A Parallel Loop Scheduling for Extremely Heterogeneous PC Clusters

Abstract

Cluster computers are becoming increasingly more common, especially with the increasing use of Beowulf systems, and networks of workstations (NOW) for parallel computing. Nowadays using cost-effective cluster computers to deal with problems which need large computing is a spreading trend. The major source of parallelism in a program is loops. If the loop iterations can be distributed to different processors evenly, the parallelism within loop iterations can be exploited. In homogeneous environment, one can divide equal amount of iterations to each processor, but in heterogeneous environment, this way is not suitable. Self-scheduling schemes, such as FSS, GSS and TSS, can achieve load balancing in SMP, even in moderate heterogeneous environment, but are not suitable in extremely heterogeneous environment. In this paper, we propose a heuristic approach to solve parallel loop scheduling problem in extremely heterogeneous environment. The experiments were conducted on a heterogeneous PC cluster.