Piotr Karwatka: [00:00:40] In today’s episode, we are going to talk about how to make the internet faster starting with satellites, then building your own web browser, ending with some thrilling stories on guerilla kind of market research in rural areas of Mexico. I couldn’t have a better guest for this topic than Peter Lepeska, who’s gonna share some of his 20 years of experience with us. Peter, I’m so glad you joined me.
Peter Lepeska: [00:01:04] Hi, Piotr. Thank you for inviting me.
Piotr Karwatka: [00:01:07] Okay, Peter, so let’s start from the beginning. You started your career in the late nineties. I guess funny times - the growing internet bubble. How do you recall these times?
Peter Lepeska: [00:01:17] Okay. So yeah, late nineties. For me, it started actually with the Y2K bug - working on that problem. But yeah, let me see... I think it’s important to get a little bit of context. In ‘97 I was actually a bike messenger.
Piotr Karwatka: [00:01:29] Bike messenger?
Peter Lepeska: [00:01:31] Yes. It wasn’t that kind of slacker bike messenger that you see in the movies or TV. I was a very intense, very kind of competitive bike messenger. And I really tried to optimize everything. Optimize my routes, carry the most packages, make the most money. And I think also I had some sort of crazy tactics, you know, I didn’t mind running down hallways and office buildings and law firms and stuff like that with packages. So I think it informs a little bit of my... it sets a good context for my career as a technologist. There’s kind of a rule breaking and optimizing quality. And then, my first child was born and that was it - I knew I needed to go look for a real job. And luckily we were in the midst of the Y2K bug crisis. So I was able to leverage my physics degree and go out and get a programming job. Actually more of a testing and code scanning job as it were with the Y2K bug.
Piotr Karwatka: [00:02:30] I almost forgot about this Y2K bug, but I still remember when I now revise it, I still remember the night 1999/2000. I was sitting in front of my computer waiting for some nuclear explosion. Nothing happened, maybe because of your job, right?
Peter Lepeska: [00:02:52] Yes, we fixed all the bugs. That’s right!
Piotr Karwatka: [00:02:54] How it was to handle the Y2K bug? How did you find all those bugs and fix it?
Peter Lepeska: [00:02:59] So it was a consulting kind of gig for a large bank in Chicago. And what we would do was: we would both scan code and run tests. So I really enjoyed it, cause I got to learn. I mean they were old languages, there was COBOL and SAS and other, you know, not the most useful languages today, but I really enjoyed kind of the challenge. I was kind of the needle in the haystack and also trying to devise tests and I thought it was fine. But it definitely got me wanting to move from that kind of work to actually building stuff with software. And that got me to my first web job. So I did eventually start working on websites. I think the biggest one was apartments.com, which was pretty big at that time and sort of around 1999/2000. I think what’s interesting about it, it was actually C++ based. So on the server side there was no scripting, you know, the scripting languages just worked fast enough. So it was C++ faith and the C++ would generate HTML. So it was sort of, you know, we had our own templating engine. And so that was where things were at that time.
Piotr Karwatka: [00:04:12] Gotcha. I was doing part scripting around this time and I remember that some folks were still doing C and C++ programs connected via CGI to observers. So it was something like this crazy, crazy times. And how did you land at ViaSat?
Peter Lepeska: [00:04:30] So I actually started working for a startup that was later acquired by ViaSat in 2000. And so that’s really when I kind of went from that consulting website and building websites and consulting “back to” kind of an optimization mindset. And I say “back to”, cause I really think of the bike messengering as sort of the same spirit as what we did in the startup. The startup was called Intelligent Compression Technologies. I worked there for seven years and then it was acquired by ViaSat in 2007. I think the way to look at ICT is sort of two main tools that we developed or tricks, tools in our toolbox. The first was data compression and we had world-class data compression, especially around kind of enterprise file formats. So of course we had good universal coders and good loss image coders that were our own. But I think the real differentiator was we could compress the heck out of basically Ms. Office files. So we did it all kinds of crazy things, you know.
Piotr Karwatka: [00:05:40] It was domain oriented, very domain oriented, and you picked the first, you know, a goal.
Peter Lepeska: [00:05:46] Exactly. So really at that time, what we were, the first customers were kind of enterprise type customers. So these were business people on their laptops. They were sending emails with attachments. And at the time the network was... We were in the kind of like the 2.5 G days. So, you know, we’re talking about maybe like 50 some kilobits, a hundred some kilobits, and sometimes as much as one second of latency, so really, really high latency and really pretty low bandwidth.
Piotr Karwatka: [00:06:13] So, I guess modem times, right, the dial up and modems.
