10 Jan UW-Madison picks scalable SGI computer to simulate weather

Madison, Wis. — To model something as big as the weather, the University of Wisconsin’s Cooperative Institute for Meteorological Satellite Studies needed a bigger computer.
The institute has begun testing its weather forecast data using the Altix 3700 supercomputer offered by Silicon Graphics Inc., a company headquartered in California but with research operations centered in Wisconsin’s Chippewa Valley. The system has been operational since the first week of December and could increase execution speed by 2.5 times on a problem domain 12 times as big.
CIMSS, an offshoot of UW-Madison’s Space Science and Engineering Center, works to improve the accuracy of weather forecasts by developing high-resolution weather simulations that cover land and sea. It focuses on charting changes in weather and environment that happen in a short time, with long-term research goals to improve the accuracy of weather forecasts. CIMSS operates two systems for weather observation: the Mesoscale Model and the Weather Research and Forecasting Model.
According to Erik Olson, assistant researcher at CIMSS, the institute chose an Altix computer because of its need for more memory and data speed over the existing systems. CIMSS’ programs had been operating on several Linux clusters, working from 24 Pentium IV Xeon processors with 48 gigabytes of memory and a gigabit switch. With the upgrade, the system now runs on 24 Itanium processors and 192 gigabytes of memory.
“The setup of the numerical weather models that we run is a task well suited for large, shared-memory type of systems,” Olson said. “The Altix expands our capabilities beyond what we could do with just distributed memory clusters.”
The system was sold to CIMSS by James River Technical, a Virginia-based value-added reseller which deals in the fields of higher education and research. SGI and James River have worked with the university several times in the past, supplying systems to the chemistry, biochemistry, and engineering departments. Last year, JRT sold two 12-processor Altix systems to the biochemistry department.
“I think CIMSS was looking for a turn-key solution that could be up in running in a timely fashion, that would not require a lot of man hours for administration,” said Stefanie Rogers, account executive at JRT’s Minnesota office. Rogers worked directly with CIMSS and SGI on the project, meeting regularly with the committee that made technical recommendations to the institute.
Rogers said that one of the things that worked out best was the scalability of the Altix system. Altix is built of “bricks” that contain a variety of services, with customers adding features such as processors, memory, routers or storage individually. This system meant that CIMSS could customize the server with what they really needed, increasing the memory past cluster limitations without adding excess processors.
“The processor count, although important in terms of building a balanced system, was secondary to their need for a large memory footprint,” Rogers said.
The addition of this memory has enhanced the capabilities of both the institute’s models, allowing them to develop algorithms for processing satellite data and atmospheric properties that will lead to higher accuracy in forecasting.