19 Burst results for "Venkat"

"venkat" Discussed on Developer Tea

Developer Tea

03:43 min | 6 months ago

"venkat" Discussed on Developer Tea

"Out.

"venkat" Discussed on Developer Tea

Developer Tea

05:44 min | 6 months ago

"venkat" Discussed on Developer Tea

"There are certainly times that i've experienced discussions with with engineers where they were bringing up the discussion as a matter of principle rather rather than a matter of practicality so in other words they didn't like that we were using a particular <hes>. Framework or something right and so so they would voice. That concern and in my mind is a manager. I'm hoping that we come away with some kind of action to take rhyme that. That's that's my hope. Is that for somebody to voice a concern. Hopefully there is either a discussion that resolves concern that you think this framework actually does work well for this particular job because of x. y. and z. Okay we're good to go right. That would be a good outcome. Another good outcome would be. Oh you're you know this you make a good point so i think there's <hes>. There's certainly you know. There's kind of these two outcomes that we could have. We could either have <hes>. An outcome where we resolve the differences. And we continue on the path that we'd planned <hes>. Perhaps you're missing piece of information i provided or i'm missing a piece of information you provided and then we resolve somehow and nothing changes other than our mindsets or something else changes. We decide to review our options again. We say oh you're you're right. This framework does have shortcoming or. Perhaps we don't know everything we need to make this decision <hes>. Thoroughly but sometimes in this is perhaps unique to engineers sometimes engineers want to have this discussion <hes>. Almost as a point of philosophical debate rather than action and. I'm curious if you've ever seen this in other parts of of your business or <hes>. With a team that you're working with if there's kind of a bike shedding the happens right where you're where you're kind of going over the subject over and over and trying to get everything right or trying to get your point made but really there's no action particularly that anybody is is <hes> <hes>. Trying to shoot for in the in the discussion. I totally hear you are being on both sides of that you know <hes>. In terms of being part of <hes>. You know discussions where it was happening with. No clear consensus and the whole thing like just festival sorta while the project doesn't go anywhere and also being on the other side where in being managed. Well i think it all came down to gladdy around the process of decision-making right. I think basically in any situation. There's going to be a big change coming up <hes>. If that isn't cloudy on who gets to own at <hes>. Who is responsible for bad transition for change for that new project to be delivered on time on the quality if it's not clear who is owning dot and then they get to make those decisions and it makes you make it very clear. Other people can contribute and add you know. Be part of the discussion but the final decision comes to the keep driving. Not and then dot is clear. Then i think things move forward on all these discussions at the end of the day makes hopefully informs the key decision maker better decision. And you know you even if you disagree you disagree income at the end and die do as a company at your best job in in giving it a shot so that we have you know not a self fulfilling prophecy or anything like that but actually have a valid the positive or negative. The places where i've seen a fail are the person introducing a very complex change but he complex kind of like re architecture type of stuff yes they are the ones driving but he has of amazonification way beyond the what they're immediately control. It changes the way how operation steam man do on call changes the way the how in a product teams just product management like it it has ramifications way beyond their immediate circle of responsibility and they really want to push through the project and they very quickly get to a point where you say look under go to a <hes>. It's your job to convince me not to write like dots man. I feel like the whole. Collaboration breaks down <hes>. And i think those are the kinds of situations which are much much cookie in a in a larger organization to deal with than your project has unintended negative consequences to a whole bunch of teams. That do that. You're not fully weighing <hes>. In in your decision making process but in most places it's wide by just making it clear <hes>. On whose responsible for making the decision and making sure that deposition only affects dot portion of the thing that they're responsible for a been dot is not clear at that cannot be achieved. Is where i think. A lot of these <hes>. You know the stereotypical shitting in our discussions happened because <hes>. You know one organization wants to make a change that has dire and a negative impact on a whole bunch of other teams.

vancouver vin
Leading A Team During Difficult Times with Venkat Venkataramani

Developer Tea

05:44 min | 6 months ago

Leading A Team During Difficult Times with Venkat Venkataramani

"There are certainly times that i've experienced discussions with with engineers where they were bringing up the discussion as a matter of principle rather rather than a matter of practicality so in other words they didn't like that we were using a particular Framework or something right and so so they would voice. That concern and in my mind is a manager. I'm hoping that we come away with some kind of action to take rhyme that. That's that's my hope. Is that for somebody to voice a concern. Hopefully there is either a discussion that resolves concern that you think this framework actually does work well for this particular job because of x. y. and z. Okay we're good to go right. That would be a good outcome. Another good outcome would be. Oh you're you know this you make a good point so i think there's There's certainly you know. There's kind of these two outcomes that we could have. We could either have An outcome where we resolve the differences. And we continue on the path that we'd planned Perhaps you're missing piece of information i provided or i'm missing a piece of information you provided and then we resolve somehow and nothing changes other than our mindsets or something else changes. We decide to review our options again. We say oh you're you're right. This framework does have shortcoming or. Perhaps we don't know everything we need to make this decision Thoroughly but sometimes in this is perhaps unique to engineers sometimes engineers want to have this discussion Almost as a point of philosophical debate rather than action and. I'm curious if you've ever seen this in other parts of of your business or With a team that you're working with if there's kind of a bike shedding the happens right where you're where you're kind of going over the subject over and over and trying to get everything right or trying to get your point made but really there's no action particularly that anybody is is Trying to shoot for in the in the discussion. I totally hear you are being on both sides of that you know In terms of being part of You know discussions where it was happening with. No clear consensus and the whole thing like just festival sorta while the project doesn't go anywhere and also being on the other side where in being managed. Well i think it all came down to gladdy around the process of decision-making right. I think basically in any situation. There's going to be a big change coming up If that isn't cloudy on who gets to own at Who is responsible for bad transition for change for that new project to be delivered on time on the quality if it's not clear who is owning dot and then they get to make those decisions and it makes you make it very clear. Other people can contribute and add you know. Be part of the discussion but the final decision comes to the keep driving. Not and then dot is clear. Then i think things move forward on all these discussions at the end of the day makes hopefully informs the key decision maker better decision. And you know you even if you disagree you disagree income at the end and die do as a company at your best job in in giving it a shot so that we have you know not a self fulfilling prophecy or anything like that but actually have a valid the positive or negative. The places where i've seen a fail are the person introducing a very complex change but he complex kind of like re architecture type of stuff yes they are the ones driving but he has of amazonification way beyond the what they're immediately control. It changes the way how operation steam man do on call changes the way the how in a product teams just product management like it it has ramifications way beyond their immediate circle of responsibility and they really want to push through the project and they very quickly get to a point where you say look under go to a It's your job to convince me not to write like dots man. I feel like the whole. Collaboration breaks down And i think those are the kinds of situations which are much much cookie in a in a larger organization to deal with than your project has unintended negative consequences to a whole bunch of teams. That do that. You're not fully weighing In in your decision making process but in most places it's wide by just making it clear On whose responsible for making the decision and making sure that deposition only affects dot portion of the thing that they're responsible for a been dot is not clear at that cannot be achieved. Is where i think. A lot of these You know the stereotypical shitting in our discussions happened because You know one organization wants to make a change that has dire and a negative impact on a whole bunch of other teams.