Peter Lepeska: [00:06:18] Yes, exactly. So for people at home, it was dial up and mobile people it was this, really slow cellular. Yep. And, so yeah, that, like that kind of thing, that the two approaches we took was one, you had to deal with the bandwidth challenge. So that’s where the compression came in. And then in addition to that, you had to deal with the latency. So no matter what, the network protocol and we, over the years we worked on everything from SMB to MSR PC, which is the protocol behind outlook exchange, a whole whole host of enterprise protocols, including specific optimizations for SharePoint and then eventually web. And when you look at ICT, we really had three big customers. Initially it was Nortel that made the first investment and they were looking at that enterprise use case. And then, you know, that’s where we were compressing ms. Office files, just one kind of interesting thing. I always thought was indicative of the kind of, I guess kind of hacker mentality we had at the time. So it turns out when Excel saves a spreadsheet. Even calculated cells are saved as eight byte doubles. And so what would happen is you would have a spreadsheet with maybe 10 or 15 pieces of data actually input into it, plus a few formulas, but that might project out for years and years and end up turning into a multi megabyte file. So we were able to go in there, understand that file format and then compress it down to, you know, we were getting things like hundred times and a thousand times compression. So we had some pretty exciting demos that, you know, Nortel and other executives got interested in. So, basically at that time with Nortel, it was really the compression that they were interested in and they invested in us to integrate that compression library into a client server proxy that would transparently do optimization in the background. So that road warriors sending emails could send them and not wait two minutes for the email to be sent, etc. So that was the first use case was enterprise, but then it did evolve over into Consumer. And Microsoft got interested in us with, if you remember MSN, which was kind of Microsoft’s version of AOL, where you'd kind of download a big client, it included a dialer that would manage the modem and you would use that client to browse the internet and included their browser. And so bundled into that was our engine, which was doing, two things again, lossy image coding. So we could download, fewer bytes transparently. And then also again prefetching. And by doing that prefetching, we were really addressing the problem of latency, you know, so if a typical website has a kind of afforded one concurrency and you have, let’s say you have 32 objects on it, which is pretty small. Then that’s eight round trips. If you can double the concurrency, then that’s, you know, just four round trips and over a GPRS connection like what we were used to deal with, which was a one-second latency that brought it down to a reasonable time to load the page. So that’s really the way the prefetching works.
Piotr Karwatka: [00:09:43] Wow. That sounds pretty complex. Like I really like the extent to which you guys were thinking of the optimizations, right? This, this example with XL, FI’s striking me because you know, it’s very risky to modify someone elses Excel file. You need to really know, like risking in the meeting, you are taking a huge responsibility to make the algorithm that’s not breaking up anything. So I really liked this pioneer approach here as if I can name it like this, do you feel like a pioneer?
Peter Lepeska: [00:10:22] I think definitely there’s a feeling of... I think of it more as just… You have to be willing to take risks for sure but it can’t be so risky that you create a bad user experience. So we always said, if we corrupt one file, you know, if we have one executive send an Excel file that ends up corrupted on the other side with some values that are different, we’re done for. So you have to try to do both. You have to try to have an open mind and kind of a somewhat hacker mentality, willing to consider things, maybe other people aren’t considering, but then at the same time, you can’t just write chaotic spaghetti code with millions of, kind of heuristic recipes and assumptions that are going to be very vulnerable to changes that you can’t control. So, yeah. Pioneer, I think at that time, I don’t know if I think I was beginning to feel it, but I think at that at the time I was still really standing on the shoulders of the founder of that company who was a real incredible, kind of a powerhouse who was both incredibly innovative, but also amazingly hardworking, I mean, someone who would sleep at the office and, you know, get six hours of sleep at night and just... really inspiring person who also was incredibly encouraging of any ideas that I had. So there was really kind of like a feeling of being unleashed a little bit in working for Bill Sebastian, who was one of the founders of ICT.
Piotr Karwatka: [00:12:00] And he's working on visual music systems at the moment, right?
Peter Lepeska: [00:12:05] That’s right. So his career is interesting from a Pioneer’s point of view I think, because he started out as a jazz musician back in the seventies and then he actually made this first visual instrument. I think it was 20 panels of light. You can think of it as 20 pixels. And he was able to control the color of those pixels. And he actually performed with a very famous jazz band, the Sun Ra Arkestra, and the Sun Ra Arkestra would play it. He traveled with them and he would play this 20 pixeled instrument. And the jazz leader would even say, you know, “Bill take a solo!”. And the band would go silent and he would just, you would just see his light. And so he started there. And what do you really want it to do, was really evolve that, you know, he was imagining VR type experiences of visual music way back in the seventies, but he needed to find a way to fund it. So essentially he took, let’s see, 20 year detour and he founded ICT and he took some of the compression algorithms that he had built for his visual music. And then when ICT was sold to ViaSat, he was with ViaSat for a couple of years and then went back to his original passion project and is now running visual music systems. Check it out. There’s just some incredible, incredible things that he’s doing. He’s playing with bands and he’s...
