A parallel multigrid method for the prediction of incompressible flows on workstation clusters

摘要

A parallel multigrid method for the prediction of incompressible flows in complex geometries is investigated with respect to the performance on workstation clusters. The parallel implementation is based on grid partitioning and follows the message passing concept, ensuring a high degree of portability. A high numerical efficiency is obtained by a nonlinear multigrid method with a pressure-correction scheme as smoother. Within this investigation, up to 64 workstations located at the TU Munich connected by an ethernet switch are used as well as two workstations located at the University of Erlangen. The workstations in Erlangen and Munich are connected by a high-speed ATM-network. Global and local communication which is required within the solver is examined for different configurations of workstations. These results are used to understand the performance of the flow solver. Moreover, it is shown that speedups can be achieved for the computation of complex flow problems on work-station clusters, although these speedups are smaller compared to other parallel architectures.