 Okay, we're back. This is Dell Storage Forum 2012. We're live. This is theCUBE, SiliconANGLE.TV's continuous production of Dell Storage Forum. I'm Dave Vellante of wikibond.org and I'm here with my co-host, Stu Miniman. Stu, we've been here all week. We've got a great guest on now. Howard Marks is also known as Deep Storage Net. That's his Twitter handle. He's a technical expert. He knows storage. He's a consultant of many Fortune 500 companies. He's been in the storage business for a long, long time and now is an independent analyst, blogger, journalist, slash all of the above. Howard, welcome to theCUBE. Thanks, Dave. Glad to be here. So, we're here at Dell Storage Forum. This is the second year that we've been here and we've been talking about the theme of the transformation of Dell, going from kind of what I called on Twitter, Car Dealer, Car Maker, meaning owning a lot of their own IP. You've seen that transformation. What do you make of it? I'm really impressed. First of all, Dell has managed to do acquisitions a lot better than other people in the storage industry have. They've took on Equalogic and when they took on Equalogic, they had a lot of channel conflict problems because for 15 years before that, they were all about disintermediation and selling direct to the customer and Equalogic was a channel driven company and several of their resellers just said, we're not dealing with Dell, you're an evil empire trying to put us out of business. But yesterday was channel day here and there were 350 resellers. Many of them had sold Equalogic and stopped selling Equalogic and then came back when they discovered Dell could manage that. Dell bought Ocarina and Exanet when they were basically infants before there was a substantial installed base and said, this is very interesting technology but it's not really a product. Let's see what we can do about rolling that in and incorporating it with our other acquisitions and some of the other folks in the storage business at best acquire companies and continue to sell the products that they've already been making and some of the others acquire products and kill them off because they just don't know how to do the acquisition business. So I'm very impressed with Dell's progression from that disintermediation into a real supplier of storage. Yeah overnight they have become relevant in the storage market when you agree as opposed to sort of not that Dell was irrelevant before but they were a channel. They resold some stuff under OEM, they resold the Clarion line, they occasionally sold a Symmetric because they had the intro into the customer but they were box pushers. They were essentially resellers in the storage business and now they bought Ecologic which is the best product in that market where ease of use is the number one thing you have to worry about and they bought Compellent which is a second generation mid-range array and have managed to not annoy people. I mean I remember going to C Drive, Compellent's user conference that this is essentially the replacement for and being amazed that Compellent's customers were the happiest campers I ever met. Nobody had major complaints except this feature isn't in the product. Nobody ever said you know it broke and they never fixed it for me which you know as you know when you see a thousand of customers somebody said with something. Yeah or the user interface stinks or there's a laundry list of complaints on the to-do list. Right. And you never heard that from Compellent customers. I had the same experience. I went to C Drive but I interviewed a lot of Compellent customers and I would ask them you know what's on the to-do list, what aren't you happy about, what doesn't work and silence. Yeah. Now you saw that, actually saw that a lot from the most sort of more modern you know virtualization, right, storage virtualization companies. Okay so let's get into it. Let's start with Flash. Flash is a hot area. It's something that you've really been studying deeply. I know you're right in the middle of a big study on Flash. We saw one of the greatest wealth creations in the history of the storage industry you know leading up to I guess the Isilon acquisition, right. We saw Equalogic and Left Hand and 3PAR and Compellent and Isilon and Data Domain. Right. I mean just huge valuations of these companies that got taken out by the whales. Yeah but none of that compared to the Extreme I.O. And then so my question is now you're starting to see companies with no product in the market. Right. You know really no marketing per se, no channel, no install base, sell for 400 plus million dollars. Just a really good idea. Yeah. And the set of engineers and enough of an implementation to say yeah that's a really good idea and it's going to work. Really smart engineers, right. But what's happening here? Are we going to see a repeat of the storage virtualization frothiness? Is it going to be bigger? I think we're going to see that. So you know traditionally we've thought about storage costs in terms of capacity. It's been dollars per gigabyte and that's really all we cared about. And Flash because the economics of Flash are completely different. Flash is very expensive per gigabyte but it's cheap per IOP turns that set of economics on its head. And now we have to think about not dollars per gigabyte but dollars per IOP for performance. And just like in the 70s we made on the mainframe the transition from doing work on tape to doing work on disk and tape for secondary storage. We're at the beginning of that same phase where we're going to do work in Flash and we're going to have disk for secondary storage. And we now need to automate that process. We need to find a way that we're going to have Flash provide the performance that Flash provides. Disk provide the lower cost capacity and