing, my direct listing gambit here. 'cause I'm really excited about this. We have a couple of more we're going to be brief. I, I'm Jay Frog I think cool and you don't think I want to know your beef with Jay Frog. It sounds like the service you use when you go to the library and you need to look up the US, the dewey decimal system and find a book it's wrong with that. Decimal System I. Love. Rap. Name. That's somehow dumber. Jay Frog is cool as. By and It's growing quickly, which matters me from sixty three, point five, million revenue in two, thousand, eighteen to. One hundred four point seven in twenty, nineteen, a grew about forty six percents most recent quarter and critically, it actually made money. It's profitable on a gap basis, which is something. We almost never say on this podcast in the startup world techcrunch dot com like it's got eighty percent plus gross margins. It's gorgeous to great business. Now to be clear, I knew nothing about what it actually does, which is like a developer toolkit service platform coat livery thing but I like it. A devops life cycle product end the real magic though is It's a huge win for the Israeli. It was just a the big VC, investors, Jimmy Israel ventures, and does scale and Safire both did later stage rounds along with a couple of others. So to me it's like look profitability. This is what's the difference between Silicon Valley and the rest of the world is in Silicon Valley people just find more ways to lose money faster. You go overseas a little bit and suddenly boom you see a profitable company eighty percent gross margins you can do it. That's my kind of magic. Okay. This is the most likely to be profitable company. That's my bet so you. Guys snowflake which loses the most money is the best and then you like on the next thing you like J. Fraga. I I struggled the square, those those two things Danny. Going to answer the question I'm just GonNa pointed out. All right. All right. Let's let's wrap with sumo logic accommodate. That's not APP SUMO, which is like I'm writing down on my post when I was covering because I'm idiot sumo logic filed I think it was early Monday morning. So when I was getting my days I was like Oh my gosh, we've an s one to start the week how exciting one new IPO for the week and then of course, delayed deluge happened really briefly fifty percent revenue growth in its last year the most recent quarter. About forty five percent growth a little bit above average for public companies. But younger is you kind of expect that blended gross margins of about sixty nine percent which was down a little bit which is not super amazing. Race a little over three hundred, million privates with a lot of money right on its success I. It's GonNa be a smaller IPO in our kind of radar. But certainly want to keep an eye on if you'd like we'll be covering the pricing and the debut. But that that ladies and gentlemen I believe is the end of the IPO. Down I think all of them how do we feel? Right now. I feel ten ten because we didn't talk about stocks and I am so tired of. You said the word they were multiples backs that were announced today and yesterday that we didn't even talk about including one that just came on our ago when we're recording here one of the senior leaders of blackberry is getting under spat game. Unfortunately, it looks like literally just filed like ago. Recording recording you know actual ducks. So I guess we can throw away early coffees now but a lot has coming in I. Imagine I mean just given the the timing I think we're going to have a lot more I. Mean I think you know whether it's from the side Dreg listing side or the IPO side as we talked about on the actual equity episode last Friday you know everyone's looking for liquidity everyone's going public. It's amazing. It's a revolution you know the the magic the last couple of years has been don't go part way out public private as long as possible. Now everyone's going the other direction it just. If, Lemons. The question is in a year when there's a pandemic, a local if global recession and an election is this year to finally find the IPO button, we shall see ladies and gentlemen but expect Llamas. Ones. Danny said expect more of US on Friday morning. No Big Liar. But more of US tomorrow afternoon when we talk about why combinator startups and then Friday morning and then Monday Equity Never Sleeps Can i?

Danny CRICHTON United States sumo logic twenty twenty Pallares Sequoia Dan Rosenstein snowflake Silicon Valley Colorado Terence Fiscal Twenty Alex I throat clearing Natasha micheals Jay Frog Department of Health and Human
Practical AI with Chris Benson

