 Live from San Francisco, California, it's theCUBE at VMworld 2014. Brought to you by VMware. Cisco, EMC, HP, and Nutanix. Now here are your hosts, John Furrier and Dave Vellante. Okay, welcome back everyone, live in San Francisco here. This is theCUBE, VMworld 2014, our fifth year. I'm John Furrier with Dave Vellante, extracting the signal from the noise. We love talking to the executives, the entrepreneurs, the VCs, all the action is here on the ground with all the tech athletes. Our next guest is Eric Herzog, the CMO, and I think you're running Vizdev as well? Yes. Vizdev for violin memory systems. Violin is, recently went public, now on a complete transformation. You're at the helm there from EMC, so you know a little bit about storage and flash. Welcome to theCUBE. Well, thank you very much. Always enjoy coming to theCUBE and doing it now for four or five years. It's been great, guys doing an outstanding job. We really appreciate it. One of the things we're excited about obviously is flash and every guest we've come up here, that's been in the storage and the periphery of storage is cloud and hybrid cloud is raving about the economic disruption of flash, the performance of flash. Flash is super hot. Now, Docker's getting a lot of the press right now because of the deal, but still, flash is at the, under the hood. That's where the action is. So what's the update? Give us the take on what's going on in flash and violin, what are you guys up to? So the big thing is flash is at that economic tipping point. So if you go back to the late 70s and early 80s, as everyone remembers, everything was tape. All the data centers were tape. Hard drives were more expensive. They were faster. And you got to the economic tipping point where using a hard drive based array was much better than using a tape subsystem than tape became backup archive, which is still great at tape. In fact, I saw from one of the analysts who track such things that tape is actually still the cheapest media. I don't see any CIO rushing to the all tape data center. So what you've got now is flash is at that economic tipping point that between the savings and storage, servers, software licensing, power, rack space, floor space, et cetera, that when you do the economic analysis, you can just literally do it with a calculator, pay us to go flash. In fact, flash is almost free these days. So certainly the economics are ridiculously amazing in terms of cost. Now on the performance side, you're starting to see some segmentation yesterday we were talking about capacity flash and performance flash. What does that mean? I mean, are they different at all? Is it flash is flash? But you start to see these conversations that are being kind of workload specific. Is that where it's going? Are we still in the flash adoption phase? What's your take on that? We're at the maturation phase. Flash is shifting away from everyone assuming it's the same. Just think of the old hard drives. You know, even today, you got 7,200 RPM, 10,000 RPM and 15,000 RPM. And it really makes a difference as you use those various capacities and the various performance mixture around them. Flash is, and it's all the same media, same media, same heads, but they make changes. Flash is doing the same thing. There is people focusing on performance flash, violin being one of those. We have one of the highest performing systems out there as measured by, not by violin, by third parties. And you got other people that want to go would all say cheap and deep flash. Not as cheap as hard drive, but let's make flash, you know, faster than hard drives, but not Uber fast. And so you could put other workloads on it that are more capacity sensitive than performance sensitive. So I wonder if we could unpack performance a little bit. So people talk about IOPS, they talk about latency. How do you guys look at performance? How should customers be looking at performance? So it's really a package. Okay, the number one enemy of most applications, particularly in mid up to global enterprise is absolutely latency. So IOPS is important, but if you don't have good latency, IOPS don't overpower that. So you need to have both good IOPS and really strong latency in order to optimize whether that be an Oracle workload, an SAP workload, a SQL workload, those types of workloads often are very latency sensitive. The lower the latency, the better the application functions and the more you can do with it. So who are the kings and queens and princes of latency? You would put you guys in that mix? We are in that category. We can guarantee under half a millisecond latency or 500 microseconds, whichever term you want to use, whether the array is empty or full. We also have some customers that have done some host-based aggregation in production and we have one of the 25 largest companies in the world with multiple petabytes in production. They aggregate on the host side our arrays and they're able to deliver two million sustained IOPS regardless of workload across all those petabytes and 0.15 millisecond of