 Live from the Sands Convention Center in Las Vegas, Nevada, it's The Cube at AWS re-invent 2014. Brought to you by headline sponsors, Amazon, and Trend Micro. Everyone, welcome back. This is Jeff Frick. You're watching The Cube. We're at AWS re-invent 2014 in Las Vegas, Nevada. This is our second re-invent. We've been to a number of summits and it's always a great show. Obviously, AWS out front in the cloud space. This is a huge, I think there's over 300 exhibitors here. A giant show. I don't know what the people count is, but it's big and there's a lot of excitement. And we're really excited to have Jason Stowell next to CEO of Cycle Computing, Cube alumni, who really is doing some very interesting stuff with software, leveraging AWS infrastructure to do things that people couldn't do before. So welcome back, Jason. Thanks for having me, Jeff. It's so good to be here. I'm always excited to be here at The Cube. This has been awesome. So let's go ahead and talk about your most recent announcement that you guys just came out with. So we had Fortune 500. It's HGST, which is a subsidiary of Western Digital. So they make the hard drives that are in most of the servers out there. And they're doing some really crazy engineering. So they have on the heads that actually read off the platters inside the disk drives, they had a million different designs. It was basically a parameter sweep of a bunch of different facets of how the head could possibly work that they wanted to simulate. The problem is it was 600 plus thousand hours of computing to do this. It would have taken a month in-house. How many hours? 600 plus thousand. So it's about 640,000 hours. So that's about 70 compute years. It's 70 and three-quarter compute years. So seven decades of computing that you would have been doing. And instead, we spun up a 70,000 core IV bridge cluster on top of AWS, ran that workload. Instead of a month, it ran in eight hours. And it got done for $5,594 on spot instances, which is just amazing, the cost compression. So one day, less than 6,000 bucks, they got to run their simulation. Yeah, not even a day, a third of a day, eight hours. Submitted in the morning, you get out at night, it's done. Fantastic. And would they have even attempted that before? Is this just something outside their own possibility, or would they have done a more clunky sort of substandard process? So this is something where they would still do it, but it would be one of those offline processes where they would maybe do less of it. And they would also essentially start it and then stop a bunch of other work and come back a month later and try to figure out where they were. And it really makes the design process a little bit more challenging. So it really is an example of taking something like cooking a turkey that used to take six hours and you deep fry it, it takes half an hour. It's that kind of transformation where it's a lot faster. It's the equivalent of the microwave. And the cool thing about this cluster is, so it had 70,000 cores. It was what they call teraflops. So this is trillions of floating point operations per second. It had 729 of those teraflops. That's number 63 on the top 500 supercomputing list. So basically for an eight hour window, we created a computing environment that was the equivalent to the 63rd best supercomputer on the planet and then turned it back in and it cost less than a single server to operate for a year. That's the kind of cost compression and time compression and what happens is when these come out, so in July, we had Novartis talk about a 10,000 server run they did. It ran about 40 years of computing in approximately nine hours at a cost of $4,300. When that happens, then we get a Fortune 100 finance company that wants to run a regulatory workload at the end of every day that they just never would have done otherwise. There's all kinds of use cases that fly out that just would never have happened. But because we're able to do this so quickly and so inexpensively thanks to spot instances and cloud computing, it now becomes feasible. And this is really what's going to push humanity forward. It's going to make it so we have safer products. We're going to have better understanding of our genome. Novartis use case was designing better cancer targets. So they basically had candidates for drugs. They found three of them in that 140-year run that ran in nine hours. They found three new compounds that could be potential cancer drugs as a result of that one run. And that's what we want. We want humanity doing things faster. We want to accelerate the human mind, get better answers faster. And cloud enables us to do that. Yeah. So like you said, I know a lot of your early customers were in pharmaceuticals and they're running simulations. But the HGST example is not pharma at all. It's manufacturing. It's just manufacturing. So with these tremendous compressions in time and cost, are people, do they get it? What's kind of their plane? For me, the Twitter moment was the plane in the Hudson where you could instantly find out what was going on on Twitter. That's right. OK, now I get it. I get it. What's these guys kind of playing in the Hudson moment? So I think this run for manufacturing will be that. We had, in life sciences, it took probably the