|
The need to speed up finite element analysis of large engineer-
ing applications leads to the development of parallel finite
element analysis. In order to achieve high efficiency in the
parallel analysis, it is necessary to select an adequate mesh
partitioning method based on the parallel solution algorithm
used. The main objectives of mesh partitioning are to achieve
balance of computational loads among submeshes (processors) and
to reduce interprocess communication. At present, there have
been a lot of mesh partitioning algorithms developed. However,
mesh partitioning is an optimiza- tion problem in which the
best solution can not always be sought in a limited time.
Therefore, there is no mesh partitioning method that can
guaranteed the best solution. As a result, for a specific
engineering application problem, it is usually better to
compare different partitioning algorithms then choose the most
adequate one. In this study, the evaluation method proposed by
Hsieh et al. (1994) is used for study of several mesh
partitioning algorithms. By expanding the mesh partitioning
software, called MeshPart (Hsieh et al., 1996), a software
evaluation environment, called MP-Lab, is proposed in this work
for the analysts to facilitate evaluation and comparison of new
mesh partitioning algorithms. Besides the Greedy(GR) and
Recursive Spectral Two-way(RST) partitioning algorithms already
implemented in the MeshPart, the multilevel partitioning
algorithm proposed by Karypis and Kumar(1995) is added into MP-
Lab and critically evaluated. The concept of object-oriented
technique is used while programming software in this work to
keep better expansion and to reduce the costs of maintenance.
Also discussed in this study are how to use the object-oriented
techniques to design a mesh partitioning software and how to
add new mesh partitioning algorithms in MP-Lab.
|