Software Engineering Daily

51:11 min | 1 year ago

Practical AI with Chris Benson

"Machine learning algorithms have existed for decades but in the last ten years several advancements in software and hardware have caused dramatic growth growth in the viability of applications based on machine learning smartphones generate large quantities of data about humans. Move through the world. Software is a service companies generate data. About how these humans interact with businesses cheap cloud infrastructure allows for the storage and compute of these high volumes of data machine. Learning frameworks such as Apache spark. tensor flow and Pie torch allow developers to easily train statistical models these models are deployed back to the smartphones and they're also used by the software as a service companies and this improves the the ability for humans to move through the world and the utility that humans gained from their business transactions and as the humans interact more with their computers. It generates more data which is used to create better models and higher consumer utility the combination of smartphones cloud computing machine learning algorithms and distributed computing. Frameworks is often referred to as artificial intelligence Kris Benson is the host the podcast practical ai and he joins the show to talk about the modern applications of artificial intelligence and the stories that he is covering on practical L. A. I.. On his podcast. Chris Talks about everything within the umbrella of AI. From high level stories too low level implementation details practical fillet. I is part of the change log network of podcasts. So you know that it's going to be a high quality well produced show and I encourage you you to check it out whether you have any interest in artificial intelligence or not. It's well done also. I want to mention that we are hiring. We're hiring both a content content writer and an operations lead. Both of these are part time positions the content writer position is for somebody who likes to write about software engineering and technical topics. The operations lead is for somebody who likes to learn about how businesses operate and is great at attention to detail. Both of these are part time positions working closely with myself and Erica. And if you're interested in working working with US send me an email jeff at software engineering daily Dot Com. Don't be shy. I'd love to hear from you and let's get on with today show Today's show is sponsored by data. Dog a scalable fool stack monitoring platform data dog synthetic API tests help you. Detect and debugging user facing issues in critical end points in applications Sion's build and deploy self-maintaining browser tests to simulate user journeys from global locations if a test fails get more context by inspecting a waterfall visualization or pivoting to related sources of data for troubleshooting plus data dogs browser tests which automatically update to reflect changes in your Ui. So you can spend less time fixing tests and more time building features you can proactively actively monitor user experiences today with a free fourteen day trial of data dog and you will get a free t shirt go to software engineering during daily Dot Com slash data dog to get that free t shirt and try out. Data Dogs Monitoring Solutions Today. The Kris Benson welcome software engineering daily. Thank you very much. You host the practical. Ai Hey I pod cast. I'd like to start by focusing on that term A. I. so there was a term big data and now that term has is basically been supplanted by a I and I think that there is in some ways. No change between those two trends but I think if you look at a technological level there have been actual technological shifts that have come to the quote unquote big data area that are significant enough to demand a change in terminology. And you know why not choose the term a I when you compare those two terms. Those two eras of Technology. What's the tangible difference between them? Well I guess yes I can offer my opinion on the reason I start that way. Is that if you if you put a room full of people in the AI space at least together and ask them what is A. You're going to get a whole bunch of different answers and almost comically. I was actually at a at a facebook broadcast that adobe was doing about a year and a half ago. There were twelve of us that were being broadcast in panel and they asked us all that and we all had different answers. So that's we could agree on all sorts of stuff but a bunch of AI. Experts could not agree they AIA. So I kinda lead off with that recognizing that many listeners may not agree with what I'm about to say in my personal viewpoint I have a fairly narrow new definition of a and that helps me differentiate what it is from big data. I think they I as almost a marketing. Term from that evolves over time as technologies change over the years and our expectations of those technologies changed and I think the term changes so the way I personally would define AI. Is I really think of it right now. In two thousand nineteen going into in two thousand twenty as synonymous with deep learning and so which is implemented as deep neural networks. And so the way I would relate that over to big data is you can use big data in all all sorts of different applications different types of analysis. And the most end. If you're thinking of how how am I going to operate on big data. You know you have an entire world of data science that that you could apply to that big data and then as a subset of that data science. There's machine learning where you have algorithms. That are self learning thing that can operate and they kind of figure out what you have to do. And then as as a specialty subset of that is deep learning where you're using specifically neural networks and Dan to my way of thinking deporting equals ai today. And if you ask me a little down the line I may give you a completely different answer. There are some specific technologies. Ogies that we could look at as hallmarks of this different a era we could look. Ah frameworks like tents or flow. We could just look at the growing volume of data from mobile devices. We could look at the growing accessibility. The growing adoption of the cloud the drop in prices of the cloud described the technology enablers. That are making a a more accessible or practical. Sure so I think we've really hit an inflection. In as technology has evolved were several. We'll necessary ingredients have come together to make it possible. And you've already called out one of those in that is cloud computing. Doesn't necessarily have to be at the cloud but it's essentially high-performance computing capabilities. Either in the cloud or that you have on the edge you know as a as a dedicated set of infrastructure. Because there's a lot of money at these days there were early early benefits being seen back in the early two thousands on you know as as we've come out the AI winter and so there's a lot of work on algorithms so that people can apply new techniques and then finally you know you you mentioned a few minutes ago is big data you have have tons and tons and tons of data to operate on. And that's really if you're finding a as I do is deep learning you. We see that. Today's deep rail. Networks require it really massive amounts of data at a level that we just up until recently haven't really thought about in a practical manner and so you know kind of going back to the National Security Agency has famously says collected all and you know from some years back and that's really what led to this wave of innovation that we're seeing in the space where you're able to take advantage of of a big data with these incredible algorithms that people are coming up with for various use cases and we finally have the compute widely available to do that in as a matter of fact that both the Algorithms and the tools to use those algorithms in many cases the data itself depending on what your sources are are either open source and or free and so that's really made it accessible to a lot of people in a very short amount of time when I talked to developers they generally seem more excited about these kinds of things like building their own models using tensor flow training their own models to do things there seems to be less excitement about the. API's like speech each recognition API's transcription API's image recognition. API's you don't hear as many people talking about the usage of the API's which are so much easier to use it kind of surprises me. Does that surprise you at all. Or is that consistent with in your experience possibly so I think the API is are are great tools from the various vendors who make them available because they've already. He done the work of the model in their infrastructure and so they're essentially offering a model as a service to you through that. Api and what that does is it. Relegates sit back strictly into the traditional world of software development where you simply make a cloud. API Call and you're able to to get the answers you need and and and that's great if your interest is in strictly on the business and if the problem you're trying to solve can be done with someone's existing API. I think the the thing that we're seeing that you're alluding to here among software developers and that's my background as well. I'm not a data scientist by background. I'm a software developer and so there has been a rather unusual turn of events as we've gotten into the age of deep learning. And that is that if you looked before this this recent the last five seven years the rise of this as a as a field that is becoming so popular. You're really looking at more. Traditional data science and machine learning Algorithms Algorithms that were out there and to use those even at a fairly basic level you had to have a fairly good understanding of how those algorithms worked and so that created a barrier to entry to their utilization and so it it it remains somewhat specialized in the data science world. And so you know prior to this modern modern deep learning moment that we're in a software developer would work with the data scientists in in many cases to say. Hey can you create the algorithms or can you show me how to use those and war together to get that work. Well something pretty interesting has happened. In the deep learning world and that is that there's the notion of transfer for learning in deep neural networks where because deep neural networks in their layers build upon each other so and as a as a brief if aside which I think is important to understand why this is the case a deep neural network has a bunch of layers of nodes or neurons that they have relationships with other neurons beside them and that creates abstractions on what different concept's in the real world is. They're essentially mapping various types of abstractions in as a simple way of thinking that if you think of your face is an abstraction as a concept and it's made up of concepts of is and concepts knows and concepts of Mao and such as that then you're getting these collections of concepts that create something bigger than that also might go in with other things to create something even larger a human face torso arms that kind of thanks. So the the neat thing about transfer learning that I was mentioning a moment ago is that it turns out that you can reuse much of a model by only keeping a certain percentage of the existing train layers. So you can have a company out there like you know. Maybe one of these leaders like Google or Amazon John or Microsoft that are outselling these services and they may have models that you can reuse without all the the intense training needed from scratch and so not everybody is a research scientist. Ai that has to go create a solution from scratch you can. If you can go use somebody else's model and say that works for a lot lot of what I need. I just need to build a little on top of that. Then you can gain the benefit of transfer learning and utilize it for yourself so going back to your original question about developers that's that's an ability to create new things that API's are not offering and it's not that hard or technical. A software developer can can fairly easily level up to take advantage of that and that opens an entirely new world of capability that they never had before. How has quote A.? I changed the dynamics of modern warfare. Well I think were going through a transitional period right now where you have the Department of the defense here in the United States along with its vendors in the defense industry I worked for one of those Lockheed Martin and as we look at the the fact that potential adversaries out there often people will note Russia and China as potential adversaries out there and they're saying And how do we always stay strategically safe able to provide the national security. That's the mission that has been specified for the Department of Defense. And so you know if you have a set of tools out there throughout the world and in all industries. Obviously it's also going to work in. It's also going went to have applicable in this industry and as you look at more and more technology basis for national security missions. You're going to see that if you you can handle faster and faster issues out there as the as the speed of conflict speech you know is constantly increasing. Ai Can play. Roles of being able to supplement and interact with humans are the primary drivers of those missions to give them capabilities his or speed up existing capabilities that they've never had before and so of course anybody just as any other industry anybody in the defense industry is going to be saying. How can this particular tool set help us? Better achieve the mission that we've been assigned. Let's say I make you the. US Director of AI. I policy what would your policy be on facial recognition technology. Well I think for civil the the person that was closely matches that that would be a man by the name of General Shanahan. Who is the leader of the Department of Defense Joint? Ai Center which came into existence last year and those are the kind of questions that they are trying to answer. If you're looking at the capability of machine vision or or any of the other common types of use this case is then you're trying to match up mission requirements that you have in the various tools that you have to implement those missions and saying how can these different different technologies such as convoluted neural networks improve my ability to fulfil a mission and so that is literally how that is analyzed and as a strategist strategist at Lockheed Martin. We have a number of people such as myself that are working on these types of questions and trying to find the answers for that which it's still a work in progress aggressive and for the moment I'm not get out in front of the Department of Defense because that's There's there's a lot of of joint effort in that but we have to look at these technologies and say where does it make sense both from a technical standpoint in terms of how it improves mission capability and also from which is of great concern to the the public at large and us is the ethics of AI. Obviously and so that is literally something that I am working on at Lockheed Martin. I'm one of the people that that drive that offer offer and eventually will probably be able to talk about that in public but right now the work is still being done and we're still working with dod and other federal agencies on on what we consider to be appropriate uses of those technologies. But let's say I'm not asking you as somebody who's familiar with military technology and military Harry Policy. Let's say I'm asking you from from the kind of world you wanna live in or the kind of United States you WanNa live in Jiu for example. Should I be able able to walk into a store. And have a security camera monitoring my face and allow me to pay by just walking in and then walking out. Should there be cameras everywhere so that I have a sense of safety as I walk around and I know Oh that there is a constant scanning of faces for potential miscreants or people who might cause me harm what what about on the consumer level the population level. Sure and I think that's I think that's one of the challenges that we have right now and I think that is a very hard question to answer. I'm certainly open to offering personal insight. I think for me it really comes down to disclosure. Obviously I wouldn't be in the AI. Field if I wasn't enthusiastic about the technology but it also brings about these capabilities. That definitely affects things like privacy perceptions of of the world however engaging with those around us so I think we're at a moment right now to where if you're going to make those those steps forward into into this to use artificial intelligence to fulfil those types of use cases in the consumer space. You need to make sure that your partners and your customers fully understand Dan that that is part of the engagement that they're going to have with you and so I think where where consumer organizations get into trouble is when they they kind of leap ahead of where the people that they're serving are expecting them to be when they don't give those people a chance to opt out or to give feedback saying that doesn't work for me. And I know I don't think there is a single answer that works for everybody so far with I'm actually. I'm both enthusiastic enthusiastic personally about and I have great concern working in this field about the potential for privacy to be violated but also people that whether or not they have any any interest in is they want to maintain strict privacy. And they're trying to figure out how to use even the tools that are out today where you have a social media tools that are able to you predict you your behavior your likes and dislikes even better than your own. Cognizant awareness allows so you know I it is. It's a fair fair thing to say that a social media site literally knows more about you than you do in the sense that they have objectively observed certain certain behaviors and that we are emotional beings and we tend to to WanNa think of ourselves in a particular way and we may not be very honest or observant about our behavior is leading. So there's it's such a complex world and one of the things I'm always telling people is these are the conversations that we need to be having literally everyone when listening to this right now needs to have think their way through what their own positions. What works for you and what doesn't work for you and are you ready as you engage consumer companies? He's to be able to stand up for your own set of beliefs and your own set of interactions at a level. That you're comfortable with and I would argue that probably most people we'll have not taken that very active stepped up and looking for a job is painful and if you are in software and you have the skill set needed to get a job in technology it can sometimes seem. I'm very strange that it takes so long to find a job. That's a good fit for you. Veterinary is an online hiring marketplace. That connects highly qualified. Workers workers with top companies. vetera- keeps the quality of workers and companies on the platform high because veterans vets both workers and end companies access is exclusive and you can apply to find a job through veterinary by going to victory dot com slash S. e. daily. That's V. V. E. T. T. E. R. Y. dot com slash. Save daily once. You're accepted to veteran. You have access to a modern hiring process. You can set preferences for location experience level salary requirements and other parameters so that you only get job opportunities that appeal to you no more of those recruiters sending you blind messages that say they are looking for a Java Rockstar with thirty five years of experience. WHO's willing to relocate to the Anti Arctic? We all know that there is a better way to find a job. So check out veteran dot com slash s daily. And get a three hundred dollar. Sign up bonus if you accept job through veteran veteran is changing the way people get hired and the way that people hire so checkout veteran dot com slash s daily and get three hundred dollars. Sign up bonus if you accept a job through battery that's V. E. T. T. E. R. Y. dot com slash. Save daily thank you to veterans for being sponsor of Software Engineering daily many any of these conversations that I have seemed to devolve into a very emotional conversation so for example the conversation around should you have a voice interface in your home the first response that I almost always get from somebody who I'm talking talking to about a voice interface. And whether or not they should have one is. I don't want that thing. It's listening to me all the time. I'm not interested. The smartphone does the same thing and and if you bring that up to these people at least in the examples that I've had they basically get a cognitive dissonance ascendants and they cancelled the conversation. They're like at. Yeah just I don't really want to talk about this. And if they were zoom in a little bit further like first of all. There's no the difference second of all if there's any difference. It's that the hardware is more secure and you're not installing random third party apps onto your. You're smart home device. Furthermore it actually provides a ton of utility so as somebody who who spends so much of their time time talking to people about quote unquote. Ai How do you navigate. These issues have become so sensitive sensitive to people. So I I really I do the same thing that you just said where I try to the point. L. The deal. If you will that people I'll have already signed up for and a lot of cases. not very thoughtfully not very cognizant of the implications. You know I personally carry around an iphone. That's it's my personal iphone. I carry around an IPAD. My personal I also have a work iphone that I'm assigned. We have our our laptops in in my house. I have Google devices. I Have Amazon devices and all of these have various listening capabilities. And they're all they can all be turned turned on remotely and it's it's something that I personally am very very aware of. I know that in finding my own choice of where to accept. Accept the trade offs between privacy and utility with these devices. You know you can in some cases go in to these settings and decide you know. I know you're not you're not allowed to to upload data you know that might be default setting on that kind of stuff and so in our family. Were a little bit thoughtful about the fact. That there are many microphones around and just technically speaking aside from the legal and regulatory constraints that a lot is possible. And and then when you not