PARA'04 State-of-the-Art
in Scientific Computing
June 20-23, 2004 (Home page)
Updated: 19 February 2004

PARDISO: The current state and algorithmic goals for the future

Olaf Schenk* and Klaus Gaertner**
* Department of Computer Science
University Basel
Klingelbergstrasse 50, CH-4056 Basel
Switzerland
email: olaf.schenk@unibas.ch
** Weierstrass Institute for Applied Analysis and Stochastics
Mohrenstr. 39, D-10117 Berlin
Germany
email: gaertner@wias-berlin.de

The PARDISO package [1] offers serial and parallel solvers for the direct solution of unsymmetric and symmetric sparse linear systems on shared memory multiprocessors. The main goal of the PARDISO development was to solve problems related to degenerate elliptic PDEs systems as efficient as possible in cases where special algorithms are not available. But dealing with matrices from other problem classes show up immediately and we will give an short overview with respect to:

• problem classes in the focus;
• basicstability assumptions and consequences for pivoting techniques;
• presentstatus of the implementation and the extension to distributed cluster of SMPs [2];

We will also discuss open algoritmic questions of different nature to generate additional discussions, for instance:

• simple extensions (aiming on bifurcation analysis for instance);
• better heuristics for minimal surfaces on graphs (improvement or replacement of the METIS ordering [3] in PARDISO);
• pseudo domain decompositions and partial refactorizations;
• saddle point problems;

The effort for developments is limited, we want to stay focused and not loose the connections to applications -- hence we are looking for further collaborations.

References:
1. PARDISO website. http://www.computational.unibas.ch/cs/scicomp/software/pardiso/.
2. K. Fuerlinger, O. Schenk, and M. Hagemann. Task-queue based hybrid parallelism: A case study. Technical report, Institut fuer Informatik, Lehrstuhl fuer Rechnertechnik und Rechnerorganisation, Technische Universitaet Muenchen, 2004. Submitted.
3. G. Karypis and V. Kumar. A fast and high quality multilevel scheme for partitioning irregular graphs. SIAM J. Scientific Computing, 20(1):359-392, 1998.

Home page