---

Transcript

• Tweet

Joshua Bixby: Hello everyone it’s Joshua again, President of Strangeloop Networks and blogger at webperformancetoday.com. Welcome to another edition of the Web Performance Today podcast series. I had a great opportunity this week to sit down with Theo Schlossnagle, CEO of OmniTI. Theo is a geek among geeks, you probably have a tab in one of your browsers right now open to an application he has helped scale, an infrastructure, he has helped build or a distributed system that he has spoken to or architected himself. He has been doing long, list longer than most people since 1997 probably longer than you and we talked about what he actually does in the day, his tech predictions for the New Year as well as a really interesting beer project where he – a big data beer project that he has been involved in, that’s very personal, I hope you enjoy.

Joshua Bixby: Hello everyone. Welcome to another installment of the Web Performance Today podcast. I am here joined by Theo on the phone, Theo Schlossnagle. Theo please pronounce your last name for me?

Theo Schlossnagle: I think you nailed it, Theo Schlossnagle.

Joshua Bixby: Theo is CEO of OmniTI, many of you will know him from his incredible presentations at Velocity, but probably don’t know a lot about what he does every day and frankly neither do I. I have spent time with Theo in Beijing, what I guess Berlin, London, Munich during the Velocity Conferences but I haven’t got a chance to really spend much time talking about OmniTI and performance. So Theo give us the, give us the digest version, tell us about what you do every day?

Theo Schlossnagle: What I do every day. I have the challenges of a CEO in a company that’s growing, I do all sorts of undesirable things like look at contracts and talk to landlords and things like that. On the technical side, I think that I split the gamut in a pretty interesting way. Most of it is related to performance and scalability of large scale web apps. Everything from infrastructure design and automation up through kernel level stuff, so I do a lot of work in the [indiscernible] [00:02:05] kernel for our datacenters and then up the stack, you know, through databases, through web servers all the way in to JavaScript to browsers, so I kind of float around from task to task with my eye specifically focused on performance and scalability.

Joshua Bixby: And the customers who you know – if I was a customer how would I know if I was well targeted to come to you guys, what would I look like, am I a Twitter, Facebook, am I you know more like an enterprise, is there a sweet spot for you guys?

Theo Schlossnagle: I think that we have three types of customers and they kind of classify outside of market verticals. We have customers that have large scale applications today that you know they are lumbering along and they are wondering if they have done everything right the technology is five years old or ten years old, they want to know if – basically they need an assessment of their architecture. A lot of times that happens for acquisition purposes as well. Then we have the other type of group that are looking to build a very large scale web app; they have a prototype they’ve tested it with a hundred or thousand users and they would like to you know take that to the moon, so they are looking at picking on 10 million or 50 million customers and they’re very concerned about their ability to address that demand as it increases the thundering herd of the internet is a scary thing.

Joshua Bixby: Yeah.

Theo Schlossnagle: And then the last group of people who have an application that has been hit with a thundering herd and been trampled to death and they come to us with a thousand pieces and say please put this together again. But in the Enterprise versus Dotcom sort of world, we have customers that that range that you know that that cover that pretty much the whole set. We service about 15% of top Alexa 100 and so we deal with very, very large websites. We also deal with a lot of enterprise accounts that most people are on, you know, there is a lot that you would find rather boring.

Joshua Bixby: I’m sure. So part of sort of establishing that streak grid because I don’t think people know a lot about you guys is to dig into what are you seeing out there I mean you guys have been doing this since what ‘97, has the world changed, I mean as you see these presentations at Velocity or people talking about performance are the same basic principles at play or has there been or will there be some dynamic market shift in how these things are architected, built you know related to performances. Is it still the same thing as it was 10 years ago?

Theo Schlossnagle: I think from the scalable systems engineering side, we are largely seeing the same thing. The question to scale out or scale up is still there; people feel like they are a little bit more confident in their answers but I think they are no more correct than they were then. They are being lured into the scale out always mentality just because software allows them to do that now when scaling things out makes things more difficult to bug, distributed systems to bugging is you know a nightmare always.

Joshua Bixby: Yeah.

