Contribution to a conference proceedings

FZJ-2015-03179

http://join2-wiki.gsi.de/foswiki/pub/Main/Artwork/join2_logo100x88.png Parallelizing Iterative Solvers for Sparse Systems of Equations and Eigenproblems on Distributed Memory Machines

Basermann, A. (Corresponding Author)

1994

Please use a persistent id in citations: http://hdl.handle.net/2128/8608

Abstract: For the analysis and solution of discretized ordinary or partial differential equations it is necessary to solve systems of equations or eigenproblems with coefficient matrices of different sparsity patterns, depending on the discretization method. In many cases, the use of the finite element method (FE) results in largely unstructured systems of equations. The main computational cost in iterative methods for solving these problems consists of matrix-vector products. When iterative solvers are parallelized on a multiprocessor system with distributed memory, the data distribution and the communication scheme - depending on the data structures used for sparse matrices - are of the greatest importance for an efficient execution. Here, data distribution and communication schemes are presented that are based on the analysis of the column indices of the non-zero matrix elements. Performance tests, using the conjugate gradient method (CG) and the Lanczos algorithm for the symmetric eigenproblem, were carried out on the distributed memory systems Intel iPSC/860 and Paragon XP/S 10 of the Research Centre Jülich with sparse matrices from FE models. The parallel variants of the algorithms showed good scaling behavior for matrices with very different sparsity patterns.

---

Contributing Institute(s):

1. Zentralinstitut für Angewandte Mathematik (ZAM)
2. Jülich Supercomputing Center (JSC)

Research Program(s):

1. 899 - ohne Topic (POF2-899) (POF2-899)

---

Database coverage:

---