"venkat" Discussed on KQED Radio

KQED Radio

01:41 min | 1 year ago

"venkat" Discussed on KQED Radio

"Death toll to thirty six the seven story building under construction filled on Friday in the coastal town of cap about one hundred miles southwest of in on him there's a growing body of research that shows a correlation between poor economic conditions and poor health a new study from the university of Pennsylvania found that nation wide counties were auto plants closed so OB would overdose deaths increase from member station W. H. Y. Y. Nina Feldman reports the researchers wanted to see how demand for few aids might be a function of economic decline a theme Venkat Romani is the lead author of the study for a long time automotive jobs were actually to the middle class the large auto plant closure effectively not only destroyed the economy of a local area but it also has implications for how the fabric of society of gable to me cancel the researchers looked at counties across the industrial Midwest and south where the bulk of the work force with employed in manufacturing they found in places where plants closed the rate of opiate overdose deaths with eighty five percent higher than it would have been otherwise for NPR news I'm Nina Feldman in Philadelphia authorities are searching for two inmates who apparently escaped from the Mississippi state penitentiary at Parchman they were discovered missing early Saturday five inmates in Mississippi have died in prison violence in the past seven days with three of the deaths in parchment officials attribute most of the deaths to gang violence I'm Nora Raum NPR news in Washington support for NPR comes from NPR stations other contributors include the crazy foundation.

W. H. Y. Y. Nina Feldman Venkat Romani Parchman Mississippi NPR university of Pennsylvania Philadelphia Nora Raum Washington
Fast Five: David Horton, President of Schweizer RSG

Aviation Week's Check 6 Podcast

08:49 min | 1 year ago

Fast Five: David Horton, President of Schweizer RSG

"The s three hundred helicopter traces. Its lineage back to the Hughes to sixty nine it received its FAA certification in nineteen fifty nine the US army bought some eight hundred of them for pilot. Training called it the osage over the years. Here's the little helicopter was owned and built by several manufacturers and in two thousand eighteen was purchased by a new Fort Worth Texas entity called Schweitzer RS jeep and. Today we're speaking with its president and managing partner. David Horton so I just wanted to talk about what's going on with the new Schweitzer and what's going on on with the three hundred thing that occurred to me. David is as we discussed. The three hundred is hardly a new design. You obviously think it's still very during competitive and I'm curious what about it after fifty years makes it competitive in two thousand twenty a couple of things. I is the opera ability. It's useful load operated do helicopters from fish potting range wildlife management the cattle mustering who law enforcement. Uh they actually use them for harvesting Christmas trees and training and one of the most robust trainers ever produced the other thing is the safety of a helicopter. I I got a substantial. Thank you reckon with best in class for the class that we operate in and I think those two things make it a very viable so helicopter. I know durable there. Lots still in operation or third in production stopped ten years ago helicopter. Belkin eighteen you know like the fifteen sixteen thousand sixteen timeframe and so it wasn't quite that long ago but there's only a handful bill but yes they have been around for a long Okon like most helicopters you replace. Parkas needed is not a complete overhaul interval. One time in operation. You can still get hard pale mainly because the retirement of the model though. The army is hard-pressed. Hell but we think somewhere around seventeen under this what we estimate. I assume helping support those as a part of your business plan part of the attraction of taking control right. It's pretty straightforward the first. The thing was is to support the existing operators so that they could depend on us to keep their fleets operational. That was first and foremost the second thing was to in the The new production going get the new production line and all the things related to that production certificate supply chain vertical integration. Take all of those things set up in order to be able to produce new helicopters. Then the third thing is to upgrade the helicopters. Currently have you know whether it's a different power plants or Avionics suite or just the vehicle management system bringing into the twenty first century on the things that the pilots in this day and age would look forward or to Saying that you know the modern helicopter when will we see taxes made S. three hundreds. We have started. Actually this week put together together. Are I e model and the ideas. We're going to probably put together four over the next several months. And the reason it's GonNa go slow as the EPA has validate eight hours. Kimberly instructions planning so to speak to produce the helicopters the FAA has worked with us very well so far and they're very excited about helping us get the next level of production support by you know helping US produce the entire helicopter I think after the first three or four or five helicopters we should be able to have have a what they call it. A limited production certificate. How many do you think you'll produce in here? We're going to win for the families flow. They maybe next year and maybe three times that much the year after probably fifty to seventy five a year would be a good average. They're actually the two models of the three hundred right it's three hundred c and the CBI correct and how do they differ. Hey Model Venkat you. Copy model of our helicopter has a higher horsepower. Engine Mexico's way to twenty fifty. And it's the helicopter that a lot of the military trainers to use around the world and so the CBI is lower macro swayed helicopter. Seventeen hundred fifty pounds and it runs the engine that is a a bit lower horsepower. Trade off though is primarily used for training helicopter although people use them for other things around the world but jacked up wedding costs. You'll possibly twenty five to thirty dollars. Our last thing speaking of dollars. What's the price of a three hundred and see beyond we think the three hundred thirteen they can be ours can be in the mid four hundred fifty thousand dollar price way somewhere in there the TV? I would vote four four or five four ten. Is it a three place. Helicopter both to be seeing is definitely a frequent helicopter when we configure it at the left hand drive it can be a left hand drive or right-hand-drive if it's bigger as a left hand drive there'd be a free fights helicopter. CPI for developed is to place helicopter Walker because the primary Pollen Command would be sitting in the right. So what's the market reception so far. You've got some motors. They we're actually able to contract for twenty seven helicopters love them. CBS and then that time we have received notes of interest. I I guess for people like listening to you are counts proxy another forty Salat orders. Considering you haven't even begun yet happens AH happens bill probably two three times a week. We get Paul's about wing to get into a helicopter. And so while we want to go faster than slower on the Production Avenue helicopters. We have to take a pay where we feel comfortable about what we're building and our partners. The best player comfortable about what we're building. You actually acquired three models from rights three models from Lackey Sikorsky. There's the three thirty three as well. Is that likely to go back into production. That's turbine helicopter right. Yes it is. Our Turbine powered helicopter. Three thirty three. We think you've got leg but we've got to do is we gotta find a customer and and have a pretty decent order a pretty good back before we start producing the three three three. We do have a demonstrator here and we will be actively working They'll That helicopter over the next several months years in order to build the backlog we think the helicopter does have some legs. We think it's gotTa play. The world especially for military trainer depend on what the market perception is. If we can get out there and sell it and we can provide something that we think is the value for things and then we will put the thirty three and when do you think you might be getting your production certificate for your operation in the field. Well we're planning on the first quarter of twenty twenty but I think you know everything has gotta go perfect in order to meet that date again. We're not trying to put too much pressure on ourselves because we really would rather do it correctly and makes it a bit you know as we start the production that there's no more kinks or any kind of issues with regards the production of the helicopters. I plan is the first quarter of twenty twenty and were dropped little hard to get that done if it bleeds into the first by the second quarter. Uh He'll be okay with that. There are two production certificates. We're talking about right isn't one from parts and the whole aircraft and all the thank certificate what they do is they limit to production tickets to what they call articles only for left currently what we have articles only production certificate once we start predicting the helicopters as the confident that we can build the helicopter sneak. Article only restriction will be removed the actual unfamiliar with this on articles. Only can you produce parts and sell them exactly what we can do. We can produce every part the helicopters for the second part of that is putting all these parts together to make flying sharing that the second part of the production certificate but yet we have the authority to build every art our helicopter. So are you in business now. Making parts and selling them for businesses to seventeen helicopters tale hard every day to manage our supply chain cups. I feel you're in a helicopter hub aren't you. You've got aerospaciale there and bell's not far away. You got some veterans around you. You know. That really helped him point of you. Know we've been able to attract

