 Live from the Sands Convention Center, Las Vegas, Nevada, extracting the signal from the noise. It's theCUBE, covering HP Discover 2015, brought to you by HP. And now your host, Dave Vellante. Welcome back to HP Discover 2015 everybody. This is theCUBE, we go out to the events, we extract the signal from the noise, check out hpdiscover.social, it's our new digital experience, that we've overlaid on top of theCUBE for HP Discover, it's got all kinds of cool social feeds and videos and other content. Vish Malshand is here, he's a good friend of theCUBE, he's the director of product marketing at HP within the storage group. And Bruce Trevarthan is here, he's the group CEO of the cloud, public cloud service provider. Gentlemen, welcome to theCUBE. Thank you. Both of you have been on theCUBE before. So why don't we start though, Bruce, with your company, tell us a little bit about it. So the cloud is a public cloud service provider. Great name. Yeah, that cloud, well that was in early 2008, so we sort of got on the bandwagon nice and early and managed to secure the name. It's become a little generic, obviously, but it's a great term to hang our hat on. So the cloud's a public service provider in New Zealand, more recently starting to roll into Canada, with the intent of obviously providing infrastructure as a service type commodity services to the small business market in New Zealand, and again, that's pretty similar in other countries. On demand and with the zero touch, so automation is key. Great, and we'll get into that, and I really want to understand how you're using all Flash, what your thoughts are. We were just having a discussion with Craig and Bruce from Alcoa, I'm sorry, Craig and Mike from Alcoa on the All Flash Data Center. Love to get your thoughts on that. But, Vish, let me come to you. So we've been tracking this Flash journey that you guys have been on. It's been quite amazing. Yes, you need. I want to make an observation right out front. This, and I've said that, I tweeted this out. This is an example, what you have done with Flash is a really good example of organic development inside of HP. And I said it's really the first blockbuster example since EVA, and it really is. It's organic development, and you guys said we're going to do this on 3-par, we're not going to go buy an array company that's going to give us an advantage in the stack. So you've done that, it's clearly been successful, but there's been a lot that's changed since we last met. We were in Barcelona. Yes. We had the PEDLO on. So what's been new since we met last early December? Sure. So Dave, I think the initial deployers of all Flash have all had very good experiences. They've been able to get predictable performance. They've been able to lower costs in many different dimensions. And feedback has been very positive. And what I'm hearing now is they want to do more. They want to do more applications on Flash. They want to do more things on Flash, right? And so it's interesting how close are we to that all Flash data center tipping point? I know there's been some in the community that have talked about this. It's clearly on the leading edge, right? But I don't think it's that far away if people start to move more and more applications and start thinking about how they deploy the Flash. David Flores says we're a year and a half away from the tipping point. Yeah, I know. I'm at a much standpoint, which is he's the only guy saying that, but I like it. I like the conversation. It's a good conversation starter. What do you think, Bruce? As far as tipping points concerned, I'm probably more inclined to think that it's going to be a transition. We're using all Flash arrays now, three HP, three power, all Flash arrays. And the key element there is performance, but you're always going to have data that is not being used, is not in flight, is not really accessed on a regular basis. So to put that on SSD seems a little overkill. But, and right now today, the price point is not there to compete with Neoline. If it's cheaper. If it's cheaper. You would put it there, if you would. You would. Unless more of a transition around pricing. Let's tape actually transforms again, right? I keep saying Flap. Flap is an interesting idea to say. It needs some development. Yeah, yeah. Okay. But I mean, you know, Dave, the way I think about the all Flash data center is, okay, let's tape is not dead, disk is not dead. I don't think it'll be dead. The question becomes, where's the center of deployment? Where's the center of innovation? Where's the lion's share of spend? In, and that's what I think the old Flash. And there's no question. Yeah, right? Where that, that the market's rotating very rapidly. And I think, let's now talk about, you know, if we look at some analogies, right? Uber, Airbnb, Twitter, okay? Those platforms all grew on the fact that the internet was ubiquitous, right? Ubiquitous on phones, ubiquitous at home, ubiquitous on the PC. So here's the question. If Flash becomes somewhat ubiquitous in the data center, then what does that baseline infrastructure enable as capability, right? What use cases, right? One that I know Dave Floyd talked about, which I thought is getting a lot of interesting play here to discover is this use case of baking copies, the data. Data sharing, yeah? And then, you know, keeping the same copy on the same, oh sorry, keeping the copy in the same array and using snapshots and using a fraction of the capacity, right? And so that dollar per 50 gig Flash that you buy today, if you can make six copies and share it on the same device, are you down to cents per gig now? And then to your point, Bruce, I mean, you know, if you look at that cost equation, does that become lower than New York? Well, it