A PARALLEL BLOCK LANCZOS ALGORITHM FOR DISTRIBUTED MEMORY ARCHITECTURES

In this paper we propose a block Lanczos algorithm suitable for MIMD distributed memory message passing architectures. It is based on an efficient parallelizaiion of basic linear algebra operations, such as matrix-matrix, sparse matrix-matrix, and dense QR factorization. We assume an unidirectional ring as connection topology and a block column wrap-around matrices distribution. We have chosen this approach to improve load-balancing, to eliminate the intersection of messages and to decrease communication The parallel Lanczos algorithm has been tested on a Convex Meta Series, a cluster of HP Series 9000 workstations running the PVM communication system. Results of the performance evaluation based on some classical parameters are shown.