 Okay, we're back at Oracle Open World. My name is John Furrier, the founder of SiliconANGLE.com, SiliconANGLE.tv, and we're inside the CUBE, our flagship telecast here in San Francisco, California, for Oracle Open World 2011. And inside Oracle Open World is the big show around IT, Oracle Database, Oracle Software. It's the biggest IT show on the planet. I mean, it's almost in size, it's CES. But one of the things that we do here at SiliconANGLE.tv is the CUBE is we do spotlights. Spotlights were introduced to go in-depth into conversations. And I'm here with my co-host, Dave Vellante. Dave, tell us about this spotlight around High Performance Data Center. John, this spotlight is focused on the High Performance Data Center. It's looking at the major trends within microprocessor design and how they're manifesting themselves in new data center designs and what it means for customers in new workloads. And we're here with David Floyer, who's an expert in that field. David, welcome. Hi, thank you. Thanks for coming in. So, what we do in these spotlights is we try to put together some visual aids and Mark will put up various slides on the screen. And the first thing I want to look at is some of the trends that are driving change in the data center. And what we're seeing is you've got a number of mega trends. Virtualization is obviously the big one. Everybody talks about huge data volumes. And David, as we know, heat density is a big problem in the service space, isn't it? Why is that? Absolutely. Heat density was rising very rapidly, 30% per year for quite some time until the chip vendors took it on and worked to reduce the heat density. And it's been rising slowly for the last few years. But it's going to become a bigger problem. They've run out of tricks. They've got it down to the minimum. There's going to be a rapid increase in heat density because fundamentally, the closer you can get chips together, the closer you can get processes together, the higher the speed of the processes of the systems. And that's what people are driving for is very high density and high performance data centers. So there's all kinds of other residual effects here. We're seeing the convergence of compute, networking, and storage. Something that we've talked about a lot on Wikibon, on SiliconANGLE. Different types of data are coming into play. I'm talking about video here and audio. David, I know you're big on video, obviously unstructured content and mixed media. And then we've also talked a lot about the communications between processor and memory and IO devices and 10 gigabit E, InfiniBand. It's coming along. We're in Oracle Open World. We know Larry Ellison loves InfiniBand. He's taken an investment in a company called Melanox. And we've now got something new here, which is with 64-bit, you've got practically infinite address spaces, which is supporting these data in memory and what we're calling data in flash, architectures. And we're seeing a move towards scale out designs. They're still scale up. We're going to talk about that in a minute, but scale out is the new hot thing in the data center. And we're seeing that in Web 2.0 and other environments, Facebook and others. Absolutely, yeah. So let's talk about the competitive landscape. John, on theCUBE, we'd love to use sports metaphors. These are the horses on the track. Yeah, and Dave, I mean, to me, it's all about knowing the horse on the track. In the data center, it's very competitive lately and competition has been heating up because we saw things like Juniper having some layoffs, Ritual and their company, the software switch is kind of coming back at the convergence layer. HP's doing really well. Some are doing bad. But what's happening is Cloud's taking over and at the fundamental speeds and feeds of it, which Oracle was all the rage about on the keynote, it's all about what horses are going to win this race. Now, we call it the triple crown. The real triple crown was the PC business. That was the initial, I think the Kentucky Derby Intel won that one, they did pretty well. And then you saw the server business in the data center, multi-core, Sir Plant, the UNIX type Spark and Solaris. Call that the preakness. Now we got the Belmont, the longest of the races. So Intel's the secretariat of the industry. And I believe that they're really going to run a great race and fly in this next race because it's going to be about mobility. It's going to be about that kind of performance. We're talking about stuff that we heard about in-memory stuff, talking about heat, management, all those tools. Intel has literally a great R&D in this area, so I think the prospects look good for Intel. So Mark, I wonder if you could expand that slide for us, because this is sort of the lay of the land here. The champ right now is Intel, the x86 architecture, I guess Intel and AMD, but Intel is the champ. The clear volume and revenue winner. This happens to be, if you can get that slide up there, this happens to be secretariat winning at the Belmont. I don't want to take the analogy too far, but we're not ready to call Intel the winner in the mobile space yet, but that's Ronnie Turcott looking back, 31 lengths. You can see the race for second place going on, if you could put that back up there. And here it is. Is between IBM, PA Risk, and Fujitsu Oracle, and Spark, and David, we're going to talk about that, but in some detail in a moment, but they're going for the largest single process of performance, right? That's the scale up that we were talking about before. Yeah, you need a combination of scale up and scale out, and from the scale out point of view, as you said, x86, and the volume in the server market that Intel have, that's a fantastic story, and that's winning. And they're going very far up in the scale up as well. But certainly IBM have had a very, very fast single processor, 4.25 gigahertz, very well constructed, and that's been the leader in terms of the largest processor, it's the largest chip and processor. They've got a 32-way, which really is an awesome box. So they're going after what are increasingly becoming niche workloads that part of the market. The main frame, yes. And then trailing the pack is HP, Itanium, it's just a matter of time. There's a big war going on between HP and Oracle right now, where Oracle is pulling support. What a lot of people don't know is Oracle's actually going to support Itanium for the next seven years, okay? So, and after seven years, they're not going to fix bugs anymore on Itanium. So there's this big riff going on, and I think that in HP and Intel, of course they've said they're going to support Itanium for 10 plus years, but seven years is a long time. Hopefully by then, most of the bugs will be squashed, but it's not like Oracle's pulling support for Itanium tomorrow, okay? That's an urban legend, all