 Live from the FIIA Barcelona Gran Vía Compensator in Barcelona, Spain, it's The Cube at HP Discover Barcelona 2014 brought to you by headline sponsor HP. Back to Barcelona everybody, this is HP Discover and this is The Cube, our live mobile coverage of HP Discover. We're here with Fish Mulsion and Lee Pendlow, Lee Pendlow is an IT practitioner at Sony and Fish, Cube alum, director of products at HP Storage, great to have you on, good to see you. Good to see you Dave. Lee I want to start with you, because you're new to The Cube, you've got an interesting background when we're talking about off air, but talk about your role at Sony and then I want to get in to share with our audience a little bit about how you got here today. Currently I'm senior director for production services at Sony Network Entertainment. Sony Network Entertainment is the entity inside Sony that offers PlayStation Network and Sony Entertainment Network, which is our online streaming videos, downloadable games, support of the game consoles like PS3 and PS4. Yeah I know it well, my kids know it well. The storage requirements there are interesting, but you came to this role through an interesting route. You started, we were talking in the 80s, developing optical storage, I was just breaking into the business at the time, and the whole buzz was how optical was going to replace hard drives, and so that's when you started as a developer of technology, talk about how you ended up here. Right, so in that time there were some really significant technological problems of actually trying to store data on optical disc, and it was through that technology that's basically solving those communications problems, which was really the same as network communications just in a small form. As far as putting data on the disc and recovering it was no different than putting it down to fiber. And in that time I moved on to actually digital broadcast, or direct broadcast satellite, direct TV, developing that, applying those same solutions to those problems, to making those products, and then from there moving actually into digital cable work in early digital cable, as well as early digital television, and of course then all the HDTV and the things that came from that. From 2006 onwards, as an outcome of that, I led a team with Sony that actually was the originator in large form, from what is now taken as de facto IPTV or streaming TV, all those early technologies working with early partners like Netflix and Amazon and others. As a consequence of doing all that work, which is really the merge of video and traditional IT, if you will, because it's through data centers and infrastructure, we had to build those kinds of facilities to do that kind of high capacity, mission critical, always on broadcast streaming, and out of that basically we moved into Sony Entertainment to handle also the commerce functions and the other things that go with it that weren't one of the IP streaming itself, but the back office that actually makes those kinds of businesses for supporting game consoles and things like that. Okay, and then so that we're going to come back and talk about the unique requirements that you have from a storage perspective, but I want to go to Vish and just, we last talked in the Cube in June, that used to be discovered in Las Vegas. You guys have been busy since, got a bunch of new announcers in the show last night. Take us through what's new since June. Sure. Since I was last on the day of June, we've seen a massive explosion in flash demand. All flash arrays very well received. Revenue for the year of 12 months, FY14, FY13, 6x growth in all flash revenues. So phenomenal reception from our customer base. Interestingly enough, the most frequent use case we see before I get to some of the announcements, Dave, is customers are sort of struggling with delivering a service level. They're burdened by complex large arrays, lots of spindles, and even after all that money and time and it's wet, still don't get the service levels that they need, right? And the flash is sort of just changing that for them, right? Reducing it down, consistent performance levels. We see that a lot. That's one of the primary use cases. So, at Discover this week here, Dave, we've announced three things, right? We've announced what we call the Converge Flash Array. We've announced Converge File, Block and Object Access in FreePower. We've also announced sort of flat backup from a FreePower Array snapshot to a StoreOnce device while we're looking at it. Okay, so give us a little more detail. Sure. On the Converge Flash Array, what's the interesting use case there, Dave, is one that we really didn't think about, but customers were asking us that, say they start off at the Flash Array, and after six months or seven months, the data ages, right? So, they're not doing tiering, they're not doing caching, they're doing all the flash, but the data ages. And what they want to do is instead of moving the data off to an archive device, they want to move it to spinning on that same array. So, we said, oh, interesting. So, is that like 50% flash, 50% spinning? They say yes. And so, the 7440 Converge Flash Array takes that into consideration now, where you can start off all flash, you've got the same performance, same features, same hardware specs. But then, if after six months you want to move a portion of your data to spinning, you have that choice. If you choose not to do it, that's fine. Leave it on all flash. How do I move it? There are ways, there are commands that you move data. You can schedule it. Now, there are also automated ways of moving data, and you get that into the realm of tearing them. Right? So, again, the customers that we talk to tend to today want to do it manually, maybe scheduled, and it's time-based more than anything else. Okay. Okay? The other one with the Converge File and Block. Basically, I think it raises the line between block and file. You can now run native file and object access protocols from the 3-poor array. Easier to manage, easier to monitor, and one-third the capacity, right? And then the last one is this notion of a flat backup. I think, Dave, for the first all-flash array now that gives you the ability to take a snapshot from a flash array, move it directly to a store-once device. And with that, you get 17x faster data movement. We only capture the differences of the snaps. When it gets to the store-once device, you can then recreate the entire full backup without having