 This is SiliconANGLE and Wikibon's theCUBE, our flagship program, where we go out to the events, extract a signal from the noise. I'm John Furrier, the founder of SiliconANGLE. I'm joining my co-host Dave Vellante. And our next guest is exciting to talk about. Earlier in the day, we had our crowd chat, Werner Vogels tweeted about a company that was breaking all kinds of records. But really what's more exciting is that this is going to be a segment that's going to talk about something that was never done before in the world, ever in the history of computing. And it's a use case of great cloud computing. Jason Stowe, CEO of Cycle Computing. Welcome to theCUBE. Thank you for having me guys, it's great to be here. You know, one of the things about cloud is they say, oh, bye-bye the drink, do all this stuff, be more agile, provision more DevOps, build your own infrastructure. What goes on and on and on. But really the promise of the cloud is doing new things in a way that you could never do before. All kinds of benchmarks, black shoals in the cloud, test and dev. But you guys did something pretty amazing. Let's go and go right into it. What did you guys talk about? Why is it so exciting? What did you do that a use case never could be done before? Talk about it specifically. So we essentially used Amazon's spot instances to create the most cost-effective super computer in the history of the planet. So we provisioned about 156,000 cores on about 16,000 servers across eight different regions of the globe in order to run 2.3 million compute hours. It's about 264 years of computing in 18 hours. And this was done towards the problem of finding materials that are more efficient at turning sunlight into electricity. Okay, hold on. Let's go slower, slow motion in the video games. Okay, let's go slow motion. So 1006,000 cores. Super computer in the cloud, never done before. How long did it take? So we did the whole run, the 2.3 million hours in 18 hours. So we did basically 264 years of computing in an 18 hour time frame. That means if you took one Intel Sandy Bridge core, brought it back to the year 1749 when Edward Jenner, the guy that cured smallpox, was born and ran it then. It would have finished earlier this year around Memorial Day. So you, it was reported that you did 100,000 cores in 30 minutes. Yeah. What does that mean? Like you just ordered it up? Did you just say dial up to the 100,000 cores? So the spot team is amazing at Amazon. They basically have the ability to allow us to acquire infrastructure in Sydney, Australia, Singapore, Tokyo, Japan, Oregon, California, Brazil, Virginia and the EU all concurrently. So all what, eight regions was that? Exactly. And bring them up all at the same time. So we were actually able to get tens of thousands of cores every 10 minutes. So you have some hedge fund back in you about what, $80 million in financing? How much did it cost? So from a practical standpoint, the computer would have been worth $68 million if you had bought it, but thanks to spot instances, it cost 30,000 bucks. Okay, I was being perceived, okay, $30,000 and the equivalent to provision was 60, what million? 68. 68 million. And how long would it take to say provision that hardware? Oh, easily 12 months, 15, 18 months. I mean, 12 months you'd be hauling ass if you got it done that fast. So what was the, this is just mind blowing, right? I mean, what's the biggest supercomputer project you'd ever been involved in prior to this? So I think the next largest one, we did a 10,000 instance one that's going to be talked in a talk with a customer tomorrow. We did a 50,000 core one with a company called Schrodinger about a year and a half ago. It's all on AWS. And this is all on AWS. What about physically? What about the physical infrastructure? You had to provision yourself. Think back over your career. So we help clients that have tens of thousands of cores on their internal environments. My first couple of customers as a business eight years ago were Hartford Insurance Group, JPMorgan Chase, Lockheed Martin, those guys have thousands to tens of thousands of cores. So why does anybody not deploy in the cloud in the HPC business? I mean, does this mark a transition point in your mind? I think it does. I mean, I think the bottom line is is that, I know IDC has reported that the number of people in HPC that are looking at the cloud jumped from in the teams percentage wise to at least a quarter of the people. And that kind of hockey stick growth and interest normally precedes a transition point where everyone goes from not using it to using it. So what we're seeing in our customer base, and we've got a majority of the top eight pharma, we've got two of the three largest banks, three of the five largest life insurance and annuity companies. We're seeing all of them looking at cloud as a way to get them innovating again. The bottom line is when you buy a fixed size HPC cluster, your questions are fundamentally