Sequential and parallel implementations of the partitioning finite-element method

With the computational capabilities of parallel computers, we should investigate new methods which have performance advantages on parallel computers even if they are not faster than conventional methods on sequential computers. One such method is the partitioning finite-element method (FEM). In this paper, we consider the implementation of the partitioning FEM on both the Gray Y-MP and the Intel Touchstone Delta. The partitioning method is shown to have many advantages over a traditional finite-element approach. On the Gray YMP and sequential computers, the partitioning method requires significantly less memory. For parallel processors such as the Intel Delta, we show that the partitioning FEM has a higher parallel efficiency than traditional FEM. EM scattering from an infinitely long dielectric cylinder is used as an example.