latency. Now that's not what we claim on an individual array that's on your spec sheet. So they're really getting it and they've proven it to us several times. So that's in the performance side of the equation. So latency IOPS, bandwidth is not as much of an issue because bandwidth obviously you can get off of hard drives and hard drives are very good for high bandwidth situations. You're not going to use all flash in meeting or attainment, applications or an oil and gas or a lot of the genomic research stuff because it's very bandwidth intensive and you could get great bandwidth off of low cost hard drives actually and create a giant NAS cluster for example is better in those workloads. But in database workloads, virtualized workloads, for example, we have a customer that on a certain physical server had 14 VM virtual machines. They then used our flash and they were able to get 50 on the same exact physical hardware, same size virtual machine, same IOPS for that, those virtual machines and go from 14 to 50 just by switching to flash. Same VM, it was VMware, same exact server infrastructure, all they do is swap the storage out. So that's an example of how A, you get the performance and B, you also get the economics because obviously putting 50 virtual machines on the same physical hardware saves you money. So I would think the big benefit too is consistency. That's what you hear from customers, I just give me consistent predictable performance. So are you hearing the same thing from customers? Yes, absolutely. So what you have when you look out at the flash world what you're going to see is certain people have a right cliff and what happens is when you hit the right cliff or they're going to have unequal performance they'll be better than a hard drive system for sure but they'll still get a sawtooth, not as dramatic as you'd see in a hard drive subsystem but sawtooth. What we do is we guarantee consistent IOPS and consistency latency whether the array is empty, half full or all the way full and very few guys in the off-flash community can do that. I want to talk a little bit about the stack. So you came from a company where you were running a very senior executive at EMC within the mid-range business, VNX, awesome stack, been around forever, a lot of value in that stack takes a long time to harden the stack. A lot of the flash guys, you guys included came out, you're solving a problem, start selling, stack takes a long time to mature. So how should we be thinking about the stack? So RAID stack is always crucial. You know, RAID is not just about performance, redundant array of independent disks. It's number one function when RAID came out, quite a bit across the bay here at UC Berkeley, was for resiliency. So that's the number one thing that a RAID stack does. The second thing it does, of course, is give you performance as well because you aggregate whether it's hard drives or flash drives or hybrids, you aggregate the performance across the pieces of media. So I think one of the benefits you're going to see from certain vendors in the flash base, we being one of them, is we have a long history. We're on our fourth generation flash configuration and we basically rev our generations every two years. So we're looking at a RAID stack that's in the eight year timeframe. Some of the other flash startups, they've been shipping for two years. You have a two year old RAID stack. An eight year old RAID stack has got much more resiliency, it's got more test time. For us in particular, our sweet spot is in the upper, mid to global enterprise. If you look at the Fortune Global 500 list, over 50 of those customers use violin, which when you're a big company is one thing. When you're a small company like us, to have 50 of the global Fortune 500 using your products, it's got to be pretty resilient in the stack or they wouldn't be using it. I mean, I was on a, I probably spoke one on one or maybe one on 20, one on 30 to over 500 customers in the first half of this year on flash. And I would ask every one of them, who's using all flash array? And it was actually pretty low penetration still, not surprising. Violin came up a lot, TMS came up a lot. I mean, not, and then pure a little bit and then, you know, bits and pieces, but violin was consistently there. So you guys did a good job early on, getting into the space. But I want to ask you about, sometimes I call it, trying to laugh, the urinary Olympics and particularly around data reduction. And so you guys are now, you know, throwing your hat into that ring. How should we be thinking about sort of data reduction, compression, D-dupe? Obviously drives pricing down. It helps create that. That's I think part of the reason why we're at that tipping point, that and, you know, MLC. How should we be thinking about data reduction? There's a lot of, you know, finger pointing, inline, not inline, post-process. Give us your point of view. So the bottom line is data D-dupe will help you in two primary workloads. Virtual desktop and virtual server. Okay, beyond