Performance Optimization for the Parallel Gauss–Seidel Smoother
Jens Gräbel, Bruno Lang, P. Ueberholz · PAMM · 2005
Abstract Simulation, e.g., in the field of computational fluid dynamics, accounts for a major part of the computing time on highperformance systems. Many simulation packages still rely on Gauss–Seidel iteration, either as the main linear solver or as a smoother for multigrid schemes. Straight‐forward implementations of this solver have efficiency problems on today's most common high‐performance computers, i.e., multiprocessor clusters with pronounced memory hierarchies. In this work we present two simple techniques for improving the performance of the parallel Gauss–Seidel method for the 3D Poisson equation by optimizing cache usage as well as reducing the number of communication steps. (© 2005 WILEY‐VCH Verlag GmbH & Co. KGaA, Weinheim)