Compaq has been making friends among the public-sector genomics gang ever since it began sharing a multimillion-dollar high-performance computing cluster with certain academic labs struggling with massive computational biology problems.
The so-called BioCluster, housed at Compaq’s Enterprise System Lab in Littleton, Mass., runs on the Tru64 operating system and, to distribute tasks, relies on Load Sharing Facility, which Platform Computing donated to the cause.
Since it was set up in May, 2000, more than one million jobs have been run on BioCluster. Various users, located in France, St. Louis, and Boston, access the cluster through the Web.
MIT’s Whitehead Institute has been its biggest customer, employing BioCluster to run RepeatMasker and BLAST queries on successive releases of the draft human sequence. In another MIT lab, GenScan author Chris Burge has used the cluster to run BLAST and perform genome annotation. Researchers at Washington University, including Warren Gish, who donated a copy of WU-Blast to the project, also use the system for various gene expression and annotation work. And the cluster has been made available to David Haussler’s UC Santa Cruz annotation team and researchers at Genoscope in France.
Why give away access to a $3 million compute farm? Compaq senior technical expert Ray Hookway, who is the BioCluster technical lead, says the system “is another public demonstration of the power of Compaq in this market.”
Adds Daniel Joy, Compaq’s business development manager for biological and chemical sciences, “It’s been useful to Compaq. We help nonprofit research and we get to understand what kinds of problems they’re working on.” Joy says Compaq claims no rights to the data run on the machines, but does benefit from observing how the system handles problems the academic labs feed it.
Although initial plans called for the system to be in place for six months, Hookway says BioCluster became so popular that its term was extended another six. Will Whitehead and others be left high and dry when this term ends? “What actually happens depends on what we see as the utility for the cluster,” Hookway says.
BioCluster consists of 27 ES40 four-processor nodes, each of which sells for more than $100,000. Of those, 25 contain four gigabytes of memory and 54 gigabytes of disk space; one has 16 gigabytes of memory and 54 gigabytes of disk space; and the ES40 file server has four gigabytes of memory and one terabyte of disk space. A DS10 login node holds one gigabyte of memory and 18 gigabytes of disk space.