FAA United States David Horton CBI President And Managing Partner Fort Worth Texas Schweitzer Belkin Hughes Army Avionics Lackey Sikorsky EPA Salat Bell Kimberly Mexico
"venkat" Discussed on Venture Stories

Venture Stories

03:11 min | 1 year ago

"venkat" Discussed on Venture Stories

"Suing. He doesn't necessarily create happiness like it's one of the things that are somewhere bar the buzzer like an pleasurable things pleasurable and they should be ready grishin but if you respectively focus on just that johnson's are you're not going to go anywhere beyond that and what does capitalism need beyond legibility <hes> take mainly leaving legibility alone like it's possible to make certain legible without trying to kill what's illegible around that legible thing and so long as you you have. That's <hes> lizzie. Fair attitudes what you make legible versus what you can leave alone to do its own thing. Think's gonna take care of themselves but if you try to kill do whatever you don't understand batsman things get problematic totally <hes> inclosing talk about the <hes> the elder blogging you were talking about earlier or act back to what is sort of the idea behind behind act to in. What is that look like. I'm still figuring it out. It's like like one of the responses been <unk>. I tend to be repealed man on twitter. Sometimes i complain about this stuff and i say things like oh. I'm slowing down. It's hard for me to get good ideas and blah. Blah people often respond with one. I think are interesting suggestions which toured all away and start again south disrupt up and i think those are the wrong answers because they are available to have a second act one which is not the same thing as having a source act too so i think it's actually interesting challenge to look at. I'm going to be forty five zero so look at forty years of life and say aren't so better worse. I've lived a certain life and twenty five years as an adult and a certain sort of bilo shit in my life that has to do with that's all for good shifts <unk> stuff the twenty five years worth of stuff and you can choose to refund in one of ribs you can wipe the slate clean and try to pretend twenty again can have redone and a do over of act one which i think is fundamentally in an interesting thing to do. All you can say all right here. I am it is what it is. Where do i go from here. What are the interesting places. I can get to by virtue chew of being where i am now and that's the challenge and it's <hes> it's something i find interesting to think about like i don't wanna i don't wanna do would've my life from twenty to forty five. I want a first instance of my life from forty five to seventy or what it is. It's been a fantastic episode bent cat rob. Thank you so much for for coming to podcasts for people who want to go deeper into your work <hes> they can follow you had v._g._a. Are they could <hes> check out ribbon farm. Where else might you want. The <hes> i have <hes> email list on consulting that is called art of gig and a half breaking smart which is when i do short podcast entertainment <unk> semi he week. I'm all over the place and as a thank you so much for the podcast. If you're an early stage entrepreneur we'd love to hear from you. Please hit us up at village global doc- slash at work catalyst <music>..

twitter johnson twenty five years forty years
"venkat" Discussed on Software Engineering Daily

