 Hi everybody, welcome to this CUBE Conversation. I'm Dave Vellante with Stu Miniman. We're going to talk about software defined storage at petabyte scale. Stu, we've got some news coming out of HP. What's the news? Sure Dave, so HP Server Group is making announcements with a few partners to go after really the capacity side of the market. We've talked a lot about the flash and performance side of the market, but really looking at big capacity and the one we're going to talk about today is Scality, who has been a partner with HP before. They've got over 50 petabytes of shared partners solutions out there, but formalizing really the relationship that you can buy HP servers, have the Scality software on it to enable really that capacity tier. So Scality is kind of an interesting company. Engineering in France, but offices in Silicon Valley. What do you know about Scality? Yeah, Dave, as you said, we talked to the CEO, Jerome LeCotte, who built the company out, has really the headquarters is in the Valley of San Francisco there. Engineering is in Paris. Started out, most of us knew them as an object company and has really matured the company to go after really the software defined storage market, looking at enabling file systems and object stores to be able to build large scalable environments to hit specific workloads, not going beyond just archiving, but other applications, some big data use cases, and other environments where really we want to change the economics of capacity storage. So Stu, why is HP making this move? HP's a big company, they got a storage division. We've covered that extensively. What's HP's angle? Yeah, great question, Dave, because you look at it, of course, HP is the leader in servers, and of course the server group just sees an opportunity here, something that compliments what they're doing already. We've written at Wikibon before about HP servers going into some big data environments for large capacity environments on the server, and Scality Software allows HP servers to be able to build not just single nodes that are controlled by some application, but build a storage environment that scales beyond single nodes out to a large pool of data fits very closely in with what we call the server-san architecture. Okay, so I want to talk about that in a minute, but just to clarify, they sort of overlap with HP's storage line. So HP's got their file base or NAS product out of Ibrics. They've got a VSA, virtual storage array. How do they fit with this announcement? Yeah, so there definitely are some places where it would compete, Dave, so absolutely the scale out NAS, Ibrick solution out there. HP does have an object store that you don't hear too much about, so these are solutions where HP can tell a lot more servers and deliver a storage solution that customers have been asking them for. Okay, let's talk about the market. You mentioned server-san, let's go to the root of this, which I like to think of as the hyperscale market. It's something that we've talked about and written about and had a number of folks on theCUBE discussing this. Many people in the industry are talking about it as well, but essentially you've got the birth of Amazon Google Facebook now seeping into the enterprise. You've certainly got hardware standards emerging out of places like Facebook, but you're seeing now enterprise customers demanding to operate at Facebook-like, Google-like, Amazon-like efficiencies, so would you agree that this is really the trend for what we call server-san that we'll talk about in a moment is really stemming from what was going on in hyperscale, let's say, three to five years ago? Yeah, Dave, absolutely, as you said, it was the huge data center guys couldn't build off of a traditional storage array or even the filers got really expensive. Many of us in the storage industry remember back when you hear of the Facebooks and Yahooes of the world, adding filers and as we tried to add a lot of photos, which was big in the social media space, it just got onerous to be able to scale that and the capacity of that was quite difficult, so you have to have a new design criteria to be able to build out massive scale and a solution like what Scality's offering allows me to buy a lot of compute nodes and scale up my storage internally and manage with that software paradigm rather than having to buy filers separate from the server and build those out and manage those. I hear you quote all the time, Dave, you say, my first couple of filers are easy to manage once I get more than 10, it's ridiculous. Yeah, so in thinking about the evolution of storage and how we got to the SAN and network storage, well, not just SAN, but NAS as well, but it's network storage in an area that you know a lot about, but you used to have big boxes that you connect and they would maybe support multiple servers and then the SAN switches came in, we saw the ascendancy of companies like Brocade and many of you might remember McData and then the SAN market emerged and exploded and the promise of SAN was the ability to share that data across multiple systems and also replicate the data and then a bunch of services grew out of that. Enormous business, it's probably, if you include all the, I mean, the hardware itself is probably 30 to 35 billion, the software is another 10 to 15 billion on top of that, so it's about a 50 billion dollar business and then a whole set of services grew around that. I mean, we've got to be approaching, I would think anywhere from 80 to 100 billion dollars of market value or revenue that was created, the market value is obviously some multiple of that. That's potentially going to get disrupted by what we call server SAN, is it not? And then so tell us what server SAN is and then we can maybe dig into some of the numbers. Sure, Dave, if you talk about those giant arrays that got built to allow me to get greater utility out of all of the components because what was the most expensive part of the system and where were the bottlenecks in the system? So you had these monolithic storage devices that I could attach through a large network and if we look at where it is today, look at those storage arrays and look at all of the components in the data center, all built off of x86 hardware. So as Moore's law keeps allowing more compute cycles to be used and costs keep driving down, doesn't make sense for us to re-architect things. So as we've talked about for Flash especially, makes a lot of sense to have my storage and compute close, but even for a capacity play, like what we're talking here with the HP scalability solution, if I have room in an architecture that allows me to build off of the compute with a lot of disk, I can now build a architecture that has better price points and much better capacities in a side of a rack that is built off of the standard x86 nodes than I could if I wanted to build an external storage array. So Matt, we've talked about this too and certainly David Floyer as well, our CTO, the notion of Flash plays in here and the idea that you can have a persistent medium closer to the processor, approximate to the processor, Gene Amdahl's famous quote of the best IO is no IO. So the closer you can move a persistent medium to the CPU, the better performance can be. Of course the challenge is always, how do you protect, how do you share, how do you create high availability? But that problem is increasingly being attacked, is it not? Yeah it is, but if we look at the challenge, Dave, it's not everything is a performance basis, there is also the capacity need. If you look at Amazon, they have performance and capacity plays, you've got S3, you've got EBS. If you look at service