does. That becomes very compelling, right? You know, we touched on tape before, tape's always going to have a place. You look at that going out to LTO 10, you're not going to compete with tape for a long term. From a cost standpoint, no way. But look, if you can get Flash down to that kind of price point, then that'll kill an airline as well and then we're done, right? Then there's less spinning disk. And of course, disk is not a, it's a moving target, but Flash prices are coming down faster than spinning disk. Yeah. So we think it's just a matter of time. It's just a matter of time. So how are you using Flash in your data centers? So I transitioned to all Flash last year, sort of end of last year, was really around performance. Now, as Vish just mentioned, if you've got the performance available, then it opens the door to do a whole bunch of things that you never really thought of. You need six copies of something and you typically keep them on different sands because of the performance impact. Well, now you can keep them on one sand because the performance impact is not theirs, it's not a concern. And because they're on one sand, you can leverage things like D-Jube, Snapshotting, all sorts of technologies that you couldn't if they were separated physically. For us, around the performance reason for going all Flash, and we're seeing a lot of customers placing bigger and bigger demands on us as a service provider that we could not actually accommodate under a spinning disk. We had a combination of spinning disk and some SSD historically, using that to augment the spinning disk and adaptive optimization sort of concept. And we still could not meet the demands of the heavy spikes that we were seeing from some users, particularly some of our users in the big data space. You know, they're ingesting huge amounts of data very quickly and then doing nothing for a few days, possibly. And so to cope with that, it takes me back to the pre-virtualization space where you're having to build out for something you do once a week. Why do that, right? So with all Flash, we can now accommodate these massive spikes and deliver on promise, right? So we can make big promises and to back them up with an all Flash environment and deliver on that promise. So what about, I'm curious as to whether or not you're able to monetize that infrastructure from a quality of service standpoint. Are you able to deliver a guaranteed level of quality of service to users and charge them for that, for example? We absolutely can. So I know another technology that's not brand new, but another great technology of HP3 Power is performance optimization. And obviously with the QoAce elements in there, we can guarantee IOPS per customer and deliver that out and guarantee that they will get the performance that they expect when they need it. Are we at the point where you can provision capacity, provision IOPS, maybe even bandwidth programmatically? Yeah, so exactly, Dave. I think going back to that copy use case, so let's say you have production and you have test dev, and let's say you had some analytics run. Let's just say three copies, right? And traditional wisdom is to keep them separate. But let's say you put them on one array now and you put God rails, right? You said, okay, test dev gets 30%, no more. Analytics gets 20%, production gets 50%. Now, maybe production's running at 30%, you give it a 20% room to spike. And now you have the protection that none of these other workloads can overrun production, right? So it's the combination of all flash, scalable performance, and then priority optimization software to give you those rails, right, and comforts. And then if you want to do that programmatically, you know, we actually announced priority optimization controls within OpenStack. Right, and so you can actually program it through OpenStack or REST API that we have available in the array as well. Are you using OpenStack or? We have installed OpenStack in our lab environment. We're definitely staying close to it. We don't have it in production yet. What do you have in production? Our environment, I'm glad you asked. Our environment, actually, with the provision I was talking about this earlier, our environment from a smart point of view, from a zero touch customer interface point of view is completely homegrown, completely custom built. And we just launched our third version of that, which we've been refining over the last eight years, which provides customers the ability to deploy servers on any of our locations in four minutes or less themselves at any time of day. And we've recently just included the support for multi-tiering or split tiering per VM, which is a key thing that the all flash arrays allow us to achieve, because now we can cut up the performance tiers and sell them at a different price point. And the best use case I can think of right now as a customer with a SQL server, they might want to pay sort of the middle of the road disk type prices for their OS volume, they want the highest possible performance for their log volume, and then whatever they can afford for the scale out requirements of their data volume. That's three different tiers of disk in one VM, rather than trying to stick it all on SSD and pay to put your data at rest on SSD and whatever. So you get granularity out of the infrastructure and you're able to monetize that. We can. More granularity can get the better. Can you set floors and ceilings? Quality of service or is it? You have to talk to the operations guys, about the actual technical detail on how that's done, but I think it's pretty smart tech. Let me ask you a question there Dave, if you don't mind about Bruce. So I mean, you know, so you said you have tiers, some flash, some spinning. So are there