Scalable and Efficient Scheduling Techniques for Clusters

Design of high performance clusters with Quality-of-Service (QoS)guarantees is becoming increasingly important to support a variety of scientific and commercial applications. Performance of a cluster can be improved by an efficient scheduling mechanism. Although a few communication-guided scheduling algorithms have been proposed recently for clusters, these schemes have not been tested on large platforms to examine the potential benefits. Therefore, the main motivation of this research is to design scalable and efficient scheduling algorithms for clusters. The proposed research addresses three related issues. First, an in-depth evaluation of the existing scheduling schemes will be done on a large cluster platform using real workloads before developing new algorithms. Second, various design issues in implementing these algorithms on the generic virtual interface architecture (VIA) will be explored. Finally, the research will examine QoS provisioning mechanisms to facilitate predictable performance in clusters. The success of this research is expected to have many important contributions in the area of cluster computing. Primarily, the practical scheduling algorithms developed in this research can be used on large platforms. The results from the VIA and QoS research should provide novel solutions to satisfy different performance and QoS requirements, and should foster further research in this area.