Addressing Sporadic Contention on Shared Computing Clusters
HPCASIA '05 Proceedings of the Eighth International Conference on High-Performance Computing in Asia-Pacific Region
Virtual Time Fair Queuing Algorithm for a Computational Grid
ICDCN '09 Proceedings of the 10th International Conference on Distributed Computing and Networking
Scheduling Strategies for Cycle Scavenging in Multicluster Grid Systems
CCGRID '09 Proceedings of the 2009 9th IEEE/ACM International Symposium on Cluster Computing and the Grid
Decentralized scalable fairshare scheduling
Future Generation Computer Systems
Hi-index | 0.00 |
We report on a performance evaluation of a Fair Sharesystem at the ASCI Blue Mountain supercomputer cluster.We study the impacts of share allocation under Fair Shareon wait times and expansion factor. We also measure theService Ratio, a typical figure of merit for Fair Sharesystems, with respect to a number of job parameters. Weconclude that Fair Share does little to alter importantperformance metrics such as expansion factor. This leadsto the question of what Fair Share means on clustermachines. The essential difference between Fair Share ona uni-processor and a cluster is that the workload on acluster is not fungible in space or time. We find that clustermachines must be highly utilized and supportcheckpointing in order for Fair Share to function moreclosely to the spirit in which it was originally developed.