NASA Goddard Space Flight Center is preparing to re-analyze 25 years of environmental satellite data in an effort to reach more accurate conclusions and deeper insights about changes in Earth’s climate since the 1970s.
The key to this project is a powerful new supercomputer the Greenbelt, Md.-based NASA field center is installing this summer. Goddard is paying $6 million to lease for three years one of the most powerful supercomputers on the market, a Silicon Graphics Altix 3000.
Goddard’s new system has 1,024 state-of-the-art Intel Itanium 2 processors , delivering five times more computing power than the three-year-old supercomputer it is replacing. The supercomputer also has 2 terabytes of global shared memory — think 1.5 million floppy disks — and the ability to store and manage another 200 terabytes of data. Once installed, the system will occupy 23 racks, each roughly the size of a refrigerator.
Phil Webster, the head of Goddard’s High Performance Computing group, said the facility chosen to house all that hardware must first undergo an upgrade of its own. “These new computers run fairly hot. You need a lot of cooling for it and more electrical power, so we are preparing the floor space for it right now,” he said.
While the new system is definitely the most powerful supercomputer at Goddard, NASA Ames Research Center near Mountain View, Calif. — the U.S. space agency’s epicenter of high-performance computing — actually has one much more powerful. Silicon Graphics delivered to Ames last year an Altix 3000 system with more than 10,000 Intel Itanium 2 processors on board. Ames has used the system — the most powerful Silicon Graphics has built — to model, among other things, design changes made to the space shuttle since the Columbia disaster in 2003.
Once Goddard’s new Altix 3000 supercomputer is on line and ready to go in late summer or early fall, Webster said he expects no shortage of demand for use of the system.
“There is a tremendous amount of Earth science research done here at Goddard, both in terms of weather and climate research, and we will support a number of projects,” Webster said. “We should be able to keep the machine full all three years it is here.”
First up to use the big machine is a project NASA calls MERRA, or Modern Era Retrospective Analysis for Research and Applications.
Michele Rienecker, head of the Global Modeling and Assimilation Office at Goddard, said the MERRA team will spend 18 months on the Altix 3000 system re-analyzing environmental satellite data collected since 1979 largely by U.S. spacecraft.
Rienecker said the re-analysis could lead to a better understanding of climate variability, particularly as it relates to the water cycle, and improved short-term and long-term forecasts.
She said a team of about 30 people, including scientists, computer operators and programmers, will be working on the computing project at least part time until 2008. Rienecker said preparations already are underway, and the team expects to take its work to the new computer this fall, but the data crunching is not slated to begin in earnest until April 2006. By the time the project concludes 18 months later, the re-analyzed data set — all 40 terabytes of it — will be available to researchers for download through the Goddard Distributed Active Archive Center.
The MERRA project was one of Goddard’s primary justifications for the new hardware. Rienecker said the team will re-analyze a quarter-century’s worth of satellite sensor data at a remarkably high resolution for such an undertaking — one-half degree, or roughly 50-kilometer-wide slices of the Earth.
Rienecker said a detailed re-analysis of such a large data record would have been untenable on Goddard’s now three-year -old and largely obsolete supercomputer. What is expected to take a year-and-a-half with the new systems would have taken almost five years on the old system — much too long to even be considered.
Goddard’s new supercomputer is one of about a half-dozen similar-sized systems Silicon Graphics has sold to oil companies, universities, research institutes and other government agencies, including the U.S. Department of Defense, according to Jill Matzke, Silicon Graphics high-performance computing marketing manager. The systems are all used for highly sophisticated modeling and simulation, she said, including genome mapping and designing and testing spacecraft, aircraft and automobiles.
