B.D. Langton, D.A. Andresen, and E.A. Bowker (USA)
Cluster computing, Beowulf, scheduling
Current systems for managing workload on clusters of workstations, particularly those available for Linux-based (Beowulf) clusters, are typically based on traditional process-based, coarse-grained parallel and distributed pro gramming. The DESPOT project is building a sophis ticated thread-level resource-monitoring system for com putational, storage and network resources [2, 3, 4]. In this paper we present our enhanced scheduling algorithm within DESPOT, our architecture for low-overhead, fine grained resource-monitoring tools for per-process network and other resource usage. This algorithm takes into account network, CPU, and memory demands to go beyond sim ple load balancing for better performance. We present ex perimental results on a small Beowulf cluster showing our performance using the multi-facetted DESPOT algorithm, an older version of the algorithm, and the MOSIX default scheduler. Our performance gain over the MOSIX load balancing implementation in many cases is significant, up to approximately 40%, and usually within 10% in other tests.
Important Links:
Go Back
