Abstract:
State-of-the-art high end computing systems (HECS) consume multiple megawatts of electric power. The operational cost and reliability of emergent HECS will soon limit system performance and scalability. This makes power a first-class constraint in HECS designed for scientific computing. Emergent HECS must reduce power consumption without impacting application performance. MISER (Management Infra-Structure for Energy Reduction) is a power aware high performance computing framework designed to address this challenge.

In this talk, I describe several components of the MISER framework including: 1) a hardware/software toolkit that profiles the power/energy consumption of parallel scientific application at component- and function-level granularity; 2) high performance power aware schedulers that dynamically adjust processor voltage according to workload characteristics; and 3) a power aware speedup model that predicts power and performance tradeoffs in HECS. This framework can be applied to identify and exploit performance inefficiencies in parallel scientific applications for energy conservation. For example, for some scientific codes MISER can achieve system-wide energy savings as large as 30% with less than 1% performance loss.

Bio:
Rong Ge is a PhD candidate in the Department of Computer Science at Virginia Tech, where she conducts research in the Scalable Performance Laboratory (SCAPE). Her research expertise is in high-performance computing, parallel and distributed systems, performance modeling and analysis, power-aware systems, computer architecture, and scientific computing. Her industrial experiences include working as an engineer for China Telecom Research and Development Center.

Ms. Ge received a B.S. degree and a M.S. in fluid mechanics from Tsinghua University, China, and an M.E. in computer science and engineering from the University of South Carolina. She is a member of the IEEE, ACM and Upsilon Pi Epsilon.