that, it doesn't help you. Compression helps you in database oriented workloads and there are certain data types that are not compressible at all. So for example, MPEGs, JPEGs and other data types are not compressible at all. They're already pre-compressed by the nature of the data type. So everyone needs to be wary that just as when you get your miles per gallon when you buy that brand new car, it will vary and it will vary by workloads. So if you've got a workload that's heavily already compressed, you're not going to get benefit from anyone's compression, including ours. If you've got a workload that's already been D-duped, you're not going to get any benefit from anyone's D-dupe. So you have to segment your workloads. I think the other thing, Dave, in addition to what's driving that price point which is compression and D-dupe is multiple workloads. So for violin in particular, our average around, we've already publicly talked about this, our average array shipping is well over 30 terabytes. That's not true of a lot of other guys. And when you've got 30 terabytes with the average database being four to five terabytes, people don't put one database on our stuff. People who sell five terabyte arrays and a recent large company just announced the new five terabyte array, they're going to put one database. With us at 30 to 40 terabytes average, people run three, four, five databases. Does anyone really buy a VMAX or a NetApp 8000 class or a high-end IBM box and run one workload on that in the hybrid world or in the hard drive world? No. But that's now that people are running multiple and mixed workloads on flash arrays, that plus the D-dupe and compression is driving this economic switchover and why flash is the right choice for your data center. Well, you guys do obviously do a lot in database, generally and specifically Oracle database. You know, Oracle's big on pushing hybrid columnar compression and trying to lock out its competitors from participating in that. What are you seeing there in Oracle environments? And again, I've talked to a lot of customers and the instances of hybrid columnar are still very limited. But in theory on the roadmap, how, what are you seeing? What are your thoughts on that? What do you talk to customers? Customers must say, well, you know, Oracle's locking you out, you know. So I say a couple of things. First of all, on the price points, it won't matter because people run violin arrays with mixed and multiple workloads already. So even if you were to buy the Oracle, if you're going to buy Oracle compression or compression to any of the database from the database vendors themselves, for us it's still benefits. We don't sell a lot of five and 10 terabyte arrays. We sell lots of 30 and 40 and 70 terabyte arrays. We can even scale our arrays up to 280 terabytes, which most of the other guys can't do. And I'm talking now raw capacity, not D-duper compression capacity. At the same time, while the database guys are trying to do that, one thing I'd encourage the end users do is just look at the list price. It's available readily, Oracle's is available. It's a pretty high ticket item. So whether it's violin or any of the other flash vendors that have compression, it won't compress as well as Oracle's will or any other database vendors. But the price is pretty high. So if you get reasonable compression from a storage vendor, it's going to be a lot less expensive than using that from the database vendor. Now maybe the database vendors in Oracle change their strategy, but right now it's a very high ticket item. And when you get it from the storage vendor and even if it doesn't compress as much, it's still a lot cheaper. So you'll have to take that as part of the financial analysis when you're looking at your database deployment. Now you made a big personal bet on violin. I mean, you and I, I was there in the front row when you announcing the latest sort of VNX, which is a great announcement. I mean, you guys ticked a lot of boxes. It was a lot of hard work and I realized that. My one big question was, what about all flash? Like, well, we have all flashed you. Well, you said all the right marketing things. And then, you know, several months later, here you are at violin. Big personal bet, right? Senior executive at EMC. Big personal bet. I know a lot of travel, but you know, pretty good life, right? You know, a lot of people working with you for you, you know, a lot of great customers. Why'd you make that choice? So a couple of things. First of all, violin's got an incredible set of customers. When they divulged the customers to me under NDA, I was like shocked. I couldn't believe who the customers were. You know, I worked at IBM as well as EMC. So of course, all the big boys are your customers and they always will be. But the number of really big companies they had was very impressive. Incredible technology. This year has been all about the software stack, which violin has been very mediocre at. Now it's got a whole set of software potential. And as you know, Dave, I've