Software Engineering Daily

02:32 min | 2 years ago

"venkat" Discussed on Software Engineering Daily

"And what you're seeing people build with rock set since they have a newer solution. So, yeah, what are the what are the problems that? You're seeing the data engineering problems. You're seeing in your customers, and how do those compare it to what you saw at Facebook. Interesting. I think people dry due largely build in a similar kinds of applications. But I think the biggest difference would be in what customer see, and what happened within Facebook dot, the the overall couldn't go, you know, the quality of the data the messy of the data is very very different. I think in the in with our customers. A lot of the time. They're trying to connect to party data sets with some internal data sets that they have right. This is a problem that is even harder than the claims data problems that I think honestly, we had it feasible because largely the data that was created with Facebook was largely created by the Facebook application itself. And it wasn't like third party. Data says at Facebook was trying to make sense out of. But here whether it is a marketing up operations applications if somebody's trying to build on marketing deligence what's happening in the market what's happening with my competitors. What products is moving moving to answer these kinds of questions, you know, enterprises have to combine external third party data sets, and and semi structured data with internal data, which is I think really interesting kind of challenge. What very different than what I think we saw Facebook. So that is something that it's been very interesting for us that that these kinds of marketing diligence applications and. Sales operations customer support operations in the world. No needs to really is a lot more complex than even what I think some of the backend up ligations that Facebook dealing with. And so those are the kinds of use cases where I think we really shine. And so those are the kind of use cases that customers are doing whether it is data signs on semi structured data. Building personalization engines on on these semi structured unstructured data sets all kinds of IOT sensor data use cases, and I say said like automation business operations, automation kinds of cases. So all of these things where I think almost all of them are are sort of in certain ways, very different, and and some ways much harder than honestly what we experienced Facebook. Vic it thank you for coming. Out software engineer daily. It's been really fun talking to you about Roxette. Thank you. Yeah. I think we have a, you know, go to dot com. That is get started for free. If you're interested in that you can get started for free. There's a feature in all you.

Facebook software engineer
"venkat" Discussed on Software Engineering Daily

Software Engineering Daily

02:57 min | 2 years ago

"venkat" Discussed on Software Engineering Daily

"Somebody needs to configure. No, you're dealing with multiple vendors and multiple hierarchical storage systems that all might not have the same kind of performance from one installation to another. And so it's kind of very hard to build a working system on hierarchical storage on Prem. It's not that. It's not possible. It's very very difficult to get it working in one setup in one center, but building offer product that would just work with a multitude of them and still give you really good performance is very very hard. That is why I think it hasn't really been done before. But then you look at it in the cloud. You know, you have all different hierarchies, and there's a very nice of cost to performance, and that is very well documented, but understood on what does it he didn't see for your local ram locally at that storage block storage aboard rock storage in SSD's and would rockstar Asian discs on all the way to things like S three Amazon S three and other blob stores and glacier. And what have you the hierarchy is? Not only deep, but also anyone understood and very will reliable that something that you can actually take a hard dependency on building a system software, and if your software can leverage it now, you can actually make let's say two copies of your data and keep the two copies in different indexing organizations. So that you can actually know autumn speed up a wide spectrum of quickies without asking too, many configuration and setup questions from the user. And again, I think it's really the cloud that the readily available cheap and hierarchical storage said was available in the cloud really enables this kind of technique and without doubt. I think it'd be very hard to do something like this. We had a show a while ago with somebody for mongo DB where we were discussing database query performance and how to speed things up or the trade offs. You can make and one of the things that the guest was exploring was the fact that when you keep more indexes over a certain. Piece of data or over sets of data the more index is you keep the longer. It can take to have consistent updates when you ingest a new piece of data because when you ingest a new piece of data, then you have to go and update all of the indexes that are over that piece of data. So if you keep too many indexes, and you have high consistency requirements over your queries than you can get into this issue where whenever you're ingesting a piece of data you're locking the database while you're spending time. Updating those indexes is that another trade off that you face when you're keeping this wide variety of indexes over the data absoulutely. What also I think what kind of indexes. So what you said is literally correct in a traditional. What is generally consider the mongo DB like Sharda databases or call term punishing databases, the term partition databases. Yes. I think the Knicks are built in a certain way where the more and more you have them. Yes. You're..

Knicks Amazon
"venkat" Discussed on Software Engineering Daily

Software Engineering Daily

03:23 min | 2 years ago

"venkat" Discussed on Software Engineering Daily

"Use it for an hour undertone into your seventy hardware provider. That's not a thing. Right. So you have the capacity plan that you have to like, you really have limited computational quarter to do anything. And then all the software optimizations make sense where you know, you have these machines squeeze every last drop out of an because that's all you're going to get for the next. I don't know. Years and the more efficient. You are the more I can extract out of this door investment that have already made an almost all of the software. That was written in the cloud data way, build without mindset. But in the cloud, what we'd be saying if you can find something, you should always something as long as you're schedulers on your infrastructure again, but it quickly grow and shrink and shrinking is just as important because you're gonna die out all the CPU and all the compute and all the storage when you're not using them. Then you see him a lot of money. So you know, if you could paralyse it, and you could without increasing the costs if your software is pretty good at doing that. Then you should not only ask for the same kind of efficiency and the same kind of cost dynamics. But it should also expect things to be much much faster than the cloud than what you're used to. Because there's no reason to just do the slow route just because you happen to have installed. Your software in the cloud and only three machines if your software. On quickly grow and shrink based on your demand, and based on your bizarre SLA's, then that means it's time to change your software in the cloud and not be stuck with slow performing systems in the cloud. So that's kind of like the fundamental realization that we had which is why I think we have done a lot of software engineering based on these kind of cloud, principles dot the absurd. And our our software architecture is built ground up in the cloud to take advantage of the fluidity of the hardware that is available. So to some degree you can use these this abundance of resources to speed things up to have a wider degree of of how you're indexing things. So for example, if you ingest a bunch of data in rock set as I understand, you're keeping a bunch of different indexes over that data. So anybody who's worked with databases if you wanna speed up database, you're oftentimes just adding an index that is. A faster way of running a certain type of query, and if you can maintain more of those indexes than you can have more performance responses to more types of queries is that an accurate abbreviation of what you're doing to some degree Iraq set. Correct. That's very very good abbreviation of what we're doing the end of the day. You know, we are a cloudy next in company show. All we need is. There are other ways that we are also very interesting, which is like we don't you don't need to describe the shape of the data all of the shape of the qualities, and we automatically indexed away. Schema your on index index in more than one way. So that a wide spectrum of your could he's fast. So that's exactly Greg. What you said the cool thing is why are we able to do this? Now, why wouldn't a multiple indexing strategies work in data centers? And I think it comes down to again being able to leverage, you know, the storage hierarchy that is available in the cloud. Really? If you were to go and build a software that acquires multiple combination of storage systems one for. I in hot data on won four gold data and.

