SGI Wins at East Carolina

BOSTON -- A powerful new computing system from SGI (NASDAQ:SGIC) has shortened the time required to run complex calculations for biologists, chemists and engineers at East Carolina University (ECU) - in some cases, from several days to just a few hours.

Purchased with a grant from the National Science Foundation (NSF), the newSGI(R) Altix(R) 4700 system is helping ECU researchers analyze computational models that were growing too large and complex for the university's existing resources. The resulting performance gains also have convinced evolutionary biologists at ECU that the shared-memory Altix platform can be a superior engine for the Bayesian inference analyses, routinely run on distributed clusters.

"When I learned that ECU was applying to the NSF for a large shared-memory system instead of a distributed cluster, you could say I was disappointed,"said Dr. Jason Bond, an ECU professor of biology. "Everyone I talked to in my field disagreed with our plan to invest in a shared-memory machine."

But after seeing complex Bayesian and dynamic homology computations soar on the new 128-core SGI Altix 4700 system, Bond was sold on the advantages of a large-node, shared-memory solution. The Altix platform allows access to all data in the system's memory directly and efficiently, without having to move data through I/O or networking bottlenecks - a common problem with traditional distributed clusters.

"We're now running MrBayes analyses in a few hours that previously would take days to run on our dual-core processor machine here in the lab," said Bond, who also leverages the Altix system to run POY, a complex sequence alignment and tree search application, and other algorithms as part of ongoing NSF evolutionary studies of spiders and millipedes.With the SGI Altix solution improving ECU's research productivity "many times over," continued Bond, "We now have the time to run lots of permutations to test our inferences. In the past, we'd run two permutations and wait days to get the results. We'd send the results for peer review, and they'd recommend dropping that analysis from the paper altogether. It just was not productive." SGI