January 2002

As three-year Compaq computer deal ends, model progress continues in MMM

Scientists in the Mesoscale and Microscale Meteorology Division are scrambling to fill the void left by recently departed computers, as a three-year arrangement that brought a tenfold increase in computing power to MMM draws to a close. At the same time, they’re grateful for research progress that would have been impossible without the loan of several dozen Compaq workstations and servers.

Bill Kuo (MMM) sits at one of the Layton Classroom computers. Behind him (from left to right) are Jimy Dudhia (MMM), Al Bourgeois (MMM), investors Gaylord Layton and Peter Baird, Wei Wang (MMM), and Kevin Manning (MMM). (Photo by Carlye Calvin.)

"The whole division benefited" from the Compaq deal, says MMM director Bob Gall. “This was a unique partnership—one we dream about in NCAR." Indeed, many of the innovations that took place during the three-year Compaq loan (see box) were cited by an NSF panel in its recent review of MMM.

The deal dates back to 1998, when Gaylord Layton, a Denver-based venture capitalist, worked through an intermediary company, Advanced Research Alliance, to finance 7 high-end Compaq servers and 42 workstations. The plan was for NCAR to host these machines—with periodic replacements by higher-end successors—until the fall of 2001. The appeal for Compaq was in discovering how its machines performed under the demanding requirements of high-resolution modeling.

Last November, Advanced Research Alliance reacquired many of the MMM machines. Not all of the hardware left NCAR, though. In a deal that was part donation and part purchase, the division was able to keep 36 Compaq workstations. Of these, 22 are used by scientists and 14 are installed in a Foothills Lab classroom for a variety of uses including intensive workshops on how to use the Penn State/NCAR mesoscale model, version 5 (MM5). The FL3 space has been named the Layton Classroom.

As for the larger Compaq machines, more than a dozen ES-40 servers have been retained by MMM. All of these are funded by and dedicated to specific projects. However, the workhorse GS-320, a shared-memory machine with 32 processors, is now gone. That computer accommodated daily forecasting with the MM5, development of the upcoming Weather Research and Forecasting model, and a variety of other research within the division. The two dedicated modeling tasks are moving to clusters of four-processor ES-40 machines, so it’s the loss of the open-ended capacity that MMM is feeling the most, says senior scientist Ying-Hwa “Bill” Kuo. “Some of the research originally planned can no longer be carried out,” he says.

With a recent hardware acquisition of its own, the Scientific Computing Division is stepping up to the plate to help. However, even with a doubling of its SCD allocation this year and a tripling next year, MMM’s total computing power will only be in the neighborhood of the level it reached last spring—and for continued progress, more is needed. According to Bill, “MMM will have to continue to search for solutions, including going to other supercomputing centers and finding resources to enhance our local computing facility.”

The scramble is painful but not unexpected, he adds. “We knew we were going to miss [those machines], and we do.”

NCAR provides twice-daily forecasts for Antarctica through a special polar version of the MM5. This season marks the second austral summer of the outlooks, which support field projects, polar research, and flights to and from the continent (including the emergency rescue of ailing physician Ronald Shemenski last April). Factors behind the success of the Polar MM5 include its accurate depiction of Antarctica’s rugged topography and its high resolution, which was increased to 2.1 miles (3.3 kilometers) this year in the vicinity of McMurdo Sound. Above is a sample 6-hour precipitation forecast on the large-scale grid of 56 mi (90 km). (Illustration courtesy MMM.)

 

•Bob Henson

Edited by David Hosansky, hosansky@ucar.edu
Prepared for the Web by Carlye Calvinl
Last revised: Wed Dec 12 16:05:08 MST 2001