National Center for Computational Sciences

PI: Patrick Worley, Oak Ridge National Laboratory

To maximize the utility of leadership-class systems, such as the Cray X1E, Cray XT3/XT4, and IBM BlueGene/L (BG/L), the performance community (performance tool developers, system and appliation performance evaluators, and performance optimization engineers) must understand how to use each system most efficiently. To further understanding of these high-end systems, this research focuses on four primary goals: (1) update and extend performance evaluation of all systems using suites of both standard and custom micro, kernel, and application benchmarks; (2) continue to port performance tools to the BG/L, X1E, and XT3/XT4, making these available to high-end computing users, and further develop the tools so as to take into account the scale and unique features of the leadership-class
systems; (3) validate the effectiveness of performance-prediction technologies, modifying them as necessary to improve their utility for predicting resource requirements for production runs on the leadership-class systems; and (4) analyze and help optimize current or candidate leadership-class application codes.