Iraq Greg
"venkat" Discussed on Software Engineering Daily

Software Engineering Daily

03:07 min | 2 years ago

"venkat" Discussed on Software Engineering Daily

"Lower because the dire. There's like a big percentage of logged in population wants to interact with that one piece of content and single server just would not be able to deal with that at all. Even if that is the only piece of content that our service serving. And so those are the kind of challenges that kind of extreme hodgepodge that happened on the server side, and huge fans that would happen on the on the client side because you're not just going logging into one server, you're looking at your asking for some small piece of content from lots and lots of different machines and very quickly and all of them come back. You're trying to render kind of Facebook newsfeed like data rich expedience and all of that required. Tremendous scaling tremendous kind of systems infrastructure, and the what in too many places we can look up to and say, hey, we should build it like that. I think a lot of the ideas had to be like infrastructure had to be kind of like by Facebook for Facebook during those years. Yeah. And this was pre cloud computing or just around the start of cloud computing. You didn't have access to really good. CDN infrastructure. I don't think or CD and infrastructure that would serve Facebook purposes. So anyway, you you were an engineering director at Facebook, eventually after nearing around the end of your eight years at Facebook, you were building data systems for profiles and friends and messages and photos, and you took those lessons from the early days of building cashing infrastructure, and the, you know this chat system, you turn this into eight years of working on data infrastructure when you zoom out, and you think about the perspective of the eight years, you had at Facebook of building all that data infrastructure, and solving all these problems and seeing in some sense the future because a lot of the problems that Facebook had in its first eight years, or in in the eight years that you were there are things that people are having to deal with on a regular basis these days, so what are the big lessons? What are the most impose? Takeaways from the engineering time, you spend a Facebook. What are the takeaways five zoom out and look at perspective? I think was blessed to work with some amazing people in those teams in infrastructure and product, and engineering dot I would say Facebook was not the full social network in the world. But it was definitely the first one that scaled to a billion active users, and there was a lot of team. I did a lot of hard work Ladoga be true, and the data infrastructure team that one thing that I, you know, I'm very proud of that is big takeaways that laboratory matters and performance matters that I think very proud that Facebook was largely up through the entire those eighty are in a dime period on all of you know, my team when necessarily for it and definitely knocked sufficient by any means, but definitely necessary. The other thing that I would say the philosophy of what we were doing at my previous team on Facebook was to make it very easy to Bronx services on Facebook. Right. There's lots of data coming in. And we really wanted, you know, product engineering to be. Bottlenecked by the creativity. Right. What does the most engaging allegation could be build on not really be curtailed by?.

Facebook engineering director eight years
"venkat" Discussed on Software Engineering Daily

Software Engineering Daily

04:02 min | 2 years ago

"venkat" Discussed on Software Engineering Daily

"Of the people started actively cutting the number of endpoint hits just from the product to the back end was high out of all of Facebook put together, and so we had to basically gone optimize that so that because every page load is explicit step from the user, except that chatmoss cannot be as expensive as a as a friend and page load. And so we ended up just figuring out tons and tons of things like that. And then, you know, find one bottleneck, and then we fix it on the next bottleneck, and it's kind of. Like telling to say, you know, we in dock mode for about maybe two and a half months or something like that. And then we cranked up. The all these specials like we had these fake specials on how many users should we should be pretentious pretend chatting right now, and how often should be should they be sending messages and things like that. And we had conjured up by the end of the second month high enough on things working very well to the point where the day we actually launched to all the all the users. We were so confident it's going to work because we are destined on even much harder higher scale than what actually happened on on the launch day. So after building chat you worked on scaling meme cashed, infrastructure and other aspects of cashing explain the role that cashing played in Facebook's early days. Why is cashing infrastructure so important to Facebook? So m cash was just amazing back in the day. I feel I often tell the story where there's a lot of services that phase. But Bill there was only one thing that actually to risque. Killed to Facebook scale. And that was McKechnie everything else was using them cash Diaz scaling kind of like layer. Right. So say you build a microbes for news feed or your Microsoft's for ads Irving, and you just be suddenly having to scale to millions of operation, we're second. But a lot of the repeating the same operations over and over and the actual aback, and that is doing that operation can't really do that that skilled, and so you use men casually as Leukocyte cash, and so it was very very important that we had a very high-performing highly available. Highly reliable cash d not just for one part of the product. But for all of Facebook in a back in the day of cash, they used to go down for you, even I don't know. Like, let's say tens of minutes most of Facebook for a lot of people would be a blank page. We can't render anything because the other back that men casually was kind of like fronting controversially take the kind of two. On the load that cash would have no problem Illingworth because it's not really trying to do too much work too much compassion to process request. But everything that it was fronting in all the backend services that was fronting put it a quest. Hoster do a lot of work, and we wouldn't have been able to scale any of those back ends doing the, hyper growth years of Facebook in the two thousand seven two thousand and two thousand nine and even much later than that without the power of men. Catchy. When I think back to that time Facebook, I think was the earliest highly interactive multimedia product that was so widely used I'm trying to think of other products in that category. But I guess there was there's kind of g mail maybe Google search, maybe some online games, but you probably didn't really have patterns to follow. Right. There weren't really any companies that had done stuff like this except maybe Google. No. That's a very good point. I used to say that as in all of my dogs back back when I was managing online. Data infrastructure Facebook where the social networking kind of backend workload was just very very different mostly because of the huge fans and the fan outs. And what I mean by that is, you know, you you have let's say a search back in and there's a lot of read only index, right? There's a lot of hard challenges in order to build a web scale search engine like what Google that both in terms of ranking problems, and in terms of just scaling the online serving index itself to be so fast. But the end of the day, you still have to figure out how to index the web and keep it replicated in many different places..

