 Welcome back to Silicon Angles the Cube here live and Santa Clara, California in Silicon Valley for a Strata conference, O'Reilly Media's big data event sold out, great show. This is Silicon Angles flagship program, The Cube. We go out to the events, I expect a signal from the noise. I'm John Furrier, the founder of SiliconAngle.com and I'm joined with my co-host. I'm Dave Vellante of Wikibon.org and we're here with Jeff Denworth, who's the CMO, Vice President of Marketing of Data Direct Networks. DDN is a company that has really specialized in storage for high-performance computing environments and really this segment is about HPC meets big data. Jeff, welcome to The Cube. First time on The Cube, so it's- Thanks guys, great to be here. First time at Strata too, so that's great. It's not actually a cube though, I mean you at least have a rectangle and some triangles going on. We're talking to our CFO about that. Oh, we don't have a CFO, but we're working on it. The set designers are working around the clock like elves right now, so we're going to get that fixed. Thanks for pointing that out. We're big fans of your company. First of all, DDN, for the folks out there, go to ddn.com if you want to find out about this company. Been around for a while, pioneering a lot of high-performance, hyperscale, high-performance computing and storage, but self-funded Dave, so I made a tweet earlier that Tim O'Reilly then retweeted. I went on a little rant about how the real rock stars in entrepreneurial culture is the guys who started with no VC money and can go the distance all the way and create its durable company. To me, that's the elite class and ones who can bootstrap and then get funding and grow and be successful. Well, particularly the one that's grown to several hundred million dollars. I mean, that's pretty rare. So, you know, DDN has a swagger to it, I like that, and also a big fan of John Luke Shudlane who came from HP, knows the space real well, understands convergence, understands the confluence of where this is all going. So, and he was also one of our first-ever cube interviews at our office when we were co-located at Cloudera, so that's great. But you guys have had a shift. So, obviously, Object Store, Cloud, Hadoop, a lot of hype, a lot of competition now from the distribution side, but at the end of the day, there's still a massive demand for scale. And Cloudera talks about the petabyte club, but this large, large use case, although small now, but will be growing on the high end. So, Jeff, tell us a little bit about what you guys are doing now, what you announced, and then how does DDN look at the high end? You guys are kind of in that storage innovation bucket with big data, and it's not talked about much at these shows. It's all about data analysts, data science, but you need an infrastructure to store the data. So, talk about the news, the company of the news, and then scale. Okay, so, you know, thinking about where DDN operates in the marketplace, we have this kind of interesting position at the top of the scalability spectrum, and what comes from that is an understanding of scalability challenges that we don't believe that a lot of our competitors have. So, if you look at, let's say, our average selling price, it's many times larger than our competition. We, for example, this month closed our first terabyte a second file system project. This is at a U.S. Department of Energy Laboratory, just to give you some comparisons versus your average NetApp, oops, excuse me, versus your average filer that's somewhere in the range of, you know, anywhere between 100 to 200 times the performance. So, that's a single file system that we've deployed. And we do this periodically, you know, throughout the year where we get and support systems for the world's largest supercomputers or world's largest web-scale computing environments. It's kind of a secret about DDN where that convergence of HPC and kind of analytics or web-scale content delivery and processing is actually happening at DDN where we're seeing a ton of really, really exciting projects where you're talking about the petabyte club, John, I would say that any more our customers are starting to talk in terms of exabytes. So we're closing fractional exabyte orders today and we've got projects on the time horizon within just two years that we'll see exabyte-sized file systems ranging from maybe HDFS-style file systems to traditional, you know, POSIX file systems. So what are the similarities between HPC and big data and what are the differences? I think over a long enough time horizon they actually are exactly the same. Our HPC customers, so DDN we've been in the business for 15 years, we operate in a number of different markets but HPC is one of our largest supercomputing. And these guys are very frugal, right? They're organizations that look at optimizing their spend so that they can move the needle as much as possible to compute so they get their jobs done quickest. So they want storage platforms that ultimately support that endeavor by delivering productive throughput and reducing their total cost of computing. And so when we look at analytics or we look at web-scale computing everything kind of converges on a common architecture where you've got clustered computers that are banging on very large data sets, big data sets if you want to use that word today. And from the perspective of the product that we're launching at the show at