second run that we did at large scale for people to be like, OK, yes, I get this. I understand why this is important. The guys, David Heinz and Steve Philpot, Steve is the CIO and David runs global computing infrastructure. So basically data center and cloud computing engineering for HGST, they get it. They are, I think, ahead of their industry a bit in terms of understanding that dynamic capacity can increase the throughput of your organization. There actually is going to be a talk on Friday here at Reinvent, 10.30 in the morning. I think it's the BDT track, session 311, where David's going to talk about that workload that he did. Last year we talked to you guys about the 156,000-core workload that ran, and after that 156,000-core job where we did 2.3 million hours of computing in 18 hours, we got a ton of new users. Now running 10 to 20,000 cores is pretty normal. Right. And that's insane. The idea that you can just borrow 10 to 20,000 cores and then give them back when you're done is just ludicrous. Well, the other thing, though, that's interesting that you just said is that it's not only simulations, but I think you mentioned some financial services firms using it for daily scheduled work before they wouldn't. So are you seeing more and more of an uptake of doing things that they would have either never done or at a frequency they would have never done using this capacity and this low cost capacity? So you have, Jeff Moore had a great paradigm around how technology gets adopted, wrote a book called Crossing the Chasm. Right, of course. You have early adopters. Yeah, you have the early majority. And then you have the late majority and you have the leadites, the guys that, the only reason they don't have a phone with a dial on it is because you don't sell it anymore. Exactly. So the thing about cloud and HPC is that life sciences was really the earliest adopter. So that happened in 2008, 2009, 2010. There was a lot of talk about it, et cetera. It started coming in. I think financial services and insurance has happened maybe a year, two years ago and up through right now. It's kind of hitting its moment. And I think what's going on with manufacturing with HGST is really the leading, you've got somebody that's visionary there. Steve actually has someone of a life sciences background in his career and he's applying these things that have tremendous competitive advantage into HGST and it's what they're doing is really impressive, frankly. And I think that that'll serve as a leader on the rest of manufacturing and hopefully we'll see a lot more of exciting work going on. Yeah. And then as you said too, about moving downscale, leveraging this capacity and this system to not necessarily have to do 70,000 cores or 150,000 cores, but to do 10 or 20 more frequently or smaller jobs. How's that adoption happening? It's great actually. So most of our customer base uses between 64 and 6,400 cores. So they're really using tens, hundreds and low thousands of bursts. What's the average duration of a job? So I don't know, it depends on the industry. But generally what we're seeing is kind of daily fluctuation where you'll have peaks during the workday and it tones down at night. And then once every week or once every month people will realize, hey, I can do a big run that'll run over the weekend on a larger core count over the weekend. I'll come back in on Monday and it's done and normally I'd never even be able to run that run. And that's a pretty typical, I'd say, usage pattern that during the day you get spikes where your internal cluster is topped and so now I want to use a lot of cloud. And then over the weekends, people are saying, you know, I could actually run something much, much, much larger than that. Right. And then eventually that leads into, hey, you know, there is this workload and the right scale, the right question is actually much larger than anything that would run internally at all. Right. I'd have to stop everything essentially. And that was the Novartis use case. That was the, you know, originally back in 2010 we did something with Genentech where we ran a 10,000 core cluster and people thought it was crazy at the time. And now we're talking, you know, 50, 70, 150,000 core clusters and the great thing about this workload honestly from a cycle standpoint we found out about it on Wednesday and it ran this past weekend. So it was a couple of days, not a lot of prep. We basically got the environment up and running and pulled the trigger on it and it ran. And so from a cycle standpoint we really look at these as very, very doable. Right. And that kind of agility when you're looking at it as a CIO as a VP of engineering or a chief science officer being able to say, well, the bar or the problem I can approach economically is now not a hundred cores. I can do problems that have tens of thousands of cores in them and I can get access to them. We got 50,000 cores from the SPOT guys at AWS. So this is Steve Philpott and I want to give a shout out to Dave Pellerin who's a BD for HPC basically helps facilitate HPC workloads on AWS. They were like, sure, go ahead dip your hand in the well and see what you come out with. So what percentage of your workloads that are done with SPOT assets? So I'd say And is that the key one of the