 Live from the Moscone convention center in San Francisco, California. It's the queue at Oracle open world 2014 brought to you by headline sponsor queue logic with support from HGST violin memory and Mark logic and Now here are your hosts Dave Vellante and Stu Miniman Hi, buddy. We're back at Oracle open world 2014. We're here in Moscone South It's a Dave Vellante actually on the solo show Stu flying in today So he'll be with me later on today and tomorrow Jeff Kelly is also at the event He's going to be coming in for a big data angle. We'll have David Fleuer on We're here in the queue logic booth This is our fifth year inside the queue logic booth at Oracle open world if it weren't for queue logic We never would have been able to sort of tunnel in to Oracle open world 60,000 people here they shut off you know shut down San Francisco Howard Street is wall-to-wall red stack It's just fantastic and a different vibe than a few weeks ago from VM world, but very good a lot of innovation going on All Oracle all the time But we're going to talk a little bit about a number of things big data fast data Performance Greg Shearer is here as the vice president of product strategy long-time queue alum with queue logic Greg Great to see you again. Thanks for having us here inside the queue. I was great to see you Dave Yeah, so we were talking off-camera a little bit about some of the things that are going on in in our world Let's let's do a little agenda here. We're going to talk about hyperconverged So we're going to start there, but we're going to talk about big data Everybody talks about big data, but you'd like to talk about fast data And I want to have that conversation with you about flash about scale out what latency means How it how it performance is affected by the application and the workload maybe we could talk about, you know Networks and storage and and scale out and how that's been affecting things and then talk about some of the underlying technologies and And no one you and I would go fast and get you all those so let's start with with hyperconvergence It's the big buzzword today. You you've seen the Amazons and the Googles and the Facebook's do a lot of the Convergence in-house and now that's seeping into the enterprise because companies like yours are bringing the engineering into solutions Well, I wonder if you could talk about what does hyperconvergence mean to you and why is it relevant to end customers, you know In a lot of respects hyperconvergence is is really building on a lot of the the practices and premises that we've been doing for Really the last decade going on two decades, you know, you remember when 20 years ago 10 years ago we talked all about Network and storage convergence about in the enterprise that was a big deal that the cloud never had that as a deal The cloud started out converge They had they had one network that was their storage and their network period and you know What we're doing now is we're moving compute into that same environment to where we have our network We have our storage and we have our compute and not just those three elements They're married together by a particular application and you know that application uses completely virtualized resources So the whole notion of whether or not it's operating on, you know What kind of a physical network what kind of physical storage that that whole notion kind of gets pushed off to the side instead of Looking at the physics of it. We look at what are my quality of service guarantees? What are my SLAs for my application require? What kind of bandwidth do I need on a per user or a per application is needed? And that plays into the latency Discussion that I think we'd love to have you know is this scale out storage or is it scale up storage? We hear a lot about big data But what about fast data because in some environments where your application really is scaling to you know Tens of nodes or hundreds of nodes. We don't have the luxury of scaling out to tens of thousands of nodes So fast data becomes really important to be as fast as possible with a smaller number of nodes So I want to come back to that but but first I want to ask you about QLogic's role in this whole hyperconvergence space Greg Because so I'm in the plane last night, and I was explaining somebody. Yeah, we're inside the cube inside the QLogic booth You're like, I've heard a QLogic. What do they do? And I smiled to the lab They said they're the glue that holds it all together. You can't get to your data without QLogic Well, what do you mean? So so talk a little bit about QLogic and specifically What your role is in this hyperconvergence training? You bet so it's funny should mention that because very often if somebody who's new to QLogic will look at the name and say oh glue logic Sometimes I'll laugh and tell him, you know what that's a good characterization You know, I think QLogic's role in this whole hyperconvergence environment if you look at the elements of a hyperconverged environment It's really convergence between our compute our storage and our networking and then virtualized resources Said another way think about a crossroads You know IO sits in the most important crossroads in the data center today Because it connects all those elements together, you know if you look at how do you how do you connect