Isoefficiency Analysis of Parallel QMRCGSTAB-Like Iterative Methods on Massively Distributed Memory Computers

摘要

For the solutions of linear systems of equations with nonsymmetric coefficient matrices, we have proposed two improved version of the QMRCGSTAB method, which were called IQMRCGSTAB method in [3] and PQMRCGSTAB method in [4]. In this paper, a theoretical model of computation and communication phases is presented to allow us give a quantitative analysis of the parallel performance of parallel QMRCGSTAB-like methods on two-dimensional mesh-based topology. The isoefficiency analysis give an useful insight into the scalability of the QMRCGSTAB-like method.