Theo Schlossnagle: So if you could scale vertically you choose to but that’s the same you know in 1997 or in 2000 when things started to get pretty serious, you always wanted to scale up until you couldn’t and then you scaled out and you know that’s pretty straightforward. I think the radical difference is that back in the, you know, late 1990’s there was so much less complexity being sent down to the world’s largest distributed computer which is, all of your consumers browsers. So the world of JavaScript running on the client side today is it, you know, its night and day compared to what it was and so that’s radically different.

Joshua Bixby: Is that a trend I mean, you have probably seen some of the post I just did recently, did one talking about the growth in not only the size of pages but JavaScript specifically and the complexity of that. Is, does that hit you know a point of diminishing rates of growth. Do we, you know, what – is that just going to keep going up, getting bigger?

Theo Schlossnagle: I think that we are still in that uncomfortable space where we do not have comprehensive development environments, best practices, IE’s and things like that for developing very, very rich applications on the client side. People are still kind of evolving their applications. They are building largely HTML applications, HTML5 leaving the JavaScript out of it and then adding JavaScript for a flare and then adding it for features and then adding it for complex features and then driving business lodging into it and it’s kind of a snowball effect and then you are left with a half a meg of JavaScript on the client side where there aren’t really the same the quality control tools that we have on the backend side. So if you are building a database and see for example or writing an operating system kernel, the landscape of software engineering practice is there, it’s so much more mature than the landscape of software engineering when it comes to shipping JavaScript to do that with the client.

Joshua Bixby: One of the things that’s always intrigued me because you work on, you know, the development and operational side, you seem to have, you know, seem to run the gamut but you spend a lot of time talking about the operational side. So for the people in the operational side of the business, what do they do when they are handed that app that has the half a meg of JavaScript. I mean are you helping people find solutions to that problem sort of assuming okay it’s a black box and this is what it comes with or are you finding that you are constantly in discussion with developers?

Theo Schlossnagle: Oh constantly in discussion with developers. I think that days operations and development and engineering being separate are, those days are over. I think they were numbered sometime ago, I think the numbers have run out. It is very difficult to compete today without shipping rather complex JavaScript down the pipe that stuff cannot really be, like any software it cannot be magically optimized. The algorithms and data structures that youre chosing because the code is definitely complicated enough to necessitate algorithms and data structures on the JavaScript side that that is not right for you know magical optimization. We have gotten as far as we will go with things like the VA chip. I mean they will steadily improve, but there is not double digits percentage there left I don’t think.

So really it’s about good software strategies strategies and you know don’t repeat things, basic software engineering premise, how do you measure that because the one thing that developers don’t really have it’s a perspective of how their code is running universally right. They can test it on their browser, they can test it on five or six different browsers but there is difference between their software and what we used to write 10 years ago is that that JavaScript is in the client because it needs to run there and it’s not just because we don’t want to run it on the server side. It’s that there is an active program running on the client side to receive sometimes arbitrary data and we are sending them very complex streams of data that were you know assimilating into something useful and you have a million or ten million customers that are all getting different data streams, it’s a very I mean the richest, you know the wealth of all of the performance analysis of this software is actually in the wild, which you want is empirical data about how your performances are, is operating, how your java script code is operating, how it’s functioning, what it’s errors are, all of the performance profiling needs to come from the real world.

Joshua Bixby: And that really changes the game for sure?

Theo Schlossnagle: Yeah it absolutely changes, I mean so we, we – most of the clients we work with have some sort of bespoke implementation of that.

Joshua Bixby: Yeah.

Theo Schlossnagle: Right there is some sort of telemetry being collected, not just like how long that they are taking down with the JavaScript. But there are you know meters and metrics being counted inside the JavaScript code and that stuff being passed back up to the server to the infrastructure being analyzed by developers and when you look at that analysis of data that looks like what ops have been doing for the past fifteen years or twenty five or thirty years on other equipment and we are looking at network switches and CPU’s and disk IO and things like that. It’s what database engineers have been looking at when it comes to you know, database performances you know like four hundred graphs for any descent Oracle DBA.

Joshua Bixby: Yeah.

