Features:

DESKTOP CLUSTERING MAY START ZESTY TREND
By Tim Curns, Editor

Late last August, Orion Multisystems Inc, a company founded by computer industry veterans, announced two models of the Orion Cluster Workstations that can be plugged into a standard wall outlet and used in an office or laboratory environment. Designed for the individual user, the Orion Cluster Workstation provides supercomputer performance for engineering, scientific, financial and creative professionals who need to solve computationally complex problems. HPCwire caught up with Orion's president and CEO, Colin Hunter, to discuss the workstations' impact at Supercomputing 2004 and beyond.

HPCwire: Good morning, Colin! This is your first Supercomputing show since your major announcement last fall regarding your workstation cluster product. What's happening here for you in general?

Colin Hunter: What we're doing is showing off to the public, really for the first time, our cluster workstation. We're showing off the cluster desktop workstation, which is a 12-node cluster, and also the cluster desk-side workstation, which is a 96-node station. Both these units can be stacked, which means you can have up to 48-nodes with the desktop, or as many as 400- nodes if you put together the desk-side. The great thing about our product, fundamentally, is that the power draw of even the desk-side can be done in a single, 15 amp wall socket. The desktop takes about the same power as a typical desktop PC. So we're essentially bringing cluster supercomputer-style performance to the desktop and desk-side.

HPCwire: Since your announcement in August, what kind of movement or momentum have you seen?

CH: We've been getting a tremendous response from people who read about the product, we're following up a lot of leads -- and we are shipping. We've been shipping since late October. We had revenue in Oct., I'm proud to announce! We're shipping, we have a lot of leads, and we actually have a backlog right now. So from a business perspective, it looks like there definitely is a market for cluster workstations.

HPCwire: So would you say you're bringing the workstation back in a way?

CH: That's a good way to look at it -- it's kind of the rebirth of the technical workstation. That category never went away in the aspirations of the user community. Everybody has always wanted the most computer performance they can get under their personal control. The problem was that PCs essentially caught up with the old style technical workstations and they were all just unit processors based on microprocessors. Meanwhile, supercomputers loomed ahead, as you can see from this show, and for all practical purposes, the word "supercomputer" means cluster now. There's really no such thing as a non-clustered supercomputer. So for really high-performance computing, you have to bring a cluster to the desktop. And that's what we've done.

HPCwire: Do you think clusters with so much personal control will make supercomputing more mainstream?

CH: We represent the first of a series of steps. The first step is to bring clusters out of the back room to the workstation format. Now the typical workstation price is around ten to one hundred thousand dollars. That's not really an individual pocketbook item. So the next step that will happen is when the inevitable price curve comes down, we can bring clusters to the PC style market. So something that costs in the order of one thousand to two thousand dollars is based on multiple nodes per cluster. That's in the future, but that will inevitably happen. And then eventually you'll have cheap laptops and things all based on clusters. I do not believe that in ten years it will be possible to buy a computer that is bigger than a cell phone, that has anything but a cluster inside of it.

HPCwire: Would you characterize these developments as part of a new zest for the workstation, then?

CH: I think it'll show some people the exciting applications that have come out in really high-performance computing. I mean people were vaguely aware of things like Hollywood-style photo-realistic rendering, which is all made possible by clusters. I think that people are becoming more aware of that. And I think what you will see is a lot of capabilities for say, photo- realistic 3- D images, will become now available to a much wider market. The workstation is the first step in making that possible.

If you look at things we take for granted now, like color 2-D graphics, which was all pioneered in the 80s on workstations. Silicon Graphics workstations were the first windowing machines that -- I mean, I saw a flight simulator back in the early to mid-80s on an SGI workstation. Now it's available on PC. And so people will see some of these exciting applications and say, "Wow, if I had a machine like that I could do some really cool stuff," and it will inevitably happen. So I think that's what you're going to see.

HPCwire: So who will be buying your product for the most part? What industries do you see using your workstations the most?

CH: Well, our main customers are the classic users of HPC and workstations. Wherever you have both a computational need and also a need for a designer or technical professional to have control of the computer. For example, I mentioned Hollywood. The whole idea of special effects, what is called compositing, lighting -- trying to adjust the lighting of a movie that has not yet been shot -- using rendering, that whole world is a classic one.

HPCwire: Do video games also fall into this category?

CH: I'm not sure video games do yet because video games are not really that into photo-realistic imaging. They are more into just something they can do quickly, plus, it's really highly optimized for hardware accelerators. But most or all Hollywood rendering is not done with 3-D graphics chips. It's all done with just CPUs.

