 Flash storage generally in flash-only arrays are completely disrupting the storage business and everybody knows it What's happened is the consumer grade flash has entered into the enterprise and Organizations companies vendors have figured out ways to make that flash more reliable as costs come down and performance improves Practitioners are starting to find ways to utilize flash that are driving massive productivity improvements We've got a span of flash implementations from server-side flash and memory extensions to PCIe cards to flash-only arrays to hybrid arrays and It's completely changing the positioning of the traditional storage hierarchy last week was flash week today is March 12th Tuesday, and we're here at Wikibon headquarters. I'm Dave Vellante, and we're gonna squint through some of those announcements I'm here with David Fleuer who's Wikibon CTO coming to us from Mountain View, California. David. Can you hear me? I can indeed well Thanks very much for taking some time and joining us as I was saying in my introduction last week We saw announcements from violin veer didn't wait in EMC had a huge flash announcement Fusion IO had some announcements, and this market is really heating up, isn't it? Absolutely We've been talking about it ever since in the announced haymaker back in 2008 and it's it's really coming to a boil and The real changes that we predicted are happening very very fast indeed Yeah, they really are and you've pointed out a number of times that the the Mechanical spinning disk is the last bastion of performance bottlenecks and you know electromechanical component of computers and and as that Wall and that barrier begins to to be pulled down We're seeing massive improvements in application performance and everybody wants a piece of that pie now you wrote a piece of Flash only arrays in full gallop. Stu's gonna put that up here. Thank you, Stu And and you did a very detailed analysis of that marketplace. You could see here It's on wikibon wikibon org all free content as usual really outstanding analysis and Stu if you could scroll down you you analyze the EMC's announcement the extreme IO Piece of it. You gave your opinion now what happened was website called seeking alpha picked up on your piece and Actually wrote a piece Quoting you or implying that it was quoting you Stu's gonna bring that up here And the the title of pieces EMC's flash storage refresh is more bark than bite argues wikibon CTO David floyer, so I got to ask you is that what you're arguing here. Did you actually make that comment to seeking alpha? I? Never talked to seeking alpha and I certainly didn't make that comment And I certainly don't doesn't reflect my views. It was a very solid announcement the Extreme IO piece of it the all-flash array was a very solid announcement I mean clearly it's not yet in full GA and it doesn't have all of the of the bells and whistles that are going to be required for full enterprise adoption, but it's driving down that route very fast and It's a it's a very sound strong Implementation and even more important EMC has the sales force that has the technical force that has the The megaphone And it has the worldwide reach And it will do it extremely well with this product in my view and it's coming in a little bit late, but it's In a horse race It's certainly not past the the first furlong so a furlong is an eighth of a mile for those of you Don't know horse racing and so David let me ask you does the announcement have more bite than bark in your opinion specifically the extreme IO announcement The extreme IO has a lot of bite absolutely it's it's a good solid implementation it uses An enclosure full of PCIe driven not PCI driven. Sorry flash driven SSEs it's got two very powerful servers driving it and it's helped It's it's scale out of all it's held together with Infinity band it's a good construction and I Personally believe they're going to be even more successful with this product than they ever believe it's going to go into places They they haven't even dreamed of it's going to go into small organizations It's going to go into research organizations. It's going to be everywhere So and with the advantages they have all their share of the marketplace and their share of the Of the channel etc. I think it's the product to watch in this space Now I just want to set the record straight so we'll come back and talk about some of that But the seeking alpha article had some factual errors that they also were attributed to you You never wrote that EMC's PCI so they not only not the flash only array, but they announced the enhancements to their PCIe Flash card and and the seeking alpha article said that it was a read-only card First of all, is that true and did you say that? It's absolutely not true having a flash card that can't be read to it is an oxymoron written to you mean You mean written to written to yeah, sorry written to it's an oxymoron So yet clearly it's a good all-naughtnery flash card you read to it You write to it. Whatever you do with a flash card you need to be able to read and write so I clearly didn't say that and It's not a very good reflection on the person writing it that anybody would write that somebody said that because it's just obviously Clearly I'm true. Well in fairness to those guys. There's a lot of confusion in the marketplace So there's a lot of nuance here. I would say that it's likely fusion seeking alpha is a Is