compute to a network? How do you connect a network to storage? How do you connect any of those to compute? It's really through the IO. So I like to think of us is really that the crossroads within the IO ecosystem That really glues together all those elements and provides seamless Communication between them. So I think that that is that's funny that that the glue the glue and glue logic But it's a good way to turn so the value prop is that seamless connection, but also Making it and this leads up to our sort of big fast data making it performant. Yes, and of course reliable I mean it's data Who's data lose data lose your job? So let's talk a little bit about you know the big data meme is exploding We all know it's it's growing at an exponential rate and analytics is becoming sort of, you know, the new watchword It's the you know driving competitive advantage But it's just driving more and more data. What does fast data mean to you? How does this fit in because the Duke generally is you know was started the big data meme? It's batch I don't think it batches fast. It's certainly not real time But now there's a big push to to real time. So where does fast data fit into big data? Well, probably the best way to look at it from the standpoint of fast data is you know You look at it what flash is doing to the ecosystem. It's really Recreating how applications are written, you know It used to be that the applications would look at what technology was capable of and say well gosh You know what? I'm really only capable of maybe several hundred random IOs per second, you know on a on a given disk array and They would really create their applications with linear data to make sure that the data was striped linearly and access it Linearly as much as possible So the applications were constrained by how the data could be delivered But now you introduce faster technologies like flash all flash arrays of different tiers of flash and the applications Themselves are actually being rewritten to throw out the old paradigm to say you know what I can actually have access to Hundreds of thousands to millions of IOs per second So it gives them not only the access but also bandwidth, you know We look at you know 8 gigabit fiber channel moving to 16 Gen 6 actually takes us to 32 gig and even 128 gig on the ethernet front We're at 10 gig today 40 gig is being deployed You know, there's a new push for 25 gigabit ethernet in the IEEE and and an industry consortium group as well as 50 gigabit ethernet We look at these trends even 100 gigabit This this is actually within the realm of endpoints supporting 100 gig this year And and that's really what I mean when I talk about big data It's kind of a combination of the plumbing becoming fast enough to transport Massive quantities of data single streams worth of data at you know up to what used to be unheard of data rates You know on 100 gigabit per second kind of data rates But also the access rates the access rates have gone up by a thousand fold if you think about that That's unbelievable when you think you go from a few hundred, you know random IOs per second to now Millions of random IOs per second that the application can take advantage with a flash with With an all-flash array so I want to ask you about that because as you all know the industry Generally in the storage industry specifically has done unnatural acts for decades because of those the mechanical movement of the disc and and the access density problem The access density problem I just summarized by saying it's too much data under the actuator if you double capacity every 18 months And you can't move any faster because it's mechanical. Well, then you're basically cutting your performance in half Which is what we've been doing for 30 years. Yes, basically and it's finally sort of hit a wall and then flash comes along So my specific question to you regarding your industry and Q logic is what unnatural acts Did you have to do that you don't have to do anymore? What opportunities does flash enable and how are you taking advantage of those? That's a that's a great question because if you look at some of the agendas and things that are moving on A few weeks ago at the Intel developer forum. There was an announcement about something called NVM F NVMe over fabrics and so you think about that and think well gosh That's an interesting marriage because now in terms of being able to throw out some things Traditional storage or block-based protocols were always about you know elevator algorithms and seek optimization to make sure that we Gained the last end percent out of the physical medium to make sure that we didn't have to move that actuator If at all possible leave it in the same spot to be able to pick up the data All that sort of thing goes out the the window when we talk about flash because there is no penalty in terms of Accessing a random location versus a sequential location. So, you know this whole notion of NVMe Over ethernet and even NVMe over fiber channel that you'll see that there's a working group now within the T11 group that's going to address that and essentially what it is is it's jettisoning some of the baggage that we don't need You know seek algorithms a sort algorithms things like that that we can just Be essentially a raw medium and go grab the data as fast as