Temporary supercomputer

CISS brings distributed computing power to Canadian researchers.

Nov 12, 2002
Nicole Johnston(nicolejohnston@yahoo.com)

By linking together 1,360 computers, a team of Canadian scientists created a supercomputer that ran for a single day. The 24-hour distributed computing project, which last week formed the world's fifth largest supercomputer while it lasted, tackled a computational chemistry problem that otherwise would have taken more than three years to complete.

The Canadian Internetworked Scientific Supercomputer (CISS) was organized by computer scientists Paul Lu and Jonathan Schaeffer at the University of Alberta, Edmonton in collaboration with the University of Calgary. Chemist Wolfgang Jaeger, also at the University of Alberta, supplied the chemistry problem that put the supercomputer to the test. "On November 4, we accomplished 3.5 years worth of computing in one day," said Lu, who created the software to link computers at 20 sites across Canada with his graduate students Chris Pinchak and Mark Goldenberg.

The project was supported by C3.ca, a not-for-profit organization devoted to providing high-performance computing to Canadian scientists. While the Canadian government has invested more than C$160 million (US$102 million) in computer resources for researchers, large-scale computing has not been a priority. The CISS project is the first time the distributed-computing concept has been tried across Canada.

The study involved modeling the interactions of two molecules in 20,000 different positions. A software program called MOLPRO defined the problem, while the supercomputing power enabled mapping the position of the two molecules relative to each other in 3D space. Each independent position of the molecules represented an individual problem. Calculating the corresponding bond energies for each position enabled Jaeger to map out energy curves for the interactions.

The CISS team's goal, according to Lu, is to designate time each month to provide Canada with computational research days, or "CISS days," that will afford Canadian scientists an opportunity to tackle research problems they otherwise wouldn't pursue for lack of computer processing power. Participating investigators would be selected by a review process to select projects best suited for the program.

Use of the CISS supercomputer won't be limited to computational chemistry problems. The team hopes researchers will use the technique to tackle physics and biological problems such as genomics and proteomics, as well. A new project hasn't been lined up just yet, but Lu would like to have a new problem up and running over the holiday season, which would provide researchers with two weeks of supercomputing time. "People like myself hate to see a computer go idle," he said, adding that he would like to see a physics or biology problem be the next one up for study.

"Innovation is key to [Canada's] future and supercomputing is part of that," said Lu, who believes that Canada needs stronger supercomputing power to compete on a global scale.

The power of distributed-computing, an increasingly popular concept among scientists lacking access to a single supercomputer, has been demonstrated by an ongoing project at Stanford University, recently described in Nature (420:102-106). "Folding@home," uses tens of thousands of individual computers around the world to simulate protein folding, a phenomenon that occurs in millionths of a second in vivo, and requires massive computing power to reproduce in silico. The project has already reduced the time needed to simulate a single protein's folding from decades on a single computer to just two years on the folding@home network.

Led by Vijay Pande, the Stanford team modeled folding@home after SETI@home, which sifts through radio signals seeking signs of intelligent extraterrestrial life. Both "at home" projects use software that runs invisibly on individual computers, exploiting idle processor power. "We now get over 70,000 computers returning work everyday," Pande told The Scientist, "with 10,000 more computers joining each week." He hopes that number will reach between 100,000 to 200,000 computers, adding that the processing power they've already harnessed is greater than all of the supercomputers of the National Science Foundation combined.

Editor's note: This article erroneously describes CISS as the world's fifth largest supercomputer. It is among the world's top 100, according to Paul Lu. The Scientist regrets this error.