By Salvatore Salamone
October 15, 2003 | FLUSHING MEADOWS, N.Y. -- What do Andy Roddick and protein folding have in common? The U.S. Open tennis tournament.
While fans spent lulls in the action discussing Roddick’s rocketing serves, IBM life scientists were running algorithms on the same computers being used to track and report tournament results.
Taking advantage of a soon-to-be-announced IBM on-demand computational technology (code-named Project Symphony), company scientists snapped up unused processing cycles from the computers hosting the Open's Web site to test protein-folding algorithms that will eventually run on IBM's next-generation supercomputer, Blue Gene.
"We're preparing the simulations well before the Blue Gene hardware is available," says Ajay Royyuru, senior manager of IBM Research's computational biology center. "We use machines that we routinely have access to." Processing cycles are the limiting factor. "So we looked where else we could go to compute," Royyuru says.
“Where else” turned out to be the IBM servers hosting the Open's Web site, which offered a real-time scoreboard, player stats, and a means to purchase tickets and merchandise. Enough capacity was set aside to handle peak activity periods, which meant there would be many periods during the Open when server processing power would be sitting idle.
IBM researchers were able to use these computing cycles via a component of Project Symphony called the Tivoli Intelligent Orchestrator. The Orchestrator allows an IT manager to tap underutilized processing capacity of commodity servers to deliver supercomputer power to applications.
The Other Fast Server
While work on Blue Gene's hardware progresses (see Blue Gene is Cool for 2006, July 2003 Bio-IT World, page 1), researchers are developing protein-folding algorithms to run on the supercomputer, which is expected to be available in 2006.
The goal of the research is to answer the basic question: "What is it in a protein that causes it to misfold?" says Jed Pitera, a staff member at IBM's Almaden Research Center. Diseases associated with misfolding include Alzheimer's, “mad cow,” and cystic fibrosis.
By tapping the unused cycles of 127 CPUs of the U.S. Open's servers, Pitera and colleagues were able to run 150 simulations of small peptides folding in water. Each simulation had 9,000 to 10,000 atoms and showed how the peptides would fold in two to five nanoseconds.
Blue Gene will be able to perform these folding calculations on 10 times more atoms, and accelerate the simulation time by a factor of about 1,000. That translates into running a simulation for medium-sized proteins of 50,000 to 100,000 atoms in microseconds.
But to get to the point where the Blue Gene hardware can be efficiently used for calculations, the algorithms must be fine-tuned. And that's exactly what the Tivoli Intelligent Orchestrator is allowing the IBM researchers to do. "This lets us develop our algorithms and test our software before Blue Gene is available," says Bill Swope, also a staff member at the Almaden Research Center. "We can make effective use of Blue Gene as soon as it's available."
The IBM Tivoli Intelligent Orchestrator, scheduled to be introduced by now, will have a starting price of $20,000.