lesser performance. And may have that all be relatively transparent to our users. So the other thing if I may, Stu, if I want to get your technical perspective and market perspective on for the last 15, 20 years we've seen function, critical function move out of the host, you know, outside the other side of the channel to the sand. Right Stu? Yeah, I know. So if I can, that's actually where I'm going with it, Dave. So one of the interesting things that I think and we've had this discussion is the transition and adoption of Flash. It's not one solution because we saw disco from kind of DAS to a storage array. And Flash today, it can be in the server. It can be more of an appliance. And there's even value putting it in the storage array. So there's there's it can be in lots of places. There's lots of ways to use it. But the thing about Flash is it's complicated by comparison to a disk drive. A disk drive is a relatively simple thing. You can read and write at the same, you know, it reads and writes at the same speed. Every time you move the heads, it takes time. You can read and write the same block over and over again without it wearing out. Flash, you can do random IO actually a little bit faster than you can do sequential IO, but you have to be really careful about managing rights because rights take longer. And rights are destructive. There's only so many page array cycles you can run a block through. So you have to look at your workloads more differently, more exactly, determine what kind of flash you're going to use and where you're going to put it. Right now I'm working on a report about using Flash in the server to accelerate access to shared storage. And that's part of the counter trend to what Dave was just talking about till two years ago. The trend was move storage functionality out of the host into the array. And so even things like volume managers and software raid and mirroring all faded away and we started pushing that out to the host. But VMware has started sucking those things back into vSphere. We've got now some storage. You'll be pushing it out to the sand. Yeah. Pushing it out to the array. Yeah, yeah. Right. But VMware is reversing that. Right. So you have storage IO control and storage vMotion so that you can have the host manage where some of that data goes. Plus, the closer you can get your storage to the processor, the less latency there is. Yeah. So if we look at HPC and big data and some of those other applications. Right. Where you say we'll distribute all the storage, we'll distribute all the compute. Everybody will have access to their chunk. It's all about sharding. But even in the more generalized compute environment where we say, well, you know, my Oracle server is running slow. And it's running slow because it's waiting for the disk. So I can introduce flash to speed that up. And I can introduce flash to speed that up as part of the disk array. I can introduce flash to speed that up somewhere in the sand between the disk array and the server. Or I can insert that flash right in the server. And there's the Fusion IO solution where we put the flash in the server and we use it as DAZ. But that's got all sorts of limitations around clustering that you, now that storage is owned by that server. And if something goes wrong with that server, it's not available anymore. If we add a software layer and we say we're going to have that flash in that server, we're going to get the 100 microsecond latency talking to that flash. But it's just going to be a cache. And we're still going to have backend storage. We're still going to use the backend storage for snapshots. We're still going to do server offloaded backups via VStorage API or some kind of snapshot management. That's a combination that makes a lot of sense to me. So we're seeing the functional shift back toward the server as a result of the catalyst of VMware. So let's handicap the horses on the track. What does that mean, that functional shift for traditional storage companies like an EMC or a NetApp? And does that mean that server companies in theory should be able to take greater advantage of that? Do you see an EMC make some moves? It means that there's plays for everybody. It opens up the field. It also opens up the field for startups because if I can do storage IO control and storage of Emotion, then I don't need to have a big monolithic array that has multiple tiers of performance. If the host is deciding where to put things, I could have an all flash array from an all flash vendor who's a startup and I could have a large capacity array from a more traditional storage vendor and let the host decide which virtual disk goes on which one and now I'm not locked in to any particular user, any particular vendor. When we start talking about moving the flash into the server, it opens up the doors for the server vendors because they own that box and you want to have something that's going to be supported in the service contract for that. It opens the door for independent software vendors. We're going to write the caching layer and the big storage vendors specifically EMC aren't blind to this. They came out with the VF cache, which is their software caching layer and then they buy PCIe flash cards from Micron and LSI and bundle it all together. But the range of products in that caching layer is enormous. There's very basic read-only cache products. There's read and write cache products. There's products that work in an operating system and therefore good for the Oracle database acceleration but if you use them in the VMWare world, don't give you the VMotion capability because the cache is locked to that host. A lot of range and a lot of places yet to go. So who are the leaders in your mind right now? Obviously FusionIO was out to a fast start. It's interesting news on