BETHESDA, Md.--The Computational Biophysics Section of the National Heart, Lung, and Blood Institute (NHLBI), one of the US National Institutes of Health, recently installed a 3Com CoreBuilder 9000 enterprise switch that is expected to expedite medical research, including genomic data exploration that will be undertaken at the institute using a networked supercomputer-equivalent that NHLBI has been building since 1997.
Eric Billings, a scientist in the institute's Computational Biophysics Section, said the CoreBuilder 9000, which supports gigabit ethernet connectivity, will allow the institute to "cost-effectively multiply the system's power, enabling us to conduct critical research in areas we couldn't before."
Billings told BioInform that one such area would be genomics. He said the decision to build a Beowulf-class LoBoS--short for Lots of Boxes on Shelves--system instead of buying a supercomputer was made with the goal of eventually performing genomic research at NHLBI. "The desktop PC has come to eclipse the major supercomputer vendors' offerings for certain applications, including genomic databases," Billings argued. He explained that "the beauty of the LoBoS system" can be attributed to geometric-scale increases in the performance capability of desktop PCs. "Every 18 months since the 1940s they have doubled in performance," he noted.
Billings said the institute has achieved significant improvement in the amount of computation it can do per dollar spent since it began building the LoBoS. "We get 10 times more work done than we would if we went to one of the big iron vendors like Cray or IBM," he said, "so we see this as a big win."
Billings said the CoreBuilder switch will enhance the speed of the LoBoS system dramatically. "At first we had 64 dual-processor machines linked in a ring topology, but as we add more machines the ring becomes a limitation for the systems we can study," he explained. Now, the 3Com switch will anchor 100 new Pentium IIs, each with two 450-MHz processors, and 3Com's Fast EtherLink XL PCI network interface cards.
Supercomputing with PCs is particularly well suited to distributed database analysis, according to Billings. "You take your large genome database and distribute it across each of the disks within the supercomputer and effectively increase your search capability by the number of processors you put on the job. This is an especially effective way to use this kind of supercomputing," he told BioInform.
The high port density and huge back plane on the CoreBuilder ensures the performance needed for the most demanding simulations, Billings said, and the switch's modular architecture offers the bandwidth necessary "as we add still more processors and upgrade communications to gigabit ethernet speeds." Ultimately, he concluded, the system could have more than 1,000 processors.
NHLBI has yet to put the system to the test with a genomic database. Billings noted that the institute mainly does "macromolecular simulations, quantum chemistry."
"We do simulations of DNA and protein DNA interactions. What we've been doing is a little bit downstream from the genomic database," he commented, "but definitely there are plans" to work with sequence data.
In fact, Billings called exploring genomic data "a big part" of the future of the LoBoS project, and one that will probably occur in the next year.