Strata it's ultimately a cluster computer that runs in Finneben that's powered by the world's fastest storage platform with analytics software built in, managed with a very simple single pane of glass. And from our perspective that's not anything that's dissimilar from what we've been doing in the HPC space for the better part of a decade. So let me just try to break this down because I know before we get down deep dive in some of the HPC stuff. So David Floyer wrote a post on wikibond.org about age-scalicing, HPC means big data. So I want you to define what that means first. And then two, we've seen appliances come and go and kind of hang around out there for specific installs but you guys are more than appliance from what I'm gathering, right? I mean, we're seeing a trend where the storage guys are becoming compute. So I want you to talk about that positioning. I mean, appliance I'm thinking like it's, you plug it to RAG, that's not what you're talking about here. It's a fricking system, right? So, or is it? So let me answer your first question. So from an HPC perspective, I mean, the whole MapReduce paradigm is just parallel programming and the distribution of processing across a number of federated compute nodes to a large data set. Whether that data lives in the node or whether that data lives on the network in parallel distributed across a number of different devices, it doesn't matter. The same net result from an application perspective is the same. And so our job is to really minimize the cost of computing, make compute nodes go faster by getting rid of all the IO bottlenecks. We've got a million lines of code in the storage fusion appliance that we've provided in our Hscaler product to really help support that. So talking about what an appliance really means though, truthfully, you get a system for us, everything is totally tuned, factory configured and it's delivered to your door with just an ETL. So you get a single pane of glass interface, you get an ETL, you point your data sources to it and you're off to the races. People can be running queries within a matter of a day. So it's not really- How big is the appliance? It's big. It's big. It's not like, oh here's a box unpack it. No, it's not like your coffee cup there. It's bigger than a bread box. Appliance for us is about a rack configuration and we sell appliance, we sell basically a scale out appliance where you build a number of these platforms in aggregate to scale to what you would see is about the equivalent of an 8,000 node Hadoop cluster today. We get to that with a lot less computing so it's not an 8,000 node cluster on our part but it's pretty big. So I saw John Luke at the Open Compute Summit here actually at this convention hall a couple weeks ago, Facebook's here and it's all about open source, open rack. Basically it's like open source hardware. But I was talking to some of the Facebook guys and we had a hallway conversation, we had a little sit down and they were saying that the amount of orders that they're doing is so massive that it literally is there's so much software projects going on that just ordering boatloads of gear. So the challenge is in that, I mean Facebook's a skewed data point but you think about large enterprises, they have footprint issues, right? HP has pods, et cetera and so there's that issue. Is that the same market you're selling into? They need all this gear so is that what you're targeting that here's a turnkey product? For our, so we have a collection of products, file based storage products, block based storage products. We sell object storage as you all know. And now we have an integrated Hadoop appliance. In the case of the Hadoop appliance, we're not so naive to think that we're going to sell to Facebook or sell to Yahoo, any Hadoop software because they've got more engineers than we do in staff, truthfully. But that being the case, we are educating a lot of these style customers around. But Facebook is not a lot of market out there for how many Facebooks got Pinterest, Facebook and a bunch of guys like that. You don't need a lot of those to make a lot of money in any case. So we have a reference architecture and we put the reference architecture forward and we say, hey, here's a better way to build a scale out infrastructure. And for people that are kind of one level down, they don't have the Hadoop staff on hand. We'll provide an appliance, we'll handle all of that from a delivery and a professional services perspective. So Floyer, I'm reading his piece. He just put it up real time in the Wiki. It's, as John said, it's called DDNH Scaler HPC Meets Big Data. Now one of the things, Jeff, he's been talking about a lot is just the notion of faster IOs and lower variances with respect to database response times. And the premise is essentially, so what he did is this great chart here, he's comparing commodity with Hscaler. And look at the cost per node. It's significantly, it's right? So you're saying, okay, just as you said before, your average price is significantly higher than a commodity. So your cost per node's 24,000 versus 17,000. But the number of nodes required, the number of cores required is significantly less. And when you just look at the hard cost, forget about the people. It comes out to about, what's that, 2.5 million versus three, about 30% less over a three-year period. Which is totally counter-intuitive. But, you know, we start from, Wait a minute. Right, right. But