CRI: MISER: A High-performance, Power-aware Cluster

点击翻译按钮获取中文摘要

0709025CRI:MISER: A High-performance, Power-aware ClusterKirk CameronThe insatiable performance demands of many critical computational science applications cannot be satisfied unless the power and energy requirements of emergent machines are reduced. High-performance, power-aware computing (HPPAC) has emerged as a promising approach to this problem. In HPPAC, power consumption is controlled by software to improve energy efficiency while meeting the performance needs of the application. The lack of power-aware cluster infrastructure of significant scale limits the type of power-aware computer science research conducted.This infrastructure effort is building the MISER (Management Infra-Structure for Energy Reduction) cluster, a power-aware cluster to enable and evaluate energy reduction techniques for scientific applications at scale. This proposed infrastructure will consist of at least 64 rack mounted servers, for a total of at least 256 processor cores. At a minimum each node will have several power-aware processors and power-aware disks. The operating system will be open-source Linux for system software development. Systems will be connected with a high-end interconnect. The project will also extend the existing power measurement and control infrastructure (PowerPack) to the new cluster.Twelve research projects, which span twelve faculty members across four departments, will be supported by the proposed power-aware MISER cluster. The participants have a history of mentoring women and minorities through the Multicultural Academic Opportunities Program (MAOP) where summer student interns participate in the project with a faculty mentor. Furthermore, the techniques will improve the energy efficiency of large scale systems, reduce waste and positively impact the environment.