only think about the fact that you have all these microphones and all these input devices cameras around but that data can be used in many different ways some of them you might think of very wonderfully and others quite various. You just have to. If you're going to be thoughtful about it you have to accommodate that realization in your life and it takes a few brain cycles to go through that which is probably why most people turn away from it but that is exactly why if you work in you know the defense fence industry or if you were earlier in the military dod and you have any sort of classified access than you literally have to leave all of your electronics out and when you have these kinds of meetings and stuff and that is why because if you do have various actor in there whether this be in the military space whether being consumer space wherever that technical capability exists so you should be aware of it. In when I think about the government's use of large data or unquote a I. It seems so nascent like I mean we could talk about but it is the amount of data that the government has relative to how much use their making of that data data even in just like noncontroversial ways. You know like Ken. My can I get a passport faster or something like you know. Can we build roads more efficiently. These kinds of things are not really I mean slowly but surely I'm sure we'll get them but I think in in some ways this mirrors the ways that legacy enterprise are are talking about their a strategy but largely. There's almost no difference between for many large enterprises. There's almost no difference between an AI strategy and just just like a data platform strategy just getting your data in order cleaning your data making very simple regressions based off of that data. Are we even at the stage where enterprises can build useful models or win. Can they build US use for models like our enterprise just kind of figuring out how to get their data in order or are they actually building models that are helping them and transformational formation ways. I think you're getting a depends on the motivation. Obviously of the people in each organization on how they're approaching it you know for profit businesses Are Out to make a profit and I think it would be naive to think that they're going to use data in a way that is not to further that goal. That's why they're collecting it. That's what they're there for are in that that would include obviously all the cloud companies. I think the play swear. Those kind of organizations get into trouble is when the people serving don't have a clear understanding of how that data's going to be used in have not given their consent for data to be used in that way. I think that is different from nonprofits and government agencies because they are there to serve the public and serve or served the constituencies that they have off and I think that presumably if they have you know if a nonprofit is in fact able to secure their their five oh one C. Three designation from from the IRS. That means because they are they have been deemed a public charity in there for the public. Good and hopefully they're operating in that manner but you know just be aware where those interactions are in specifically. Don't give consent to use data unless you're fully cognisant of how those organizations are going to be because they can learn about you you and your your various behaviors that may influence their for profit operations. And if you're okay with that then you know. Go ahead and do that. And I've had to consider that myself off recognizing that by opting into certain social networks in search companies that I know they're going to use that data and they're going to mix it with other people and try to get not only views of me as a person but also larger societal views that can further their business. So you really have to. You really have to just recognize whether or not you part of their transformation process. What's the most creative application of generative adversarial networks? That you've seen seen the most creative I would say we're still trying to figure that out created from like I got a chance to write me a a movie script or something I I just made like from a technical perspective. Well I mean I don't even know that it is the technical perspective. That's the most interesting I it's you know. It is a particular approach approach to model development. You know and we and we may want to define that where you have essentially in Layman's terms. You have when model that's generating possibilities ladies and you have another model. That's kind of discriminating. You know what's real and what's not and in that feedback loop let them together as a team produce better and better content towards whatever that end goal is and so you know there have been some pretty amazing areas in terms of music creation that I've heard in terms of art. Were you probably are familiar with the fact that there was a group in somewhere in Europe. I don't remember the specifics. But they had created painting probably about two years ago now roughly and they were going to sell it at Chris. Christie's auction house. Now think that roughly they were thinking it might go for something like ten thousand American dollars. It ended up fetching somewhere in the neighborhood neighborhood of half a million at the end potentially because it was the first of its type to use. Scans in this kind of big way out of how that would affect pricing going forward but you know it was interesting moment because it was the moment where we I realized with some of these innovations that were literally in the world. Are that yes. Hey I can be creative yes a hey. I can generate things in an artistic sense that we head up to that moment reserved for human intellect and thinking that would be the last bastion of or humans would be dominating for some time to come so now that we understand that a it has a creative ability in the model itself. It creates you you know. Where do we want to go with this? There's a lot of ethical issues. You know what what is valid to create. How does that Inter operate with humans? You know you can be created for simulations so you know the sky's the limit on the technology and so I ended that myself on the practically I- podcasts. My Co host Daniel White Neck. We are eagerly watching what happens week after week and month after month and kind of calling out some of the more novel things that we see along the way. Okay but that piece of art that was not actually like a milestone in like computer creativity like computer creativity. You can say you know some auto. Cad Design or or some data visualization like a really nice line graph like. There's really not much of a difference between that and a work of art created by a Ganz. I'M GONNA disagree with you right there and I'll tell you why I don't even think it's it's what is the actual allow put itself. I think it's the fact that if you're going to open up a CAD program and work on it still requiring the creative element element you know that provides tools for us to realize this vision that you as an artist may have but at the end of the day. It's not the CAD. Software is only enabling. This thing that you have in you as an artist what Ganz have done in this concept of of creating this painting and things that have come subsequent to that with Music Dick and with photography and everything is that it's no longer arising out of ourselves as human. I I actually do think not because of the technical Output but because of the source it was truly a touchstone moment for humans more than Ai. In that this thing which we always always expected to be the last thing to go to computing. was there already and so we've had to kind of readjust go You know that that that I thought we'd hold in reserve is different. It affects humans in a fairly deep way in that way. So I would if you're looking at it just about the technical output of the art itself. I I totally see your point but it has changed. What may happen for instance in the creative world's going forward at this point if you're a marketing agency and you have client? Do you have a staff of designers going forward that are creating the novel things that you're trying to sell a fortune five hundred company in terms of branding and in terms of the various marketing campaigns. Or do you have a bunch of supercomputers that are running thousands of Gannon instances and generating ideas. Whether did they beat music or imagery or stories and dialogue. Whatever that you're GONNA turn around and sell it? I think it's actually a really big moment. As businesses become more integrated with their software than ever before it has become possible to understand the business more clearly through monitoring logging end advanced data visibility. Sumo sumo logic is a continuous intelligence platform that builds tools for operations security and cloud native infrastructure. The company has studied thousands thousands of businesses to get an understanding of modern continuous intelligence and then compiled that information into the continuous intelligence report which is available available at software engineering daily Dot Com slash sumo logic. The sumo logic continuous intelligence report contains statistics about the modern world of infrastructure. Here are some statistics. I found particularly useful. Sixty four percent of the businesses in the survey were entirely on Amazon web services which was vastly more than any other cloud provider or multi cloud or on Prem deployment that's a lot of infrastructure on aws another factoid. I found was that a typical enterprise uses fifteen. Aws Services assists and one in three enterprises uses aws. Lambda appears server. Lewis is catching on. There are lots of other fascinating. Statistics in the continuous intelligence report including information on database adoption Cooper Netease and web server popularity go to software engineering daily dot com slash sumo logic and download the continuous intelligence report today. Thank you to sumo logic for being a sponsor of software engineering daily. It just drill into this a little bit further. So the autocad example with autocad or with data visualization like an excel model. You're often drawing relationships But tween data points using interpretation and then you are saying okay. Based off of this set of data points and and the interpretive relationship that I have described to you computer please extrapolate beyond this model that I have described gripe to you and then the computer has been able to do that for for a long time. I mean our interpretation extrapolation has gotten better the only difference with the Ganz ends stuff is. You're saying hey here are some works of art and here are some parameters that I'm going to define that I want you to look at those works of art in terms of and here's the interpolated relationship. I want you to think about between those works of art based on those parameters that I described to you now please extrapolate from that relationship and it's really just the same thing thing I mean. I think it's great. I think it's wonderful but I think it's just iterative. I do agree. It is iterative in one sense but I think the thing that distinguishes the two is that in the case of the first where you're using software as is a tool as a human. You're still talking about that. The the kernel of that creative vision came from the person. It came from the human as you you look at model generated are in. It doesn't have to be art by the way it can be all sorts of different tasks in the world and south. Then you're looking At a capability coming about where you didn't have to to start with an artistic talent in the inhuman from the origin of that. And even as we're looking as as you start multiplying this new capability in the computer world out across many industries. That's why we're arriving at this moment. Now where we have to decide. What is this relationship that we're going to have with computing? Since since in many cases says that that nugget of creativity that nugget of insight did not start with human it started with a model somewhere. And you know if you run run that out on any given industry and you say well we started. You know. We're going to see down the it may be that. Instead of building on the shoulders of human giants you may be building on on the shoulders of model giants which is why we as humans a need to start rethinking how you interact with going forward. What is the right right relationship because it will change the business models if you are if you're a business person out there in one of these these companies that we've that we were just alluding to a few minutes ago with a profit motive then you have an option between do I use humans for this going forward or do I use models for this going forward or some mix of the two and if it's a mix what is that mix? What does that mix look like? And what are the moral ethical considerations that we need to apply to those as well as as organizations and citizens of of the world at large so I think we're a moment of really trying to figure out a a lot of the implications where we may be starting small and I think you know you Kinda pointed out that that's that evolution but you also crossed the line in that evolution that's led to a whole bunch of of new things coming in the world before us. Yeah I mean whether or not iterative whether or not there is you know. A total hodel. Adjacency between somebody building excel model in defining the parameters in their excel model versus somebody defining a art generation Komo model whether or not that is Discrete or continuous relationship. Between those two paradigms 's I do think that whatever watermark remark we crossed recently. Maybe it was. You know. Literally what what you're describing this we've crossed some you know indefinable. definable moment in computer generated creativity. Or maybe it's just You know we define it in terms of just the amount of applications ends of interpretation extrapolation quote unquote too deep. Neural networks. And these these other these other things that are are are to one you know from one point of view just iterative but you know you could also say like Whoa Yo. We're basically just like we've just interested beyond vacuum tubes. I mean clearly like there have been a significant inflection points in the road back from vacuum tubes to you know multi-million dollar Christie's art sales from you know Ai. I generation. I do think something really significant has happened recently. And there's just so much going on I think that's that's what your your podcast covers you. Do you have a mix of the of the technical and the cultural and I think. That's that's important because there's a lot of people who sit between I mean everybody experiences the cultural impacts of of Ai Everybody's experiencing this on on a day-to-day basis basis and the technical elements. You know people have varying degrees of fluency with people have varying degrees of desire to attain that fluency see and I think by exploring both of those sides of a I you are covering a lot of ground. Why the name practical a I of all the adjectives that you have chosen why practical? Ai So we kind of back to our way into that to be perfectly Klay honest and it came with vision on. Even you know wasn't even specific to the PODCAST But the Vision on what Daniel and I warranted to do. In the world in the Serena and that is there's so much hype around. You know you see it. It's not even it's not you know. Once upon a time it was the kind of thing where when you read technical articles on line or you know back when things were still in paparazzis and stuff it was you know you'd see it there but in the if you look at the last five years you go back ten years ago and it was very rare that you would have a you know in the media you know that you would watch you know. CNN or are Fox or whatever your whatever your preference would be and see a regular set of stories but you know we are now in an age where it is normal. You know mass media conversation and with that there is a lot of hype and there's a lot of marketing that goes with that and there's a lot of misinformation on what what it is at any given point in time and so when Daniel and I we were already friends and we came from the software development world actually where we got to know each other but in this Ai. Space We said you know we've been learning and we're recognizing that there's a lot of misunderstanding and misinformation out there the wide we help ground people. Why don't we help? Provide a medium where people can come and just see what it really is. And what are the real capabilities is and what are what are based on those you know even without the high. What is the implication to the world around us in society of those technical advancements that were saying you know much like this conversation? We're having right now. We want to have this conversation with the entire world. You know if possible and so we deliberately chose to. It makes it practical in a sense of. It's both practical. It's very accessible to people. I commonly say if your grandmother wants to learn about Ai. They should listen snark podcast because if we do talk about technical topics but we we take the time to define the jargon along the way so that anybody can understand it and in doing that it makes it productive in their lives even if they're not an AI person even if they have some job that that is in their minds very different from. Hey I just want to understand. What is this thing? And how will it affect their lives so we like to say that we make a practical productive and accessible to everyone. And that's why at that very practical connecting you and your life whatever that life is with this thing that's out there changing the world in a lot of ways but often not the hyped ways that people are worrying about all right. We're nearing the end of our time. I got a couple more questions for you. Let's say I give you two million dollars cash. No no attachments. Except you have to leave your company and you have to start an AI company today you have to spend at least a million dollars building that company what would that company be honestly. There are so many opportunities the real answer I get people asking me on a regular basis. I WANNA go to other organizations or or create new companies from scratch. I'm in Atlanta. And I participate in Atlanta startup community. We actually host a meet up at one of the Georgia tech incubator. And so I have that opportunity and I have this conversations. I don't don't have one that I see I. I can think of of so many. I don't know what's right for me right now but to be perfectly honest. What the kinds of work work that? We're doing at my current employment which is Lockheed Martin is just so freaking cool that I honestly don't WanNa go do the startup right now. Maybe someday but right now I am so busy learning and seeing what's possible at a company that is pretty quiet about its. Ai Stuff off because of the industry it's in but we have so much cool stuff out there that it really puts us in that top tier and very few people are aware of that. That if I get to a point where I'm not learning anymore and things aren't moving and I don't feel that then maybe our consider that but you know there are big problems in the world that hey I can be a component to solving some of those include medicine and health care which I'm keenly interested in one of the things that I love doing is animal advocacy animal rescue type stuff off and I started a nonprofit to do work with AI and other advanced technologies in animal welfare animal advocacy type issues. Because there wasn't the profit motive for so for that moment. Staying at Lockheed Martin for the foreseeable future learning as much as I can taking my spare time to use these technologies. Oh Jeez to save. Animal lives as an animal lover. And then maybe someday if the time is right I'll look elsewhere but if they're just so many opportunities I I would challenge you if you're in farming their capabilities in farming and things that you would never normally associate with using CNN's to look at crops and can give him just the right amount and that's not even knew that's not cutting edge. That's been around for several years. There's really not an industry out there that you can apply this technology to an incredibly productive away so it's really pick what your passion is. And then say. How can I take this great tool set and utilize it in that way last question? What's the difference between podcasting asking in journalism? That is a very very good question so I guess it really is. You know you can podcast cost and you may or may not be a journalist and sometimes I feel like a an amateur journalist to his podcasting and sometimes I don't really depends for me. It changes day to day. Not sure what the right definition. I feel like the the journalist I at least I tend to associate with kind of the classical news person who is objectively actively trying to get to the truth of everything and that you can count on their objectively. I will confess that. I don't always do that myself as a podcast. Maybe I'm not bound announed to that level of kind of searching for the facts I think I probably am more the conversationalist and I just like talking into smart people who are doing a really cool things not sure that would rise to the level of what I think of his journalism but I know I have a whole lot of fun doing it and hope to continue for some time cheers to that. Thanks a lot. Thank you Chris. been great talking to you too. Thank you if your product has dashboards and reports you know the importance of making those analytics products beautiful lodge analytics gives you embedded analytics and rich. Visualizations you don't need to be a designer car to get great analytics in your product. According to the Gartner analyst firm the look and feel of embedded analytics has has a direct impact on how end users perceive your application good analogy analytics dot com slash S. e. daily to to access seventeen easy changes that will transform your dashboards that's L. O. G. I. Analytics Dot Com Slash S. e. daily lodge analytics is a leading development platform for embedded dashboards and reports and logic gives you complete control to create your own analytics experience. logi analytics has been a sponsor software engineering daily for a while. And we're very happy to have them so thanks to lodge analytics analytics and go to L. O. G. I. ANALYTICS DOT COM slash. Save daily to find seventeen easy changes that will transform warm your dashboards you can get better dashboards and reports inside your product with embedded analytics from lodge analytics

