PACIFIC NORTHWEST NATIONAL LAB: Scientists to Test Disaster Scenarios With Supercomputer

June 2, 2003 (Computerworld) As powerful computers are harnessed together to create even more powerful supercomputers, scientists continue to find new ways to exploit this strength by conducting experiments they could only dream about in the past.
At the Pacific Northwest National Laboratory in Richland, Wash., a massive Linux supercomputer that will be completed this month will be used to help model some of the most challenging and dangerous environmental cleanup scenarios in the world.
Scott Studham, technical lead of molecular science computing facility operations at the laboratory, which is part of the U.S. Department of Energy, says the goal was to create a supercomputer that would allow scientists to model disasters that wouldn't be feasible to test in real life.
One such disaster might be an accidental spill of radioactive uranium, Studham says. Using the new supercomputer, scientists will be able to model a wide variety of substances, including enzymes, in their search for materials that could help in the cleanup.
"That would be an example [of testing] that's too dangerous or expensive to do in a real lab," Studham says. With a supercomputer, such high-risk experiments are safe because the testing is theoretical.
Using a powerful custom chemistry modeling program called NWChem, the lab's researchers can peruse particles to search for better ways of cleaning up environmental hazards. "We can solve problems here that can't be solved anywhere else," Studham says.
Other programs can model different types of environmental disasters, including oil spills. The core research areas for the machine are chemistry, subsurface modeling, biology modeling and regional climate experimentation.
When completed, the machine will include 1,900 Intel 1.5-GHz Itanium 2 processors in 950 nodes, which is believed to be the largest Linux cluster in the world. The supercomputer, which is designed to perform 11.7 trillion floating-point operations per second, will cost $24.5 million.
Two problems had to be sorted out as the system was being built. One was figuring out how to make 256TB of total hard-disk space across the 950 nodes appear as one seamless storage area. This was solved by using and tweaking the Lustre global high-performance file system from Cluster File Systems Inc. in Mountain View, Calif. Also difficult was creating an air-conditioning system powerful enough to cool the massive, hot-running machine.
Barton Miller, a professor in the computer sciences department at the University of Wisconsin - Madison, says this kind of supercomputer is just what scientists need to test the stability of stored nuclear bombs, for example, or drug interactions in genetics research, he says.
"They can't test these things out in a meaningful way [in a lab], so you do it using simulation," Miller says. "It's changed the way people do research."