a site that basically anybody can come in and write on a lot of investors and it very likely could have been a bear an EMC bear somebody was short the stock just wanted to try to keep driving it down But we don't know that for sure who knows I mean it's it's possible But it really it's kind of inexplicable But the the real issue that I want to talk about David is some of the nuance here because you basically in your article Called out EMC and said you were somewhat surprised that they spent so much time focusing on Company F, which of course stands for fusion IO Your your commentary there was really around if I can interpret it correctly the fusion IO plays in in quite a bit of a Different market they go after the hyperscale market there. They're really driving software innovation and EMC's PCIe card and an all-flash array are participating in a different market Can you add some color to that and just clarify your statements there? Yeah, sure In our forecast that we've done we we separate out storage into three groups There's an additional this space storage, which will still be there in very large volume because it's much much cheaper to Capacity to store it on the hard disk There's the flash only arrays or flash predominant arrays Those are the ones that they announced with the extreme IO and those run on the same fans as They do today They they are much faster, but they're down in the millisecond area obviously Distries are down the tens of milliseconds. These ones are down in the millisecond. So the third area is server-based storage traditionally DAZ as disk drive can be SSDs as disk drive, but More and more PCIe cards directly on the bus are taking over that marketplace being very very close indeed to the server And those are three very separate Marketplaces and three very separate architectures that applications can fit into but now EMC was showing benchmarks that very clearly showed that The their product their PCIe card will outperformed Fusion Ios both in terms of IOPS and latencies. They had this kind of tongue-in-cheek example where they were powering The volts of volts electric vehicle with the fusion IO card and an emc card and the emc card out Outrace the the fusion IO card. What do you make of those benchmarks? Well, Dave, you and I have been around benchmarks a long time, you know And our days at IDC if you remember we were analyzing benchmarks on Skyline from Hitachi and IBM Dell FDR and all sorts of things like that So I have a lot of experience of doing those both of the living and of commentating on them And and there are two things about a benchmark which which are important. The first is that it's Reliable that you can repeat it In another place and the second is that it's valid. It's actually measuring something that's useful Are these particular benchmarks are allowable? I strongly expect they are Do they mean anything in the real real world that we it's been deciding between cards? I'm not sure. I don't I think that when when when it all comes down to it We'll find that a PCIe card used the way that they're using it PCI card is a PCIe card There'll be little ups and downs because of technology refresh cycles there'll be improvement people make in the software drivers, etc but The and the benchmark presented to me Ask more questions than they did give answers. So so let me ask you What what software was used what was the environment, etc? And in particular they did an analysis saying as a result of that they were 58 percent more expensive That that analysis is in that they said who was who was 58 percent more expensive the fusion I always 58 percent more expensive. Yeah, sorry the EMC were 58 percent cheaper. Okay, so So that that analysis they did on on the costing of that was was clearly Not done by experts in service and so that particular piece in my view Was in error, but the benchmark that they may be right, but still does it matter? Is it the right benchmark? Is it the right work? Time will tell I suspect that it's a wash well Find that out. Well, so I'm gonna follow up on that In the spirit of full disclosure you're you're very bullish on fusion IO you have been for a number of years You've written a lot about yeah, I talked about fusion IO. I like that technology but but I'm a very straight shooter when it comes to benchmarks and You know, you know the Oracle benchmark everything every Oracle benchmark They're twice as powerful as everything else on the planet or three times more powerful You also know that it's Not true they have taken a benchmark and compared it with an inappropriate Problem, okay, so I you've written a lot about atomic rights Yeah, and in particular talked about how fusion IO has software to enable that and there's not a lot of work in that area One of the points you've made is that atomic rights essentially not only bypassed the spinning mechanical delays of physical disc, but they also Basically reach around or eliminate the protocol disc protocol issues like SCSI, which is a chatty protocol You've made that point a number of times when you've you've said that that's delivers orders of magnitude If not, you know one or two orders of magnitude greater performance So I'm curious as to your comments just now about well time will tell it's probably a wash Do you believe that the the the solutions are comparable from a performance standpoint? Well, what I was talking about there was if you use the same software on both