AUTOMOTIVE

High-resolution numerical weather model simulations covering large areas of land and ocean are an important part of the research activities at CIMSS. Their broad focus is improving weather forecasting accuracy while their specific focus currently is providing observation capabilities to monitor the evolution of weather and environmental changes that might occur within a short time frame. More accurate observation capabilities can impact planning of a variety of space-related missions and many commercial activities. The models, software, and algorithms CIMSS is now developing using the SGI Altix system will apply to not only geostationary hyperspectral measurements, but also other types of similar remote-sensing measurements. "Running such numerical weather models requires a large amount of memory and is heavily dependent upon data transfer rates between processors," said Dr. Allen Huang, principal investigator and senior scientist at CIMSS, Space Science & Engineering Center at University of Wisconsin-Madison. "SGI Altix, with its global shared-memory design, was the best choice for these requirements and complements our existing hardware. This computing capability will allow CIMSS to support federal government research efforts and maintain a leading role in the field of atmospheric research." Funded by a U.S. Navy Defense University Research Instrumentation Program (DURIP) grant, CIMSS purchased the SGI Altix 3700 system with 24 Intel Itanium 2 processors and 192 gigabytes of memory. The Altix system at CIMSS currently runs both Mesoscale Model version 5 (MM5) and the Weather Research and Forecasting model (WRF). WRF is a new weather model that is rapidly being adopted by the weather and atmospheric research community. While CIMSS has used cluster solutions in the past, the high-performance computer SGI Altix was chosen over these clusters because it is more capable for both MM5 and WRF modeling efforts. "Dr. Huang asked us to find the right tool - and an affordable tool -- to get this particular job done," explained Erik Olson, Assistant Researcher at CIMSS. "We looked at clusters from a variety of vendors, did some experiments and performed some benchmarks. We concluded that Altix was something we could afford and that it would handily solve the specific problem of running large weather models. We run these models over a large area at a fine resolution, which means that it uses up a lot of memory and it needs a faster interconnect to communicate between processors quickly to do the job in a reasonable amount of time." The question of how long it would take to be up and running was also of paramount importance, Olson noted. "A main factor for our choice of the Altix system had to do with calendar time: how long would it take us to be able to run our MM5 model on Altix," continued Olson. "On a cluster solution, our estimation of how long it would to take to get the MM5 model running well -- at the resolutions that we want -- would be on the order of six months to a year. To get it up and running on Altix, we figured it would take two to three weeks, and that's about the length of time it actually took. That was a big factor in our decision, as well as the support that we got from the people at SGI who had solid experience with MM5." With the new SGI Altix system, Huang and Olson estimate a 2.5 times increase in model run speed and a 12 times increase in model domain size capacity. "SGI's solid history of delivering high performance computing in weather research is now enabling CIMSS to achieve their goal to run high spatial resolution models to support future design of next-generation satellite instruments, and to meet the requirements of NOAA, NASA, or the Navy," said Dr. Ilene Carpenter, SGI Weather and Climate Business Development Manager. "We welcome these distinguished scientists to the rapidly growing family of Altix supercomputer users in the meteorological field, and throughout the scientific research community." The SGI Altix system was sold to UW-Madison by the Eagan, Minn., satellite office of James River Technical, Inc., a SGI value-added reseller headquartered in Glen Allen, VA, focused on the higher education marketplace. "James River Technical has a long-term relationship with various scientific departments at UW-Madison. For CIMSS, we worked closely with SGI applications engineer Dr. Gerardo Cisneros to address their main concern, which was how much memory they needed to get the job done," said Stefanie Rogers, account executive for James River Technical, Inc. "The large memory footprint, at 192 gigs, and the fact that SGI Altix can scale to meet future demands, plus the benchmarking assistance Dr. Cisneros gave, coupled with compelling pricing and their confidence in our long-term support -- and almost two years of serious discussion -- all led to Dr. Huang and technical lead Ray Garcia's decision to purchase this great leap in SGI high performance computing power for their important research."