Shuttle disaster spurs NASA to buy overdue computing power
The space shuttle Columbia disaster in 2003 pointed up NASA's fallback in high-performance computing.
'Over the last 10 years, with flat budgets and competing priorities, we have not made the investments in high-end computing that we made in the past,' said Walt Brooks, head of advanced supercomputing at the NASA Ames Research Center at Moffett Field, Calif.
Current systems had fallen so far behind that they could scarcely execute some of the more complex tasks, particularly modeling complex fluid dynamics, Brooks said. Planning for the upcoming manned missions, as well as tasks in atmospheric and oceanographic modeling, all require more computational power than NASA currently has at its disposal.
In late July, NASA announced procurement of an SGI supercomputer dubbed Kalpana, to be located at Ames. The acquisition was spurred by the Columbia's fatal crash, Brooks said. Needing to understand what went wrong, the space agency decided to model how loose insulation foam had damaged the shuttle.
'We needed to look in detail at the very complex fluid dynamic flows around the shuttle,' Brooks said, using a modeling program called Overflow. 'We had to stop everything else we were doing because we had so little computing capacity,' he said.
It took six months of dedicated computer time to finish the modeling.
Ames has an assortment of computers for its largest tasks, including 512- and 1,024-processor SGI Origin 3000 clusters, as well as a 16-processor Cray X1 and a Cray Opteron system.
NASA hopes Kalpana will supply 10 times the computing capacity of anything it now has in-house. Once completed, the new machine will have a theoretical peak capacity of about 50 trillion floating-point operations per second.
It will consist of 20 interconnected SGI Altix clusters running Linux. Each cluster will have 512 Intel Itanium 2 processors, forming a 10,240-processor array connected to a 500T SGI InfiniteStorage storage area network.
Brooks said the agency chose the system largely because of the Origin's single system image. That architecture can run multiple nodes under a single instance of an operating system, rather than requiring each collection of nodes to have its own OS.
The computer costs an estimated $160 million through the Scientific and Engineering Workstation Procurement II vehicle, although leasing arrangements and reuse of existing equipment will cut NASA's initial investment to about $45 million.
NASA had not budgeted for the computer but moved money from other departments, Brooks said.
Ames has installed the first three nodes. More nodes will be added through the summer and into the fall.