Facebook Google Microsoft McKechnie Bill Irving Illingworth Diaz
"venkat" Discussed on Software Engineering Daily

Software Engineering Daily

03:28 min | 2 years ago

"venkat" Discussed on Software Engineering Daily

"Tara money. You are the CEO and co founder of Iraq set. Welcome to software engineering daily. Thank you. Thank you for having us. You were at Facebook for eight years you start in two thousand seven and I'd like to spend a bit of time talking about scaling Facebook. Because you were first on the team that built the back end for Facebook chat. And then eventually, you got into cashing infrastructure data infrastructure a lot of different things. But let's start with that. I chat product that you worked on. So Facebook chat was a unique product in relation to something that people would build today. Like if you build a chat product today, you wouldn't have scale from day one, but Facebook was already a popular product. So when you were building Facebook chat, you knew that this had to be scalable from day one. How did that affect the architecture way back in two thousand seven oh was a like things about a lot of old memories? Yeah. Thanks for the you know. Also known Cizek trip. It was a very challenging project. Mostly because it was a completely web based chat right back in the day. Ajax was just the thing. And we were doing these long poles, and that kind of pardon was not well understood there were lots and lots of issues both from the client side jealous. Crypt on the Balkan. Scaling. I would say yes skinning was a we spent a lot of time making sure that the product would work up on launch. And so we came up with this idea to sort of do a dark launch, which is really how we did scaled this thing because there's no Mon the workload that we could create our own sandboxes that could mimic that point forty fifty million monthly users all actively using the product. And so the way we actually did scale destined back then was to launch the product on all decline side without actually any UI and sending kind of like randomly pick a friend and sendo fake kind of a message or that somebody's actually typing on sending attack message. And then tracking how many? Of them were successful and whatnot. A scaling kind of like we call the dark lunch and doing dark launch was extremely important. Because I think we thought we were Don thought it would work, and then we did an arc launch and realize nothing really worked and we spent about two months from the navy Todd. We were ready to debut would actually ready. So you had you basically had the client you, deployed the client. But it just wasn't rendering anything on the UI. Yes. And then randomly you would have a message that would be sent technically from my account to maybe your count or would send a bunch of messages, and we would have a conversation, but you're just mimicking conversation, sending crite or. Hello. I'm I'm Hello. I'm thinking, and and and you just did that at enough scale to find out if it would work without actually revealing to users that that was happening. That's pretty interesting. What kinds of issues did that reveal? Oh my God. I mean, every bottle of the stagnating really worked at we thought. Everything would work. And nothing really worked before I did the dark launch. I mean, everybody in the stack that we're some skill liberty shoes, and I'm gonna concurrent connections that we thought we would any for the memory Ord had connection suddenly crashing a bunch of Buchan servers. We bought we were mostly on patches end up HP stock back, then the Facebook kind of like the HP application running and the for message, or what had processing overhead was way too much that we realized if enough people started chatting, maybe even like a small percentage.

Facebook Ajax Iraq liberty shoes CEO HP Cizek navy Todd sendo co founder Buchan Don eight years two months
"venkat" Discussed on Software Engineering Daily

Software Engineering Daily

02:59 min | 2 years ago

"venkat" Discussed on Software Engineering Daily

"Advertising technology were creating fast moving semi structured data. The user base of Facebook was growing the traffic was growing and the volume of data was growing and the popular methods for managing this data or insufficient for the applications that developers wanted to build on top in previous episodes about data platforms. We have covered similar difficulties as experienced by Uber and door dash incoming data is often in Jason which is hard to query for large data science jobs. So the Jason data is transformed to a file format like Parque and this requires an ET L job. Once it is in Parque file on disk in a data lake the access time. Is slow to query the data efficiently. The data must be loaded into a data warehouse which can load the data into memory often in a columnist format that is easy to aggregate. Imagine being a developer at Facebook or Uber or door dash and trying to build a simple dashboard or machine learning application on top of this data platform, where do you find the right data? How do you know, it's up-to-date? And what if you don't know the shape of your queries ahead of time? And you haven't defined indexes over your data. The access speed will be too slow to do exploratory analysis the rar many steps in this process. And each of these steps creates friction for applications developers that want to build on top of big data. It also creates opportunities to lose data or have time discrepancies between those pieces of data since even Facebook was having trouble. Managing this problem of the data platform, vinca figured that. There was an opportunity to build a company around solving the data platform for other software companies vinca is the CEO of rock set a data system that is built to make it easy for developers to build data driven apps in rock set. A piece of data can be ingested from data streams data lakes and databases rock set. Creates multiple indexes and schemers across the data because there are multiple models for querying rock set can analyze an income inquiry and create an intelligent query plan for serving that query vinca joins the show to discuss his time working on data at Facebook. The untapped opportunities of using that data the architecture of his company rock set. And also the changing economics of cloud resources. How it makes it profitable to build a? Application like rock set where it might not have been possible before cloud services made storage and compute so much cheaper. We are conducting a.

Facebook developer Jason Advertising technology CEO
"venkat" Discussed on KOMO

KOMO

01:42 min | 2 years ago

"venkat" Discussed on KOMO

"Romero, KOMO news spending, hurricane on the east coast already having impact on blood supplies here. The northwest. Dr Jim AVI schon who heads bloodworks northwest has already shipped thirty four units of blood to the Carolina in Virginia areas. He says that will further depleted already perilously, low local blood supply come in and donate today. So that we can bring our inventories back up to where they need to be and support areas of the country affected by the hurricane pieces. There's a particular need for typo blood along with the be negative AB negative. Dr Shawn says, if they had more inventory, they could share more inventory, but he stresses that the primary need is here in Seattle challenge to Washington cyber stalking law will get another day in court after an appeals panel says a federal judge was wrong to dismiss the case. Komo's Ryan Harris talked to one of the attorneys arguing the first amendment case man making the challenge. Richard ranier. Bainbridge island. Brought the case after community activist Clarence more walkie claimed right near Sens. Repeated criticisms of his positions were harassment and got a protection order against Ryanair Aersson Ranier since attorneys Venkat Balasubramaniam says now the federal judge in Tacoma will have to hear their claims that is a public figure criticism of Mara walkies positions as political speech, essentially, our client, just voiced his disagreement with the subjects treatment of a particular topic. And it wasn't even a personal topic attorney general Bob Ferguson sent KOMO news. A statement that reads in part, quote, the Washington state legislature passed the Washington state cyber stalking law to protect Washingtonians from harassment. Online has attorney general my job is to defend laws passed by the legislature. Brian harris. Komo news. Komo news time eight thirty four triple A traffic.