possible and present it to the application So I want to follow up on that and ask it to dumb it down for people like me a little bit So so you're talking about a lot of things that you had to do in In the industry had to do in the protocol to optimize the seek so you didn't have to move So you so I presume something is Conceptually simple as reordering if you if you could if you could serve us something while it's there do it and then reorder At the back end and then make it all work. That was the hard part. Yes Actually, so you didn't lose data and so You created a lot of value as a result of that Does that value go away now and where's the next white space for you actually? This is where it doesn't actually go away It's just it's just augments the current value proposition because if you think about it while I'm very bullish on flash I think flash is is definitely a game changer in terms of how applications are written It it's still if we were talking off camera earlier If we were to use a hundred percent of all the the nan flash capacity that exists in the world today and apply it Just to the enterprise storage market It could it could take advantage of maybe 10 15 percent of the overall storage requirements the nan flash Growth is growing at about 40 percent capacity per year, but the storage requirements are doubling every year So you think about it another way every year. There's less nan flash that can satisfy the enterprise Capacity that's right. I mean prices are going to stay up for a relatively long period of time Maybe not relative to spinning discs. They're coming down faster than this, but there's still a an under supply At this current pricing level. That's right and and and to to look at it just from a pure capacity standpoint The capacity means that that the tiering of data is becoming more and more important So my my hot data my data that I need available right now Versus, you know, some folks that have even termed the notion of cold storage, you know That this is data that I can't afford to lose but like HIPAA requirements I may not look at that data ever again, but I'm required by law to keep it So now the idea of tiered storage my storage demands continue to grow But I don't have to have everything in my fastest tier of storage We're going to have Mike workman on tomorrow Oracle announced you remember of course pillar data. Yes And it sort of disappeared after Oracle bought it and because they were heads down to an R&D for two years So what Larry did is he wrote a big check to Mike workman and said go compete with you know, whoever, you know All the sand guys and they announced this way you talk about tearing it brought it up Something that I'd never really seen before this sort of chameleon array I don't know if you've been able to you probably haven't said just announced it But it's sort of you've got cheap and deep and you've got all flash sort of all in one block-based device Versus what I think of us is sort of this tiering Automated tiering approach and I'm wondering you know what happens to that automated tiering approach. Is that still alive and well? Are we going to see this sort of new model? Do you have any thoughts on that? Well, I think we're going to see many different models and and Hopefully over the course of this week stay tuned for a press release from us What what I what I can say is is is Q logic powers? Literally a hundred percent of Oracle storage and so I'll leave it at that and leave it leave it up to your imagination But so we're we work on both ends of the wire in terms of both at the server side as well as the storage array side and So we have deep relationships for many many years and I think you know a lot of the the current industry is still moving and you know You know block based storage I think what we're going to see over time is more file an object Yeah, you know moving in because frankly, that's the whole virtualization side of this hyperconvergence where you know block is still a very Important aspect, but more and more applications. That's going to be hidden behind something else You know whether that's file object or or some new capability down the road Well, and I think for Oracle. I'm oracles. I don't know how many block based sand providers There are in the Oracle marketplace Let's say the six Oracles probably five out of six in terms of market share So they need a product and that's what this is about so they can they can do oracle can do well just by gaining share in its own Base, which you would think with it with its bundling power. It should be able to do I want to ask you about Latest One of the things that Oracle said last night is we're of 9x faster than the emcee. It's class It was classic Oracle, you know bench marketing, but the focus was on IOPS But there's also an emphasis these days on latency when you talk about all flash arrays It seems like you know latency getting getting under, you know Some threshold to get consistent performance is also very important I wonder if you could talk about sort of the IOPS versus latency discussion how it relates to workload Why should anybody care? Oh you bet and and this is where it gets very in some respects Very specific to the application I'll give an example if you have a fluid dynamics application where you can't do your next step Until you