Theo Schlossnagle: That is immature on the JavaScript side and it’s evolving rapidly and what we have noticed is that like all of the tools are very, very similar and it’s actually valuable to have all of that data just, not just you know the systems and network information but also the database information, the distribution database infrastructure information and the JavaScript performance profiling stuff, all of that really needs to be in one tool because they are all tightly interrelated.

Joshua Bixby: So I had a really interesting discussion with a gentleman who remained unnamed out of Microsoft really senior architect who basically said ops guys are used to dealing with tons of data and devs aren’t and what we are seeing is a culture where devs are being exposed to big data which has always existed on the ops side. Do you think there is some truth to that?

Theo Schlossnagle: I think it’s a really interesting – actually I had a discussion like that today. I think that that’s true but that sword has two edges because developers hadn’t been indoctrinated with crappy tools, their eyes are not, they are not winking in pain. So you look at network operations, you look at systems operations they are used to collecting telemetry up with every machine they have, you know they sample everything every you know one second or ten seconds or sixty seconds and they have got graphs coming up their ears and you give that to a developer and they would go wow that’s really useful. They don’t stop there, they ask the next exact, you know that there is the next question which is well I don’t have enough data, actually I need all of this extra data, I need this deeper data, I need this to be more dimensional.

Joshua Bixby: Yeah.

Theo Schlossnagle: And then you know the OP guys are like well that’s how the way these tools work, which is sort of interesting because then the software engineers go and they build the tools, right cause that’s what they do. So what we have been doing for a while now with one of our companies Sekonic is trying to take those more evolved telemetry analysis tools, the ones that engineers have designed to answer their questions and push them back into the ops group to make the decision making capabilities faster and more rich.

Joshua Bixby: If only the world could see that there is all this you know incredible value and collaboration, I mean I keep coming into organizations and admittedly more enterprise type organizations where it feels like there is just a complete separation of the worlds; there is the velocity world which is big data iterate, fails quickly, fails cheaply that whole lean startup mentality and then I get transformed into this world of almost I shouldn’t say this dinosaur like, waterfall like approach to the world where none of this is really happening, so it was interesting when you said you guys work with you know 15 or the top one 100 or do you also find that there are just stark cultural differences between these businesses where some get it and some you are just, you are looking at these people thinking I can’t even start to explaining this to you because you don’t understand the first principles, I mean I pull my hair out over this stuff.

Theo Schlossnagle: I think that all of the organizations actually understand the value of data. They all understand, when you go in with a value proposition around understanding how your systems and your business are working I think you basically would be asinine not to understand that right like that’s clear. When it comes to the tactics for are actually running the organization, but you know as it relates to technology how technology is deployed, how it’s developed, how it’s managed, I think there is an enormous cultural divide but I think that it’s reinforced with a vernacular that the Velocity group will now let go off to this idea of move quickly and fail fast.

Joshua Bixby: Yeah.

Theo Schlossnagle: Enterprise, you will let, you know, to say failing is never an option, it’s just reality, right. So they don’t want to embrace the fact that they are failing right they want to avoid it. But that doesn’t preclude very similar tactics by this whole waterfall model, we were consultants, when we go into a place we don’t try to disrupt their language, if they are with Linux , we stay with Linux; if they were with Solaris, we stay with Solaris, you know, Java, Python; they are using waterfall, if they are using AdRoll, we fit in and try to add value inside the framework that they have established. So we have worked in a lot of you know large contracts that have been waterfall driven and all of it comes down to how well a project manager or scrum master understands when to apply the model and when to not and that’s the success of project, it’s not really AdRoll is better than waterfall. It’s that leaders, some leaders are awesome, some leaders aren’t.

In the waterfall load you can actually orchestrate it so that you do a deployment everyday right, we will see this idea of dark features for example right. Waterfall says I am not going to release my product until a certain date well I didn’t release it, I mean at the point it’s not released, right. So now it’s all mincing words and when you maintain that exact same facet that the enterprise expects you have a lot of flexibility in the way you implement things. So the problem is, is that in the velocity group when we talk about our implementation we try to push those tactics in that mentality all the way up into management and they don’t want to use that mentality.

