Preview

PDF, English Download (1MB) | Terms of use

Abstract

In the last years the power consumption of high performance computing clusters has become a growing problem because number and size of cluster installations raised and still is raising. The high power consumption of the clusters results from the main goal of these clusters: High performance. With a low utilization the cluster hardware consumes nearly as much energy as when it is fully utilized. In these low utilization phases the cluster hardware can theoretically turned off or switched to an lower power consuming mode. In this thesis a model is designed to estimate the power consumption of the hardware with and without energy saving mechanism. With the resulting software it is possible to estimate the cluster power consumption for different configurations of a parallel program. Further energy aware hardware can be simulated to determine an upper bound for energy savings without performance leakage. The results show that is a great energy saving potential for energy aware hardware even in high performance computing. This potential should motivate research in mechanism to control the energy aware hardware in high performance clusters.