taken all that energy and time to do a full backup. So I seed the backup once, and then I'm taking the changes. Yes. And I'm creating an index. Correct. Correct. So I'm not taking zillions of copies. Correct. That's the beauty of flash. Correct. But then when I recover, I'm not recovering full backup and then applying incrementals on it. I can take the full backup. So it's a faster restore process. And the software that is doing those periodic changes is what? Yes. The software that moves the data is called Store-once, Recovery Manager Central. And then within Store-once is something called Express Protect that captures the snap diffs. And it's three-par snaps. It's three-par snaps. They're doing the continuous data protection, essentially. Correct. And it's also within vCenter, for example. The first version runs in VMware environments. So from within vCenter, you can click on a data store and set up VMs and say, hey, protect those VMs. And automatically makes the snaps VM-consistent, moves it automatically to this. And I can schedule the granularity of when I take the snaps. Yes. Every hour on the hour if you want. 15 minutes if I want. 15 minutes if you want. I mean, the array supports, I don't know, a max of 20,000 snaps. So plenty of snaps. OK. All right, Glee, I want to come back to you. And I know you're not supposed to endorse the products or anything like that. But based on what you just heard, I saw you sort of nodding about the tiering. And I want to get your reaction to sort of how you approach these things. But let's step back a bit. Talk about your unique storage requirements. You talk about PlayStation environment. Lots of users, lots of demand, gaming. Talk about that environment. And what's so critical to you? For us, the critical element is our commerce functions. People playing video games want instant gratification and being able to purchase from their machine. They don't want to go to another place or go to a PC. They want to do it while they're doing it to new calls by. So they want to be able to transact right then, right there, instantaneously. So that's a mission critical function. I may not get another chance to get that customer to come back. They think about it for a while. They may go and find some other entertainment opportunity. So for us, that was the basis of our business. We had a system that we put together back more than four years ago. Made investments in very, at that time, very high end, tiered storage. Three tiers of storage as our block storage for our commerce database system. And we were seeing that it was being forced into an economic end of life. Even though the system was still quite performant as a function. We're running out of capacity. We're running out of compute and we have been pouring money into it year over year. But we weren't really getting return on investment from that additional incremental investment because it was just running out of steam. With the introduction of PlayStation 4, our business went up so dramatically that time scale was advanced rapidly forward and we had to quickly come up with a new solution and do something bold and something different. As opposed to just buying another array. Right, you know, just start racking and stacking more here. Basically, we were beyond that ability to do that economically. So we had to move to something else. Flash had reached a level of maturity as far as parity with cost parity in general with spindle that it was worth looking into that. So we started an activity, quite rapid activity to evaluate different vendors of all flash arrays to find if there was a solution that would work for us. And we came to three parts. Okay, so you essentially brought in a three-power all-flash array to replace a tier one system or multiple tier one system? Well, it was a three-tier system and had a small layer of incredibly expensive SSD because it was bought back in the day and then had both SATA and SAS drives. We spent a lot of time migrating things around to optimize storage on that system which was a huge labor cost and you always wound up with the data you seem to need most on the slowest elements here, constantly playing that shuffle that Vish was talking about. And that was a manual exercise or it was quasi-automated? Well, it was purely manual. You're constantly reevaluating, looking for hotspots, looking for stale and cold data and constantly reshuffling them. You're in the CLI bang in the way. Okay, basically, okay. And how much capacity are you talking about? Roughly. So that system was running about a half a petabyte? Half a petabyte. And you brought in an all-flash array of similar capacity here? Oh, much larger. So that was the whole point as we were looking at for a system that would provide us immediate additional capacity as well as performance, but also gave us room to grow because we didn't want to have to go through the same exercise in a year, in two years. We're growing exponentially right now. So what we were able to do is take, in our data center, the space of seven racks in our data center occupied by that product and replaced with just one rack of three bar. And in that one rack of three bar, it's only 25% occupied. So I've got tons of room to grow by just adding drives and expanding that capacity. Okay. And how long was that array? When did you... How old was it? Yeah. It was only three or almost four years old. Okay. But relatively... Yeah. It should have been performance from a service life, but the business was such... Business growth was such that we had outgrown that system and couldn't keep up and there's no further investment we could make into it. Okay. When I saw what they were doing right, it looked like they took... They had so much drives for the level of performance, right? To get that same level of performance completely correct me if I'm wrong, they reduced the footprint. But they said, look, I like to reduce footprint, but I still got capacity I need to grow to. Yeah. Okay. And so why three bar? You got a lot of choices out there. What's the deciding factor? We brought in a product from all the major manufacturers and we literally did a babe off side by side in our data center with real workloads and we were evaluating performance, scalability, resiliency, and total cost of ownership. Those were our criteria for evaluation. And, you know, quite honestly, all the different machines effectively stored data, right? We were able to access data on it. But when you started looking in