Joshua Bixby: Yeah but as you say the tactics can be very similar almost indistinguishable in many ways.

Theo Schlossnagle: Absolutely, but there are distinct advantages too, for example launching code every day or every hour all right so if I am forced to develop in a model, even if that model – you know even if the approach to management is waterfall, but I am forced to develop code that’s not allowed to break the existing system right I am not adding features, even if they are not released I am putting things in and I am not allowed to break the current regression tests. That means I can deploy the code, no one is going to run it.

Joshua Bixby: Yeah.

Theo Schlossnagle: Right, at the end of the day when you are ready to launch your product you have removed that, you know that two months death cycle and death march of deploying that code that has never been deployed before that’s a typical problem in enterprises that they don’t embrace operational mechanics of the release until the very end when it’s suppose to be launched, which is disastrous.

Joshua Bixby: Yeah and then they wake up and think and wake up and realize that software is you know made by humans and errors and when an error when you know because it, because of errors they are in trouble and they haven’t got you know they have got a long cycle to get that stuff out. I am interested on the data side the last time when we were in Beijing last year, was it Beijing? No it was Berlin last year you were trying to drink a different beer every day or your goal was to have if I remember correctly three hundred and sixty five different beers drunk am I remembering that goal properly.

Theo Schlossnagle: That is correct.

Joshua Bixby: And can you tell me what you achieved, that was last year right so 20, end of 2011 as a goal.

Theo Schlossnagle: Yeah I measured those things, Thanks Giving to Thanks Giving so my dead line is tomorrow.

Joshua Bixby: Okay so what were you last year, set the table for me what were your – what last year?

Theo Schlossnagle: I think I hit three seventy eight or something like that.

Joshua Bixby: Amazing, so for those who didn’t spend time in -.

Theo Schlossnagle: But I haven’t achieved this year.

Joshua Bixby: You have.

Theo Schlossnagle: I am at 715 total, so whatever that – that is.

Joshua Bixby: Okay and those are 715 unique beers.

Theo Schlossnagle: Unique beers.

Joshua Bixby: So can you find any left, I am just thinking as you were talking about data and the process I am thinking if there is a data project that we talked about are you at the end or do they import more varieties?

Theo Schlossnagle: They make more varieties every year and there is usually you know some of it’s been a 100 and 250 new beers, new recipes that come out every year, the micro brew scene in the United States is incredibly vibrant, but there is probably about 8,000 to 10,000 beers, unique beers on the market today that I could go buy.

Joshua Bixby: Okay so you got, you still have ways to go, you could still hit, you still hit six or more, six or seven more years with this?

Theo Schlossnagle: Absolutely, I have enjoyed every minute of my weight gain.

Joshua Bixby: That’s brilliant and what is the – not only weight gain but you are also in the business of changing what hair and facial hair quite readily, what is your current, what’s the current configuration?

Theo Schlossnagle: I would say that I am in unkempt San Diego style.

Joshua Bixby: Okay describe that?

Theo Schlossnagle: I had the, I had been told I look like the person who does the voice of doofenshmirtz.

Joshua Bixby: Okay so for those who don’t specialize in such arcane references, can you describe that for us?

Theo Schlossnagle: I have the goatee with no mustache and a head of hair.

Joshua Bixby: And color, color of hair?

Theo Schlossnagle: Oh yeah brown.

Joshua Bixby: Okay, I think the last time you saw you were blonde were you?

Theo Schlossnagle: I had dyed my beard, I had -.

Joshua Bixby: Oh that was it yes.

Theo Schlossnagle: I had an enormous beard and it was that black, I was going to go as Brian Wilson for Halloween.

Joshua Bixby: Oh nice, nice, for those who don’t know Theo there is a constant change, is this, there is just a testing, there is a multi variant testing process that you undertake with your looks.

Theo Schlossnagle: I wish I had the discipline to you know take a picture of myself with my laptop every day.

Joshua Bixby: You should.

Theo Schlossnagle: Because that would have been brilliant but now alas I haven’t.

