A New Approach To Task Decomposition For Parallel Particle-In-Cell Simulation
Nasser G. Azari, S.-Y. Lee · International Journal of Modelling and Simulation · 1998
The particle-in-cell (PIC) method is one of the particle simulation methods, which is predominantly used in areas such as plasma physics, semiconductor device physics, and global climate modelling. Its long execution time and large memory space required, especially for 3-D simulations, have been its major drawbacks. In this paper, a new task decomposition scheme, called hybrid partitioning, is pro-posed for the PIC simulation. Given a grid siie and the number of processing elements (PEs), it provides a set of possible decompositions of which two extreme cases correspond to the conventional partitioning methods, that is, partitioning grid or particles only. The usefulness of the hybrid partitioning stems from its ability to choose a proper size of subgrid (rather than a fired size) for the possible maximum speed-up depending on system and problem parameters. The proposed scheme lias been developed for both message-passing and ohared-me mory systems. Through experiments on the iPSC/2 and iPSC/860, BBN Butterfly (GP100 and TC2000), we demonstrate the advantages of hybrid partitioning.