we start from being an engineering company and you have to recognize that Hadoop wasn't really built by storage people. So as you get to a certain level of scalability spectrum, weird things start happening. Like system failures start dragging down the aggregate cluster performance. You have Amdahl's Law problem. The other thing is, counter-intuitively, people think with Hadoop, that the processing is always shipped to the data. But in large-scale systems, the processing actually happens 30% of the time on a node that's external to where the original job is started. So as a result, there's a lot of efficiencies that can be done by really building a decoupled storage system that can grow independently where you grow compute and storage, both in whatever dimension you need, not what you actually get through the whole scale-out paradigm. And then finally, by providing equidistant access to all these nodes to a very, very high-throughput, centralized storage system that can sustain performance even when things break, that's where the benefits really come in. We're making Hadoop nodes faster and we're making it easier to deploy. The easier to deploy is kind of the black magic number that Dave didn't even put into his calculator there. Yeah, he said this does not include the soft dollars or anything like that, which is, by the way, always the most expensive component of the TCO, right? Yeah, but people don't think about it like that. You know, when you're making your purchase, they look at it and they say, okay, I need to just assess my acquisition costs. And that's where we started the project. We said we had to be more affordable than a commodity cluster and the value proposition is it's simpler, but that doesn't mean that we get out of competing with the market. So we think we have a very compelling product. So where are you at with the beta program or are you in production yet? We're in beta right now. We'll be out of beta by the end of the quarter. So we'll be shipping very early on in Q2. And early feedback, I'm sure it's been good, but what's on the to-do list? There's a lot on the to-do list. He's not going to say, oh, feedback has been horrible. Yeah, I guess I say. I answer for him. No, no, so the feedback's been great. We've actually been showcasing it ever since last November. We brought out a preview platform to the first trade show that it made its debut at. It's here at Strata. I think that ultimately where we're at is we think that we can drive a lot more efficiencies into the platform and drive down the cost of computing further through making more optimizations at the Hadoop layer as opposed to in the storage layer. We've done our job there. So we're going to start climbing up the stack and really making IO pipelining, IO scheduling, and ultimately the cost of your Hadoop cluster a lot less. Stuff that, you know. Yeah, well, Dave and I have always been saying since the beginning of Hadoop as it starts to grow, we're going to start to see a lot more maturation in the actual subsystems and it has an operating environment. So, you know, we see this trend obviously on the high end developing fast. My question for you kind of the segment out here is what are you guys seeing at the show here? Obviously you're here to do some business. You're talking to some folks. What are you seeing? What kind of conversations you have with folks here? So the common thread is this whole, I don't know who created the ratio but the thread of, you know, out of every five Hadoop projects going on right now only one of them's in production. And so from our perspective, that's a gift and a curse. One, it's creating a bad name for the technology and the marketplace because people are understanding it to be too complex to deploy or just taking considerably longer than they expected. On the flip side, it's an opportunity because we're kind of at that cross the chasm moment and appliance vendors are ultimately going to be the ones that get this type of technology into the enterprise. You know, there's a lot of talks about shadow IT and as much as the CIO doesn't like shadow IT, it's a reality in a place where people have to do business and if you can just deploy a toaster into an environment and say, here's your answer engine. That's a good thing. So we think there's a lot of opportunity but I think Hadoop's at this inflection point where it really has a perception problem that it needs to remedy. Yeah, and we're seeing that too on the cloud. I mean, all the data that I talked to says, you know, Hadoop's terrible on the cloud yet Amazon Elastic MapReduce from my sources is really doing well because of the ease of use but they're charging a lot. Yeah. They're charging a lot for it. So not an optimized environment but people are using it. Yeah, well, the cloud has just levels of simplicity that make it easy to use. Yeah, it's hard. Okay, we're inside theCUBE with DDN again, self-funded startup over hundreds of millions of dollars now. It's not a startup anymore but founders are still around, you guys are growing and really had a nice positioning over the past couple of years into a really fast growing big data market from, you know, I don't say niche storage but you know, storage business is now center of all the action and big data. So congratulations. This is theCUBE Silicon Angles flagship program. We've got the events. Do you try to see them from the noise? We'll be right back with our next guest.