some separate arrays or? So historically we were dealing with the whole multiple offerings of storage out of our environments by augmenting near line and SAS for the low end and augmenting SAS and SSD for the upper ends using adaptive optimization. But there was still a limit to that because we're dealing with spinning disk. The introduction of all flashes meant that we now use the priority optimization, performance optimization technologies to actually govern the IOPS, knowing full well that we have an abundance of them. And so we're no longer limited by the actual physical capabilities of a disk. When did you first install All Flasher A? The All Flasher A was installed in October 2014. End of last year. 2014, so prior to 2014, 100% of your primary storage spend was on either spinning or maybe hybrid. It was hybrid. Yeah, so we were running three piles with a mixture of near line SAS and SSD. Okay, so 0% at that time was All Flash. Great. What would you say, look out 24 months from now. What percent of your spend do you think will be All Flash versus sort of hybrid? It's pretty easy to answer that because we have a very prescriptive build and so we know exactly what our stack looks like. So from a ratio point of view, we would be spending three quarters of our storage budget on flash and 25% on near line. So we're no longer by spinning disk at the middle SAS disk. So three quarters of your spend by 2017 will be All Flash. IDC numbers say, and I saw this at a recent conference, that 17% of the revenue for the entire storage market will be All Flash. And that struck me as conservative. Yeah. What are you guys, how do you guys see the market? I mean, you know, other guys do the forecast, I know, but I know every forecast is wrong. It's always hard to forecast things like this, right? I think from my observation, Dave, I think the forecasts have been a little conservative. I think it's gone beyond everybody's wildest expectations. It's exceeded everybody's expectations, right? They had a pretty aggressive forecast, but it's even exceeded that now, right? So that's really a big groundswell and a lot of momentum building on itself. Has the uptick surprised you? You know, it has, it has, it absolutely has in terms of how aggressive it's been and how it's sort of feeding upon itself, right? These things, in our experience, go on an S-curve, O-Gyve, O-Gyve curve, when you get the steep part of the S-curve, you know. Inside the tornado. Yeah, the return is so much greater than the effort and then, boom, it just takes off. And a lot of times the market, I've done a lot. So I've committed this crime because you get these straight-line forecasts. Yes. You know, things die like this. Oh, right, right. You go like that. It's hard to predict sometimes, but we think that we are on the cusp of that real steep part of the S-curve. And I think, Bruce, you're seeing that too. We are, absolutely. I think one thing that's going to play into when this is actually going to hit this tipping point is obviously the refresh cycles, right? Because you put in a sand, you've got 36 months until you need to refresh that or 34 months until you need to refresh that. So we will be replacing all of our environments with front and center flash as they refresh. Our next refresh is August and then not until next year. So it will take a couple of years for us to be 100% across all our sites. Yeah, Bruce, that reminds me of a customer I spoke to this morning, Dave. And he said that, look, we have multiple arrays and now we do attack refresh every year because every year, one or two arrays will come off Lee so we'll be ready for refresh, right? So he says we're on a continuous refresh cycle for the next five years, right? And every time we do that, we are evaluating all flash. And he says interestingly enough, we did one set of purchases. The next set of purchases are so much lower because the technologies are going up, prices are coming down. Are you primarily an HP shop? So yeah, we're World of War HP on all of our environments with the exception of some of the networking and firewalling elements. So did you not look at competitive all flash arrays? We have, we've certainly spoken to a number of competitors in the market and we have actually had in the lab environment some competitors sans for child purposes. It's interesting, you know, when you compare that and having talked to the operations team around the experience, right from getting it out of the box to running it up, it just never compares with what we've had over the last five years with 3PAR. You know, 3PAR is the gold standard for standing up infrastructure. I mean, that's always been the claim to fame there. Okay, so I don't even have to ask you why HP you pretty much just answered it but I want to ask you about the stack. So one of the advantages that we've always talked about with HP is that you didn't choose to go buy an all flash array startup. Sure. That didn't have a hardened stack. You're building on top of 3PAR now. A lot of your competitors at the time would say, oh, that's a bolt on. Well, that's kind of what you guys used to say about thin provisioning that was bolted on. So. Well, others are right, yeah. But you always said we used to have David Scott and he said, no, no, you don't understand. The architecture is such that, and who knows? You know, when top guy says that, maybe it's marketing, you don't know until you actually see the proof, you know, as a customer, you know, just never know. It appears that the architecture was very able to accommodate that, you know, no question. So how important is that stack to you as a customer? That robust full set of storage services? Basically the fundamental reason for our success I would argue, if you think