providers, often they are buying a performance tier. We've looked at SolidFire as one that's providing an all flash array to give service providers an all flash tier and they also will need to have a large capacity environment that might be an object or scale out NAS solution to be able to build that capacity tier. So we're really seeing that kind of two tiered model building out in the biggest environments and driving down to the enterprise where customers are going to potentially build those solutions, whether it be a flash solution for your performance and a capacity solution for the rest of what you're doing. So okay, so I kind of like that scenario. You've got the high performance data, the metadata running in flash, maybe a lot of block storage running in flash and then you've got the bit bucket. And that bit bucket, I could even see a third category of sort of long term retention, almost like an Amazon Glacier. But the latter two are moving toward object, are they not, I mean it's kind of a get put model? Yeah, so Dave the funny thing is if you talk to everybody in the storage world, we all say if we look at where storage will be in the future, it's going to be object. That's where the cloud guys are going, that's where it has to go. But we've got this big air gap because all the applications today are written for either block or file, so we need to be able to port that over and that's what Scality matured with their solution. They do have the file offering to be able to give my application that file interface to be able to take advantage of the architecture. So Stu at Wikibon, as you know, we like to look at disruption and so we've got some data on the server sand market that I wonder if we can pull up here, Andrew. So if you take a look at this forecast, it's the traditional enterprise storage, the hyper scale server sand and the enterprise server sand on the left hand side you've got dollars and as I said today that read that sand market is around 30 billion, just above 30 billion of and that includes sand and NAS. And then the blue area is this enterprise server sand and then the green area is the hyper scale. So Stu, you can see this huge hump that now, of course it looks dramatic but that hump really doesn't start declining until sort of a few years out here, 2016, 2017 and then dramatic decline. So you and David devised this forecast and you show huge cagger for the blue area, the enterprise server sand and of course the green. Now describe the blue and the green. What are those two segments? Sure Dave. So the green is really the hyper scale environment. So these are you know, Google, Facebooks, Amazons of the world that are you know, buying the discrete components, building a single application to go on that environment as opposed to the blue is the typical enterprise that are going to buy solutions from companies that are building sometimes they're called hyperconverged, what we call server sand, or you know really the software defined storage solution where it's a compute based, highly scalable architecture that scale out architecture which this HP scality environment would fit under. And some of that blue could seep into the green, could it not? I mean maybe we see that green accelerate, I mean there's a gray line pun intended between the blue and the green, is there no? Yeah, I mean usually we say the green are kind of those you know, 10 biggest web companies out there as opposed to the service providers or everything underneath that and the enterprise also is big piece. So you know, absolutely they'll be a little bit of blurring as to you know, who fits into those biggest of the bigs but you know, the blues are the ones that are going to build off of solutions and there's some building block solutions out there that could blur that line definitely over time. Look at where open compute's coming out, it could definitely blur those lines. Yeah, so now let's take the customer perspective, if I've got you know, billions of dollars you know, okay or my company I got millions of dollars invested but the market certainly has tens of billions, hundreds of billions of even of process and people built up around the traditional sand and nass world, what do you do with that? How do you get from point A to point B? Am I looking at a rip and replace? Am I looking at massive disruption? Is it more of a workload based kind of model? Let's talk about that a little bit. Yeah, Dave I think mostly it is workload based as you say and it tends to be customers that aren't being solved with their traditional nass solution today are kind of the easiest ones to pick off here and say that just the economics of adding new filers and how I manage that whole solution and grow onto it becomes onerous and if you just look at some of the offerings that I could buy today instead, it's going to be a pretty clear return on investment for putting in some new architectures. I mean you know, you spent a lot of time in Silicon Valley and the real buzz there is around the cloudification or the sassification of business and you're seeing it everywhere. Certainly companies are moving toward the sass model and you see that with the likes of the work days and the service nows and the salesforces but companies on premise are also trying to replicate that so is this in your view the storage architecture of the future? I mean the forecast that you have there is pretty aggressive. It's more than a niche according to your numbers. Yeah, Dave as funny you mentioned the hashtag we're actually using for the crowd chat is future of storage so we think that if object is where the ball, where we need to go to if it's where we're skating to the puck it's solutions like this that are going to help enable file access to that object store and allow me to build that scale out architectures. I mean Dave the future's all about distributed systems and new software solutions that can help leverage standard products like x86 servers from HP and build a software layer on top to enable those new distributor architectures are where they need to move to. So we're talking scale out, I mean storage is scale out systems are scaling out network hopefully someday we'll scale out so the network's always dragging us down here too. There's people working hard to solve that problem Dave. But in thinking about that and that sort of distributed network environment from a customer standpoint what's the business case here? We've had people on from an object perspective that have talked about managing very large objects whether it's files or a video or pictures in one case where they cut their costs literally by two thirds. So one third of their initial sand and NAS costs is what the new implementation was. Is that what people can expect going forward? Yeah Dave it's just got to be that the cost you know dollar per gigabyte's got to go down and how I manage that entire environment has to change. The Achilles heel really for the sand is that I spend way too much dollars to be able to implement that solution and keep it up and running. So I need to have a solution if this is a true scale out capacity play that really mostly manages itself once I've put it up and running. I shouldn't need to be sitting here optimizing a bespoke infrastructure and it needs to be an infrastructure that adds pretty simply and manages with a lot of ease. So thank you Stu that looks like the future is x86. It's moving stuff closer to that processor what we call server sand distributed environments scale out so great perspectives really appreciate you coming on. Thanks Dave. All right thanks for watching everybody. This is the cube, cube conversations we'll see you next time.