limited to what fits in that cluster and will execute in a reasonable period of time. That's not what you want. You want your smartest people asking the right question. You want them asking the question that'll change your business, that'll change your research, that might help cure cancer. And the thing that we're getting out of AWS is the ability to create the infrastructure to answer the right question rather than trying to fit inside what we were able to afford last year. Do whatever you can do, just because that's all you can do. That's the big one. So we have our own DevOps team. We're pretty hardcore DevOps developers. We have Amazon, we work with Hadoop, HBase, all that stuff as well. And we push the envelope. Stuff breaks, as Mark Zuckerberg would say. So I want you to share to folks, what did you learn, what happened? Obviously, great economics, dialing up a use case that never before could be imagined before now. What did you learn? I mean, what did you stumble upon? Because you probably said, hey, push the new code on the stack. Was there stack issues on the software updates? What did you guys stumble into? I'm sure there was some speed bumps. Just share with us, be candid about some of the things you learned. Absolutely, so there's a couple of things that worked great that have worked great in the past. So we used Opscode Chef to do the deployment of various software packages on the infrastructure. It's done a great job in prior runs. It still does a great job. So there really wasn't a change on that front. But the hard part for us in this run was essentially resiliency and dealing with scale. So we actually did a progression of test runs up until this 150,000 core. And some of our test runs were larger than most Fortune 100's clusters. So we did 125,000 cores as a pre-run of this one. So just to give you an idea of the scale testing, the hiccups that we saw were essentially in being able to distribute millions of hours of compute across that many cores is a very hard problem. We actually wrote a scheduler, essentially a way of distributing that job workload on ourselves to reach that scale. We call it Jupiter, because it's the largest clouds in the solar system. But practically speaking, it was able to do something that no other scheduler really is able to do right now. We can get hundreds of thousands of cores doing millions of things. And if we shut down any individual machine, data center, or region, everything would still finish. So it's like Netflix. It's chaos monkey proof. Right, right. From a practical standpoint. You might be, it's like running a marathon to train for an Ironman. It's about your little test case there. Talk more about cycle computing. Talk about what you guys do, your business, some of the products that you guys offer. So we're the leading software company in utility HPC. We have a large number of Fortune 500, government agency, academic customers that use our software based on one fundamental belief. And that is that utility access to high performance computing, the stuff that engineers use, that life scientists use, that clients use, is going to be the largest single accelerator of human invention over the next couple of decades. So our presence is predicated on the fact that we think that this utility model, what AWS really helps us do on the infrastructure side, by making it so that that can turn into a supercomputer at the push of a button. That's going to help invention broadly across life sciences, across manufacturing, across quantitative finance. There's all kinds of areas that benefit from this kind of model. And we write software that fundamentally tries to connect smart brains to up to super computing class infrastructure very easily. How is this changing the way, Jason, in which your customers think about data? You know, data, you talk about the data explosion, and five years ago, data was this big problem, and now it's, with projects like this, it's becoming an opportunity. I wonder if you could talk about that data opportunity and how it's changing your customers' businesses. Yeah, so let's talk about the researcher in this case. So this was a guy, it was Professor Mark Thompson, he was out of the University of Southern California. He works in affiliation with various industry. Essentially, he's got a branch of research on trying to find more efficient materials to make sunlight into electricity, so as part of solar. So fundamentally, what they're trying to do is find materials that are cheaper to manufacture and more efficient at generating electricity. That's what the goal is. So to do this, if you were to look at an individual new material, it take a year or somebody's time to even figure out, all right, how do I make this material? How do I make enough of it and isolate it so I can experiment against it? And then I get the properties, and at that point I know whether my year of time on average was worth something or not. And a lot of times