ENGINEERING

To achieve faster responses to wildland fires and better forecasting for prescribed agricultural burns, the Desert Research Institute, in Reno, Nev., will soon be delivering twice-daily forecasts of fire weather and smoke behavior to meteorologists and firefighters in California and Nevada via the Internet. To compute the mountains of data needed to deliver meteorological products from a mesoscale meteorology model, the Desert Research Institute (DRI), a long-time Silicon Graphics customer, has added a 32-processor SGI Altix 3000 supercomputer and 1.7 TB SGI InfiniteStorage TP9100 Fibre Channel storage array to its arsenal of SGI high-performance compute (HPC) systems. Mesoscale modeling provides meteorologists a level of detail that scales down to individual cloud cells with dimensions of 1-20 km and life spans less than one hour (mesoscale-gamma). These forecasting models help determine the profound effects from subtle variations such as the wind-funneling effect of a terrain and the temperature and humidity influence in the formation and distribution of fog. As a partner with the newly formed California and Nevada Smoke and Air Committee, or CANSAC, DRI purchased in March the now fully implemented 64-bit Linux(R) OS-based SGI(R) Altix(R) 3000 and SGI InfiniteStorage system for this specific CANSAC project, part of the Climate, Ecosystem and Fire Applications (CEFA) program (www.cefa.dri.edu) at DRI. Six CEFA scientists and graduate students at DRI are assigned the mission of providing fire weather, fire danger and fire behavior products, as well as smoke movement and air quality information, and research studies to support the product development. The Altix(R) system runs predictive models that help meteorologists, scientists, managers and fire specialists visually predict fire weather conditions for up to 72 hours, such as how a fire might burn under certain conditions, how the smoke will disperse and who it could impact, such as local hospitals, schools and others in a community. CANSAC comprises approximately a dozen federal, state, county and local agencies within the two states. CANSAC members include the U.S.D.A. Forest Service, Bureau of Land Management, National Park Service, U.S. Fish and Wildlife and California Division of Forestry and Fire Protection. The California Air Resources Board, and San Joaquin Valley Air, with other agencies such as the County of Los Angeles Fire Department are expected to come on board soon. CANSAC is also a member of the U.S.D.A. Forest Service National Fire Plan Fire Consortia for Advanced Modeling of Meteorology and Smoke. CANSAC members will be evaluating and assessing DRI's regional forecasting products this year -- in both test runs and possibly in real-life wildland fire incidents -- and fully implementing widespread use of the products next year. The SGI Altix 3000 supercomputer, based on Intel(R) Itanium(R) processors, and SGI InfiniteStorage system use the Pennsylvania State University/National Center for Atmospheric Research (NCAR) mesoscale model version 5 (MM5) which is designed specifically for regional forecasting, to simulate and predict mesoscale atmospheric circulation, calculating weather every 108 seconds for three spatial domain areas. These domains are three different-sized grids, with varying ranges of grid points. The first, and most highly detailed, has a grid point every 4 kilometers over all of California and Nevada, and a small coastal part of the Pacific Ocean. The second spatial domain area contains grid points at every 12 kilometers, and covers roughly three times the area of the first. The third domain has grid points every 36 kilometers, and covers three times the size of the 12-kilometer grid-point domain, including a major portion of the eastern Pacific Ocean, the entire western U.S. and parts of southwestern Canada and northwestern Mexico. In addition to calculations within all these grid points in the three spatial domains, the SGI Altix is also required to make calculations at 33 different levels in the atmosphere, from the ocean or ground surface up to approximately 60,000 to 70,000 feet. "And this one machine, SGI Altix 3000, is doing all of this in remarkably good time," said Dr. Tim Brown, director of DRI's Climate, Ecosystem and Fire Applications program (CEFA). "We're actually able to achieve all this -- the 4, 12, and 36 kilometer readings and the 33 levels of atmospheres -- in two hours of real clock time, with additional time taken by pre- and post- processing. Our minimum run is 7GB of data and we're doing it twice a day. But we're thinking about a set of runs that would actually be about 45 gigs a day. "When we first began developing the project, we knew that we were going to need a lot of crunching power so I initially proposed a 128-processor PC cluster," added Dr. Brown. "When we began testing this system, we ran the models on a couple of different SGI machines, including a 40-processor Altix we already had within DRI. We also tested on PC clusters and although the timing results were about the same, I preferred SGI because we know the MM5 model runs really well on it. We also know we don't have to spend a lot of time tweaking or doing adjustments on the Altix system, which I consider a stable and reliable system. We feel we have a better sense of scalability on the SGI system than we would have on a PC cluster. I like the fact that on Altix, because of the SGI NUMAflex global shared memory architecture, we can run the computer with all of the processors sharing the job or we can distribute the job to individual processors." At DRI, the 40-processor SGI Altix 3000 is part of a National Science Foundation EPSCoR project called ACES (Advanced Computing in Environmental Science). Other faculty at DRI run meteorological models on a 16-processor, SGI(R) IRIX(R) OS-based SGI(R) Origin(R) 2000. The first server system for Dr. Brown's group was an SGI(R) Origin(R) 200; it's currently doing some real-time processing, and will be handling some post-processing from the output of the new SGI Altix 3000 system. Currently, using the new SGI Altix 3000 and the MM5 model, DRI produces two-dimensional maps using a graphics package called RIP from the National Center for Atmospheric Research, specifically designed for MM5 meteorological model output. Data is currently visualized on a variety of desktop systems and workstations. Eventually the output will be linked to Web-based software that will allow end-users, such as incident meteorologists and fire behavior analysts, to interactively add layers on top of fires for real-time field calculations and decision-making. Future plans at DRI include adding terabytes of storage to its SGI InfiniteStorage TP9100 as well as a tape-based archive. DRI is currently assessing an SGI(R) Reality Center(R) as the graphics delivery system for a planned virtual reality six-wall visualization facility for research and development. "Making as precise and accurate forecasts for agricultural burning and wildland fire is a very critical issue in the western U.S. and SGI is pleased to have our much-lauded SGI Altix 3000 chosen to power this important Desert Research Institute and CANSAC initiative," said Dr. Jill Matzke, manager of environment marketing at SGI. "The SGI Altix family offers unparalleled performance in the HPC market, and we can ask for no greater validation of that statement than when our nation's top scientists and researchers continue to choose SGI products to achieve their goals."