 Live from San Jose, California, it's the CUBE Happy Adaptive Flash launch brought to you by Nimble Storage. Now here are your hosts, John Furrier and Stu Miniman. Welcome back to Silicon Valley. We're at Nimble Storage's Adaptive Flash launch event here at their headquarters in San Jose. They've got events all across the country, and actually from London, New York, Chicago and here in the Valley. Joining me for this last of our interviews is Varun Mehta, who is the co-founder and VP of Engineering on the program. I always love to talk to the founders. Varun, thanks so much for joining us on the CUBE. Sure, Stu. It's good to be here. So, you know, I think back to when Nimble launched, you know, there were so many things. Flash was actually kind of early. I mean, you guys started in 2007. Most people weren't thinking about Flash in the enterprise. We have companies like Fusion I.O. and EMC that came early. And you and some of the other founders came from NetApp and Data Domain. Things like storage efficiency and, you know, really helping to transform storage services. So can you give us an intro as to, you know, where we are today for Nimble Storage? You know, you've IPO'd. You've got almost 700 employees worldwide, you know, in great momentum. So first of all, congratulations and, you know, tell us some of the journey. Sure. So going back, you're right, that a lot of us came from the big companies in the valley and my own personal background was at NetApp and then at Data Domain. And at each of those companies, I had this opportunity at NetApp of working in a really good primary storage company and Data Domain working in this category defining backup company. And the thought was when we were working at Data Domain, my co-founder and IOMesh was, hey, what if we could combine the best attributes of a primary storage system and a backup system because people had this box proliferation going on. And it just technically wasn't possible till flash came along. And in 2007, as you said, I remember opening up a copy of the Wall Street Journal and seeing the first ad for an SSD and it blew me away. You know, it was like way faster than anything that anybody had ever shipped. So I called up OOMesh and said, hey, we can do something with this. And that was a genesis of what we set out to do. So just early on, the thinking was, hey, let's have this, let's build this thing, this device that can do backup as well as primary storage really well. So we set out to build that, but then we got this bonus along the way, where the performance turned out to be way faster than we had thought. And then, of course, we had really good dollars per gigabyte. So we managed to get it all, data protection, which was our original goal, but also really low dollars per gigabyte and really high performance. So Varun, there's been waves in the storage industry of companies built around a feature. So things like theme provisioning or scale out NAS. Many of those can build good product lines that can grow. But it's tough to predict the future. You look at, in 2007, we knew server virtualization was pretty big. Cloud wasn't on most people's lifts for that. How do you build an architecture that's not just a feature, but a sustainable, durable architecture that can grow and expand into the future? Very, very important. Actually, there were two things that we looked at. One was just market entry. Where do you go where you're going to find the least resistance? So that was very important and that defined our initial product. But then you look beyond the market entry to five to ten years out and say, okay, how can I have the largest footprint? And as you said, we were one of the earliest flash companies to get started. And what that allowed us was to stake out the largest area. So we do believe that the combined flash plus disk market is going to be the vast majority of all the storage that is going to be shipping for many, many years hence. And so we were able to stake out this market. And we could see that companies that came after us decided not to butt heads against us directly. So that's one of the advantages of coming in first. You get to set the position. And then about the market entry, we could have started up actually at the very tip of the pyramid, very high performance, because we had the field wide open. What we decided instead was after studying storage companies, we looked at companies like Blueark and Threepart. These were companies that had started at the high end. And it took them a while to get traction. And it's because if you go head to head against the big incumbents in their sweet spots with the large companies, it's very expensive and very hard to get in there. So we said, okay, let's build a product for the mid-size enterprise. And we put in all of these features. We leave open our options to move into the high enterprise when the time is right. And so that's what we did. We built a product with the right capacity, the right performance, and the right pricing to enable us to go into the mid-size enterprise. And it's really hard to defend when you have hundreds of thousands of potential customers. So that's allowed us to have success in the mid-size enterprise. So Flash has really changed the design of applications, hasn't it? I mean, it's changing it now. When you look at databases, for example, they're starting to add the Flash-specific features. Like Hannah, for example, is now loading and unloading. It becomes the bottleneck. And so they have to have Flash there. And SQL 2014 has got some very nice features. And obviously there's Flash in the service as well. They are going to have to be integrated into that. How do you see the applications moving over time? Did the new applications which assume that Flash is there? And how is that going to make the storage market different? Great question. It's actually interesting. I've been waiting for the application vendors to start managing Flash on the server. I think it's a really, really powerful concept and something that will have huge benefits to customers, because having Flash close to the application is very helpful. Now, and it's great that they're jumping on board. You