Time and/or space sharing in a workstation cluster environment

The clustered parallel computer (CPC), based on a workstation cluster, is becoming popular as a choice for high-performance network or parallel computing. However, operating system overheads, network protocols, and higher message-passing latency contribute to a lower overall communication performance in a cluster of workstations, increasing the likelihood that timesharing of parallel jobs can be used to improve system throughput in a workstation cluster. The traditional means by which the CPC minimizes user job turnaround time (JTT) is through space sharing, in which user jobs are given exclusive control over clusters of processors. The objective of this study is to examine methods by which system utilization may be increased by giving timeshared access to parallel jobs without sacrificing the primary goal of minimizing user JTT