Analyzing the Energy-Time Tradeoff in High-Performance Computing Applications
Although users of high-performance computing are most interested in raw performance, both energy and power consumption have become critical concerns. One approach to lowering energy and power is to use high-performance cluster nodes that have several power-performance states, so that the energy-time tradeoff can be dynamically adjusted. This paper analyzes the energy-time tradeoff of a wide range of applications-serial and parallel on a power-scalable cluster. The authors use a cluster of frequency- and voltage-scalable AMD-64 nodes, each equipped with a power meter. They study the effects of memory and communication bottlenecks via direct measurement of time and energy. They also investigate metrics that can, at run time, predict when each type of bottleneck occurs.