Piotr Karwatka: [00:13:34] Yeah, that’s amazing! For this kind of AR technologies, right? Perfect! Let’s go back to your career, Peter. So it was ICT, acquired by ViaSat. Do you remember your first days in the new company? What were your challenges, you continued to work on those tools you created for optimizing the web experience, or maybe you switched your focus? How was it?
Peter Lepeska: [00:14:06] We did. So we kept up some of the enterprise focus for a little bit, but really soon enough pivoted to web acceleration. And what ViaSat wanted to do was take that technology and embedded into ViaSat modems. So as you know, ViaSat’s an internet service provider, a broadband service provider, and really this sort of, you know, the one issue with satellite geosynchronous satellite networks is the high latency. And so they really purchased ITT in some ways for right. I think the web acceleration technology, but then I think beyond that for some of that pioneering kind of innovative spirit.
Piotr Karwatka: [00:14:46] Peter, can you maybe say something more on this new satellite network? How does it work? I know that ViaSat already has like two or three different approaches. Maybe not approaches, but iterations of this network. I think that would be also great if you could tell our listeners a little bit on this, how this works, where does latency come from? What does it mean “high latency”, it’s like second?
Peter Lepeska: [00:15:13] Right. So if you just look at the speed of light, which is a constant we can’t change, right? And you look at the distance you have to travel to get to a geosynchronous, orbiting satellite. Basically like going to the satellite and back takes about 500 milliseconds. So there’s sort of a physical minimum to the round, the round trip time associated with a satellite, a geosynchronous satellite network. So, that’s really where the latency comes in now. Over the years, latency has become less and less important for most applications. It’s still significant for some certainly, some real-time gaming is the example that usually comes up and it still affects web browsing. You know, you really will get a different browsing experience on a very low latency network. At the same time, the source of latency in many applications isn’t always the network, you know, so sometimes it’s the server it’s a poorly written or an overloaded server.
Piotr Karwatka: [00:16:14] In most cases.
Peter Lepeska: [00:16:16] Yeah, it can very often be. And what’s interesting is that the effects are the same. So that's what's kind of cool about the technology we developed is if you’re doing prefetching, you’re sort of minimizing the risk of the impact of any of those sources of latency, but, back to ViaSat -So I think. No, that’s. I think what was really great about bias that was, you know, it’s very much kind of an innovation oriented culture. I would say that the biggest book, that the most influential book at the company is, Clayton Christianson’s innovator’s dilemma, which is really about sort of, you know, how do big companies avoid being disrupted? Or how do incumbents avoid being disrupted by innovators? And it’s sort of the, almost the idea that you have to disrupt yourself. And I think vice has done a great job of that. So our very first satellite ViaSat one had more capacity than all other satellites that were in orbit at the time, including, including those that our own customers were using. we had, we had purchased capacity from, from another vendor.
Piotr Karwatka: [00:17:22] What's the capacity of, you know, average satellite. How do you measure it? Is it like in a number of concurrent connections? Gigabytes?
Peter Lepeska: [00:17:33] it’s a great question, because basically the way we use it typically is just bit gigabits. Let me see, I actually am looking up the numbers here. Here we go. Vice that one was 140 gigabit via set. Two was 260 gigabits. So we almost doubled the capacity from, from the first generation satellite to the second. And then ViaSat three is it is a terabit now, I said three is three satellites. so, and, and the three satellites will have global coverage. So not only is the, is the capacity satellite, jumping, doubling, or more, but the coverage area is increasing. So with ViaSat one, it covers mostly just, continental us, virus. I said two covers more Latin America. And even across the Atlantic to Europe to cover those. transcontinental flights and ViaSat three is, is global. So it’ll cover the entire world, but I do think a really important part of the innovation with ViaSat three, that I think your, your listeners might be interested in is that the capacity is very focused bubble. So essentially we don’t really want to put capacity any place where there are no people who want to use it. And so the example we often give is we can kind of focus a capacity on an airplane. That’s crossing the Atlantic and not actually focus any capacity outside of that, or, you know, put the there’ll be no capacity in the Sahara itself only in the population center.
CTO of Acceleration Research Technologies Division at ViaSat Inc.