AI Software Engineering United States Ai Amazon sumo logic Lockheed Martin Department of Defense Google Kris Benson CNN Christie Dan Lockheed Martin E. T. T. E. R. writer
Kubernetes at Cruise with Karl Isenberg

Software Engineering Daily

1:12:09 hr | 1 year ago

Kubernetes at Cruise with Karl Isenberg

"Cruise is a company that is building a fully automated self driving car service. The infrastructure of a self driving car platform presents a large number of new engineering problems. self-driving cars collect vast quantities of data as they are driving around the city and this data needs to be transferred from the cars onto cloud servers. The data needs to be used for training machine learning models. These models must be tested in a simulated environment which provides more data to be integrated created back into the self driving car system which is deployed to the cars as the cars drive around the city. They can communicate with custom cloud services to get information nation about traffic navigation and weather. Cloud services are also used for internal tooling that can help with automotive diagnostics. Configuration changes Deployments and Security Policy Management. The software platform used to manage infrastructure at cruise is a combination of cloud products open source tools and custom built infrastructure. That is mostly deployed to coober. Netties Carl Isenberg is an engineer at cruise and he joins the this show to talk about the engineering requirements of building. A self driving car service as well as cruises approach to platform engineering. Carl has a great deal of experience in an building platforms. He has worked on Mesas and he's also worked on cloud foundry so he brings a ton of experience to the conversation. We are hiring during a writer. If you are interested in writing about software engineering or computer science we would love to talk to you. This is a part time role. But you'll be working closely with myself and Erica. Were also hiring a operations lead if you want to figure out how software engineering daily works and help us improve. Then you can apply for this part time role as well both of these roles if you're interested just in email jeff at software engineering daily Dot Com. Tom Love to hear from you. And let's get on with the show as businesses become more integrated with their software than ever before it has become possible to understand the business more clearly through monitoring logging ended advanced data visibility. Sumo logic is a continuous intelligence platform that builds tools for operations security and cloud native infrastructure restructure. The company studied thousands of businesses to get an understanding of modern continuous intelligence and then compile that information into the continuous tenuous intelligence report which is available at software engineering daily Dot Com slash sumo logic. These sumo logic continuous intelligence agents report contains statistics about the modern world of infrastructure. Here are some statistics I found particularly useful. Sixty four percent of the businesses in the survey were entirely on Amazon web services which was vastly more than any other cloud provider or multi he cloud or on Prem deployment that's a lot of infrastructure on aws. Another factoid I found was that a typical enterprise is uses fifteen. Aws Services and one in three enterprises uses aws. Lambda appears service is catching on. Aw there are lots of other fascinating. Statistics in the continuous intelligence report including information on database adoption Cooper Netease and Web Server Server popularity go to software engineering daily dot com slash sumo logic and download the continuous intelligence report. Today thank you to sumo logic for being a sponsor of software engineering daily. Carl Isenberg welcome software engineer daily. Thanks for inviting me onto the show. Cruise is a self driving car company. That's where you work. Describe how Kuban Eddie's fits into the the architecture of cruise sure so crews has cars driving around San Francisco mostly autonomously and the back end that the ride tailings service works on and other other miscellaneous pieces that that cruise works on runs in the cloud on communities so we have a lot of different things a wide variety of workloads and many of them run on Cuban and he's but not necessarily all of them so just a sampling of workloads. We have some some job processing and in some sheen learning and batch jobs and micro services and sort of caches and semi stateful services and databases and all sorts of stuff and so basically Crews needs a lot of processing power and one of the way we can enhance developer velocity and productivity is to give the engineer's a platform that they can build on top of to sort of accelerate their deployment in getting production readiness. Does it ever make sense. To deploy a Kuban as cluster into a car. A we don't do it. I'm not sure that I would would trust communities to do that at this point. But hypothetically in the future maybe it's not something where currently investigating largely we use it for large distributed systems and while they are people using committees for IOT systems distributed. There's a high requirement for quick lick processing in. The car is more so than some other areas so we use a real time operating system for our cars running mostly like C. Plus plus the the architecture of Kuban as though it can. Can you imagine a world where some of those architectural features are useful for the car Dr Operating System so in general now this. This isn't quite my area in Abia Engineering. But in general cars have multiple computers on them. These days is and so they need to have some sort of coordination for software installation management and stuff. We don't do over the air updates because we're running the cars in sort of a fleet fleet that that they can come back and get software. UPDATES in the garages so we don't have a live update requirement for that so we don't necessarily need to have distributed systems uh-huh that we can roll out rolling updates to they. We just load the car software on there and that's sort of a safer quicker way to do that but if you wanted to do it it live like that. You might have a system that you needed to have super high availability so we have a backup car computers and back. GP US and there's a whole bunch inch of testing that goes into that but we don't use for that functionality. The other sort of options. Either you get clustered on the car. or a cluster. That has nodes snowed in the car and Cubans wasn't really designed for the use case where the nodes are sort of intermittently collectible so that the scheduler model that committee is uses and the distribution is not really designed for that use case before you worked a cruise you worked at pivotal and Mesa sphere. Both of this companies spent lots of time developing platform as a service products or argument. Just open source platform products. You saw a a lot about the design requirements that enterprises need for platforms since joining crews and seeing. What I presume was an entirely new domain of company that had its own requirements for platform infrastructure? What has surprised you what has been fresh or novel as you are getting acquainted with building platform infrastructure for car company so I've been doing container platforms for about six years at those companies and then even before that I was doing software platforms for e-commerce stuff? So there's a general requirement almost everywhere therefore platform services internal to the company or external depending on the size of the company so a lot of the compute requirements are effectively effectively universal by one of the things that is special about what crews is doing is is the wide variety of workloads that we run which is unusual from my I experience with other customers I worked at other vendors so for example we do a bunch of AV processing probably you know a significant amount of our time and compute process is taking the data off the car and processing chunking different units and then feeding that into different systems whether that's machine learning algorithms or whether that's data infrastructure giving mapping data for our applications were for for developers to then go use to replace an areas so that they can optimize their algorithms or trying their systems so there's a huge amount of investment asmat behind that sort of idea itself but also we have the standard micro services that everybody has a standard applications. Everybody has they need to be highly scalable to you. Know Handle Internet traffic in the future when we have customers in our cars and requesting rides and then we have a whole bunch of sort of behind the scenes processing using and tooling internal to engineering a lot of companies have so like CIC D. and testing infrastructure. I think one of the biggest challenge. We have isn't just getting the car to drive safely and quickly but also to to be able to test that to be able to simulate faster than we can drive in the road and in having simulation platforms that can do that kind of thing at scale using thousands of machines or even tens of thousands of machines to to get as much compute power as we need to during a reasonable timeframe. The idea of platform engineering. This has been done for a long time at Google. The goal. It's obviously been facebook for a long time. Net flicks has a great reputation as a platform engineering company. These companies that have large amounts of resources great revenue streams great engineering teams companies like cruise. It make sense to have a platform engineering team. Now we're here cube con. There are a lot of companies that have different requirements. Different sets of constraints companies like big insurance companies banks. It's on gas companies where they do have compute intensive requirements they do have complex software platforms. I wonder if you have any general perspective on who should have a platform engineering team. If I'm a bank if I'm an insurance company should I have a platform engineering team or should I sort of say look are from engineering team is. Aws or apply for caring team is like we don't have of the volume of engineering resources. That netflix has we should not try to do. Platform Engineering. Like wind should accompany have the platform engineering team. So this is actually a question. I had asked for myself when I was. You know like in between jobs trying to figure out where I wanted to go. I wanted to work in a place that had an investment in platform team. Because that's what I've been specializing so from what I've seen the companies that have a platform team whether they they should or not tend to be sort of medium to large size And you don't usually see small startups sub two hundred with platform teams. Because they just don't have enough people to specialize that much so as companies grow they get more specialized in what their individual people and teams do so there. There seems to be a turning point to me somewhere between like two hundred thousand. where I you start with like either an infrastructure group that has has to like manage giving hardware and then there's sort of another layer where you focus on engineering productivity or C. I. C. D. that kind of thing and then the next layer after that is a platform group where you're building tools? That are accustomed for your use cases but all of these groups are are doing effectively the same thing they're not necessarily building platforms from scratch worshiper. Not It's more like a lot of integration and then bring your own components and integrate them into the whole so so I usually talk about our crews paths as constellation of applications and components and services that all sort of provide vied value to our engineers rather than something that completely abstracts and hides all the stuff underneath it because more often than not the engineers need to treat them. uh-huh sort of like onion layers. Where they they peel back to the layer to get the functionality that they need so some people will start at the top where they want like an application pass or a function as a service solution? Because it's the least amount of work to configure but more complicated solutions will need to peel back those layers and go directly to the containers and some systems systems will need to go directly to the VM's so we tried to build from from the bottom up there and our general policy accrues is to only build if we can't buy four back end components specifically that doesn't work for the car because that's a brand new tech but in platform space. There's a lot of tools off-the-shelf that we can buy and me personally having experience working at vendors that did platforms in the past. I have a very good idea of what's available available and what its abilities and capabilities are. I did a talk. A few years ago on the container orchestration wars and basically didn't an audit got it on all of the different platforms available while I was at Mazuz fear sort of key comparing measures pie against them. And obviously at that point Cooper Netease was one the ones in the mix so I'm very familiar with the edges of where those tools and but it's not always obvious to people so you have to kind of look at the tools and figure out how you're going to integrate them and who's going to do that integration work. If you don't have a platform team then you have people doing it themselves and if you have a lot of them doing it themselves than they're doing different things and it's a little chaotic and inconsistent so really a platform team is to sort of bring the chaos together and in have some rising tide that lifts all boats this is not directly related to cruise. And you don't have to answer if you don't want to but do you think coburn as deserved to win the container orchestrations horse was there. Something architecturally that it did wonderfully or was it. A combination of funding and marketing being and open source tactical savvy that caused Cooper Nettie to triumph over Mason nomad and all the other ones I think technology that becomes popular is not just because of its technical merits. I think there's always Some other aspect to it whether that's marketing or some popular vendor or some someone who popularizes it on the Internet with a fancy video or something. There's always a little bit more to it than just technical value and I see that as coming back to the root cause of that is that people don't have time to go research all of the available options and make the best choice. That almost never happens instead. People do a little bit of investment in what they know about and they do a little. RND and research on the Internet and they might proof of concept like one or two solutions and make a decision because they need to move forward so sometimes that bias on the things that do one thing and sometimes they biases on the things that do many things because you needed all those things covered and you get one tool and you invest in that. And it's faster than I'm getting something else. So I think Cooper Netease does more than Mazo said. I don't think anyone would disagree with me on the scope of what it's capable. I love there's more components there's more features. API's and so just from that comparison alone like people would take coober netease because it had more capability and and you wouldn't have to invest as much but if you compare it to something like data center operating system that that measures fear offered Mazuz spheres. Solution had a lot lot of components possibly more than Cubans does internally at least at the time and tried to satisfy more of that so there was there's kind of a step value function in like either. Either you're getting the kernel of your operating system or you're getting like a full operating system where you're getting pro mix in the middle and so if you provide more of the stack in a single solution. It's attractive to people that don't have a solution yet but it's less attractive to people that already have something because it's harder to integrate because it has it's more opinionated so when you talk about cloud foundry or DCIS's opinionated stacks whereas Cuban is of less opinionated and less complete than those full stacks but also offers different value so in terms of whether deserved to win or not. I don't know but I don't think it's super matters. It's what one one in terms of popularity and so what what people are using right. Well one thing. That was interesting if I remember D. C. O. S.. Had a really good story around and we're going to help you get cough deployed. We're going to help you get sparked deployed. We're GONNA help you get. These awesome. Open source frameworks deployed and it turned out that people wanted cloud providers like