 This is theCUBE, SiliconANGLE Wikibon's production of HP Discover. This is day two, we're here live in Barcelona. We have a special segment now, Vish Malshan is here. He works in HP Storage Group. We're going to unpack some of the things that are going on in Flash. And also, Pete Robinson is here. He's an IT practitioner with ExactTarget, and we're going to talk about how he's applying Flash and what impact that's having on the business. Gentlemen, welcome. Thank you, guys. Thank you. Actually, Pete, I want to start with you. Tell us about ExactTarget. What's the company all about? So we are a digital marketing company. We provide a platform through our private cloud for marketers to reach out to their customers via a multitude of methods, email, social, web, and many others. So you help brands sell stuff, really? Yes. Ultimately, yeah. Okay, Vish, let's talk a little bit about the three-par all-Flash storage approach you guys chose not to go out and acquire a company. You chose to do it in three-par itself. I talked to a shook last night, and he was saying to me, explaining sort of why that was, but I wonder if you could sort of reiterate that. Sure, Dave, be my pleasure. So from our perspective, Flash is disruptive. And depending on your starting point, that either means a new storage architecture or incorporating Flash as a media. And most have had to adopt a new storage architecture to accommodate Flash. And this unfortunately proliferated a fragmented complexity, silo, as additional complexity, and requires compromises, right? When you bring a new storage architecture, it always requires some sort of compromise as an architecture matures. And so if one could adopt Flash in an existing architecture where Flash becomes a new media, yet still deliver the Flash acceleration that is required for a new media, that would be ideal, right? And I think the challenge is it's hard to do. It's hard to bring an existing architecture and bring a brand new disruptive technology and not suffer from compromises, right? And I think that's one of the things that the approach we took with three-par because customers can avoid the risk, they can avoid trade-offs, and they basically can get a continuum of data services and they can adopt Flash when they're ready. Okay, so you're, when we talked about this last summer, you really emphasized, hey, we got a full stack. We started developing this thing in the early 2000s and we've hardened that stack. We're going to apply that now to Flash. You see a lot of the Flash players don't have a stack. I mean, TMS is a good example. IBM will use the SVC and they'll use that stack. You see EMC with its acquisition trying to build up its stack. All the, all Flash guys, you see what Violin's trying to do with Symantec. And so, so obviously the stack and a robust stack is a challenge. So that was your big bet is that you're going to, now the obvious question is, okay, but can you bring the same value proposition as the startup guys? Yes. With regard to performance and, you know, IOPS and latency and scale. Yeah, and I think, Dave, there is a difference between a traditional storage architecture and a modern storage architecture like 3-par, right? And, you know, early on in our offering, we actually been through a couple of media transitions. We've been through a couple of technology transitions and the robustness of the architecture allowed us to bring Flash in as a media without the compromises. Now, Flash is different. There's no doubt about it. And we do handle the Flash differently. We recognize when it's in the system and we operate with it differently. Okay. Now, Pete, we were talking last night, you and I, and you were sharing with me that you came to ExactTarget. You were the first storage admin at ExactTarget and you had, I think you said you had no experience with 3-par. This is before Flash was the big trend. And so share with our audience what you found, what you discovered, what the experience was like. Sure. You know, when I came on board, went through the same learning curve that a lot of customers do where literally our training was a two hour morning session and then we went to lunch. I honestly, I was looking at our SE going, seriously, what else is there to, come on, it's not that easy. It can't be that easy. And, you know, we come back from lunch and we were provisioning and managing arrays after lunch. And you were able to manage a lot of storage pretty much off the bat, right? Yeah, I mean, I had a storage background. So once you understand the terminology and the concepts, it's very easy to start applying it. You know, to give you kind of a reference point, when I was hired, we had one individual managing over two petabytes of 3-par and he was not a storage guy. He was a developer that became a sys admin who became a storage admin. He grew with the company as we grew and he was really learning as he went along. So when we came on, when I came on board, I was really, you know, initially shocked how, number one, how easy it was. Number two, how one individual could manage an environment this size. Yeah, we were talking about last night, I think I've seen some data from, I think it was Gartner that the average is well under a petabyte. And I think I was sharing with you guys that I was at a customer council last week and one of the folks on the council was very proud of the fact that they were managing a petabyte per person. You're saying it was over two per person. Even though I've said publicly that, I think petabytes per person is kind of a funny metric. It's an easy one to track, but it doubles every 18 months. Volumes is probably a better indicator of complexity, but nonetheless, it's hard to... Well, and you mentioned volume. I mean, you look at chains, right? You