Workshop: High Performance Numerical Computation and Applications

High-Performance Computing and Networking

Volume 1593 of the series Lecture Notes in Computer Science pp 1052-1060

Date:

A parallel implementation of the block preconditioned GCR method

  • C. VuikAffiliated withFaculty of Information Technology and Systems, Department of Technical Mathematics and Informatics, Delft University of Technology
  • , J. FrankAffiliated withFaculty of Information Technology and Systems, Department of Technical Mathematics and Informatics, Delft University of TechnologyCenter for Mathematics and Computer Science (CWI)

* Final gross prices may vary according to local VAT.

Get Access

Abstract

The parallel implementation of GCR is addressed, with particular focus on communication costs associated with orthogonalization processes. This consideration brings up questions concerning the use of Householder reflections with GCR. To precondition the GCR method a block Gauss-Jacobi method is used. Approximate solvers are used to obtain a solution of the diagonal blocks. Experiments on a cluster of HP workstations and on a Cray T3E are given.

Keywords

approximate subdomain solution parallel Krylov subspace methods orthogonalization methods