it isn't, a lot of times these materials are worthless. Essentially what's happened in this case though is we were able to take 205,000 of those materials, and in 18 hours, determine how applicable they were to being efficient solar panel material. Previously you would have had a, what, model it, sample it? Synthesize, experiment, yeah, that's right. How about space exploration? You got any projects going on in terms of just... So on Friday we were fortunate enough to have three companies talk on our behalf. So one of them is HGST, which is a Western digital company. They made the original inch hard drives that were in the iPods. They just bought Verident. Yeah, so they're amazing. And they do amazing hard drive research. Well, they're designing their next-gen hard drives on AWS because it's more agile. Aerospace Corporation, who are the guys who designed the rockets that brought man to the moon and curiosity to Mars. They're using AWS as burst capacity in GovCloud in order to be able to run simulations. And then we've got Novartis talking as well about a very large run that they did and some of the things that they're finding as they move over to a utility HPC model. And all of those guys are able to take advantage of AWS's cluster compute instances, the high-performing instances. They're also taking advantage of spot to reduce costs because many of the forms of science that are becoming pervasive now are Monte Carlo oriented. They're pleasantly parallel. And they work really well on spot instances. Pleasantly parallel. Jason, I just tweeted, I like this guy, Jason Stowe. And it's the guy from Texas A&M. And you have the Jason A. Stowe Twitter handle. Hey, it's Jason Stowe from Texas A&M. I said, I like this guy. So tell us about your background. How did you get here? Because one of the things we always talk about on theCUBE is when you have these transformation markets, like big data, like what you're doing, you got to think differently. It's a mindset, right? And I want to extract this from you right now is that what's your background? How did you get here? I mean, is there a unique path? I mean, really creative. It's kind of like the Remington microblades. I was a user first. So I'm not just a vendor in this space. My original background was as a person that used high-performance computing environments. And then later, before starting Cycle, I worked at a Disney movie production that had number 84 on the top 500 list and wrote a software stack to help people manage that environment. And essentially Cycle is an extension of that work trying to connect people to computing. Did you bootstrap the company? Yeah, we did. Yeah. Yeah. It's been awesome. Another tech athlete, entrepreneur. I always tell my entrepreneur friends we bootstrapped our business to companies now. Awesome. To profitability and that's the pinnacle. People think like, you know, I got to get financing but to bootstrap your own company, very difficult. And it's the highest honor. It keeps you focused on the customer. That's the thing, right? It's a discipline. You have no choice. Exactly. Do you have an absolute focus in our case on enabling science? So I got to ask you, I got to ask you since the big brouhaha with the big use case you nailed, what's the craziest thing that's happened to you? What's surprised you? The press? Has it been the parade? The people kind of patting you on the back? It's been how awesome you guys are. The crowd chat you did? You guys have been more welcoming than anything. Thanks for coming to the crowd chat. No, definitely appreciate it. This is awesome. What's the craziest thing that's happened besides coming to you? This morning I had six people that I met in a conference in Germany about a month and a half ago who I had emailed and said hi to but hadn't necessarily heard from. All called me in the space of an hour this morning after seeing that and our aesthetic had published our articles. So that was a little surreal. We saw Quinn Hardy who's a deputy out of the New York Times walking around a friend of ours, I'll get him on the case. But I think the thing that I like and I want to get your final comment on this is that this is about the new way to do things. It is. And if you could share the audience out there, your perspective to the camera about this new modern era of computing. What is it going to look like? What's it take skill set wise? Minds that share your vision. Yeah, so what is the world as we wish it could have always been, right? Don't say steroids era. That's not the money ball we want, right Dave? No. That's interesting. I think it's really simple. And every brain, every device connected to whatever compute power is needed to solve whatever problem they're trying to solve. I mean that's in our space in technical computing, quantitative finance, engineering, et cetera. You want people unlimited by computing. You want them to be unshackled by a fixed size resource. You want them to be