Another section would be financial analytics. Not widely realized, but on Wall Street there is a tremendous amount of high-performance computing in real-time derivative training and things like that.

There's another one which I think people are reasonably aware of right now, which is the whole life sciences industry. Bioinformatics is the fastest growing are where computers are used. And they are being used in modeling, they are being used in DNA sequencing analysis...there is just a tremendous amount of high-performance computing applications. And then you have pharmaceutical engineers designing what is called "rational drug design," and they use it the same way that CAE and CAD have always been using electronics. They're saying, "let's move this molecule this way and see if it's a better drug. It's all being done with computers now.

Classic areas like computational fluid dynamics for airplane design, automotive design, analysis for modeling mechanical structures, architecture, anything mechanical. I'll bet this device here [pointing to tape recorder] was modeled first in a fine element analysis to see if it would be strong enough. You want to make things as strong as they need to be, but no stronger. If something is stronger, it's expensive. So you use modeling programs to see the mechanical stresses. All this stuff is classic stuff, it's just much easier to do the modeling that you have HPC.

HPCwire: Are there any untouched markets you'd like to tap into?

CH: We're still early! I'd like to be able to delve into all these markets. The usual estimate is that the HPC market as a whole, in the price range we're talking about (between ten to one hundred thousand dollars) is about a $2.5 billion market this year. Bioinformatics alone is going to grow into a $2.5 billion market in a few years. So I think there is a pretty good market just addressing taking existing HPC applications and moving to workstations. But, once these workstations are out there, I expect to see clever programmers come up with new applications that will enlarge the market. I don't necessarily want to predict what they will be myself -- I don't think I'm clever enough to do that, but it happened with workstations. When workstations got started, they were taking UNIX out of the lab and putting it in front of people, and those people came up with cool things to do with it.

HPCwire: You are relying on Transmeta chips. What's the advantage here?

CH: We have standardized on the x86, the x86 architecture. There are three vendors that sell x86 compatible chips, and we have good relationships with all of them. Every few months, we do a shoot-off, based on everybody's road map. We selected Transmeta because the Transmeta Efficeon is currently the highest performance per watt chip that we can use. And we care tremendously about performance per watt. By the way, I want to point out that everybody cares about performance per watt, if you notice, the whole world is moving toward multi-core. Multi-core is all about the most performance per watt. People finally realized that the way to get higher performance is to have multiple units, as opposed to trying to run one unit so fast it's like a nuclear reactor.

So we'll be able to take advantage of that, too. In other words, we care about the performance per watt, per CPU. We just want to have as many CPUs as possible inside the box.

HPCwire: Is there any space within the Grid arena for Orion?

CH: Yes, Grid is a buzz word that means a bunch of different things. I think fundamentally what it means is the ability to take advantage of unused computing cycles. You have computers sitting around, it's useful to be able to somehow suck these cycles and give them to other people. Most of the Grid solutions I've seen are things we can easily support. In other words, we would run Grid software on our box, and if one of our boxes is not being used 100 percent, the spare cycles can be administered centrally and given out to whoever needs it. This is the general way people find they use workstations. The whole idea of the workstation is about the control of the engineer. But in the old days, with the Sun workstations, everybody would login over into someone else's machine, and I think Grid is kind of a systematic way of doing that in a friendly way!

HPCwire: How do you further plan to advance the use of clusters?

CH: I think by making them easier. Here's how you get a cluster now: You call up a cluster vendor and say, "build me a cluster!" And then a few weeks or months later, equipment starts showing up in boxes. Eventually, the engineers to assemble it show up, and they take all the computers out of the boxes, they rack them up, screwdriver them together, and test them, get them wired up, and get the software running and you pay them a lot of money. Finally, your cluster is in the back room, you've got the air-conditioning right and it's all working.

With us, you order it, it arrives in the mail, you take it out of the box, you plug it in, press the on switch and you're up and running! I think that's the main thing in terms of making it easier to use.

Plus, there's no doubt that clusters are exciting right now, but it's a big transition to go from having a PC to trying to build a cluster or have one built. Just making the process so similar to buying a PC will make clusters seem less intimidating.

HPCwire: Well, we look forward to Orion's impact on the market, Colin. Thanks so much for taking some time to speak with HPCwire today.

Colin Hunter was a founder of Transmeta Corp and served as the company's vice president of Software Engineering. He also co-founded Ready Systems (VRTX OS) and Hunter Systems Inc, and served as President and CEO of both companies. Catch Orion Multisystems at SC2004 in Pittsburgh, PA this week at booth # 1146.