Understanding Power Measurement Implications in the Green500 List
Authors:
- Balaji SubramaniamBSBalaji Subramaniam
- Wu-Chun FengWFWu-Chun Feng
- 1 December 2010
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 245-251
Abstract
For decades, performance has been the driving force behind high-performance computing (HPC). However, in recent years, power consumption has become an important constraint as operational costs of a supercomputer are now on par with the acquisition costs of a supercomputer. Even though we face major energy issues in achieving large-scale performance, there is still a lack of a standardized power measurement methodology in the HPC community for energy-efficient supercomputing. In this paper, we report on our experiences in updating the run rules for The Green500 List with a particular emphasis on the power measurement methodology. We use high-performance LINPACK (HPL) to study power measurement techniques that can be applied for large-scale HPC systems. We formulate experiments to provide insight into the power measurement issues in large-scale systems with the goal of improving the readers' understanding of the measurement methodology for the Green500 list.Keywords
This publication has 7 references indexed in Scilit:
- Energy Profiling and Analysis of the HPC Challenge BenchmarksThe International Journal of High Performance Computing Applications, 2009
- PowerPack: Energy Profiling and Analysis of High-Performance Systems and ApplicationsIEEE Transactions on Parallel and Distributed Systems, 2009
- Power efficiency in high performance computing2008 IEEE International Symposium on Parallel and Distributed Processing, 2008
- The Green500 List: Encouraging Sustainable SupercomputingComputer, 2007
- Power and Energy Profiling of Scientific Applications on Distributed Systems2005
- Blue Gene/L torus interconnection networkIBM Journal of Research and Development, 2005
- The Bladed Beowulf: a cost-effective alternative to traditional Beowulfs2003