people wanted like look if you look at cruise infrastructure. You're all in on Google cloud you kinda you want Google cloud sequel. You want the whole Google cloud stack and a lot of people want full. Aws Stack that may or may not have a design opportunity for spark or Kafka or whatever open source source thing you want to install it you kind of want to be all in on a cloud provider in many cases when it comes to compute availability. There's a a lot of cloud providers give you that you can opt in later where you can. You can initially opt into some low hanging fruit like got an infrastructure as a service solution that has compute and storage and networking and then beyond that. Then you can pull things off the shelf that are valued to you and getting on a cloud cloud provider that is a platform itself. Gives you the ability to to pull from the service catalog and that provides value. I don't know that that's like a fault of of Messier's comparing I think Messier's solution was targeting more of the data infrastructure requirements and end people that had on premise requirements. And then there was then later a cloud and hybrid focus as well and not. Everybody has all of those requirements. So will our general approach is like if we can get it for less work than that's what we want to do as long as it satisfies a requirement so we'll biased towards a SASS product if if it satisfies requirements and then when it doesn't will go on premise. We don't have the same sort of data locality requirements that some people have in the EU. And so we don't necessarily need to do all our data processing on premise. And because of that we aren't trying to do hybrid. Ncube Annetta is we have a hybrid network spans multiple clouds and data centers. But we don't have hybrid clusters clusters in individual compute areas doc at cruise. You vetted the different cloud providers you tried the different managed carbonates instances you try and I believe some some of the other ones. You wound up going all in on Google cloud specifically well based off of a lot of Google cloud related services. What what was that? Vetting process like the the cloud provider vetting process. So when I came into Cruz two years ago some Cuban Eddie's it was already in use so I was put in charge of making sure that became a platform that could be used not just a bunch of different snowflake clusters however they were also using or we were also using rancher the one because it was easy to deploy for a small team at the time cruise news has been in hyper growth for the last four or five years or something so the team was much smaller when those choices were made and it seemed like a good idea at the time because it was smaller and we get more bang for our buck and as we acquired more clouds specialists and people who would be on a platform team they came with opinions and investments and ideas about where to take the platform. So we did some investigation of cloud platforms but we didn't necessarily try to audit the whole system. I mean like I had done some of that previous job. Site took that knowledge with me and some of the other people had done similar things. So it's not like we went and did a proof of concept of all of them. We did investigate. Some E KS for example was not really mature enough to us at the time two years ago it was like barely out of infancy. Didn't have any upgrades upgrades. There are a little further along now but they're sort of behind a s and K E and we were using a bunch of aws at the time and we still have some in aws investments. So I wouldn't say we're all in K.. We maybe default decay e sorry all in GDP we default the GP. But we'll use other reese cloud resources that are valuable eligible to us because they're better for some reason and we have the hybrid network to support that and make that possible. I like that approach and I think that's the only they're coming in. I think I've talked talked about that. That kind of platform strategy where you center around one cloud but you keep yourself open to being able to use the other clouds actually. I've talked to number companies. The had this kind of infrastructure but thumb tack comes to mind. I talked to thumb tack about because just because they started on. Aws kind of before cloud was really mature a little bit earlier in the time line crews and they built a lot of their business logic on CBS and then they've built. They used a lot of data platform services on Google. But so you're talking about like cross cloud networking infrastructure across cloud. Whatever tooling what does that look like? How much infrastructure give to put into you having cross cloud functionality a lot so eh blog post recently with Collaborated on with one of the network. Guys a accrues and it turns out that there's a lot that goes into creating hybrid cloud and in ours specifically we had a high bandwidth requirement for getting getting data off the cars into the cloud and so we ended up having to invest in some some private fiber to and some some Internet exchange point of presence so that Whole setup is is pretty expensive and not everybody is doing it but if you talk to the bigger enterprises a lot of them are doing something similar just to get have neck an on on premise and a cloud data and similarly if you're using sort of Iot or edge cloud stanchions yet to have some way to get data between and clouds quickly? So there's there's a lot of complexity that go into setting up the interconnects and routers and the cloud or to connections and managing propagation increasing clouds and then handing out IP force base managing those that kind of thing so the platform team at cruise doesn't do that part specifically who we do that for Ark Guberman. He's clusters but we have a whole other like core infrastructure team that handles a lot of that and then a networking team adjacent to that handle a lot of the sort sort of making it possible for clouds to work together. So there's definitely some some concerns around having to be close to your compute power so your your storage in compute. Ideally you want to be in the same place if they're the same application but if you have other application that doesn't talk to the the you know the main data laker whatever it can it can run somewhere else. Many might have some on premise. Things of czar like big use case I was mentioning. Was Getting data off the car so we have an on premise installation clusters for that. And then they upload into the cloud and then the cloud takes over most of the big processing for compute requirements. Apache Cassandra is an open source. Distributed database that was first created to meet the scale ability and availability needs of facebook Amazon and Google. In previous episodes of software engineering daily we have covered CASSANDRA's architecture and its benefits. And we're happy to have data stacks the largest contributor to the CASSANDRA project since day. One has a sponsor of software engineering daily data. Stacks provides data stacks enterprise a powerful distribution of CASSANDRA created by the team that has contributed the most to Cassandra Ndra data stacks. Enterprise enables teams to develop faster scale further achieve operational simplicity ensure enterprise security and run mixed next workloads that work with the latest graph search and analytics technology all running across hybrid and multi cloud infrastructure more than four hundred companies including including Cisco capital. One ebay run data stacks to modernize their database infrastructure improve scale ability and security and deliver on projects such it just customer analytics Iot and e commerce to learn more about Apache Cassandra and data stacks enterprise go to data stacks dot com slash. Save the daily. That's data stacks with an X. D. A. T. A. S. T. A. X. at data stacks dot com slash ESI daily. Thank thank you to data stacks for being a sponsor of software engineering daily. It's a great honor to have data. Stacks is a sponsor and you can go to data stacks dot com slash. Save daily to learn. Learn more in the blog posts. You talk about some of with the internal tooling that you've built and you already alluded to a little bit of the strategy around how you make build verse by decisions. Tell only some of the maybe one or two examples of build decisions times where you had to build some particularly clearly Specific internal tool where the cloud infrastructure the third party vendors for whatever reason were not ought supplying a product that met your needs sure so we have a number of open source projects that we've released in the last year or two. I think our first one one out the gate was called are back zinc and generally. It's really just an integration tool time together two different pieces of software that weren't talking to each other and in this case it was cooper netease and needing to use Google groups to manage permissions so we would have the platform T. managing the role binding. And then we would. Delegate relegate group management to like engineering manager at a tech lead to make sure that their people were in their group so that they all had permission so that that added a little bit of self service and just tying those pieces together required a little piece of software that that we put together. Google groups isn't that like email lists tool. Yeah so Google Groups oops is Part of G. Suite and we use because we're integrated educate E. N. G. C. P. G. Suite for our identity provider and then we use OCTA for single sign on and duo security for for two factor authentication but that's like user identities and so because G. suite we has google groups. You can put people in Google groups and then he can do role bindings. GCP does this negativity you can. You can do role bindings against Google groups and then put people in those. Those groups are adds another layer of abstracts. Awesome so we do that. And build that into Taku Bonetti's that to me that kind of integration the integration with the human identity email stack and your infrastructure seems like one of the most powerful opportunities Google cloud relative date of us. Yes I'm what it'll be. US has a pretty good system for for system authentication of their own tools but in terms of single sign on usually had to go somewhere. Someone else you can get all sorts of identity providers for user authentication. I think the the frontier for for pushing forward is actually for service accounts. And there's there's a lot of sort of missing missing tooling around identity providers and policy authorization for for service accounts and then integrating them together. One example is in our back sync application we. You allowed putting service accounts into groups which Google groups doesn't natively support. What is a service account so serve guys what services use the talk to each other? Basically so so they have different means of authentication. Usually so user account you'll us to factor authorized user password or something and then service account will usually use a signed certificate from from a trusted authority or temporary issued a token or a J.. Wt Jason Lead Hogan again and so that allows the identities to or sorry the services to know who they're talking to know who's talking to them. How does a service account and service identity relate to sort of like I've done a couple of shows about the spiff inspire projects which are for? I think assigning application identity are well. There's a SPEC spiff is the SPEC. Assigning application identity inspires and implementation of it. And I believe that these are useful. Because if you if you give applications identity then you can assign certificates. Security certificates deal S. certificates based on and that correlate with those identities. I think that that's my understanding of why that that identity system is useful. Could you tell me more about those projects or how that pertains to to the things that you've built around service identity. Sure so I worked a little with a with WWF. Back a few years when I was measures fear and he had this Cio Zia idea that there wasn't a solution for service authentication authorization that was open source. Everything everybody had their own solution. And there was nothing that you could provide to like integrate with Kuban Eddie's natively for example if you have an open-source ecosystem like the cloud native foundations tools. There wasn't a solution in a pull off the shelf for that. And so he spent a bunch of time defining the piff specification to try to solve that problem. And I think when we sort of started started Dr Platform Team at Cruz. Two years ago that was in its infancy. It was like zero point four version or something. Actually I'm tired. I'm confusing the two. This spire implementation of that a specification was in early stages and so it wasn't quite ready to be pulled off the shelf and used and because the specification itself didn't have have more than one implementation and the one that had wasn't mature yet. There wasn't a lot of vetting on it so I think there's been a lot of progress on that in the last two years but it's not necessarily as integrated as stack as if it was available two years ago. So we have some ways to do workload entity primarily with vault for example oh so vault can do certificate signing and basically hand out identities and also then put in subject alternative names into those certificates for validating and. That's one of the ways that that sort of stiffy connects things is with us or ticket sands. So we have some compatibility with that in in our tooling but it's not a primary function of how how we operate is definitely a place that needs improvement in the ecosystem in general and we're trying to move that forward a little a bit but we haven't yet quite been participating in the sort of species fire crew kind of waiting for that to mature a little bit. Tell me about the release release process at crew. So there's different types of releases obviously so people could deploy cloud services. They need to release process for that. People also need a release process for deploying software to the car to this some kind of rollout process. There can you just tell me about. Cic De broadly at Cruise. Sure so platform team doesn't own all of that ourselves. We have other teams that that focus on some of those pieces but just in general I think we have two different different flows for most software. And you can think of them as either the deployment of the car workflow or the deployment of the cloud workflow. Oh so we have a couple different. Is We use primarily like circle. Ci and Build And those usually do the testing testing for you of the software and building. Most of the time we have other tooling that does more complicated build pipelines as well. That's kind of home grown and we have of a homegrown image builder that triggers off. Get Hub Web Hook so we can basically build the image and then tested NCI and then we connect through due to continuous deployment. tooling we use. We spinnaker for most of our deployments now and we wrote a spinnaker operator around it for managing being the sort of multi tenancy of that one of the complexities with running Cubans multi. Tanna is that a lot of the sort of constellation of tooling and components around. It are not necessarily. He has multi tenant. As you'd like. SPINNAKER has a rolls enroll bindings and so what we do. Is We create accounts with our operator for those for breach namespace and then in each cluster. So when you go into spinnaker you're in this group and this group has permission to deploy that his name's Base Group also has permission to use this account in spinnaker and that account has permission to deploy to the namespace too so you can you know. See what you see. In Kuban Eddie's as you can also see and deploy to that with with continuous deployment and spinnaker and then spinnaker usually gets triggered with service account from CI or some automated basis. Assist depending on your workflow and so then spinnaker has a pipeline that eventually deployed into the cloud usually varying here uniformly about about cloud services. We can't tell anything about the plummeted. The car process. So that's not my area of expertise but mine understanding is that we we build in are on custom build pipeline run testing and there's a large simulation platform that runs. What we call the Matrix to simulate in San Francisco and do a bunch of runs of the car before we ever put the software on the car itself? And then I think that the deployment strategy for getting this off around the cars proprietary. I Terry so I can't really talk about that but it's effectively similar in that it's a CIC d flow but because of the heavy testing requirements that effectively end to end tests. We have have to run a lot of them and there's so many of them we can't