ERCIM News No.22 - July 1995 - CCL

Practical Partitioning Methods for Unstructured Meshes

by Chris Greenough and Ron Fowler

The use of computing systems with some type of parallel architecture has grown significantly over the past few years. These systems are increasing being seen as the major way by which sufficient affordable computing power can be provided for the advanced engineering simulations now being attempted in fluid dynamics, stress analysis and electromagnetism. These simulations typically use meshes or grids containing 500,000 nodes and use the process of automatic grid refinement during the solution.

Many of the parallel machines being used are based on an MIMD form of parallelism where the memory of the machine is distributed over a network of processors. A consequence of this is that the program and its associated data must be distributed between these processors. In finite volume and finite element methods this leads to the problem of how to distribute large unstructured grids and meshes initially, and how to redistribute them subsequently.

The PPMUM Project intends to address the development new partitioning and mesh redistribution methods for the parallel implementation of finite element analyses.

This research project intends to address and advance three elements of this problem:

Software implementations of the new algorithms developed will be integrated into a large existing applications program to test their effectiveness.

The Figure shows the result of appyling one of the these meshes partitioning methods to a three-dimensional mixed mesh. One result of the software implementation will be a library of routines which will be made available to other research groups to enable them to take advantage of the new methods developed.

Please contact:
Chris Greenough - CCL
Tel: +44 1235 44 5307
or Ron Fowler - CCL
Tel: +44 1235 44 5243

return to the contents page