IMPLEMENTATION OF THE PARALLEL ARNOLDI METHOD IN THE IBM SP2 DISTRIBUTED MEMORY SYSTEM

Abstract

The following article discusses the Implicitly Restarted Arnoldi Method used for solving large sparse eigenvalue problems. It presents the parallel implementation of this algorithm for a distributed memory architecture developed by Maschhoff and Sorensen ([1]) and included in the P_ARPACK library. The article gives results of performance tests of the P_ARPACK subroutines in the IBM Scalable POWER2 (SP2) parallel system installation at the Academic Computer Centre TASK in Gdansk and describes some technical problems concerning use of message-passing libraries (particularly the Message Passing Interface (MPI)). as well as communication subsystems available in IBM SP2 with the discussed software package.