Komo Dr Jim AVI schon stalking Richard ranier Washington harassment Dr Shawn Romero Brian harris attorney Bainbridge island Ryan Harris Ryanair Sens Bob Ferguson Venkat Balasubramaniam Clarence Carolina
"venkat" Discussed on Software Engineering Daily

Software Engineering Daily

02:48 min | 3 years ago

"venkat" Discussed on Software Engineering Daily

"And for that, we have things like azures fear and how we can integrate some of those concepts into the edge paradigm itself is something that's super exciting for me. What other kinds of tools around IOT. We've got a couple different tools today, so I t- hub this fish thing. That's a gateway between your cloud and your deployment. We've talked about the edge platform talked about virtual Hewlett. These are some very specific technologies. What are the kinds of tools doing perhaps tooling around machine learning at the edge? What are the other areas of tooling the expect to see in the next five years? Some of the tooling, I would say this to make it more accessible the edge computing itself. So again, if you look at it from developer. Perspective, this could be just extensions and tooling for for building these modules edge computer self, and to make it really easy to be able to talk to the cloud. And that's something very interesting. A lot on, we have a whole bunch of visuals to the accord extensions with IOT edge. That's only getting more and more important. We want to bring in other technologies, visual studio into the as well as well as more sophisticated debugged to right so that debugging monitoring. So these are kind of tools that we feel can be improved or need to be almost retard in some cases to manage deployments when something goes wrong, how do you get notified that something's not functioning the way. A lot of the tools around that and one of the tools around debugging issues as they come up. Right? So these are two important things. I think that we will invest a lot of time on and as an industry, we will need to look at as compute scales, not just in the cloud, but also these various devices on the edge gala. Thank you for coming out software engineer Taylor. It's been really fun talking to you. Absolutely. Thanks. Thanks. Test collab- is a modern test management solution which integrates with your current issue manager out of the box for all development teams, it's necessary to keep software quality in check. But testing is complex. There are various features to test various configurations and browsers and platforms. So how do you solve that problem? That's where test collab- comes in test. Collab- enables collaboration between software testers and developers. It offers wonderful features like one click bug reporting while the testers running tests, collaboration on test cases, test excecutions, test automation integration and time tracking. It also lets you do a detailed test planning so that you

cloud developer Hewlett software engineer Taylor five years
"venkat" Discussed on Software Engineering Daily

Software Engineering Daily

04:18 min | 3 years ago

"venkat" Discussed on Software Engineering Daily

"Learn any new API's or any new technologies around are not and not hub or any of those technologies can use your existing knowledge on deployment now start adding edge pieces to it as well. Got it. So in the cloud, you have you would have a coup deployment and Cooper as deployment in the cloud that you're operating against has virtual culet and the virtual cube it interfaces with the IOT hub in the candy factory, and the IOT hub is fulfilling the interface, the coober Netease interface that you are the your programming against in in the cloud. And so you don't even necessarily need a cougar. Netease deployment at the edge in that example, is that correct? Absolutely. Yes. You don't. So as as we know, Kubina deployments can be a little bit in ward for good reason. You want higher will ability you want to be able to, you know, sprint how these kind of consensus systems being set up. But once you have an edge device, this is a really easy way to now not really need Kuban. It is on the edge but really get all the benefits of using software. So in that world, is there any advantage to actually having coober netties running at the edge? Yes. So one of the advantages high availability, right? And that's really the Britain butter of cluster schedulers. Where if you are deploying onto a mission, critical edge infrastructure, you cannot have a single point of failure and what we have at the moment, something that we are building as we speak, but what is out there and most edged solutions out there are really focused on one device, but you want to almost think of offered as a service that you can deploy deploy to the cloud, deploy to highly available cluster on the edge and now on the edge. When you have those high availability scenarios, it may very will make sense to have Kuban be the scheduler that is managing that device or that cluster. He. Yeah, yeah, that makes sense. So nowhere near the end of our time, I want to get a little bit of perspective on the future. So kind of showed my hand the little bit earlier about what I'm excited about in terms of the the stuff involving connected cars and drones, and and these sorts of things and being able to control these things from software, give me your perspective on the future. Like how far are we from that kind of stuff that I described or do you have a different picture of where we go from here and what I f software deployment will look like in the near future? Yeah, definitely. So the way I look at it in the future is almost not tie it just not at all rate in the sense that it should think about the problem at a at a larger skit, it's just that you have this huge amounts of data that is being generated on the edge, and that's only getting more and. More true as we as progressing to the future and you want you want to be able to apply compute to it. So if you look at how Satyarthi or put sick, you have the intelligent cloud and the intelligent edge. That's it's not the intelligent. I o t its intelligent edge. So what I'm excited about is some of the technologies on the platform, the edge Runtime the containerization, these various open ecosystem that is building around the intelligent edge to be almost a free standing concept by itself. What we have currently is yes in the context of not. But in the future, as things become more capable computed cells becomes more capable being able to think of it in a broader terms to be able to apply compute to this wasp number of systems that come online right on the close to date is that is really exciting to me, and this is like the full rein to it. But there whole bunch of other pieces like we talked about. Where you have to talk about high availability, you have to talk about the other manageability of these things. Curiosity is a huge concern and how do we secure this?.

