by Steven Witucki, assistant editor LIVEwire
Dallas, Texas — Most research that requires massive computational power goes unanswered because high performance computing has been priced far out of reach. Parabon Computation, Inc. has developed a solution to this problem. Parabon’s Frontier computing platform hopes to provide the power of distributed computing to any developer or independent software vendor.
The Frontier software development kit was introduced this week at SC2000 in Dallas, Texas. Steve Armentrout, Ph.D., the President and CEO of Parabon, attended the conference. Joining him was Mark Weitner, Vice President of Sales and Marketing for Parabon. The two took time during the conference to speak with HPCwire.
HPCwire: Is Frontier available via the Internet?
WEITNER: This is downloadable from the web. Obviously what we’re trying to do is encourage folks to understand our API (and) understand our software development environment. To be able to process it via our Internet network requires a subscription. Individuals essentially get charged based on the amount of power that they process across our Internet network.
When programs are submitted, they’re actually tasked out to individual computers and workstations. Right now I think we have about 15,000 registered downloads around the world. Individuals have downloaded this application which we call Pioneer to their workstations. This is a Java application. When they download this, they get the application as well as a Java virtual machine. So when they connect to the Internet. . .under the coverage, their machine will connect to our server. If it has completed tasks that have already processed while the computer is idle, it will send those tasks back to us. We register it and log it in, (and) put it in a cue for any individual customer who has submitted programs.
HPCwire: Tell us about some of Parabon’s philanthropic endeavors.
WEITNER: We support a program called “Compute Against Cancer” and up until recently, the majority of paths that were processed across our network were cancer related research projects. We’re working with the National Cancer Institute that’s doing a cancer cell study. In the beginning, the majority of folks that downloaded Pioneer did it to help support cancer. We’re actually in the process of signing up a series of non-profit cancer organizations to become affiliate partners, and we hope to have our first one signed up early next week.
HPCwire: Tell us more about Frontier.
ARMENTROUT: The real value of the (Frontier) platform is being able to scale so quickly. I can go anywhere on the planet to write and run codes across Frontier. We’re trying to bring to the desktop – with the same ease with which you would send an e-mail – the ability to drive ten or ten thousand machines. (This has) profound implications for the consumer, because this is the first time they’ve ever been able to buy (power) “by the slice”, so to speak.
HPCwire: So the goal would be to bring supercomputing power to anyone who has access to the web.
ARMENTROUT: Yes. Absolutely. And to change the way we think about buying and selling computation. Right now, you have to have an enormous up front (cost) if you want to get in the game. That limits the liquidity in the market. “I can’t get in because of this up front cost until I can really justify to my organization that $500,000 is worth it.” Well, if you bring down those barriers, if you let folks get in at small increments, suddenly you open this up to a market that has never had access before. So we’ve taken away the access barrier – you can sit at your desktop and do this – and we’ve taken down the monetary barrier – you can buy it by the drink. So we’re really trying to bring high performance computation to a market that’s never enjoyed it before.
An important point is that we don’t own hardware. We’re not in the business of owning hardware. We rent or lease or buy time from individuals and companies that wish to sell us idle time. So therin lies the power of the model. If you go back to the earliest days of computing, they would sell you time on the mainframe. This is the same model, except we don’t want to own the mainframes. There’s no reason to. The value is in intermediating this market that right now has all these “frictions” in it. So you bring those frictions down, you bring supply to the demand that exists, and you create added value for everyone involved.
People in organizations need bursts of computational power. They need a lot, they need it now, and (then) they don’t need it for a couple of days. You can see this. . .with your own desktop machine. Most of the time your machine is idle, but occasionally you sit down and fire up the word processor. . . and that CPU works like crazy for a few nanoseconds and then back down. Clock that through time (and) what you see is very “peaky” CPU usage. Your demand is very peaky, and yet what do computers deliver to us? They deliver sustained, flat power. So some of the time, it’s not meeting your demands.
HPCwire: Are you aiming for a particular kind of client?
ARMENTROUT: The platform is completely horizontal as you might imagine, but our target protocols are the biotech industry and the finance industry. Those two verticals are in desperate need of more computation. We’re actually delivering not just the platform, but applications to serve those markets as well.
HPCwire: This sounds similar to [email protected], which involves using idle computer power to search for life in outer space.
ARMENTROUT: [email protected] is an academic-based single point solution. What they created there was a screensaver to do a single calculation. To use that as a point of contrast, when you download Pioneer, Pioneer operates on your system like a screensaver. . .but it’s general purpose. One minute we might be doing stock prediction, the next minute weather prediction.
HPCwire: It depends on the particular application being run at the time.
ARMENTROUT: That’s right. Pioneer takes the next task it’s handed (and) works on it as it can, opportunistically.
HPCwire: Can clients see what kinds of computations are being performed on their machines?
ARMENTROUT: We have feature projects because of our philanthropic work, but on a moment to moment basis they do not. We give our clients anonymity. We need to do that in order to give everyone in the system security.
HPCwire: Tell us more about the benifits of Frontier.
ARMENTROUT: What it gives companies is the ability to use resources that they’ve already invested in. They’ve put, for example, 6000 machines in their building for all their staff to use, and right now those machines aren’t doing anything in their idle moments. The organization has research needs. If they put Pioneer on those machines, suddenly they have an instant supercomputing capability they didn’t have, and they didn’t have to lay out any extra hardware costs. So within an enterprise context,it gives an enormous return of investment.
Another advantage – and I think this is a significant one – is that Frontier is Java based. We’re able to run native code bases, but if you look at the productivity analysis of the industry, people using java are about twice as productive as those same individuals using C, C++ or Fortran. I was a C++ zealot my entire career until I found Java. It was like getting religion. We get a number of big wins for using Java, not the least of which is productivity. People will write and run applications and get results a lot faster (with Java). It’s a big win.