run them all every time we test every time we build because it takes too long as too expensive so we had to like run. Samplings blinks of them to make sure that it's valid and do a statistical measurement to make sure that the car is safe. And then you know do a live test as well. Cool the cloud services so are there cloud services that the car is talking to on a regular basis or is the car. Mostly you think of it as an air gap module the car is not air gapped however it can function without connection. We generally don't like to operate without a connection because it removes some capabilities ability. But it's definitely safe enough to like you know pullover or continue going down the block if you lose connectivity. We also have a very redundant system so we're across many any zones and many different ways to get information into the car so we have a lot of cloud systems. You can think of like the even like Uber and left have similar scenarios where the APP that the person is using is constantly talking to the cloud for mapping data or traffic redirection direction or just real world updates about what's going on and and then if the car gets stuck or something you might have to phone home and figure out what's going going on so th there's all these sort of systems that it needs but doesn't need a hundred percent of the time right. That's actually a really good analogy. Because I have been playing situations where you're the the drivers phone is like doesn't have complete connectivity like I R- I know there's been situations where my phone has connectivity. The drivers phone does not have connectivity sitting in the back seat I'm I can see their phone and like their map. Looks Different Than Mine. And I'm like why are you going straight here like you should be taking a right. He's like well. I guess my phone is broken or whatever but but there's a graceful degradation and it's not it's not catastrophic. Life goes on multi tenancy. That was your middle name until yesterday when you're deploying this Large volume of services. You WanNa think about how to slice up your infrastructure to use it men in ideal fashion and you got plenty the of data intensive workloads. You've got plenty of workloads where you need. Lots of redundancy and so thinking about how to deploy Oy the different containers the different services in a multi tenant fashion where the containers are not suffering from a noisy neighbor problem. That's GonNa Starve them from resources to important to to think through this tell me about workload isolation at cruise sure so that was a large focus of my mytalk earlier. Cube gone all the different layers of isolation that's kind of required and optional for for Multi Tenant Platform I tend to think of a multi Z.. As sort of a spectrum of requirements that you can opt in to depending on what your specific workloads need. So there's sort of a base level of isolation around authentication authorization and there's also some more isolation around Resource resource isolation using using CPU and disk and stuff and then there's more isolation around the integrations that are required for the whole system them to operate and then you can get all the way to system isolation which is effectively a multi instance or single tenancy and that that would be like full full system isolation so different workloads have different requirements for that and different workloads have different requirements for availability for example job workloads and batch workloads blood's might be more interrupt -able than web applications or something that has low latency requirements. So you might want to group your applications in a way aware. The ones that are similar to each other are isolated from the ones that are different from them so that they don't cause disruption on the other hand. If you have ones that are dissimilar from each other that usually means you can get higher utilization of your compute nodes. So if you're packing in jobs and services onto the same mm nodes in the same node pool. In the cluster then you can get higher utilization out of your. VM's that you're paying for in the cloud so there's is a trade off either way you go and there's a lot of concerns around what you want to isolate and what doesn't matter as much so we tried to pick off the low hanging fruit through an isolate those and try to get towards logical isolation where the tenants can't see them so each other which is difficult to do in a Cuban Eddie's environment but thankfully awfully this isn't sort of hard multi tenancy scenario where we don't trust the people who are using the system. If if something is broken we can figure out what it is broken with metrics metrics and go find the team person responsible and get them to to fix it. which would be sort of less viable in a full month multi tenancy scenario uh-huh where you have say GDP having us as tenant if we do something poorly they don't usually go and ragged on their customers and tell them to stop doing stupid things? So so there's a little bit more influence availability to to make sure that we're getting to cloud native and that's also part of the reason for doing multi agency is to find out where those edges are in our case before we have customers and be able to solve and remediate and add isolation and add security layers in order to make a better product and more production ready faster and sooner. So we're sort of optimizing for that. But when we as we get closer to sort of production of customers in our cars and using the back end software in anger we might value availability more than you know experimentation and getting our requirements in a row so as we have reached scale peaks where Cooper Netease will cap out in certain scale aspects. Ex- we have found that we needed to pull out domains of workloads and put them on different clusters different names bases or even just different nodes appending on on those workloads so those like spawn clusters or those. Those secondary clusters are not necessarily single tenant. They're still multi tenants like a group of teams teams that work closely together. That all need to talk to their. Their systems need to talk to each other and collaborate but they might have five or six names bases on that cluster. So it's just a smaller multi tenant cluster Kuban as cluster management. A couple of things. I don't know the answer to one. How many governors is clusters? Do we want if I'm a big company. And there's a bunch of people throughout the organization. Who are doing cooper netease proof of concepts do we eventually want to merge those coober netease clusters to have all the containers in one big cluster? Do we want to keep these kupuna urbanized cluster separate. Do we want to have some glick. God Kuwaitis clustered that federate S- into those other clusters that question is also a question of the pets versus cattle analogy. Does that extend to Cooper says clusters. Do we want disposable Kuban as clusters or can they be pets. Those are two questions you can take them as unrelated or related. I think he largely depends on your use cases so I don't think there's a one-size-fits-all answer to that question. In Our case what we've chosen to do is do the hard thing I up front so that we know what the challenges are for scaling and we can test that sort of scale capacity better with multiple workloads rather than trying to run synthetic workloads clouds a lot of the scale tests. That you've seen people give numbers out for like three thousand or five thousand nodes or whatever tend to be synthetic workloads and a real collection of workloads will follow her because of something else. That wasn't in this indebted workload so we specifically biased towards that to find the rough edges and and fix them and that strategy means that we are defaulting to the multi tenant. And then if there's some reason why that isn't working then we separate that out into a another cluster domain whether that's single tenant or a smaller multi tenant with less neighbors so that strategy is both from like a cost savings perspective because we optimize for the thing that saves us the most money until we can't do that anymore. And then we spend more money to do another cluster but also it just realistically speaking from an operational sense when we started two years ago the platform team was like five or six people and five or six. People can't operate one hundred third clusters realistically without a an extensive amount of tooling. And for the most part those. Those tools don't really exist off the shelf. You can get things like Tara for them to do some of the tooling but you end up writing a lot of the tariff arm yourself like Cuba. ADM is not a full solution. It's only part of the solution. And and then once you get to the like you know tax two hundred x two thousand x scaling clusters. You have to write some pretty beefy automation to do that. And there's a couple of companies trying to to do that but I don't think there is a well baked solution for that and they're definitely wasn't two years ago so I think we're getting to the point where we need more clusters and we will invest in cluster automation. And we would like to get to clusters cattle but I don't think we have a requirement for our production clusters just to be super disposable. They need to be disposable for a couple reasons primarily like if you make bad choices when setting it up or things change leans over time since when you set it up so like our prime example is like if you allocated too many or too few. Ip's to the cluster. And that's that's a hard thing to change range. After you've created the cluster so you might need to create a second cluster and migrate your applications and delete your first cluster and then replace it. And ideally so you'd have some sort of multi cluster ingress which people are still investing in now in order to like do load shifting to another cluster so that you can replace clusters but but it's not like your discarding them when they fail the same way or discarding pods because it's a self healing system. It's highly available. It's pretty large and complex complex. For the most part the cluster will heal itself so I don't need to delete it and replace it so it's really only when you're like root configuration chains. And you can't do alive upgrade or rolling upgrade on that there is another use case where you want disposable clusters and that's usually for ephemeral development environments if for example us on the platform team if we want to spin up a new feature in our has sometimes we have to deploy an ephemeral cluster. We experiment on it and throw it away. And now we deploy into the data cluster. Our other engineers are using development to testify against some real workloads. That aren't you know don't have as much availability requirements. So so there's a developer velocity argument to having distinct clusters like that but we also like I said have a have a cluster to provide a space for engineers that darnall pass team to have a target Dev environment that they can create a namespace in and go do so. We're getting to the point where we can do self service namespace creation for developers person. And where we're GONNA make those disposable in the near future but we haven't quite got to the disposable clusters yet kinda sounds like maybe I misunderstood. He said but it kind of sounds like the fact that we cannot just throw all of our production workloads into a single Cooper Nash. Cluster or throw all of Dr deb workloads into a single Cuban as cluster. That's always it but it's kind of sounds like a failure. Failure of Cooper Netease the fact that Kuwaitis cannot itself figure out the the resource constraints of these workloads to the extent that it it could spin up new machines or put these things on new machines or whatever like in an ideal world. Seems like you would not have to manage a hundred Kuban. Eddie's clusters there's with six inch and I realize you're not doing that anymore but like you wouldn't have to even be something on the horizon like Kuban as it self should give you this one. Big Cluster that you can operate from the Default Cooper Nettie Control panel and nobody should have to write this amount of tooling. Should all just be at the Cooper nighties level right aspirational. I'm not sure. I agree with that having worked in different complainer container platforms uh-huh for for six years. I've seen a lot of different customer use cases for these things and there's marketing that says you know you just I just have one cluster and it solves all your problems but in reality that never works out so everybody is using multiple clusters. There's like nobody is running. Just one Cubans cluster. They might have one. Production Kuban Eddie's cluster. If they're small or they're multi tenant even us with our like gigantic multitask detained cluster. We have multiple production classes. Because we have like you know them in separate regions because spreading your nodes across regions means you have cross regional traffic. Do Your your API server. And that's really slow and expensive. And if you have to do like data exfiltration across boundaries like that that gets expensive too so there's always going to be a scenario where people need multiple clusters even just for a same environments whether that's developmental production so spinning up. Clusters is is a basic basic requirement of using coober netease but that doesn't necessarily mean that the ecosystem has gotten to the containers as cattle sort of solution yet. I think there's still. Some investment meant to be made their log log. DNA allows you to collect logs. From your entire Cooper netease cluster in a minute with two. QB C. T. L. commands whether you're running one hundred food or one hundred thousand containers you can effortlessly aggregate and Parse and search and monitor your logs across all nodes and pods in a centralized log management tool each log is tagged with a pod. Name and a container name and a container ide- and a namespace and a node log. DNA is logging. That helps with your cougar. Netease clusters there are dozens of other integrations with major language libraries in. Aws Ask her Roku and fluent D.. And more logging on Kuban indies can be difficult but log. DNA simplifies the logging process of couponites. Clusters give it a try today with a fourteen day trial. There's no commitment. There's no credit card required. You can go to software engineering daily DOT DOT COM slash log. DNA to give it a shot and get a free t shirt. That's software engineering daily Dot Com slash log. DNA thank thank you to log. DNA for being a sponsor of Software Engineering daily cruises Uses Hashi court vault for secret management. Is there anything particularly hard about managing secrets at a self driving car. Company so crews has a strong bias and priority for safety and security because they're they're tightly intertwined when people's lives risks at risk in cars so our cars can't be compromised. Our back end can't be compromised. So we spend a lot of investment investment in security and validation and testing and that means that We need to do things in ways that are robust and we'll get as has towards production readiness so I think that requirement alone while maybe not the same as someone who has a web application you know might be more along on the lines of people that are processing credit cards. They have similar requirements so we have a an infrastructure security team and actually a whole sort of security group organization in crews and they manage our vault clusters for us but it's effectively highly available system and I know how Gore has invested tested in in making that a little bit easier recently but it used to be a little more complicated. I think now you can group them with console and get them to be a highly available cluster cluster but it still doesn't necessarily scale horizontally for load and there's there's a lot of complexity in managing the permissions themselves so for example. All volts has a lie for allowing changing things in an API that you can do that with but there hasn't been as much investment in the permission mission management using that API so like we have a get ops tool for having the configuration defined in get hub and then we can push it to vault to update update and now works for maybe seventy five percent of the vol. API But there's some of the Isn't really designed for that kind of item. Potent usage setting setting up route. See as or something that has to be done off. Line or manually. Once and you don't want to reapply it later so it's complicated to Make a secure system in general and so we as as the cornerstone for our secrets management but then just because of all is something that does secrets management. It doesn't mean you've solved it with vault. He's still have to do