A dynamic partitioning self-scheduling scheme for parallel loops on heterogeneous clusters

Abstract

Loop partitioning on parallel and distributed systems has been an important problem. Furthermore, it becomes more difficult to deal with on the emerging heterogeneous PC cluster environments. In this paper, we propose a performance-based scheme, which dynamically partitions loop iterations according to the performance ratio of cluster nodes. To verify our approach, a heterogeneous cluster is built, and two kinds of application programs are implemented to be executed in this testbed. Experimental results show that our approach performs better than traditional schemes.