|Fulltext||PDF (requires Acrobat Reader)|
|Fulltext||PostScript (requires a PostScript Reader)|
|Article title:||Bucked-Based Modification of the Parallel Recursive Coordinate Bisection Algorithm|
|Abstract [en]:||Most finite element methods used nowadays utilize unstructured meshes. These
meshes are often very large and require substantial computational and memory resources.
Hence they are a reasonable application for parallel supercomputers.
One of the steps of finite element methods is mesh partitioning.
In this paper recursive coordinate bisection (RCB) algorithms for mesh partitioning are studied and a new heuristic for selecting the cut plane is presented. The heuristic aims to improve partition quality in terms of the total cut size of the resulting partition.
Both the original and modified algorithms have been implemented using portable Message Passing Interface (MPI) library. Two different communication strategies have been implemented and used for experiments.
Experiments have been done on Parsytec GC/PowerPlus and Cray T3E supercomputers for large unstructured meshes. The obtained results demonstrate that the new heuristic achieves better partition quality for complex-shaped meshes. However, the running time of the modified algorithm is longer on average than the running time of the original algorithm.
|Publisher:||Linköping University Electronic Press|
|No. of pages:||47|
|Series:||Linköping Electronic Articles in Computer and Information Science|
|REFERENCE TO THIS PAGE: