A RETROFIT BASED METHODOLOGY FOR THE FAST GENERATION AND OPTIMIZATIONOF LARGE-SCALE MESH PARTITIONS - BEYOND THE MINIMUM INTERFACE SIZE CRITERION

Mesh partitioning is often the preferred approach for solving unstruct ured computational mechanics problems on massively parallel processors . Research in this area has focused so far on the automatic generation of subdomains with minimum interface points. In this paper, we addres s this issue and emphasize other aspects of the partitioning problem i ncluding the fast generation of large-scale mesh decompositions on con ventional workstations, the optimization of initial decompositions for specific kernels such as parallel frontal solvers and domain decompos ition based iterative methods, and parallel adaptive refinement. More specifically, we discuss a two-step partitioning paradigm for tailorin g generated mesh partitions to specific applications, and propose a si mple mesh contraction procedure for speeding up the optimization of in itial mesh decompositions. We discuss what defines a good mesh partiti on for a given problem, and show that the methodology proposed herein can produce better mesh partitions than the well celebrated multilevel Recursive Spectral Bisection algorithm, and yet be an order of magnit ude faster. We illustrate the combined two-step partitioning and contr action methodology with several examples from structural mechanics and fluid dynamics problems, and highlight its impact on the total soluti on time of realistic applications on current massively parallel proces sors. In particular, we show that the minimum interface size criterion does not have a significant impact on a reasonably well parallelized application, and highlight other criterion which can have a significan t impact.