Parallel Multilevel Fast Multipole Algorithm for GRID computing allowing Full-Wave Electromagnetic Simulations

This work is concerned with the parallelization of the multilevel fast multipole algorithm (MLFMA) for large scale electromagnetic simulations on distributed memory computers. We focus on techniques suited for low-cost clusters of workstations (COWs) and computational GRIDs. These consist of a number of nodes interconnected through a Gigabit Ethernet network and a fast switch. The algorithm is implemented on a method of moment discretisation of a two dimensional TM electromagnetic scattering problem. Because the parallelization of the MLFMA inherently relies on heavy communication between the nodes special care is taken to avoid congestion of the interconnection network by means of an asynchronous algorithm, i.e. some nodes are communicating while others are performing calculations. This allows us to evaluate large problems such as passive optical devices or electromagnetic compatibility simulations of complex high-frequency electronic equipment with up to several millions of unknowns. Keywords—Electromagnetic scattering, parallel MLFMA