FEATURES & COMMENTARY
Morrisville, N.C. — From angular, flat-panel displays to Lifesaver colors, Christina Dyrness reports that the latest computer equipment is the best designed – at least that’s the idea being promoted these days by the people who sell it. But don’t try telling that to the guys at Computational Engineering International.
In an air-conditioned room in the middle of their Airport Boulevard offices are two machines with supercomputing power. One is a mainframe with a sleek, Grape Kool Aid-colored case, operating with a low hum. The other is a silent, gray metal rack of 16 off-white server boxes wired together with unadorned cords. Guess which one is the high-performance computer of the future? Hint: It ain’t purple.
“It’s cheaper than that one, and it can run the same applications,” CEI Vice President Anders Grimsrud says as he points from the $400,000 Onyx 2, a supercomputer from Silicon Graphics, to the $130,000 rack of generic-brand servers.
For Grimsrud, as for computer scientists all over the world, it’s a simple case of cost efficiency. Consumer-grade hardware has matured to such a level that a cluster of ordinary servers working together can run virtually the same applications that a huge mainframe computer can. CEI’s setup, called a Beowulf computer cluster, was installed to support a $1.4 million contract awarded to CEI last month. The grant pays CEI to research the use of clustered computers to visualize large-scale models for the Department of Energy’s Accelerated Strategic Computing Initiative, or ASCI.
The Beowulf cluster gets its name from the sixth-century Scandinavian hero who was famed in verse for taking down a monster named Grendel. The computer version, which was designed to slay the monster mainframes, was invented in 1994 by Don Becker while he was at NASA’s Goddard Space Flight Center. The idea was to use the Linux operating system with regular hardware linked to create a high-performance computer cluster to handle large tasks, often number crunching.
CEI, which was spun off from Cray Research, a Minnesota-based supercomputer maker, in 1994, makes software to analyze computer-aided engineering and scientific data. That software – called EnSight – takes the reams of data created by Beowulf’s number crunching and plots it in a visual way.
The company’s only customer in North Carolina is R.J. Reynolds, which uses the software to help design its cigarette production facilities. But another of CEI’s regular customers is the Department of Energy, which started ASCI as a means of replacing the actual testing of nuclear weapons with computer simulations.
Grimsrud and other developers are working to modify ASCI’s software to work in a cluster environment dividing computer tasks to run on separate machines rather than one, big mainframe. “This is new territory for everyone,” said Kent Misegades, CEI’s president. Red Hat, which makes the Linux version that CEI is running on its Beowulf cluster, has long supported clustering technology.
“It’s a specialized market where Linux has a huge amount of customer adoption,” said Paul McNamara, vice president of products and platforms at Red Hat. “The total number of clusters worldwide is relatively small, but there are a lot of professional services that go around the cluster.”
Mike Grosvenor, an industry analyst with Giga Information Group, said: “They’ve been saying for years that clustered computing would replace mainframes.” And will it? “It sort of becomes self-fulfilling prophecy,” Grosvenor said. “[Clustering] steals mind share from other new developments.” Grosvenor said clustering setups like Beowulf are especially popular with start-ups and in the educational arena.
In the end, the potential savings will push clustering into the mainstream, Misegades predicted. “Traditional users of big computers are fairly conservative in buying hardware,” Misegades said. “[Clustering] is a little bit new for them, but the cost will come into play.”