PARALLEL ADAPTIVE MESH REFINEMENT AND REDISTRIBUTION ON DISTRIBUTED-MEMORY COMPUTERS

A procedure to support parallel refinement and redistribution of two-d imensional unstructured finite element meshes on distributed memory co mputers is presented. The procedure uses the mesh topological entity h ierarchy as the underlying data structures to easily support the requi red adjacency information. Mesh refinement is done by employing links back to the geometric representation to place new nodes on the boundar y of the domain directly on the curved geometry. The refined mesh is t hen redistributed by an iterative heuristic based on the Leiss-Reddy [ 1] load balancing criteria. A fast parallel tree edge-coloring algorit hm is used to pair processors having adjacent partitions and forming a tree structure as a result of Leiss-Reddy load request criteria. Exce ss elements are iteratively migrated from heavily loaded to less loade d processors until load balancing is achieved. The system is implement ed on a massively parallel MasPar MP-1 system with a SIMD style of com putation and uses message passing primitives to migrate elements durin g the mesh redistribution phase. Performance results of the redistribu tion heuristics on various test meshes are given.