PARALLEL MATRIX SOLVERS FOR MOMENT METHOD CODES FOR MIMD COMPUTERS

Abstract

Parallel algorithms are presented that are suitable for the solution of the system of linear equations generated by moment method problems on local memory Multiple Instruction, Multiple Data (MIMD) parallel computers. The two most widely used matrix solution algorithms in moment method codes are described, namely the conjugate gradient (CG) method and LU decomposition. The underlying philosophy of parallelism is briefly reviewed. Suitable parallel algorithms are then described, presented in pseudo-code, their timing behaviour analyzed the- oretically, and timing results measured on a particular MIMD computer- -a transputer array --are presented and compared to the theoretical timing models. It is concluded that efficient parallel algorithms for both the CG and LU exist and that MIMD computers offer an attractive computational platform for the solution of moment method problems with large numbers of unknowns. [Vol. 8, No. 2 (1993), pp 144-175]