Efficient Pipelined Communication Design for Parallel Mesh Refinement in 3-D Finite Element Electromagnetics with Tetrahedra

Minimizing communication latency is essential for designing cost-effective parallel finite element methods (FEM). A new, efficient pipelined communication strategy is presented for parallel 3-D finite element mesh refinement with tetrahedra. A Petri nets-based model is developed which simulates the interprocessor communication costs for both the target mesh refinement algorithm and parallel architecture. The potential benefits of this approach for optimizing utilization of the system resources are demonstrated. Performance measures derived from the discrete event simulations show that the new pipelined design yields improved communication speedup over a range of problem sizes and different numbers of processors