A Comparison of Decomposition Techniques for Structurally Symmetric Sparse Matrices for Parallel Incomplete LU Factorization

Authors: Buket Benek, H. Dağ (Istanbul Technical University, Computational Science and Engineering Program, Institute of Informatics)

Abstract

Incomplete LU factors are widely used as preconditioners in conjunction with iterative methods. As the size of the equations at hand gets bigger a parallel iterative solver needs be used. In parallel environments, however, a balanced load distribution and less communication are desirable for the whole solution process. This paper explains the effect of decomposition of the general and large structurally symmetric sparse matrices on communication and computation time for the parallel algorithms for finding Incomplete LU(p) factors of a matrix.