Sandia computing team wins SuParCup Gold Medal for parallel crash simulations
A Sandia team received the gold medal (first place) prize in the SuParCup high-performance computing competition last weekend (June 12) in Mannheim, Germany. The prize is awarded for "an outstanding contribution in the field of parallel computing."
There were 34 entries from researchers in Germany (17), the US (7), Switzerland (3), Austria (2), Russia (2), Belgium, France, and the UK (one each). The award was presented during the Mannheim Supercomputer ’99 Conference.
The award celebrates the Sandia team’s solution of "one of the toughest problems in computational science, the development of a highly scalable and efficient massively parallel crash code," says Bill Camp, Director of Computation, Computers, and Mathematics Center 9200. "Crash dynamics is an important area in nuclear weapons development and safety as well as in transportation safety." The Sandia entry was submitted by Steve Plimpton (9221) and Bruce Hendrickson (9226). The team also includes Steve Attaway, Kevin Brown, Martin Heinstein, Kurt Metzinger, Jeff Swegle (all 9117), David Gardner (9221), Courtenay Vaughan (9226), Jeff Gruda (5337), Michael Neilsen (7841), and Ed Barragy (Intel).
Computations were completed on Sandia’s teraflop machine, originally supplied by Intel, with a current peak performance in excess of 2 trillion floating point operations per second. Judges included senior figures in parallel computing from Europe and the United States. US judges included Jack Dongarra of the University of Tennessee and Oak Ridge National Laboratory and Horst Simon, director of the National Energy Research Supercomputing Center at Lawrence Berkeley National Laboratory.
The winning team received a check for DM 5,000 in addition to the gold medal. Sandia team representative Dave Gardner accepted the award from the Lord Mayor of Mannheim.
Sandia’s winning work was titled "Scalable Parallel Crash Simulations." It represents work being undertaken by Sandia for DOE’s science-based stockpile stewardship program.
The capability to carry out 3D crash simulations has been under development for a number of years at DOE’s Defense Program laboratories. Codes include PRONTO-3D from Sandia and DYNA-3D from Lawrence Livermore — and their commercial counterparts, ABAQUS and LS-DYNA. These codes are used within DOE to simulate weapon dynamics. Their use in the auto and aerospace industries to assess design safety is one of the most successful uses to date of super- computing by industry. "To envision what is involved in crash simulation," Bill says, "consider crushing a soda can under your foot. As the can buckles and crumples, its motion is constrained by the many contacts of the crumpled metal with itself. So, to simulate this kind of crush-up, we must not only treat nonlinear, elastic-plastic mechanics and failure, but we must also detect and deal with the many metal-to-metal contacts that occur during the crush simulation."
Says Bill: "Over the past several decades, successful two- and three-dimensional contact algorithms have been developed at a number of institutions including Sandia. However, to take full advantage of 3D simulations for crash dynamics, we need the power of massively parallel processing (MPP) computers.
"Even though MPPs have been around for a decade, until now no one has been able to use them effectively to solve in detail the really tough crash problems that we and industry would like to address. This is because the computational techniques needed to get the mechanics and contact detection to simultaneously scale up to thousands of processors seemed intractable. Indeed, at least one top academic researcher predicted that we would never be able to develop effective scalable crash codes on MPPs.
"The Sandia team found sophisticated parallel algorithms for crash dynamics that solve very large problems on thousands of processors with high efficiency."
More details about the competition can be found at http://sc.rz.uni-mannheim.de/ suparcup99/.