ABSTRACT
Laxmikant V. Kale´, Eric Bohm, Celso L. Mendes, Terry Wilmarth, Gengbin Zheng Department of Computer Science, University of Illinois at Urbana-Champaign
The increasing size and complexity of parallel machines, along with the increasing sophistication of parallel applications, makes development of parallel applications a challenging task for petascale machines. The National Science Foundation has planned the deployment of a sustained petaflops (PFLOPS) machine by 2010, which is likely to have several hundred thousand processor cores. The Blue Gene/L already has 128K cores, and some future designs with low-power processors may have over a million cores. Machines in the intervening years are slated to perform at over a PFLOPS peak performance. Further, as portended by the Roadrunner project at Los Alamos National Laboratory (LANL), some of the large machines will have accelerators along with commodity processors.