sets of cards and the software You know appropriate. Let's say that EMC has got an Extreme SW product if they tested it on on there that their software Then it's reasonable to say Constantly care compare the performance, you know, as long as the the microcode and everything else have set up correctly So that's one one use scenario in that particular example They would have been on caching software. So that's a benchmark for caching software The other benchmark so that fusion IO have done have been do it do for example A year ago to do a benchmark which produced a billion IOs in a rack of eight service Which is clearly Many many many times more powerful than the the existing one They've recently did a benchmark which showed that they could do nine million Per IOPS per second from a from a car from a single car. Okay single car So again, that doesn't mean it's nine million times faster than or you know, a hundred thousand times faster than the EMC card it needs a different environment and fusion IO are focusing on this server marketplace They're there their key marketplace is helping the Hyperscale people like Apple like Facebook like other Large organizations of that sort and they're doing that by doing something called atomic rights a lot of things they're doing it with the software that BSL software which provides a connection a virtualization of the Office of the right spaces as they're doing it with direct FS, which is a Linux based file system which again Works with the DSL and works with the card and they're doing it with the card themselves those three things make up a little ecosystem and in that environment they They instead of having to write it using scousy. They are writing it using Memory protocols are DNA Directly to memory and they can write that in a hundred nanosecond as opposed to a millisecond And you said one or two all of the magnitude that's for all of the magnitude faster and when you look at specific workloads where Right latency is critical. Those are the databases. So those are especially the big databases the more complex that database the The more that's going on The more you want to put into that database the bigger it is The the more the constraints on it is the locking rate and the constraints on locking rate from a technology point of view Is IO performance right IO performance? Okay, so so let me interrupt you there because this is a complicated situation for a lot of people So you're basically saying atomic rights allow you to directly write to the memory correct bypassing bypassing disk protocols right and that's delivering up to four Four orders of magnitude performance benefit, right? So my so my question is can't EMC do and others do atomic rights? Of course There are some initiatives going on at sneer Around what's called NVME around this area of allowing communication direct communication between the Processors and the and the flash and Intel is behind that correct Intel is one of the big drivers of that and it clearly they are a flash provider So so why doesn't why doesn't Intel and their ecosystem which is very well established and has a lot of money Why won't they just you know crush what fusion IO is doing? Why won't they roll over fusion IO? We're gonna get it out first. You got to get people to buy into it The name of the game is can you get ISVs to buy into any particular way of doing it? and the biggest hurdle to adoption of this particular technology is Adoption by the ISV so where is NVM where is NVM in your opinion in terms of you know time to market? Well, if we talked about the furlong furlong before they're probably in the first five yards of that furlong They're still in the starting gate is what you say Just out of the starting gate, and you know I Would everybody wishes them well and it would be good for the industry that better they do it But fusion IO have been working on this for five years since 2009 So they've got a head start now, you know it it would be great for the industry if if NVM makes it out into the marketplace, but Not even a not even a prototype. There's just talk. So how much of a lead does fusion IO have in your opinion? I don't two three years And it's moving fast and and they've got the The opportunity for fusion IO if they if they can get the They already have got for example for Kona one of the distributors of my sequel to distribute a Atomic right version if they can if they can get that in general adoption by By all of the database market and and the no SQL upcoming database market They are going to have been a very very strong position Now obviously ISP's will do more than one Implementation and adding on the second implementation isn't there at the end of the end of the world, but They will begin with be using fusion IO Well, you mentioned my sequel in Percona, but doesn't doesn't fusion. I have to get Oracle and and and sequel server from Microsoft. I mean some of the larger markets or Absolutely, but but you just imagine that now My sequel operates at twice speed of anything else with far less resources You're going to feel a bit of pressure in that marketplace certainly Microsoft role, okay, so I want to come back to the EMC announcement EMC basically indicated that it's no longer doing thunder what thunder was was a if I understand it correctly was a shareable, you know protectable server side flash using PCIe but a cluster-like capability much more Toward the direction if I understand