Joshua Bixby: Two more things before we end. First we are at the end of the year getting close to the end of the year and this is we are getting close to prediction time, so I am not going to ask you for a super bowl pick but I am going to ask you to tell me about two things that will surprise me next year?

Theo Schlossnagle: Two things that will surprise you next year.

Joshua Bixby: Ideally tech but you know I am open.

Theo Schlossnagle: Okay that’s a great question. I think that the convergence on distributed database technology, people feel like it’s coming and they will be surprised that it did arrive.

Joshua Bixby: Okay.

Theo Schlossnagle: Yet again -.

Joshua Bixby: Yeah that, I have heard that predicted a few time so okay but the negative side of that prediction is probably a much safer that okay that’s a good one, what else?

Theo Schlossnagle: Yes, let’s see I think that the android marketplace, the you know the devices running android, the android app market, the ecosystem around the android has hit the you know the corner of the hockey stick and I think that it will absolutely explode next year.

Joshua Bixby: That’s fascinating because we -.

Theo Schlossnagle: That’s been steady this year but I think that the tides will shift.

Joshua Bixby: So I use strangeloop as an example we have you know fifty, sixty people here, we went from the Blackberry to deep, deep iPhone in penetration and where I am not seeing my dedicated iPhone users all move to android and it’s happening on mass it’s amazing, it’s like a virus that’s taken over.

Theo Schlossnagle: Yeah and I think the trends that we see in the tech community like that always you know they immediately pressure end consumer activity that that was in with that so. I think that, I don’t think it will happen this year and we have less yet but and I don’t actually think it’s going to be the nexus four seven ten series but that really does it and I think that will raise the awareness necessary to make that you know that tidal shift.

Joshua Bixby: And lastly Velocity Beijing you know I had a chance to be there with you last year, I understand you are running it this year, so tell us about quickly about running it and finding who is listening why they should go?

Theo Schlossnagle: Beijing is awesome, sure opportunities see the great wall, it’s a forbidden city, Beijing is a fantastic town, you can combine technology with good site seeing. I would say that if you are not fluent in Chinese that there are probably other velocity conferences that you will extract more value from. Most of the talks are given in Chinese and the translation there is always challenging because almost live translation and technical topic from English to Chinese is difficult. Planning here has been a very interesting trying to recruit talent, it’s a very different demographic that I normally talk to so being in the United States and finding conference in the United States it’s kind of assumed that everyone speaks English and they do and so everyone who works at an American company basically speaks English.

So the criteria for finding speakers at the companies that I would like hear from so my usual approach to finding speakers is I find that you know some really interesting architectural problem that’s been solved identify the company that encountered it and then I try to find some member of that technical staff, management staff that that really understands what that solution look like, what were the challenges involved with, there were personnel challenges or technology challenges or what have you. We are limiting that to people who are not limited, we are finding our good number that were fluent in mandarin, posed on interesting challenge though I had a lot of help. But it worked out great so now we have a speaker from Google, a speaker from Facebook, a speaker from Twitter, one speaker from Etsy and there is me.

Joshua Bixby: Who is coming from Etsy?

Theo Schlossnagle: Sam Hapkins.

Joshua Bixby: Okay I don’t know.

Theo Schlossnagle: He is going to be talking about a continuous deployment.

Joshua Bixby: Oh great well those guys know that stuff.

Theo Schlossnagle: They know that stuff.

Joshua Bixby: It is, thank you for taking the time the day before Thanksgiving to chat, I appreciate it, enjoy your meat and bird over the next two days and let’s connect, I guess we will see each other definitely in June at Velocity, but hopefully before that.

Theo Schlossnagle: Yeah excellent.

Joshua Bixby: Thanks man.

Theo Schlossnagle: Looking forward to it.

Joshua Bixby: Good take care.

Theo Schlossnagle: Alright thanks.

Joshua Bixby: Thanks for joining me today and thanks again to Theo for making the time out of his incredibly busy schedule to chat. Be sure to check out my blog webperformancetoday.com/podcast for all the links what we talked about today. If you have a suggestions for a future podcast fire me a note at joshua@webperformancetoday.com, hope to see you again thank you