FusionIO yesterday made an announcement with Cisco. We talked about who's a potential server guy getting into storage. We've been watching Cisco for the last decade of his space. You see Fusion, you see violent memories raised $160 million, got an $800 million valuation. SolidFire hasn't come to the market yet but they're doing some interesting things. EMC you mentioned made a bunch of announcements. You're hearing Dell talk about it, IBM, HP, they're all talking about it. Help us squint through that. Who are the leaders right now in your mind? I think for the server side you've got to say EMC is the leader because they've got the channel. Their product is very basic. VF cache is what they could get out on the schedule that they wanted to get it out. It's just a read cache. It doesn't let you do VMotion. But they got boots on the ground pushing that product out and they're going to have that first player advantage. FusionIO snapped up IoTurbine who had a more sophisticated caching layer for VMware. It does let you do VMotion with the limitation that when you move a workload from one host to another the cache is cold when you get to the second location. And then there's SanDisk bought up FlashSoft and there's the seven dwarves of ISVs with new products. The interesting thing about the UCS announcement and LSI made a similar announcement last week is the first generation of UCS blades only had one mezzanine card. And I looked at that and I said well there's one mezzanine card and you have to put the network card, one mezzanine slot, you have to put the network card there. There's no room for anything else. And so all of these announcements about PCIE Flash for UCS are dependent on second generation UCS blades that have that second mezzanine card. So that opens up the possibilities as well. But both HP and especially Dell with their latest generation of servers now have front mount PCIe slots specifically for SSDs. So you don't have to open the server, derack it to get to the slot. Storage guys really like the idea of hot swap. SSDs are going to fail, they may not fail in the same way or as frequently as hard drives. If I can pop it out of the front of the server I'm a lot happier with that. If it's a PCIe connection I'm going to get the same performance as if it was a card. And NVMe and SCSI Express are working on standards for that connector so I can buy that SSD for multiple vendors as well. How about a couple of like Violin, I mean I've looked at them and said they're kind of like the data domain of Flash and that you can just plug it into your existing environment and they've got this huge valuation and Don Bacillus raised a bunch of money. What do you make of that? Violin's done a great job raising money and they're in the all Flash array category so they compete with Whiptail and Nimbus and Kaminario and a couple of others. And I did a blog post on network computing last week or the week before where I classified the all solid state array vendors into the dragsters and the Lamborghini class and the BMW M class. We're running low on time. I've got one follow up question I'd like to ask you. We said it's kind of early days. There's a lot of technology work that needs to be done here, a lot of new startups. Where are we in this trend and how long before most customers, there's going to be technology available that most customers are going to be able to hop on? I think right now we're kind of at the end of the first quarter. There are some products that are worth having. They're all very preliminary versions and none of them have the full feature set. The truth is what we really want is something that will accelerate all of our workloads that will work in the Vemotion or Hyper-V or Zen environment and let us do Vemotion with a hot cache so that you can move a workload from one server to another and have that cache preheated and not take a performance hit when you do it. What we really want to do is turn on DRS. We want it to happen automatically. No, no. I thought Carter George summed it up great yesterday when he said what we really want is a knob I can just turn to 11 and it takes care of everything. So what's the... And that's sometime next year. What's Dell got to do in Flash? Last question. Well, the RNA network stuff that they're calling Hermes is really promising. They need to do a better job with Flash in their arrays. Right now, both Ecologic and Compellent use Flash for automatic sublun tiering. I'd rather see them do some additional caching with Flash as well because I think the combination of the two is worth a lot more than either of those features alone. I think on the server side, they've got that interface in the front of the Gen2 servers that makes a lot of sense that Hermes software, if it really delivers read and write caching and the mobility that they're talking about is going to be better than anybody else that's out on the market and they've got the potential to be able to deliver a real winning solution. So your study is out this month or next month? It's out in July and people can get more information about it on the projects page at www.deetsstorage.net and I think this is a very exciting piece of the market and the last thing is I've heard people say that server side caching is just a Band-Aid and even if that's true, Johnson and Johnson makes a lot of money selling Band-Aids. If you're bleeding and you have that performance problem, a Band-Aid is the solution for now even if it's not the 10 years from now solution. Excellent. All right, Howard Marks, Deep Storage Net. Thank you very much for sharing your independent perspectives. Thanks for coming on theCUBE. It was great to see you. My pleasure. All right, keep it right there. We'll be right back with Brian Payne and we'll talk servers. All right, keep it right there.