know, VMware has this VFlash initiative, and now Microsoft and SAP are also jumping in. I'm waiting for Oracle to jump in. My SQL is also a custom extension. Right. So that is great. It provides data much closer to the application, which is good. Now the problem it doesn't solve is your resiliency and your data protection problem, because you still have this need to take snapshots, for instance, to replicate your data off-site. All of these things, these data management functions, which are really hard to integrate into the server. So that's why you still need your traditional storage array to be this ultimate and very reliable repository of all your data, and it handles all your data management functions like snapshots, cloning, replication, DR, all of that. Now the interesting thing there is that the traffic to the array is going to be changing quite a bit, because all the read-intensive data is going to be absorbed right in the server. Applications will go much faster because of it. Big DRAMs and big Flash. Right. What happens is that to the back-end array, their traffic profile changes considerably. It becomes a lot more write-intensive. And this is one of the things that we anticipated when we built CASL, because with sequential writing to disk, we can absorb very, very high frequencies of writes. In fact, our new product, the CS700, we did some really nice things to improve write speeds even further. The write speeds on this new product have gone up actually a little more than our read speeds have gone up, because we believe that's where arrays are going to get hit now with Flash moving to the server. So I was looking at an exchange system which had four locations and four DAGs in each of those locations. And the read-write ratio was 75% writes in that environment. So is that the sort of environment that you're anticipating in the future? That's exactly right. So this is what is happening. That the servers themselves are getting so good at caching that you could already see the traffic flipping over to be write-intensive. We see it for VDI also, 80% write traffic at steady state. So it's really, really important to be able to deal with actually random writes. Those are the killer. Now, traditional Flash and certainly traditional disk could not take advantage of random writes. It just imposes a huge load on them. And this is one of the places that we really designed CASEL for to be very, very write-efficient. So how do you acknowledge writes? I mean, all of the key aspects of a database, for example, are getting the log out, getting the write committed, those two-phase commits. How do you design that in this very high-write environment? Because the classic ways of doing it, protecting it in the controller, they're not sustainable long-term. They aren't. And here's the challenge has been, if you look at actually up to the entry point into the array, everything looks similar. People have used NVRAM for a long time. Now what happens with legacy architectures, architectures that were created before Flash ever came about in 2007, as we were saying, is that they did not have this advantage of this huge buffer. So their architectures tend to be very write-intensive, very seek-intensive on disk. So if you look at any pre-2007 file system, its disk's usage is much more seek-intensive than newer architectures. And this is their problem. Now changing that becomes very hard. So one of the S in CASEL stands for sequential, so cache-accelerated sequential layout. So we took huge pains to make sure that we always write sequentially to both Flash and always write sequentially to disk. And this allows us to get really, really high throughput to both Flash and disk. Now the thing is that when you're writing purely sequentially, there's a lot more fancy footwork you have to do on indexing and managing the index, especially with when you have snapshots going on at the same time. So that was a lot of where our innovation was. So Varun, this is a great conversation. Before we get the plug pulled, I need to ask you one question since you're the VP of engineering. When you look forward, we've had a lot of innovation in the storage in the last few years and in Flash. What excites you going forward about, what's it about this time right now? What should we be looking for from this? What should people watching look at the storage industry in general and to nimble specifically? I'm not asking you to share too much roadmap. We hear things, fiber channels come in and things like that, but what's exciting? What should we be looking for? Right, right. From a customer's point of view, this is very interesting. It reminds me of there are these periods in evolution where they call it, for example, the Cambrian explosion where lots of animals sprouted forth, but only a few succeeded down the road. This is what is interesting for customers right now. They have a lot of choice, but what happens is over time, a few winners will emerge. That's the thing that I think customers will have to watch out for. There's a lot of choice right now, but a few years later, there will be some winners that emerge. The thing that I think they should look out for are the most broad-based architectures. What are the architectures that can handle the maximum amount of workload? The things that you said, what is coming down the road, that could be exciting. I think it's part of what you referred to, Dave, which is the increased involvement of the application vendors. We've heard about things like OpenStack, which is great, but I think the application vendors getting into the storage provisioning business with architectures like VWall and SMIS from Microsoft, that is really interesting. Managing flash on the server along with the backend storage, that will be interesting. Alright, well, Arun, unfortunately, we're out of time. We look forward to continuing the conversation with you on another session in the queue, but at some upcoming event. Thank you once again, co-founder of Nimble, VP of Engineering, Varun Mehta, Stu Miniman, David Fleuer. We'll be coming right back with a quick wrap-up from Nimble Storage's Adaptive Flash launch here in San Jose.