a lot of management of the robot vending in the groups and the policies and integration and the validation things using volt thankfully has a pretty good API for that the service Mesh Abstraction. Is this something that everybody that uses coober netties also needs so we're doing an Sto co-pilot now to sort of answer. The question of whether we can roll it out to everybody. So the biggest. I think drawback to the service this Mesh as implemented by Sto or even liberty is that you have a layer seven proxy load balancer Sir or a reverse proxy or proxy as like a Sidecar Damon next to your application as opposed to doing layer for IP switching Kuban Eddie's does is natively with Ip tables or B. E. B. PF so the labor force solution is just inherently faster. But it also doesn't doesn't allow you to gather the metrics doesn't allow you to do quality of service so much so there's a lot of investment in this space by various players to try to make the network layer provide vied more value so many times you can get more metrics and performance from an overlay than you could from like an underlay because underlay does. Less work has less hooks into everything and the same is true. With the sort of difference between using IP switching the therefore versus a proxy at layer seven layer seven can gather all sorts of metrics and enforce quality of service and inspect headers. And get in your connection and do connection pooling and things like that that you you couldn't do it though lehrer therefore level or is much more difficult so I think I know like Google is investing in trying to make some more functionality available in coober nineties in the ecosystem is as as well because I think Talk Tim Hawkin. Did a coupon recently was Talking about how he already is kind of a service meshes just not of the feature complete one and so other alternatives provide more functionality. And so that's kind of why. Do as popular as it has a lot of problems. Salem's base that it's trying to solve for some of the things we want to get out of it is metrics for Ingress egress quality of service isolation between gene appplications and tenants and also the ability to do sort of a canary rollouts or also the circuit breaking so unions skins are decided the client which dependencies are accessible at any given time and then remove them from the load balancing list whereas if you have a reverse proxy proxy doing all of that then it doesn't really know enough about the client to make that decision and every client is then making the same decision as opposed to moving into the client side where you can have a more dynamic system that is based on where the client where the services and where it got bounced too. So there's there's more capability four metrics and observe ability and making decisions on the fly dynamically but that also means that there's a latency impact so not all applications can tolerate that blatancy impact. So there's been a lot of work ideo to minimize that by removing mixer which was involved with the metrics aggregation. And then you can sort of decrease the scope of addresses or names that each application or service workload knows about sort of by by blacklisting thing everything by default. And then you have to white-list dependencies and the things that depend on you in order to set those up that requires more operations but also is more secure and it makes latency hit less bad because you don't have to propagate everything everywhere let's have an engineered crews. I wanted to play a new service. What is that experience like today? Is it turn key in as beautiful as you'd like it to be smooth. Does you'd like it to be. Hey you're smiling or do I need to like sync up with a bunch of platform people and devops specialists and like how. How much do I need to get this thing out the door so I mean I can say that? We're not where I would like to be yet. But to be honest I've never worked worked at a company where they were where they wanted to be. A great yeah so like do you what people have carte blanche to like launch a service internally internally so I think there's there's a couple different way that's a permission question I'll get to in a second but the first question is is sort of answered by win. We onboard engineers. There's we actually have a class that they go through where they learn how to use Kuban Eddie's basics and deploy an example service and so we do some of that right out of the gate and we'd like to do some more training around integration of CD. We'd also like to do some more work around making it trivial to create an application and Abba integrate a great with all the things because a lot of times the integration points and learning all the tools and becoming proficient at all those tools enough to use them is take time rather than writing code necessarily. So it's pretty easy for our developers to come in and in their first week or second week go like us our clusters but they might not know the full extent of functionality. And we'd like to get a good experience. Yeah so we. We'd like to get to the the point where there are better defaults and there's better tutorials and people can get further faster but they're they're you know already in the right groups I have the right permissions get attached to the group. That's that's their team is in and then can deploy to their deb namespace to play around with and get a field for what their team is doing. Yeah one of the reasons I asked. Questions is the three at a five companies that I've worked at before software engineering daily. My I like probably two months were spent understanding the Java monolith that I was working on and understanding the internal systems that were built built around this Java monolith and I don't think that is even the minority I think maybe the majority of developers spend their first two months in some kind of painful on boarding process learning legacy systems and INTERNAL SNOWFLAKE systems. And it's it's not it's not a smooth process and my sense is. The industry is improving moving. And part of the reason it's improving as for him because of standardized platform systems. Maybe yeah sort of I think because of the popularity of Kuban. Eddie's we get people that either want to know coburn as because it's good for their career or they already know it from somewhere else that was using it or they don't i WanNa know what at all and would really rather right software code and then deploy it and not have to worry about how to deploy it. So I think there's usually kind of two different camps there and so us having some Shared tools there and open source tools. That are familiar to people when they onboard lowers the barrier to entry as opposed to if we had a full abstraction that completely hid Ed Kubo Netease and all the components. We could build something like that. It would be a lot of work and we could sort of hide all the the defaults and make them all smart so that you have you know ten line command or a yellow file in your Ribaud to deploy but that would mean they have to learn that interface and then that hides the underlying interface race from them and all the tools that are going around. And if you do that and you your abstraction is sort of imperfect which invariably it will be and then you're allowing your developers to be to know less about the system which is great for the initial developer productivity. But if you're in a devops model where those developers are operating that code it will follow over. You know later. And they won't have known the tools to get there so there's a little bit of frontloaded pain in understanding the system that helps you later when you're operating it especially in a in a devops model like we have because you're not just putting it over to an operations team operations team knows guber netease like We haven't been doing that. Model because of the standard benefit of devops is that your customers are not just the people that user application your operators are also your customers because they use the code more than you do after you've written as an application developer so devops sort of treats those both as customers and allows the developer of the application to have more empathy for both of those use cases. So if you are also the operator then you you optimize for code to make your operations easier and really. That's what the platform team is trying to to make better because if we all the teams do it or if all the teams had to do it themselves it would be a little bit of chaos. I think in general most most companies he's will try to identify chaos and reduce it in an iterative pattern. Where you say I have ten teams and they all need to do something similar? They're all we're going to build their own solutions and then when we've identified what the common pattern is then will will abstract that and make team to handle the abstraction and build some mm tool that solves a for everybody and then they'd have to do less work but there's always a little bit of growth of chaos to create the pain that motivates the software development in the new abstractions. Well the the platform abstraction of G K E or AK ASS or whatever like the Kuban as it cloud things this seems like a pretty good abstraction for us to be working with as an industry where you can go from Google from work on Google Cooper nattiest Eh installation to Amazon carbonized installation have some motive consistency. And then you you know you just have whatever. Don't like a cloud specific traffic permission systems and cloud services and you have the consistent layer of coober netease and you kind of the best of both worlds. And if you want to you can just spend your question Western so as school. I had a conversation with her. Somebody Harajuku recently. Where they're kind of thinking like will they have this kind of Harajuku platform right which is very great to work with you? Know if you're a small team then there are some big companies on her Oku But like coober Nettie is sort of. You know they're thinking about how to look. What do we do with Kuban do we do we try to make her Roku like a coup grenades platform or do we try to like you said? Hide the gears and wheels and stuff underneath Hiroko Hiroko. Whether it's Cuban Or not and just keep the experiences smoothest possible. It's kind of an open question anyway. We're up against time so I I want to ask you about one other thing. Do you think we're in an industry. Mass hysteria about micro services. Do you think that like why not have just a couple monoliths at each company. Why don't have like a monolith per team or a monolith for fifteen people or something like that? Are we going crazy with micro services just to sell more software well. I don't think it's to sell more software. I think obviously with any sort of a Hyped term like that. You're going to get some marketing. From companies. That that motivates behavior in somewhere another that might not be ideal for the entire fire industry but I think the intent of micro services was always to separate sort of development lines lines across teams and to to really to align to maximize velocity by. Not Having to you. Have everybody work in the same code base so if everybody works in the same code base say like how Google runs their software internally. You have to have a lot of tooling to make that possible and to improve the developer experience of that and I don't think a lot of companies have really invested that way the way Google has so get hub for example which I think is one of the primary reasons why open source has had a big resurgence in the last ten. Or fifteen years is modeled for you know a unit of of Code and deployment and management at the Repo level and so you can have have multiple repos and then have different teams working on those repos and they don't have to step on each other's toes and they can release software faster because they can release a smaller chunk of it at a time. And if you try to do that in a monolith you end up building your own building release platforms or frameworks or tooling to slice and dice. The repo up to be able to do that. So it's more work or expand on the automation rather than using existing tools that are at the repo level in order to do that so you can easily pull off. Most see is available on the market now and get them to build repo code for you like run a make file or something and and build a container with the doctor file and then publish that to registry and then can you know rapid into Kuban. Eddie's pretty quickly for most small micro services and that availability of tooling that you can get you know not necessarily early for free but relatively inexpensively to integrate allows for a developer velocity that you might not get if you have to build all the tooling yourself for a monolith. So I think there's still value still exists. I think there is a little bit of hysteria. Around making everything monolith. The choice risk right up front is like do we do Melissa. No that's crazy that's old school in legacy. We WanNA do micro services. Because that's the new thing and we've gotten past that now that everybody's like oh I want to do like functions as a service. And that's that's the new thing that's like my Nanna service and deploying that and so there's always a risk in taking taking things too extreme and I think I think monoliths have their place and they're useful for some things but they're not necessarily useful for everything micro services have the same thing they're good for you know Dividing along team lines or development lines released lines really so that you can release things in more granular chunks and then functions as a services is the next level up swear functions as a service works really good if you have like a small set of functions but once you get into the thousands of functions you. Have you have a devops problem where you you can't release and deploy an iterative on those in any reasonable manner because every team has a thousand each and so you spent all of your time doing the release management and deploying razz like micro services. You'd spend less doing that and in Melissa spend even less doing that. Provided you ship at all as one unit by think when people say monolith they mean different things they sometimes they mean I had a repo and I shipped one binary and Iran that one binary and some people will just mean like sharing a repo and building only a lot of different artifacts and then deploying them separately sort of like micro services or service oriented architecture Carlisle. Thanks for coming on the show. Great conversation thanks bricks reverie. I remember the days when I went to an office every day. So much of my time was spent in commute. Once I was at the office I had to spend time going to meeting meeting rooms and walking to lunch and there were so many ways in which office work takes away your ability to be productive. That's why remote work is awesome. Awesome remote work is more productive. Allows you to work anywhere less DP with your cats. I'm looking at my cats right now. But there's a reason why people still work full-time in offices. Remote work can be isolating. That's why remote workers join an organization like ex team. Ex Team team is a community for developers. When you join ex team you join a community that will support you while allowing you to remain independent and ex team will help you find find work that you love for some of the top companies in the world ex team is trusted by companies like twitter coin base and riot games x DASH team dot com slash S. e. daily to find out about ex team and applauded joined the company? If you use that blink ex team will know that you came from listening to software engineering daily and that would mean that you listen to a podcast about software engineering in your spare time which is a great sign or maybe you're in an office listening to software engineering daily. And if that's the case maybe you should check out X.. DASH team name DOT com slash. Save daily and apply to work remotely for Ex team at ex team. You can work from anywhere and experience a futuristic culture culture. Actually I don't even know I should be saying you work for ex team. It might be more like you work with ex team because you become part of the community community rather than working for ex team and you work for different companies you work for twitter or coined base or some other top up company. That has an interesting engineering stack. Except that you work remotely ex team is a great option for someone who wants to work anywhere with with top companies maintaining your independence not tying yourself to a extremely long engagement. which is the norm with these in-person companies and you can check it out by going to ex dash team dot com slash? Save daily thanks to ex team being sponsor of software engineering daily uh-huh