about the need to put in an environment and a hand on heart and know that everything's going to interoperate well and cohesively and that the vendor is going to give you end to end support across that entire ecosystem and outside that ecosystem in terms of your overall roadmap. That is the success of my business is their partnership with HP. Yeah, now, what are you hearing in the field in terms of how much of a differentiator that is? Oh, that is a huge, huge factor from many customers, right? The fact that it's one version of the same feature set, same operating system across their different infrastructures. Yeah, big, big factor. The fact that they can buy different size, flash arrays, another big factor. We're also hearing from our customers that they go all flash that after a year and a half they'll say, look, we may have a use case for spinning. They want to add spinning to it as a tier. They have that option as well, right? And... That happened to us exactly, right? So we went all flash in one of our facilities obviously at the end of 2014, 100%. The only spinning disk was the backup stores, right? So there was no production spinning disk at all in that environment. And we did that because the price point on flash had gotten to the level where even at one-to-ones and no D-Drup, it worked for our business model. We did not have to change our go-to-market pricing at all, so it was a no-brainer. We got some D-Drup, so that's a nice win for us and the overall scalability. But as we got, you know, sort of up in the 70, 80% utilization on the sand quicker than we expected, we suddenly thought, well, actually, let's take a look at what's been stored here and we found there was a huge amount of data at rest. You know, we've got customers with three and four terabyte file servers and they access maybe 10 or 20% of those files in any given month. So for us, it was a no-brainer then to then think about spinning disk still has a play right now. And so we all meant to be going. Unless the CapEx crosses over, then... Right, and we'll get there. We'll get there, but as you say, that could be 18 months to two years. So you're using data reduction, D-Drup. We are, and then you'll flasher it. And what are you seeing for results? Yeah, we're seeing sort of in the 1.5 to 1.8 range, depending on the CPG, depending on the workload. Bear in mind, we're a service provider, so it's a very mixed workload. We have no idea what customers are going to do on any given day, what they're going to upload or what they're going to change. So we're still working with that. It's only been sort of nine months on the environment and there's a lot of tuning that can be done. We've measured the install base, Dave. We are seeing, on average, two to one across the install base. And we are seeing cases where clearly VDI is one where you get good three to one, four to one, five to one, sometimes even higher, depending on the workload. Interestingly enough, there is also a lot of zero data that shows up. That continues to show up. So even in VDI environments, but yeah. And that's part of the two to one. On average, two to one, yes, but in cases in VDI where it's higher than two to one, like three or four to one. Yeah, that's part of it. And I think calling it out separately from the D-loop. So you're talking about three to four dollars, gigabyte raw, and then you're getting that down. You're cutting that. Well, at $2 a gig, usable. We were at about six and a half, seven raw. Yeah, okay. We're bringing that down now. We're at $1.50 usable, right? And we've seen some customers going up to 10 to one. I'm doing VDI. Do you have tools to help them predict like you did with thin provisioning? Yes, we actually have tools that let them take a thin provision volume, for example, and run some D-loop analysis and says, okay, how much duplicate data do I have in this volume, right? Especially if they're running on a spinning array and moving to flash. Then we can say, okay, here's how much of a use case you have a duplicate data and how it can justify your spend. And you have a hero report at the back end? Like you used to. So we still do that. The telemetry data that comes back and the aggregate data, the two to one measure I gave you was aggregate telemetry data. We should let Foyer whack at some of that data. It's all metadata, right? It's no customer information. We keep all the customer specific things out. It's just generalized data. Love to take a look at that and see because you're getting there where you have enough data now to start analyzing that. Well, you know, one thing that's very interesting, Dave, we looked at the size of the blocks, the IO blocks, and the amount of data transferred. And, you know, it's interesting that 20% of the IO's above 16K generate about 80% of the bytes transferred. Okay, and so all that D-loop benefit comes in that large bytes transferred, right? So it's actually validated our 16K design as the right sweet spot for D-loop in the array. All right, we're out of time, but Bruce, I'll give you the last word. Talk about cloud. You know, your business, the future, I mean, where flash fits in, but what's ahead for you guys? What's ahead for us? So, rolling across Canada, you know, Canada's interesting. Segmentation-wise, it's identical to New Zealand, so we can just do what we're doing well, replicate it in a much bigger market and scale up and use that growth to keep doing things, innovating and keep doing things better. It's great. Well, thanks for coming into theCUBE and sharing your story. Vish, it's always a pleasure. Thank you for having us. All right, you're hearing the transition to the all-flash data center, supporting the digital economy. This is theCUBE. We'll be right back after this short break.