Heuristic Load Balancing for CFD Codes Executed in Heterogeneous Computing Environments

Abstract

A graph partitioning-based heuristic load-balancing algorithm known as the Largest Task First with Minimum Finish Time and Available Communication Costs is modified to take into account the dynamic nature and heterogeneity of current large-scale distributed computing environments, like Grids. The modified algorithm is applied to facilitate load balancing of a known CFD code used to model crystal growth.