Google Cooper Netease Aws Kuban Production Kuban Eddie Software Engineering engineer Sumo logic US cloud facebook Amazon Cooper Nettie Carl Isenberg San Francisco
AWM Insights #26: Finding the Next IPO Unicorn?

Athlete CEO

06:39 min | 5 months ago

AWM Insights #26: Finding the Next IPO Unicorn?

"Hey everybody. Welcome to awm insights. This is a quick hit each week on timely investment topics. I'm Brandon April partner with a w m capital and today we're joined by co-host Justin dyar chartered financial analyst and awm chief investment officer without we're going to jump into a conversation to help you maximize your net worth Justin. Welcome to the show. Thanks Brandon. It's an honor to be a part of this podcast. I'll I'll do my best not to go too deep into the nerdy esoteric as it were that chartered financial analyst like to a live in awesome. Well today, we're not going to make that change on. Yeah, we're going to jump into a actually a pretty complicated topic. We're going to talk a little bit today about what it's like to take a company from private to public. There's been a lot in the news recently about spam Bots. We've talked about that in the past awm insides, but we just got some news on Direct listings and then there's always the ever-present IPOs that you know, the media just loves to blow out of proportion. So, you know, I think God They will be good to break some of that down for those that you don't know I P O as an initial public offering and it's been the most traditional way to take companies from private public. But we also know when we really dig in a CFA that chartered financial analyst side. The most of the value for companies isn't actually created at IPO or when they go public most high values actually created in private markets. And so just and I know this is an area use the word plot into I mean we saw we've seen a lot of companies go public. Can you just touch on a little bit the the data behind this and talk a little bit about where that value creation is actually made certainly so off yet like Brandon said the IPO or private to public life cycle or step in the life cycle rather super important within investing in the capital markets and in general those IPOs get a ton of headlines and and coverage in the media. But when you look at the data in general participating in IPOs, right out of the gate generally haven't favored the dog. Bester over the ensuing six to twelve months most of the time there's some under performance relative to the market like as an example looking at a. From 2001 to 2008. IPO is under performed by a little over 2% versus the market as a whole. So in general, they haven't been great Investments on their own now, there are some unicorns as the phrases bantered about which are the very very popular private companies of the world, you know, Spotify was one there's a recent company called palantir coming coming up to go public is another and these are the headlines offers for sure, but net-net on average IPO participation generally doesn't bode. Well cuz most of that value creation is Brandon alluded to happens in the in the private marketplace where Venture capitalists are reaping the vast majority of the reward and they're going public to liquidate their Investments and and reap that reward. And so again kind of keeping most of the gains to themselves as well. You mentioned palantir and they're going to take a little bit of a different direction and going from private to public is what I was reading. I mean, we still have some companies. I think it's an interesting time right now. We've got five companies that I know of taking that traditional IPO route snowflake Unity software jfrog Sumo logic and am well, but we're going to see a couple of companies take a different approach between a sauna and palantir that are going after this kind of direct listing and be able to raise capital in a new way. I know Bill Gurley the former Benchmark Capital Partner has been outspoken about the support of this and being able to create, you know, a raise money through this direct listing. Have you read much about that? What's your take on Direct listings? Yeah General. I like the concept and in the concept of motivation behind direct listings is super similar to this fax or special purpose acquisition companies that you guys have touched on in the past it you know, essentially you're trying to cut out the middleman. Yep. Right, where in the past maybe one could argue that part of the reason IPOs happened performed is because there is a middle man. That is somewhat, you know taking a piece of the pie. The direct listing is off of a direct to Market type approach to get getting your company traded on a public exchange and in general if you can do that within an industry or a Marketplace, you're generally going to take a trip some rewards from from that activity. So I really like the idea of direct listings now, will it improve the performance of public listings going forward? Who knows we'll have to wait and see but it serves as a more efficient way of companies getting access to the public markets. Now one quick comment traditionally people haven't companies rather haven't gone that route because it's not a way to raise additional Capital. It's just the way to put your company down the public market as is if you will, but recently the SEC just passed last week actually the abilities for companies to raise additional Capital through direct listing platform and that certainly took Probably going to take the attraction of that method going forward. I think that's great to know and and now these rules change obviously, you know, it's important to kind of keep in mind all the new avenues that you can get on investing. I think the big message here is just that you know, it's no different than what we've hit on in the past. The public markets are super efficient, you know, IPOs are certainly not withstanding or you know, the information now becomes public and so the information Advantage is somewhat gone. I'm sure that indirectly is led to some of these IPOs not performing as well as they probably should have I think kind of going back to if you have the ability and you have the desire to try to find out performance all of this just hits so much more on try to find your way into the to that private side and we've had the private investing side quite a bit. But remember there that the reason why all this value creation is made there is because it's also a place where a lot can go wrong. So you want to make sure you're investing with the right people and that dog I'd so Justin great debut here in the your first awm insights. Hopefully this has been a big value for everybody. It's a complicated topic. We definitely encourage you to reach out if you have more questions, I mean, we got forty four billion dollars worth of start-ups going from private to public right now. I think is the estimate so welcome everybody to visit awm insights.com to learn more about this topic and to listen to other episodes and certainly if you want to be the first to hear the next up, so please go on and sign up for our newsletter. And with that we're going to sign off here. Just remind everybody should stay hungry. Stay humble and always be a pro.

financial analyst Brandon Justin dyar awm awm insights.com awm Benchmark Capital Partner partner Bill Gurley Sumo logic chief investment officer Spotify CFA Bester SEC forty four billion dollars twelve months 2%