Life sciences company Bristol-Myers Squibb Co. may be able to increase its computing power by a factor of 100 by moving to a desktop grid computing model.
Supplied by Toronto-based Platform Computing Inc.
Computing Inc., the solution will take advantage of unused desktop cycles and contribute them to Bristol-Myers research efforts.
“”We view grid as an innovative way to take advantage of the resources that we currently have so we can improve are research capabilities,”” said Richard Vissa, executive director of technologies, located in Bristol-Myers’ Hopewell, N.J., office.
“”Where we really need an order of magnitude — more compute power — is for our early drug discovery initiatives, which tend to be computationally resource-intensive. We’ll hopefully have better candidates that come out of that research.””
Bristol-Myers has been in trials with the grid for about a year. Results indicate a 100 per cent increase in computing power, but Vissa said a 1,000 per cent increase is certainly possible. A $1 million supercomputer, from the likes of Sun Microsystems or IBM for example, would only yield a five- or tenfold increase.
The grid lends itself to tasks that can be broken down into smaller pieces, said Platform CTO Songnian Zhou. “”You have management software sitting on a machine that takes this request to do this big application and then sends out pieces of work to the PCs, gets the results back and assembles them together,”” he said. “”The user never knew it was run on a desktop instead of a big server.””
The application only commandeers the desktop to the extent that it’s available. “”During the day it can happen, during the night it can happen more,”” explained Zhou. “”Whenever and wherever resources are available, you soak it up.””
However, the desktop grid isn’t universally applicable, said Vissa. It’s best suited to a type of research that can be broken down into component tasks — including some kinds of chemistry and bioinformatics research, as well as virtual screening of compounds and drug targets. Tasks that won’t be moving to the grid include those that require half a gigabyte of memory or contain too much data to be stored on a desktop hard drive.
The development side of the house (moving research into clinical trials), for example, will remain on Bristol-Myers’ existing supercomputers and Linux servers for the foreseeable future, said Vissa.
Bristol-Myers will move beyond beta testing the grid to actual research in the next two to three months. Several thousand desktops in the company’s five campuses in the Northeastern U.S. will be used to form the grid. That number may increase to five or ten thousand in a year’s time.
The grid will be limited to the main campuses. “”It’s harder to justify increasing the bandwidth to accommodate the grid (in smaller offices),”” said Vissa, but he didn’t rule out run geography-specific grids in other parts of the U.S. or in other countries.
Mike Swenson, an IDC analyst based in Minneapolis, said the life sciences industry is a prime candidate for this type of grid computing. “”They work quite well and you can see remarkable speed-ups over traditional servers, even some supercomputers for certain kinds of problems,”” he said. But most companies will view a grid as complimentary to their existing technology rather than replacing it, he added.
Desktop grid computing has been successfully used in the SETI project (Search for Extraterrestrial Intelligence), said Swenson. According to Zhou, there is interest in the technology from financial, insurance and mechanical design companies.