Performance evaluation of load balancing algorithms for parallel single-phase iterative PDE solvers

Abstract

We review and evaluate the performances of six data mapping algorithms used for parallel single-phase iterative PDE solvers with irregular 2-dimensional meshes on multicomputers. We provide a table that compares the six algorithms for eight measures covering load balance, interprocessor communication, flexibility, ease of use and speed. Based on the comparison results, we recommend the use of the simplest and fastest (P/spl times/Q) of the six algorithms considered for sequential compile-time mapping of 2-dimensional meshes.