detail on those four criteria, there was significant differentiation between the different brands and we found three bar basically hit all four of those points, hit the ball apart. So that's high level. Now, the one thing about three bar that I've always observed is most all flash arrays have a very immature stack, I'll call it, that hasn't been hard in over a decade. Three bar's got the stack. It didn't have the all flash array. Then you guys introduced the all flash array. That was an advantage. How much of the performance, scalability, resiliency, and total cost of ownership relates to the majority of the stack, if I can use that. Resiliency. So that whole idea of maturity, one of the things we did, because this is real world and a mission critical system, we'd actually go in as part of the test and break the system. Start pulling drives out, start pulling controllers out, start pulling cables out. A system that's not mature. We saw systems fail, fail catastrophically. We were actually informed by certain manufacturers, you can't take anything you want out of the system, well, not that one. Okay. We went to this three bar and we actually could remove things like a controller, which would be one of the most catastrophic failures you could have, but we actually saw no degradation of performance. So that's the other question I wanted to ask you. Relative to each other, were you able to discern a significant performance advantage for any of the systems that you tested? Yes. For just simple, you know, read, write, flashes, flash, and they in general would perform the same. But when you're starting to do lots of churn, where the flash, there's a lot of garbage collection going on in the background to rebalance the system, that's when we saw a big differences between systems. So when the thing filled up and you had to move things around, and did you see that with the three bar array? Well, we knew the three bar was in the background working, but we never were able to see that effect of performance where on other systems we actually saw the system perform out. The consistent performance is what you saw there. Correct. All right, I'm going to ask you about the migration. Because migration, we've studied this in the Bibi, migration can be very expensive, as you know, any array to any array. So what was the migration like from the tier one system, or the tier three tier system, the model of the system to the all flash array? Well, this is something that we were really concerned about because we did this right before Black Friday. Really? Yes. We didn't have time. So very recent? Yeah. Two weeks ago. Two weeks ago? Black Friday's two weeks ago, right? We completed it two weeks ago. And one of the things we were concerned about was that migration data. And the tool set that comes with the three bar array performed all that migration from that old system to the three bar array basically hands off. So one thing I can say very clearly is it was the most drama-free integration we've ever had. So there must have been a lot of concern about doing this just prior to... There was a lot of concern. Right? Why did you take that risk? Why didn't you wait? And what did you do to de-risk it? I couldn't wait. Again, we were out of space. So it was either do something now or, you know... Move business, right? Move business to holidays. Yeah, holidays. You know, the time frames for acquisition, purchase, receipt of materials, all these things that go on. We started the project in July and it took until October to get it done. So, yeah, it was... And your experience is what you expected, consistent performance, resiliency, processes that you've set up. It exceeded my expectations and our storage engineers were absolutely delighted with the experience. It's seldom that you see them smile and speak positively about vendors, but they're absolutely tickled to death. And what are you doing around quality of service to ensure quality of service? But how does that all sort of fit in here? Are you pinning quality of service to the app? So, for us, the quality of service is managed by the applications. And in that case, you know, we're trying to set a bar for quality of service and basically moderate the system around that. That's the floor of threshold. It is. And I was saying, in our current, our new system, the performance of that system is so much better. We're seeing a 5-10x increase in performance that actually we've been able to reduce our investments in application servers and application compute because the hidden choke point was storage. You've reduced your number of cores that are required. Correct. Well, we were actually planning to double the number of cores that we were running as part of this expansion for the holidays and we actually were able to defer that investment. We don't need it now. In fact, we can actually retire some of the nodes we're running now. What database? Oracle Rack. So, Oracle, of course, prices per core. So, your license and maintenance costs. Way down. That's awesome. Let's weigh down. Double digit percentages. 50% reduction in Oracle license and maintenance costs. We were going to have to double our investment. That's awesome. I love that story. You had that story, right? It's great to see customers, actually. When I was talking about the total cost of ownership, there's the initial acquisition cost of hardware, right? And, you know, 3-par was competitive in that area, but when you look at total cost of ownership, it's not just the checking right, it's the checks that you don't write. So, I didn't have to write checks for new licenses for database. I didn't have to write checks for additional servers. So, actually, you know, there was a net savings by going to that route. That's an awesome story. I'm getting the high side for about five minutes, but it's such a good story. I'll give you the last word, maybe summarize the... You must be very excited. Dave, you know, these great customer case studies that we are starting to see now, and I think the delivery of the service level, right? And David talked about it in his keynote, service-defined storage, right? I think that's so key. So many people are basically struggling with just getting that service delivery and seeing these stories very gratifying to see how we can help make a difference for these customers. All right, Lane, Vish, thanks very much for coming to Cube. Great to talk to you. Keep right there, everybody. John and I will be right back after this word.