it correctly that fusion IO is going first of all is that correct and second of all Is my understanding correct that EMC is not No longer pursuing thunder That's both of those statements are correct One of the ways I would describe it is I mean you've got the traditional sand and you've got a service Sand which is you know using very very high-speed interconnections RDMA Obviously clearly in a very small space a hundred nanoseconds is a hundred feet That light travels in so you're talking about extremely much smaller spaces much more condensed environment So you're talking about a totally different marketplace and different set of technologies than the traditional sand to make this work okay, so it basically by Canceling thunder EMC is essentially said that we can attack That marketplace that that that fusion is participating in with a combination of our PCIe cards and a combination of the The extreme IO all flash arrays now if I look up your forecast that you published back see back in 2011 I think you've updated it what you said is that you break break it down in three categories flash on server Flash only arrays and then arrays and I assume hybrids are included in that and you have basically a very small proportion of the marketplace in terms of terabytes as The flash on server 3% but in fairness you've got 20% of the spend You've got most of the spend in either the flash only arrays or the the traditional arrays almost 80% yeah 80% of the market 35% flash only and 45% with the arrays and the hybrids So is that essentially your take on what EMC is saying that they can compete that that market really isn't big enough today that Facebook and and you know Facebook's of the world are essentially a Niche and and not the mainstream of EMC's market opportunity. Is that do you believe that's a? Correct interpretation of EMC's position and do you agree with that position? I? agree that it's a Good assessment of where where EMC's mind is at the moment. Do I agree with it? No For two reasons first of all, it's only three percent of the capacity sure I mean the major amount of the capacity will still be on this But it's about 20% of the spend And that spend is at the leading edge and it's high value and the potential There is for very high margins as you go down the stack the margins You can make if you're doing a just the standard disk driver pretty small very low indeed You'll make more in the flash array, but that's high end is where the real potential for margin is in my view in that marketplace and and that's Exactly how EMC has gone about the market. They went into the tier one market They went after the IBM mainframe at the tier one level because of exactly those reasons Yeah, you feel so that that high that high value that 20% is the database market That's the high value market and you feel like that's that's critical for thought leadership and and and margin leadership Do you you mentioned in your piece that this notion of what EMC's calling directed availability? you felt like that was a Tiptoe strategy to shield its Vmax business I'm not sure if EMC has commented on that or not But can you help us understand what you mean by that you feel like extreme? IO is competitive in block-based environments to Vmax correct Sure, and I mean Vmax is a good product and They have fast VP which is tiered storage which allows you to have Direct flash towards specific applications So it's a good product very sound product is you know, they're obviously for a long time It has very high margins from the emcee point of view. So, yeah, they they They don't want to for example to bring in extreme IO Find that it has not got a feature that's required like for example replication and then See that the business is given to somebody else. So from a from a practical point of view That's that's a reasonable thing to do from a marketing point of view calling it directed Availability, I think it's a little you didn't like that term, but but well, you know Okay, but having said that isn't this a case of emcee eating its own young before somebody else does I mean you got to give emcee Props for actually going out buying extreme IO and positioning it Because now it can get to control the transition of its customer base. Can it not? absolutely, I'm not Debating at all that they are going after this marketplace properly and and if they put the full wood behind this arrow Get the pricing strategy correct They as I said before they can be Incredibly successful with this product. Okay, so just in summary. So you feel as though the emcee extreme IO Announcement legitimizes the marketplace. You got the biggest whale in the sea now competing for that space You expect them to do very very well that was that was clear But if I understand it correctly, you're not buying the fusion IO knockoff You feel like fusion IO still has a lead and and it remains to be seen whether or not NVM and Alternative approaches to fusion IO can Compete at the same time you're saying the fusion IO has to move fast and sign up the ISVs and get them writing to their Software development kit in their API. Is that right? That's correct. Yeah All right, good. Well David floor. Thanks very much for hanging with us here a lot of nuance here You're a real expert in this area really appreciate you helping us squint through it. Thanks for watching everybody This is Dave Vellante at Wikibon with David floor and we will see you next time