Cluster Architekturen und der Windows Computer Cluster Server 2003

Abstract

Advanced scientific and industrial applications are characterized by an ever increasing demand of computing power, which usually exceeds the capabilities of a single computer. Hence, companies like CRAY or NEC have developed new, however expensive and proprietary solutions for high performance computing. To overcome the tremendous costs of proprietary supercomputers, cluster computing aims at achieving a fair amount of computing power at an affordable price by coupling large numbers of commodity systems, like conventional personal computers or workstations. Even more important than the hardware is the software which coordinates the resources of a cluster and manages the cluster applications. Recently, Microsoft launched the Windows Compute Cluster Server 2003 to meet this trend and to provide its customers with a Windows-based solution for cluster computing. Like other cluster solutions it is capable of running MPI and OpenMP applications. The objective of this master thesis is to analyse and evaluate the Windows Compute Cluster Server 2003. For this purpose we have set up a small experimental cluster and performed a number of performance tests, including performance experiments with the Linpack benchmark.<br />Furthermore, a comparison of the Windows Compute Cluster with a Linux-based cluster solution based on the Parallel Knoppix Cluster software has been made, and the main differences with respect to performance, configuration management and functionality have been studied.