Abstracts for the Conference on
Grid Adaptivity in Computational PDEs,
Edinburgh, July 96		Next
Previous
Index

3D parallel mesh adaptivity : implementation and analysis

P M Selwood , M Berzins, J M Nash & P M Dew

School of Computer Studies,
University of Leeds, UK

Abstract

The need to solve ever larger CFD problems has made it necessary to use distributed memory parallel computers to achieve acceptable solution times. For transient problems, one is still forced to consider mesh adaptation to retain the efficiency of the solver as the solution develops. The use of serial adaptivity causes problems in that serial bottlenecks are introduced which make the adaptivity expensive. Also, the meshes being handled in such problems are also often so large that they cannot fit into a single processor's memory. There is thus a demonstrable need for parallel adaptivity.

We will present a 3D tetrahedral parallel adaptivity algorithm based on so-called `$h$-adaptation' using refinement and derefinement of an initial mesh. We will consider parallel data-structures and issues of data consistency across multiple processors.

The way the mesh is partitioned is also crucial in order to balance the computational load whilst minimising interprocessor communication. We will present a novel cost model for a finite volume solver which indicates how the mesh should be ideally partitioned.

[1] W. Speares & M. Berzins. A 3D Unstructured Mesh Adaptation Algorithm for Time-Dependent Shock Dominated Problems. Submitted to Int. J. of Num. Meth. in Fluids.

[2] D. C. Hodgson, P.K. Jimack, P. Selwood & M. Berzins. Scalable Parallel Generation of Partitioned, Unstructured Meshes pp 665-672 of `Parallel Computational Fluid Dynamics - Implementations and Results Using Parallel Computers' (editors A. Ecer, J. Periaux, N. Satofuka and S. Taylor), Elsevier Science 1996.

ABSTRACTS
NextPreviousIndex
Last modified Fri Jun 21 19:19:16 GB-Eire 1996 (DBD)