Process/processor mapping based on similarity analysis in heterogeneous grid computing environment

This paper presents a direct approach to partitioning for resolving process/processor mapping problems in a heterogeneous grid environment. It employs a scheme of analyzing similarities in graphs to cluster processes and select processors. Based on a phase for pre-analyzing the underlying network topology, it takes the information on the underlying network topology into account to select processors during the process-partitioning procedure. During the refinement phase to add certain capabilities for jumping out of the local optima and satisfying the requirements for steady performance, we used a two-stage Kernighan-Lin (KL)-like approach in our current study without introducing any randomness. Experiments were done to assess the performance of this algorithm, and compare it to the popular KL-like algorithm with a sequential-mapping approach and a greedy-based approach to search for the initial solution. The new proposal is capable of finding better solutions than these other two approaches in most cases.