Computational & Technology Resources
an online resource for computational,
engineering & technology publications |
|
Computational Science, Engineering & Technology Series
ISSN 1759-3158 CSETS: 2
PARALLEL AND DISTRIBUTED PROCESSING FOR COMPUTATIONAL MECHANICS: SYSTEMS AND TOOLS Edited by: B.H.V. Topping
Chapter 12
Parallel Adaptive PCG S. Blazy, R. Diekman, U. Dralle, F. Neugebauer and T. Romke
Department of Computer Science, University of Paderborn, Germany S. Blazy, R. Diekman, U. Dralle, F. Neugebauer, T. Romke, "Parallel Adaptive PCG", in B.H.V. Topping, (Editor), "Parallel and Distributed Processing for Computational Mechanics: Systems and Tools", Saxe-Coburg Publications, Stirlingshire, UK, Chapter 12, pp 236-251, 1999. doi:10.4203/csets.2.12
Abstract
The work described here takes place in conjunction with an object
oriented re-design of the parallel FElibrary PadFEM. The environment of
PadFEM includes a graphical editor for specifying domains with boundary conditions,
automatic mesh generation, automatic mesh partitioning and mapping
onto the processors of a MIMD-system. The parallelization is based on PVM and
MPI which guarantees portable parallel code.
We describe the PadFEM environment and OO-data structure briefly. The solver module of PadFEM includes a parallel conjugate gradient method, a parallel multigrid solver, and two different parallel preconditioned conjugate gradient methods. The preconditioning is based on domain decomposition according to Bramble and Borchers. The FE-method works on unstructured triangular meshes which are adaptively refined using error estimates of the solution. Several different refinement strategies are implemented. To preserve load balance after refinements, a distributed load balancing algorithm based on diffusion is used. The load migration strategies try to optimize sub-domain shapes which is essential for the convergence behavior of the sub-domain solvers. All implementations consider the concept of programming frames which allows the easy generation of user-specific code. purchase the full-text of this chapter (price £20)
go to the previous chapter |
|