have the the data that you pulled back from your database because your your calculations are dependent upon the data that you've pulled out Well, you could have a an unbelievable compute farm that's essentially stalled until you get your answer back That was pre-stored from a previous calculation So in that case absolute latency is the most important thing So if you can complete your your IO in say 50 microseconds instead of 5 milliseconds be you know to eliminate the seek time Massive massive importance because you think about the number of CPU cycles that that you could do in 5 milliseconds worth of wall clock time. It's unbelievable and this is again where those applications have had to be written to be massively parallel Because if you think about 5 milliseconds 5 milliseconds is literally a lifetime of computing for you and I If one cycle per second is what is what we can perceive that would be a lifetime of computing 100 years worth of computing and 5 milliseconds worth of time So so these applications have had to come up with massive parallel schemes also to take advantage of all the cores But in those types of applications You know latency absolute wall clock latency per IO is very important However, there's other applications where they'll they'll use the same word. They'll say that latency is important They don't mean latency what they mean is is that they want to offload the CPU to where they can use those CPU cycles that used to be used for Computing that the IO for handling the IO they want to use those for other purposes For example, like the file system. I know this is probably the wrong show to talk about this But Microsoft has done a lot of interesting work with their SMB direct file system, right? You know SMB over over RDMA And in that environment Lowest latency really isn't a factor. What is a factor is is lowest CPU usage So if I can load my CPU, you know to five or ten percent and still fill a 40 gig ethernet pipe and Have massive amounts of data going back and forth, but essentially free the CPU up to where now it can do other things That that's another aspect of latency. And so I think it there and there's there's every color in between in terms of you know CPU offload versus absolute lowest latency And I think depending upon the application you pick what your tolerant what your tolerance is So it's a really horses for courses strategy Yes, I love having you on because I can ask you all these Colombo questions and you can explain them in terms that I can understand I mean, I'm not a technologist, but I've been around a long time. So I know all the buzzwords I want if we could talk about scale out Compute scales out storage now become scale out networks are still very hierarchical and rigid Are we going to see scale out? Networks and and when and how's that going to manifest itself now that again, I see I love being here because you ask great questions But you know a lot of the scale out of network I think we're starting to see it it it began maybe two years ago where it began with the thought of virtualized network So you'll hear people talk about overlay networks some sometimes tunneling and The basic ideas is to take the physical network and completely virtualize it to where I don't really have to know Where this node and this node are physically located in order for them to look like they're on the same subnet? Because I can just put it put an envelope around it and make them look like they're in the same You know environment the way I'd like to think about it is think about inter-office mail If you're if you're a single office and and you want to send something to the second floor You take out an envelope and you put it in the envelope You put your piece of paper or whatever it is you want to send and you just write, you know Joe office number XYZ building Y and You're done thinking about it and the mail delivery system gets it to where it needs to go Joe takes it out and takes it out of the envelope and he doesn't care where it came from He has the the contents and he knows who it came from it was written down on the outside and You think about virtual networks now expand that to where I have a worldwide enterprise And I can still use inter-office mail just as if I sent something to Joe on the second floor But now Joe is really in Taipei Or Joe is an Iceland or you know God only knows where The the sender doesn't have to know where Joe is that the network will encapsulate that put it in the right envelope and Ship it to where Joe is going to get it. So, you know the networking world port utilization is always the big metric But what I'm hearing from you is it's more than just that it's the flexibility. It's yes It's like everything we talked about cloud. It's the agility and the flexibility That's so true. And you know, I'd like to think about it is you know about Well, gosh, this started out with IBM mainframes years ago that the whole idea of virtualization was really a you know an IBM You know 390, you know serious and you know virtualized workloads and and then we brought that to The commodity servers and x86 is and that's really what vmware has done incredibly well Um, we look at at hypervisors, you know and what they've done to a standard server You know that we used to think of a single operating system and a single