IBM upgrades NWS supercomputer

In a move that will significantly improve the accuracy of weather forecasts, the National Weather Service awarded IBM Corp. a $35.6 million contract to supply the agency with a faster supercomputer. IBM will lease the agency its massively parallel RS/6000 SP supercomputer, replacing the current Cra

In a move that will significantly improve the accuracy of weather forecasts, the National Weather Service awarded IBM Corp. a $35.6 million contract to supply the agency with a faster supercomputer.

IBM will lease the agency its massively parallel RS/6000 SP supercomputer, replacing the current Cray Research Inc. C-90 vector supercomputer used by the NWS' National Centers for Environmental Prediction (NCEP).

The new IBM supercomputer will crunch mathematical models used for weather forecasting five to 10 times faster than the current supercomputer, said Lloyd Treinish, research staff member at IBM's Thomas J. Watson Research Center, Westchester County, N.Y.

"Current forecasts can run faster, which would enable the [NWS] to make more timely assessments," Treinish said. "The [NWS] can also do more ensemble forecasting— running multiple forecasts and doing comparisons between them. Ensemble forecasting is done now, but when you talk about a machine that is five to 10 times faster, you can do more of it."

Higher-Resolution Models

The new system also will allow the NWS to run higher-resolution models to produce forecasts with more detail and accuracy than before. For example, the NWS currently uses a regional model that provides forecasts covering a 32-kilometer region. The new system will be able to zero in on regions 5 kilometers to 10 kilometers in size, said Wayman Baker, acting director of NCEP central operations.

Providing more forecasting detail will enable the NWS to predict more accurately where a hurricane will reach land, Baker said. "The National Hurricane Center would be able to reduce the length of coastline that's warned so less people have to be evacuated, so there is a clear economic advantage," he said. "We will also provide airlines with more accurate models for flight planning...which means reduced fuel expenditure."

A more powerful computer also will allow the NWS to introduce more complex models. "There is much research that goes on to improve models," Treinish said. "There are specific codes and simulations that the [NWS] uses that will continue, but they also have the ability to introduce new codes or to enhance current codes to produce better results."

The NWS has been using the current Cray system since 1994 but has outgrown it. That system uses 16 processors in parallel, but the IBM eventually will provide 2,048 processors. "We will also increase storage in terms of disk and tapes so it's a balanced system," said Bill McCracken, senior computer specialist at NCEP. "As the model resolution increases, we're processing more data points and need more storage to retain those data points." Peak performance of the IBM should reach 2.5 trillion floating-point operations per second.

In moving to the IBM system, the NWS is moving to massively parallel processing, in which processing is split among many processors and memory. The current Cray system is a vector supercomputer.

Although the vector supercomputer has been the system of choice in the weather community, using massively parallel computers to process weather applications is gaining acceptance, said Brett Berlin, president of Berlin Consulting Associates Inc., Alexandria, Va.

"NCEP has come of age in the operational side of the business equation. This award puts IBM into the game at the National Weather Service— something it was unable to have for over 30 years. It's an important message."

NEXT STORY: Data for Hire