Kuban IOT Netease Cooper Britain Satyarthi
"venkat" Discussed on Software Engineering Daily

Software Engineering Daily

01:39 min | 3 years ago

"venkat" Discussed on Software Engineering Daily

"Real world like interfacing with your oil rig or your candy factory? These can have very serious physical impact on the world. So what does it do from a security perspective? Is that change the the network architecture. Yeah, it does not only network architecture. I will say you wouldn't change the change the platform architecture. For example, if you think about security center, right? So there you can rely on that particular were to be in a secure location that is guarded twenty four, seven, and is not really difficult to get physical access to, but an NGO I could be a smart device that is sitting on the factory or someone could just disconnected, put it in their bagging walkaway. So the platform that we have again as flexible to go across the entire spectrum of requirements around security as part of the edge architecture at the edge platform, we have something called the edge security demon, which is a native component and that ROY abstractions over radius hardware security modules we have. So we drink to provide flexibility to for various kinds of deployments. We have something guard a standard promise, and then. Have a more secure promise, and you know. One where where you have more hardware based security also again, based on your requirements, you can have, for example, some of your security that is stored not on the disk, but you have you started in hardware security module like a trusted platform, not PM that you store the.

"venkat" Discussed on Software Engineering Daily

Software Engineering Daily

04:16 min | 3 years ago

"venkat" Discussed on Software Engineering Daily

"Machine learning stuff is obviously interesting. What about a more basic example? So let's say there's a, there's a big candy company like candy company comes to you and says, we've got our candy factories. They've got computers in them. They've got conveyor belt computers. Maybe we've got some security camera computers, some of its old software, some of its new software help us modernize our factory. Where would you begin with that? Because this is kind of an an edge computing domain. You have a bunch of connected things in a factory. How do you get them on a platform like I o edge? Or if you want to talk about coober Netease like what is the software stack that the candy factory should adopt. The the baseline, they would probably the first thing they need to select an device, right? So and they would do they would select to device doctors conducive to their their use case, their price points TRE on that edge device where we would start is for running. I o d edge you would. You have to take a back on containers. And so by the way, this will be like a server that they would actually just put in their building on on the factory floor, whatever. Exactly. Yeah, this is just a solar desktop class, whatever you can, you can even be depending on what their scenario, but it can span across all of this hardware. Right? So in the end, so then they would talk about, okay. Is this a greenfield brownfield as we call it? Right? So greenfield is where the have the luxury of starting from scratch, starting from a modern framework using containers. Coober TH are not right and brownfield has where they have existing softened already that they need to containerize. So again, because we take a big bet on containerization, what we would do is try to figure out in their all stack, what they can containerize. Usually if it is a Lennox basis, DM almost always as possible to containerize their legacy code base in and run it as part of a container then they would integrate their legacy good with some of the SDK's that we provide as part of Azure -iety. Now these decays are available in multiple languages. So in most cases have a language that they can use and integrate the data that the produce out of their legacy software using the and now. Talk to the new architecture and wants to have this new architecture off to the races where they can either send that data directly to the cloud, do some preprocessing on the edge using other custom-built modules or modules they have bought off our marketplace and pretty much able to modernize their entire solution. So that would be a pattern that we would recommend for a brownfield solution. Brownfield situations. I don't know a whole lot about enterprise software for factories, but I imagine some of it is proprietary binary that were not built by the candy factory. Are these kinds of applications things that can be containerized or are there just certain parts of their build that they're going to have to leave outside of a container? Depends again on this show, but yes, depending on how the legacy software has birth. For example, if it's windfarms application that has some glee. Dependencies, right? Which cannot easily be containerized. Then in those scenarios, it would be more difficult to do something like this where they containerized pretty much lift and shift there also Lucien moved to the new one more what you really do as then kind of take a phased approach where you start moving various parts that do that can be containerized that that can be modernized somehow building translators to translate between the legacy stuff and new stuff with an all goal of moving everything over. But then again, take a phased approach where you have some pieces that are running on legacy hardware. For example, age computer age device could be running or in hardware with the office tack. And there are some translation that happens to an edge device that is now more modern..

greenfield TRE Lucien
"venkat" Discussed on Software Engineering Daily

Software Engineering Daily

03:19 min | 3 years ago

"venkat" Discussed on Software Engineering Daily

"Here. We've done a few shows recently about edge computing. We've talked about sensor networks in an agricultural field. We've talked about machine learning at an oil refinery. We've talked about a cruise ship that operates its own data center, and I would classify each of these areas as edge computing. You have lots of computation in these environments that have historically not had very many computers involved. How do you define edge computing? Sure, yes. So all the examples do gave are definitely a good examples of computing. The way I like to look at it is really coming down to two key phrases so that and that is data gravity. So wherever you have. Lots of data being generated automatically that actually attracts compute to the data right when you're talking about the edge, especially when you think about things like I'll refineries are think about, for example, connected stadium, which has cameras that is looking at audience so that to make sure everything is going fine. That's generating huge amount of data. When when you have that much amount of data, you apply some compute to it. But the concept of data gravity tells us that the moderate, I so much that actually attracts compute to it. Right. And all of these scenarios that you talked about the scenarios that we're seeing in the future, and also the president is does data gravity at the edge and that is attracting computer to it. And when computers attracted to that data, and that's what I would call h. compute that data, gravity, that idea. I want to understand the economics behind that because people have come on the show and they've said this idea that it's oftentimes cheap. To bring compute to the data as opposed to moving the data to a place where the compute is. Why is that explain that in more detail? One of the primary reasons is of course, the fate, in the sense that is just a moment of data that there is right. So let's take a canonical example of smart camera now that is no, it is. Most of these cameras are ADP somewhere even four k. so it's generating a huge amount of data every second. So there do options here. You either take all of these video frames and ship them off to the cloud, and you and you have the cost of ingesting the data, the bandwidth costs, etc. And then you can apply the last computer vailable in the cloud, but really what you really looking for is that one key frame a one key frame. If it's a security camera or someone is better because detected a human where it shouldn't be right. So that is the key frame that it is really looking for. And the rest of the ninety. Nine percents of the date has not. Interesting, but with the concept of data gravity, if you can bring compute to the data and now have a closed loop where the data is being evaluated at the edge. And then when the key frame happens right, where the human is detected, that key frame is then sent to the cloud. It's a lot more cheaper to actually get the insights you want. So on start key frame was detected. You send it to the cloud that can set off a whole bunch of actions using..

cloud ADP president four k