able to use things like spots and have a cost effective way of pushing humanity forward. That's what we want. What's next for you guys? What's the big, what's the next Everest you're going to climb? What, how do you top this? What's the next? So the funny thing that happens with these workloads is they get other workloads. So what happens is we started off doing a 10,000 core cluster for Genentech in 2011. We then did a 30,000 core cluster for Novartis. Later that year, we did a 50,000 core cluster for Schrodinger, which is a 200 person science organization. They used $44 million of equipment. And if they ran that on spot today, it would cost somewhere in the neighborhood of 800 bucks an hour to take advantage of that resource. So what'll happen is we're going to all forget about this, but other people are going to be thinking, you know what, I have this huge problem that would totally change the way my research is going, my engineering, my science, and I can now approach it. I don't have to be a government lab to get access to this. I can dial up Amazon and do it. It's easy. I got to ask you about the large scale question. So one of our friends of theCUBE, Carson Schwann, runs the Georgia Tech large scale computing lab advisor to us. Good friend. And we were just talking about the next gen large scale problems. What's around the corner technically that needs to be worked on? That's white space for folks, whether they're getting their master's degree or just in general, if Amazon continues on this trajectory, the world's going to look different. And again, the problems do shift and more opportunities come. What are those next problems? Was it just overall network traffic? Is it coordination? Is it software? What do you see there? Well, so on the science side, I think quantitative finance is going to get a lot more interesting. I think genomics, material science, engineering are going to get a lot more thorough than they were before, because we can be now. From a technology stack perspective, I think I'm really excited about the work we're doing and the reason we're doing the work is because I think it's an answer to this problem in scheduling. So I think being able to tell larger and larger amounts of processing power to be able to do larger and larger numbers of analytics and simulations and what have you in the face of failure, because when you have this much infrastructure outstanding, something is always in a bad state. So it's actually been identified as part of the exascale problem. So exascale is the next set of three zeros on supercomputers. And one of the things that they're identifying there is that we have to deal with failure, because when you have this much infrastructure working on a problem, there's always going to be one part of it that isn't quite doing well. It sounds like the world. Mathematically, it's a certainty. It's a certainty, yeah. Sounds like the world's a mainframe now. It's a global operating system. It's got to need to be less than a brain. Notification, checking devices, and the cross-do it. But this must also change the funding model, right? I mean, you must have, right, because the funding process in your world has been painful over decades. Yeah, I don't own a server. We don't own any servers. Right, we got people lining up now to change the world. And get much more efficiency and productivity out of their dollar. And that's actually where I think the spot part of this really deserves some attention. So the thing that's awesome about Spot is if you think about the cloud, it's really like insurance for your home. So when you insure your house against fire, you're amortizing the fire risk across everyone that owns a home. When you do infrastructure in cloud, you're amortizing the utilization risk across the entire body of users. So what that means is Amazon has a float that's available to basically help ensure that that amortization works. And what Spot does is it actually makes that available to scientists and engineers and other users to do real work at reduced cost because they don't necessarily care if you take all the infrastructure away from them. That's okay, I'll do the science tomorrow. You know, so those kinds of use cases with Spot I think are an absolute game changer. Jason Stowe, cycle computing, bootstrapped entrepreneur, tech athlete, awesome interview, congratulations on your success. Put a bumper sticker on the car on this event. If the event's a car, I want to see a bumper sticker. What's it saying? What is the Amazon re-invent conference about for the audience out there? Put it in a bumper sticker. For technical computing, it's my other cluster is all of EC2. That's a little long bumper sticker. Okay, we're live here in theCUBE. Jason Stowe, Jason A. Stowe with an E at the end. That's his Twitter handle, great success. Just showing you the power of thinking differently in a whole new way. This is theCUBE, we'll be right back after this short break.