done seven startups. Five of them have been acquired. And I can smell a stinker. This is not a stinker. So it passed the fume test after doing seven startups. So it, you know, feels like the right thing to do. What was that attraction? Obviously the IPO went off without a hitch, in terms of at least going public, but it stopped and climbed. It was a little hitch. Excuse me. Yeah, absolutely. I'm being a little like, Dave, come on. We'd like violin. Emerging player, obviously the market TAM is huge. Yes. So that's, I mean, one, market opportunity. So with that kind of the IPO stumble, if you will, you still came on board. Yes. That was not an issue for you. Like, okay, I'm gone guns blaring. Well, in addition to doing seven startups, I've done, this is my fourth turnaround. And all of them have ended up very well. IBM bought one of my turnarounds. I was at MacStore as the senior VP of marketing when Seagate bought MacStore. That was another turnaround, although be it a very large company, obviously MacStore at five billion at the time of the acquisition, but done a number of turnarounds as well. So it's an attractive thing to do. It's a fun thing to do. You felt you could really do this. Not go out of the park. I know I'm an old man, but I'm not that old yet. It's pretty straightforward. You get the customers, give them some good product, collect some cash, do it again. I mean, It's all about execution, you know, and violin did a lot of really great things. They did really well by the customers. Customers love them. Great tech support, great field support, the SE teams, even a group of consulting engineers and all the consulting engineers actually are ex Oracle and Microsoft guys. You know, they're learning story, but they know all about the database community and we got a couple guys from actually actually VMWare guys as well. So that's a big thing, but I think the key thing is you got to execute on all cylinders and we had a great technology leadership group that did the first set, got the company to the first hundred million, but it wasn't the right guys to grow the business and make the business. And by the way, you guys interview VCs all the time. So you know, it was very common. You get to a certain point and then the founding executive team sort of needs to move aside. Great technology guys, but not the best businessman and that's a strong attraction. Well we were just talking about some VCs up here, some tier one, Greylock and NEA. We had the question that came in over text that Dave was texting me that we wanted to ask was, you know, at these big valuations of private companies it's hard for the employees to make money. So the silver lining in your opportunity is there is a lot of growth opportunities and money making opportunities for the management team and investors, right? So that's a good position to attract some talent. Yeah, well that's the appeal. You know, when you think about, there's certain guys that are really good at IBM, EMC, Microsoft, HP, VMware and they're never going to do well in a startup. Yet other guys that are hybrids can be big and small company. And the attraction for those that can do both is you can bring the season management that you learn at an IBM and EMC and Microsoft, a VMware, bring that to the small company which has great technology but often does not have the discipline and rigor that a big company does. And what you have to do is bounce the drive for new technology and new customers with the business model and not become overly bureaucratic. And that's the attraction of a turnaround as well as guys who do lots of startups is to be able to do that and grow the company. And the key thing is you got to grow it properly and that's the upside you get. Your track record is phenomenal. We've been following your career. Tech athlete for sure. Now Wall Street, you got to kind of do the dance and you know, keep nice and get these guys back to snap them in line, right? That's kind of the key focus too as well, right? Yep, it's about financial execution right now. We brought out a whole bunch of new products. Our Windows flash array, inline duty compression, a whole class of, I'd say, unmatched enterprise class data services in the all flash array space. And you've got to be able to leverage all of that. And that's a key thing. You've got the technology. If you don't execute on the business side, you know, you go out of business and we've got the right team in place now to take the technology where it needs to deliver the business value to the shareholders and the stockholders. Eric Herzog, CMO, violent memory systems. You know, my philosophy and my experience, although, you know, not as extensive as yours is, in a growing market, a few missteps can be rewarded with great products. So you guys have certainly a good product. You get a mulligan with a growth market, wind behind your back. So congratulations seeing things on track and really exciting to see. Good company. This is theCUBE here at VMworld 2014. You're right back after this short break.