A parallel SSOR preconditioner for solving the problems of electromagnetism in frequency domains

Abstract

Several approaches to the parallelization of a symmetric successive overrelaxation (SSOR) preconditioner in the Eisenstat modification are considered. This preconditioner is used for the iterative solving of linear systems resulting from the approximation of the corresponding variational problems of electromagnetism. The parallelization of the preconditioner is based on a computational domain decomposition (including the algebraic one) with a joint matrix reordering. Special efforts are spent to ensure a high performance of the preconditioner on NUMA-architectures. The numerical results obtained confirm a high level of performance and scalability of the proposed algorithms.