application You know running behind that operating system And and now people don't think that way Overlay networks are going to do that what a hypervisor did to a server overlay networks are going to do to a network They're they're going to virtualize it to the point where you won't know or care where your workload is being carried out And and that's sort of the real power of the cloud And I think what a lot of larry was talking about this whole idea of being able to seamlessly migrate your workload From an on-premises, you know application configuration to exporting it to the cloud You know where you have a similar, you know environment being run You know in order to do that and then export it back if if your needs change I thought that was pretty interesting talk last night. A lot of times larry's talks on sundays are kind of Duds, but I thought last night he was pretty sharp um, the the value proposition of being able to do that and he talked about It was sort of interesting when he said well, we got a bunch of customers on mini computers We started oracle database version two. He said there was never any version one Because he knew that nobody wanted to buy a version one So they started with version two But he said at the time it was mini computers and mainframes and then when client server came out Customers came to oracle and said hey, we want you to run We want to be able to run our applications on client server and oracle said just rewrite them The customer said no, we want you to you know put that technology in place so we don't have to rewrite them Now he's putting forth the notion that it's happening in the cloud, you know yet again That's a pretty powerful concept, you know, and I think in the case of oracle it works. It just it comes at a price Um, you know customers complain about that but they keep lining up To buy and that seems to be a pretty viable strategy is we have java. We have middleware We have our database. We now have cloud You can move it in without changing a line of code and move it out Assuming you have the red stack. So that's a very compelling value proposition for a lot of customers Before we go getting the sort of the hook here. I want to ask you about mobile You talked about a hundred gigabit at the end points Yes, it strikes me that mobile is driving, you know, a lot of demand at the edge So I wonder if you could talk about mobile how it's changing your business and you know, what's your forecast for the future? Boy, I mean some of the statistics are just unbelievable when you think about what's happening in mobile You know just you think about even even personally, you know, not that many years ago I remember just being aghast at the idea of my wife and I went to a restaurant and you know, they were It was a family, you know, they bought one of the first, you know iPads and they were actually streaming Netflix You know to keep the kids quiet in the restaurant and thought oh my god Do you know what this is going to do to network traffic? If this catches on and now shoot, you know, I I venture to say that there's a lot of folks Streaming us live right now on on their, uh, you know iPhones galaxies, whatever they've got And just this explosion is unbelievable in terms of the demands And it's not just the this is probably the most powerful thing is it's not just the enterprise that's creating this demand Matter of fact, I'll say it's not the enterprise. These are consumers. These are kids, you know, that are that are using Either laptops iPads or iPhones Whatever other technology that they're using to go ahead and pull down rich content and You know, you look at the Now at hd was a big deal not that many years ago to be able to to look at streaming hd video And now we're looking at 2k and 4k video and you start looking at that and thinking 4k video That's 4x hd and you know, where are we going to get that bandwidth? And certainly the codecs are getting better at the consumer level, right? But we still have a plethora of data and then where do we store that? There are replication sites to make sure when there's a big demand when You know, one of the starlets does something that everybody wants to watch On their mobile devices and all that generates huge huge demand in the data center, which is Fabulous news for those of us that are data center suppliers And into the moving data business right and keeping it reliable It's just it's amazing to watch a third of the the the internet's traffic in in prime time No, two thirds is netflix. Yes, which is just an astounding fact We have a couple years ago Maybe a year ago we bought a new car my wife's car And it came with those video, you know things in the back headrest and I don't want it It just came with it The kids don't even need it anymore You know, we don't even know how to use it because they're on their devices, you know and We're in the early days. So Greg it really always a pleasure having you on Thanks very much for helping us squint through all these trends and After the announcement this week, maybe we can have you back and chat about that a little bit. Love to. All right. Thank you Thanks again. All right. Keep right there, buddy This is the cube we're live from the Q logic booth at oracle open world 2014 in mosconi and san francisco And we'll be right back after this work