know, how many requests are coming in? That really is that indicator of how efficient can you be. Today, we do most of our provisioning via scripts and we're extremely efficient at it. We're starting next year. We're going to be automating our provisioning even further to the point where we're literally push-button deploying our applications using the interfaces that 3PAR provides. Yeah, so 3PAR has always been kind of the gold standard of simplicity of high-end storage. I mean, that's been well understood. Okay, so talk about your adoption of Flash. So the summer 3PAR announces this HP, the All Flash Array. When did you bring that in? Were you an early customer? Talk about that a little bit. Sure, so we had a chance to get an early look at it this summer before Discover. And we, you know, in our history looking at 3PAR, what we found is that the arrays behave remarkably similar to each other. You know, as every new release comes out, you know, it's, you know, faster CPUs, bigger pipes, you know, more drives, but from a performance perspective, from a scale, scaling that performance, it's a very linear. And the only difference really is is how high or low do I start and where do I end up? But that line tends to be very similar across the different product lines. So by the time the setting 450 had come out, we had already made a fairly large investment in Flash drives on the P-10000 series. So we already had a chance to see the 3PAR Flash in production. And what we were seeing, again, really matched what we were expecting from it. We were seeing the very low latency on the back end. You know, one of the, I guess, what we called a catch-22 with 3PAR is the benefits that we've gotten from the white striping and the multiple controllers. We've really been able to outlast or out-drive what a standard hard drive normally handles in a standard storage array. So we were already at that bleeding edge that we needed to decide, do we just start buying more and more hard drives or do we look at creating a performance tier strictly on Flash? And after seeing what the P-10000 was doing, after seeing that that performance tier was not only performing on how we were expecting it, looking at a truly optimized array running only Flash, it was a no-brainer. So what were your concerns prior to bringing in the all-Flash array? What were some of the things that you said, I'm going to be watching? What were some of the things you were skeptical about? Well, the one thing that we've looked at as we looked at all the different Flash vendors out there, the one thing that stands out is really capacity at scale. Does my application fit into whatever platform that you're selling me? And at 90 plus percent of the vendors that come in here and try to sell us on Flash storage right off the bat, my application doesn't fit on Flash. So when we looked at a 7450, the first question we had to answer was, does my application fit within the footprint that the 7450? You mean, can I fit my entire app? Exactly, the data, does my data fit into that space? The surprise that I had, Dave, as a vendor producing this equipment, I was surprised at the capacity people were putting on these 7450s, right? And I asked customers like Pete and others, and their point was simple, if I can't lay out my data, I can't use it no matter how fast it goes. I need a sufficient amount of capacity to lay out my data. So paint a picture of the apps that you're using for Flash. Well, so we rolled out, this particular application was a VDI deployment. It was a brand new project that we had not deployed VDI before. So there was some unknowns around what the application size was going to look like. We had some good estimates, but estimates are just that. They're guests based on what we think the application's going to do. As we look at the 7450 use cases, we're looking at doing more with it. We have a very large database in a VM or environment that I want to start testing on it on all Flash Array, but that capacity has always been that really limiting factor. So when three part came out with the 400 gig MLC, we finally hit that tipping point and capacity where now my application starts to be more feasible and fitting into that disk space. That's the announcement here. Well, no, that was a June announcement. That was a June announcement. So now they're coming out with the... 480 and the 920? The 920, so the capacity's continued to increase. And as the capacity increases on these Flash Arrays, the portfolio of applications that have a potential to now fit into that size increases as well. So you talked about VDI, and VDI is always a challenge. I mean, with the boots or storms, you mentioned sizing. It's really tricky to size VDI. And so what are the results been the VDI that I want to talk about the database? What did you say? It was exactly what we expected. We've rolled out dozens of three part arrays over the years with really no issues. HP installs it. They hand us the keys to this new array and we go to town provisioning storage. We had, I believe we had, we had Luns provisioned off of the array the day that it was installed. And our project, which had an extremely aggressive timeline, rolled out actually slightly ahead of schedule. And users are happy? I mean, performance, I mean... Performance has been, well, better than expected. Our average service times that I looked at it before I came to the conference, we're averaging about 300 microseconds regardless of what the workload is right now. Okay, and when you think of all flash arrays, of course you do take a database. You mentioned you got a virtualized system. Is it Oracle database? We're running SQL server today. Okay, so you got SQL server and you're going to put the entire database on an all flash array. Are