Argonne Plans Double Dose of Computing

announcements

Argonne National Laboratory, which serves as a lending library for scientific computing, is doubling down on supercomputers, installing two of the most advanced units International Business Machines makes and linking them together to work as one.

In a deal to be announced Thursday, Argonne will get a computer from IBM that performs 445 trillion calculations per second, the 445 teraflop Blue Gene/P system. Put in human terms, every one of the 6 billion people on Earth would need to perform 70,000 calculations a second to match this supercomputer.

Researchers at companies like Pratt & Whitney and Procter & Gamble, as well as major universities, use Argonne's computing power to solve problems that are inaccessible to ordinary computers. For example, P&G researchers use computer time to simulate the molecular basis of bubble formation to develop better detergents and also improved fire-control chemicals and environmentally friendly consumer products.

Argonne is already installing a new Blue Gene/P that is slower than the 445 teraflop model due for installation next year. When the two are combined, they will operate at 556 teraflops. The lab also operates an older Blue Gene/L model that will continue to run separately at 5.7 teraflops.

"By the time this project is complete, Argonne will be home to one of our country's pre-eminent computing facilities," said Rick Stevens, associate laboratory director for computing and life sciences.

Indeed, if Argonne's combined Blue Gene/P machines were running today, they would easily be the world's fastest. Prices for such a computer range from $50 million to $150 million, depending on configuration, an IBM spokesman said. Specific terms of Argonne's deal weren't disclosed.

While some scientists will use the computers for their research, most of the new computing capacity will be used by labs across the country. High-speed connections enable researchers working hundreds of miles from Argonne to run programs as if they were next door.

Jet-engine science

Pratt & Whitney researchers have used 750,000 hours' worth of computer time this year to simulate conditions inside a jet engine combustor where fuel and oxygen combine. Their goal is to reduce jet engine emissions by 55 percent.

"Modern combustors couldn't be designed without their computer tools," said Peter Bradley, a Pratt & Whitney computer scientist. Argonne not only provides computer time to the company, but also supplies advice and expertise to help use the supercomputer for basic research, he said.

"We don't use the supercomputer to design engines," Bradley said. "We use it to study the science and physics that provides us with tools we use in designing engines."

The U.S. Department of Energy, which funds Argonne and oversees the program that doles out computer time from Argonne and other national labs, at first focused on academic research, but recently has expanded to include industrial scientists, said Herb Schultz, IBM's supercomputer division marketing manager.

"We like that because it exposes the systems to more applications," said Schultz. "It shows how much more science can be done. A lot of times, a supercomputer comes on the market and people think it's just academic and esoteric. We want to get more people using them."

Software development

Part of IBM's deal with Argonne includes a collaboration to develop more open-source software for Blue Gene machines to expand the applications available. Argonne computer scientists will also provide feedback to IBM to help in designing future machines.

"Very large machines have unique challenges in making them operate efficiently," said Ray Bair, director of the Argonne Leadership Computing Facility that houses the Blue Genes.

Most work done using Blue Gene machines are simulations intended to provide insights into how fundamental processes such as formation of soap bubbles or combustion of jet fuel works, Bair said. "The vast majority of very large problems are in the domain that requires a supercomputer," he said.

IBM and Google agreed last month to work with a group of university scientists to expand "cloud computing," where large numbers of server computers will handle tasks presented by millions of users. This need grows out of the popularity of Facebook, MySpace and other social computing platforms.

"Programmers have usually been taught to write for a single computer or a few," said Dennis Quan, chief technology officer of IBM's high performance computing and software group. "They're not taught to write for tens of thousands of machines. But levels of parallelism and complexity are advancing to where in a few short years, this will be very mainstream."