keys to the economics? Yeah, so SPOT is a big one. And I think what's going on is there's two classes of workloads. There's the ones that have to get done and so on demand and SPOT are used together in a mix. There's the ones that are essentially like can we run this? And if it's anything shorter than 30 days that's great, but if it's shorter than a day then that means I don't have to stop thinking about the problem I can come back to it. We have a lot of workloads like that where SPOT fits in great because there isn't necessarily the economics for those otherwise. And then we have workloads that are more regularly planned where reserved instances make a lot of sense and we want to use reserved and then we might want to use a mix of either on demand or SPOT on top of that. And the thing is AWS gives you a lot of different options there and I think the the economics side of it though to get to your point is the thing that's enabling people to do science that they never would have done before. You can now ask questions at the right scale not the scale that fits on the cluster you could afford a year and a half ago and that's a big deal. Especially when it comes to life sciences when it comes to product design when it comes to risk management those are the areas where we all want the safest possible car. We want the safest possible annuity so that we get our retirement funds back. We want you to be thinking about cancer treatments that you might not otherwise think about. Those are the things where SPOT I think enables a lot of science that just would not happen otherwise. And is most of the SPOT purchased from other people's unused capacity or does Amazon have that much extra cores laying around that you can go in and put in an order for 70,000 on Wednesday and you're up and running on Saturday. We find out about the workload on Wednesday you know when you're dealing with really large things that are above your AWS instance limit you know you have to let them know so we let Dave Pellerin know and Steve Valiant know and Joby and a bunch of other guys on the AWS team and they did whatever they do to make it so that our limits were high enough and then yeah we just ran it. We didn't talk to them while we were running it at all. Actually we didn't even do much manual work that much at all. Anyway our software kind of just went in grabbed all the core counts ran the workload. Shut itself down. And again your software really is the management layer that enables the orchestration of all this hardware under the cover. So we have basically Cycle Cloud does orchestration Cycle Server with a feature we call Submit Once where you submit it once and it will run in any region. Does the distribution so that's that ran parts of the workload in US West 1, US West 2 and US East in this case we ran and then we have Dataman which does the data movement between internal environments and external environments and kind of handles that process. So yeah but I think the key thing getting back to where you're positing is Spot definitely is an enabler of experimentation and that's the mother of insight and the mother of invention. That's experimentation is what you want and because we were able to go in get 70,000 cores run it at 700 bucks an hour get the 63rd most powerful computer on the planet at the time to do it. It's just amazing. That's really amazing and I think it's going to be transformative to what happens in science and engineering and risk management. A lot of fields basically everybody now is doing technical computing. I would actually say that people don't worry about websites and emails. How do I deploy that? It's a CIO I don't care about that anymore. I don't care about as much about the ERP and the CRM systems. Sure deploying SAP properly is an important goal but the workload that's hard now is the tens and hundreds of servers that are in analytics and big data and simulation and HPC all of those areas are the areas the technical computing workloads are the new enterprise workloads. Which is really because the economics enable it in a way that wasn't possible before. They had a lot of those questions they just couldn't afford to answer them or the ROI wasn't necessarily there and the competitive advantage you get out of understanding this is the drive head that will enable me to get more data off the platter faster so I can have a 20 terabyte drive that does an amazing job like that kind of science yields business benefits and that's really what technology is all about is enabling us to be able to solve problems we can never solve before. Awesome, well Jason thanks for stopping by again we're getting the hook again Jeff Frick here at theCUBE we're at AWS re-invent 2014 we'll be here today Wednesday Thursday going wall to wall getting the smartest people in the room asking the questions that you wish you could ask them and then bringing it to you live. Yeah and if I can just ask one question Jeff are you guys going to do March Madness again this year? I think, where's Greg? Yeah we're doing March Madness. And Jason won of course as you may have hopefully I can sit from the sidelines this year because it was competition, it's fierce. It was good March Madness check it out on siliconangle.tv just google CUBE Madness Jeff Frick signing off