you doing that today? Or is that the plan? We're not doing that today. That's one of the options that we're looking at down the road. Again, looking at the capacity, we have countless, several databases that are tens of terabytes in size. So when I look at putting a database that continues to grow, I have to know how much capacity can I grow into before I literally have to move it off to something else. Right, okay, and today those databases obviously leveraging spinning disks, right? Which is this mechanical bottleneck. So what do you expect to see when you move to an all flash? Well, we rolled out, so we rolled out on the P-10,000s with AO and Flash. And what we saw was on the back end, we're getting a consistent 600 microseconds on actual flash drives across the platform. And if you look at the percentage of IO that we're pushing to that tier now, we're now averaging 50% of that IO for my database environment in one use case in particular, all being driven to that 600 microsecond service time compared to the other 50%, which today sits on hard drives that are running between eight to 15 milliseconds. So when I look at my database, I'm getting a larger percentage of my IO requests now being serviced from that low latency tier, which is what we would expect. Right, now, are you going to evaluate other products or have you evaluated other products? We have. And usually I tell them the same thing. The first thing I hear is we can do a million IOps, two million, five million, 10 million. Billion? Yeah, okay, it doesn't matter to me. It's more of a question of use case, scalability, manageability. You know, I can't afford to chase after every little golden nugget out there and create 10 different silos of storage because at the end of the day, when we have to come back and now manage that, the complexity is what's going to kill us. That's the attraction to you, predominantly. Yes. You're looking for, you're coming from wrong, but it doesn't necessarily have to be the absolute best performance on the planet if you can get comparable performance with the stack and the manageability and the same point of management. The same ease of use. I don't want to compromise. Is that a fair statement that I made? Yes, I don't want to compromise on the features that I have today. I don't want to, I'm not looking. What features? What are the things that you want in that all flash array that you're leveraging heavily today? Well, as an example, we're getting more and more heavy into data protection. So snapshots, replication, the automated provisioning, all these different things that we're starting to roll out work really, really well when we have that unified platform of a 7450 or P10,000 or 7400, where we don't have to spend extra time re-engineering and redesigning that wheel. There's one wheel and it works regardless of where the application sits. What about the cost? Everybody used to focus on the cost. I want to have that discussion because you guys got to pay the checks, write the checks. So what about the cost factors? Well, cost per gig is always the first question that comes up. But my argument here recently has been that cost per IUP, dollar per IUP is becoming equally as important from a planning perspective because if I look at the performance that I need from my application, yeah, I can go buy seven racks worth of spinning disk and get the performance I need. But now my performance is where I need it and my efficiency is there. But from a capacity perspective, I'm hugely inefficient now from a power and rack space perspective. So when I look at the dollar per IUP at a more equal level as the dollar per gig, all of a sudden the higher dollar per gig cost on a flash drive starts to really come back in line with what you see with a hard drive. Because of the business value. But now, as well, a lot of the data reduction techniques being applied to flash lower the cost. Now some of the competitors, Vish, are going to say, well you guys don't have, just take an example of Ddupe. A lot of guys are doing Ddupe inline compression and that's how they're competing from a cost standpoint. You don't have some of those features. How do you respond? So, you know, Dave, I think several aspects there. So first of all, with flash, you want to be very right efficient. Not only from a capacity perspective and cost perspective, but also from a media endurance perspective. So many of the same white striping capabilities that we've done on the three power arrays, which were predominantly in the past, back end, this loaded, that white striping across the controllers and the ports, allows us to write very effectively onto the array. Now, the other thing I can tell you, Dave, zero detection. We use it a lot in VMware environments. That minimized cost. Thin provisioning still applies in the flash environment. And for things like Ddupe and compression, stay tuned. Yeah, so, okay. So, but you're comfortable where you are today in terms of your ability from a storage reduction standpoint to compete in the marketplace and saying you got more to come. And can you confirm that? I mean, based on your independent evaluation of other suppliers, HB from a cost competitive standpoint is there in your view? I believe so. I mean, we can always, you know, my former director always like to say, you know, the first one is always free, right? So, cost is always, you know, a key indicator of, you know, what we look to buy, obviously. All right, Jess, we're getting the hook. So thanks very much for coming on and sharing your story, Pete and Vish. Really appreciate it. Thank you very much, Dave. Appreciate it. We'll be right back after this word. Keep right there, John Furrier and I. We'll be right back with our next guest. We're live from HB Discover.