Server Clusters Divvy Up Research

Clusters of servers offer computational economies.

Until recently, the most computationally intensive research in the life sciences field was done by major institutions that had the money to buy expensive, high-end systems.

However, smaller companies in the space now are able to harness the computational power that had been primarily the domain of their larger brethren. That is due to the advent of clusters of smaller, cheaper commodity servers equipped with increasingly powerful Intel Corp. and Advanced Micro Devices Inc. processors and running the Linux operating system.

In addition, server and PC vendors such as IBM and Gateway Inc. offer preconfigured or hosted clusters that enable companies to get the computing power they need, when they need it.

Moores Law has led the drive for cheaper, more powerful processors, resulting in systems that offer higher performance at a low cost, said IDC analyst Mike Swenson. Throw in the fact that applications such as BLAST (basic local alignment search tool) for genomic sequencing can now run on Linux, and smaller companies find they have high-performance systems that can run at 70 or 80 percent of the computational power of the larger SMP (symmetric multiprocessing) servers for substantially less money.

"People are willing to make that trade-off," said Swenson, in Minneapolis. "People might say, So I might not be able to do this as quickly as with the larger SMP systems, but what theyre doing is not so time-intensive that they cant wait a day or two for a run to get done."

A 2002 study published by the U.S. Department of Commerce indicated that 58 percent of biotechnology businesses surveyed had fewer than 50 employees, and 53 percent said research costs were a major barrier to their success. The study also illustrated the growing role IT has in the field: The fastest-growing occupation in the industry was computer specialists focused on research and development.

It can cost as much as $1.7 billion to bring a drug to the market, so biotechnology and pharmaceutical companies welcome any way of reducing the time—and cost—to do that. Much of the work to reduce the time to get through the initial target identification and verification phases of the process is being done on industry-standard systems, such as IBMs Intel-based xSeries and BladeCenter servers.

Apart from the systems it sells, IBM, of Armonk, N.Y., is also offering life sciences companies cheap computing power through a hosted grid.

Electro-Optical Sciences Inc. uses photonic computers for early detection of cancers and other tissue diseases. The Irvington, N.Y., company does this by searching through thousands of mathematical algorithms and observational data to determine if a particular lesion is cancerous or benign. Recently, when EOS has needed extra computational power, its engineers have turned to IBMs hosted grid, said Marek Elbaum, chief science and technology officer.

"The attraction to us is the trade-off between the cost and the need," Elbaum said. "There is always a trade-off between using the IBM grid or buying hundreds of computers and doing it in-house. Its an issue of economy." EOS has the engineers capable of running a computer grid, but IBMs hosted grid provides a cheaper and faster alternative to setting up something in-house, Elbaum said.

For its part, Gateway is using its own PCs to power newer life sciences applications. The Poway, Calif., company in December 2002 rolled out a computing grid fueled by middleware from United Devices Inc. that takes advantage of the untapped computing power of up to 7,000 systems on display in its 180 retail stores. Life sciences organizations ranging from Inpharmatica Ltd. to the American Diabetes Association have put their applications on to the grid to get the kind of computing power they had limited access to before.

Gateway in the spring will upgrade its Gateway Processing on Demand grid to Version 4.1, which will include network optimization enhancements. There also will be hardware improvements enabling users to run Linux workloads at night and then reboot the systems into Windows in the morning, a move particularly important to life sciences companies, which tend to work more with Linux.