Parallel performance characteristics of ICEPIC
Peter J. Mardahl, Andrew D. Greenwood, T. Murphy, K. Cartwright · 2004
Fast, efficient results from the ICEPIC (improved concurrent electromagnetic particle in cell) code are key to the Air Force Research Laboratory's efforts to design high power microwave sources for electronic warfare and nonlethal weaponry. Parallelization of ICEPIC allows the use of DoD supercomputer assets to perform device simulations which would previously have been impossible, and also to obtain these results quickly, with little waste of expensive CPU resources. We explain the parallel design and implementation of ICEPIC. Its parallel PIC loop exploits the possibility of computing and communicating between CPUs concurrently, reducing the waste and overhead inherent to parallel computing. ICEPIC's static and dynamic load balancing features ensure that computation is divided evenly among the CPUs, a necessary condition for efficient operation. We show scalability results from both test simulations and actual device simulations, illustrating the effectiveness of ICEPIC's parallel design and implementation.