Building supercomputers powerful enough to run complex simulations is a challenge that the National Nuclear Security Administration’s (NNSA’s) Advanced Simulation and Computing Program (ASC, formerly ASCI) has met since its beginning in 1995. The ASC Program has been remarkably successful meeting its goals, and its machines are the workhorse computers for running complex two-dimensional and three-dimensional physics codes for the nation’s Stockpile Stewardship Program. As the capabilities of ASCI machines have grown, so have the sophistication of applications and number of users. Costs have fallen as technologies advance with each machine, but vendor integrated systems using workstation processors and proprietary software have now matured to where cost per teraflop improves at only a slow exponential rate dictated by Moore’s law. Lawrence Livermore National Laboratory (LLNL), as part of the ASC Program and its own Multiprogrammatic and Institutional Computing Initiative, is looking to new technologies that can be exploited for better cost-performance. This talk presents an overview of current and planned ASCI systems, and explores some promising approaches, including open source software, cluster architectures, and system-on-a-chip designs that promise cost-effective platforms to run the next generation of complex scientific simulations.
Steve Louis is Assistant Department Head for Integrated Computing and Communications at LLNL. He has recently completed a one-year assignment in Washington as a technical advisor for the NNSA ASC Program Office for new computer procurements and software development. From 2000 to 2002 he led installation, integration and early use efforts for the 12 TeraOP ASCI White platform at LLNL, and has been the Principal Investigator for the LLNL Problem Solving Environment Program since 1999. Prior to these positions, he was Project Leader for High Performance Archival Storage and Group Leader for File Storage Systems at LLNL’s National Energy Research Supercomputer Center (NERSC) for ten years. He received an A.B. in computer science from the University of California at Berkeley in 1974 before joining LLNL to become a founding staff member at NERSC, one of the first supercomputer centers in the nation.