right? Itanium is, if you're an Itanium customer, you got some time. So don't freak out. And in addition to that, the HP and Intel parts of this, they're doing every other cycle in terms of performance. So they are going to be behind the others. It's just a matter of time. And then you've got the fall over here in the picture is ARM. ARM, absolutely. It's really a future player in the data center. Everybody used to laugh at Intel. Ah, Intel is just for PCs, it's a toy. Well, people are maybe saying the same thing about ARM, but Facebook gave a hard-looking ARM, right? Absolutely, a number of the people have looked hard at ARM because it is so going back to heat density. It is so efficient from a heat point of view. So putting these much smaller, cheaper chips in very close proximity was one architecture that's being looked at. And in my view, we are going to see servers at the low end in particular using the ARM technology. They're going to be coming in underneath the X86 umbrella in the server end, but they're not going to have a significant impact on the medium and high end of the marketplace. So I think they are a fall. David, you wrote a piece on Wikibon this week, actually, saying that Oracle was economical with the truth, I believe, is how you said it. Very British way of saying they're stretching the truth and maybe downright misclaiming things. But anyway, here are the facts. The big X is exologic, exadata, exolytics. Exolytics. Run on Xeon today, correct? Correct, yeah. Okay, now having said that, Oracle announced the Spark cluster, the T4 Super Cluster. Yep, Super Cluster, which is the exologic equivalent. Why is that important? Because the key thing that Oracle want to maintain is their customer base. They've got a very loyal customer base running on Solaris. A lot of their install base, 70%, I think they said was running on Solaris and on Spark. That's a lot of customers and a lot of goodwill. So they need to keep those customers. It is expensive and risky if they try and convert them to exologic. It's risky because it's x86 and they can go anywhere. And it's expensive for the customer. So by far the cheapest and most effective way of keeping those customers happy is to keep on turning the crank on the Spark system. That's exactly what HP didn't do when they had their own chip set. They went to Itanium, they left their customers in the lurch with a big conversion issue. So Oracle have got it right in the sense that they've got to put money, got to put effort into keeping that large customer base happy. And then just again to go through the facts. And this is important because Oracle claimed like a zillion world records and number one in everything and every category. And when you wrote a piece, you squinted through there, it's really not the case. It's not the case. Oracle was comparing its latest and greatest with a somewhat outdated configuration. It was using Flash. The other configuration wasn't so it's really tuned. So you have to be careful with these claims. But having said that, the T4 performance is much, much higher than T3. It's about twice. About twice and sometimes it'll be more than twice. And it's about on par with IBM. So in this game of leapfrog, Sun or Oracle has caught up. It's caught up on the processor. Now it's got to put more processes together. But the problem is that now it's IBM's turn to jump. Exactly. So they've got to keep going and they've got to keep going hard. Larry has promised a T5 next year which is going to double the number of, it's going to be an increase in speed and double the number of sockets. So that'll take them on their way. So, but IBM has a very substantial lead in terms of the largest single processor at the moment. Sorry, the largest single system at the moment that can be going. But Oracle can still be successful by keeping their current customers happy. If they do that, then to me they've won a lot. So competing with IBM for that single scale up is the mainframe class. And it's actually conceding the broader market to Intel, right? Yes, yes. All right, here's the watch list. All of the things that are on our watch list. These new workloads are emerging. Hear a lot about them, Cloud, Big Data, Web, Video, they're real, they're certainly real at companies like Facebook. We had Fusion IO on earlier. We saw some amazing action with Facebook there. These are, and others, Apple. Apple announced iCloud availability today. And so, we're really seeing these workloads hit the mainstream data center and managing those workloads will become increasingly important. How you take all these commodity distributed servers and allocate workloads to them is becoming more and more important, isn't it? Absolutely, yes. And then heat density, as you pointed out, is on the rise again. Asheri data shows it flattening out. You have said that's wrong. That's not correct. Yep, very clearly, to me. They've had four or five years of the chips, Intel and everybody else working at this problem, turning things off, trying to cut down the amount of power. A lot of it driven, obviously, by the mobile marketplace. So they've been very successful in that and it's been around 5%, 10%. But it's going to go up again. And these systems that are being produced by Facebook and Apple, et cetera, they need the highest performance possible to deal with these huge databases that they're putting in place. And these databases need to be really real-time. So if I'm using Apple and I buy an app and my wife is on the other side on the same account of the world, she wants to see her app coming down at exactly the same time or the tune that we've bought together. So this type of very, very high-volume, high-speed database is going to take real architecture and real compute power to put together. And it's things like Fusion IO, very flash memory, and putting things very, very close together. They're going to make this possible. And we're seeing Rom Lee come out shortly, so we're watching that. One of the other things we're watching is how CIOs are going to fund the transformation from point A where they are today with a lot of legacy, diverse infrastructure, to point B, which is this new scale-out, new workloads. That's not trivial. There are certain things you can put on Greenfields. There are certain things you can't. That's what we're all about at Wikibon.org. It's peers helping peers solve problems. So come please check out the site, contribute your knowledge, or if you got a question, ask it, go to siliconangle.com. Go to SiliconANGLE.tv. Lots of resources there to help you through these transitions. Thanks for watching this spotlight on the High Performance Data Center, everybody. As I say, check out the sites, check out the resources. You've been a great audience. Keep the questions coming, and we'll be right back from Oracle OpenWorld Live 2011.