COMPUTATIONAL RESULTS FOR PARALLEL UNSTRUCTURED MESH COMPUTATIONS

The majority of finite element models in structural engineering are co mposed of unstructured meshes. These unstructured meshes are often ver y large and require significant computation resources; hence they are excellent candidates for massively parallel computation. Parallel solu tion of the sparse matrices that arise from such meshes has been studi ed heavily, and many good algorithms have been developed. Unfortunatel y, many of the other aspects of parallel unstructured mesh computation have gone largely ignored. We present a set of algorithms that all th e entire unstructured mesh computation process to execute in parallel- including adaptive mesh refinement, equation reordering, mesh partitio ning, and sparse linear system solution. We briefly describe these alg orithms and state results regarding their running-time and performance . We then give results from the 512-processor Intel DELTA for a large- scale structural analysis problem. The results demonstrate that the ne w algorithms are scalable and efficient. The algorithms are able to ac hieve up to 2.2 gigaflops for this unstructured mesh problem.