We discuss insights gained from an effort to develop a versatile parallel PIC code (PICARD – PIC with arbitrary rebalancing and decompositions). Because of the highly dynamic nature of PIC, it is sometimes necessary to tailor decomposition strategies on a problem-by-problem basis instead of solely on the basis of the algorithm. The dynamics of a problem may dictate the particular strategy to be used. Thus, finding a general operational method for parallel PIC applications is important. We find approaches to partitioning the mesh and particle domains which provide some trade-off opportunities, but the relationship enforced between the partition layouts in addition to the way one chooses to manage the problem's dynamics defines categories of strategies which are independent of the partitioning technique used. The overall strategy enforced is as important as the individual techniques applied to partition the domains. Three strategy groups arise from the analysis and form the basis of widely disparate parallel PIC operational methods. These distinct solutions are obtained from rather simple parametric choices. We introduce the concept of rubber boundaries as a means to bring a compromise between what are commonly referred to as Lagrangian and Eulerian particle mappings. We show several parallel PIC strategies that are as yet unexplored. © 1997 John Wiley & Sons, Ltd.

Objectives of this study were to develop techniques and methods for effective analysis of large codes; to determine the feasibility of parallelizing an existing large scientific code; and to estimate potential speedups attainable, and associated tradeoffs in design complexity and work effort, if the code were parallelized by redesign for a distributed memory system (NCube, iPSC hypercube), or straight serial translation targetting a shared memory system (CRAY2, SEQUENT). MACH2, the code under study, is a 2-D magnetohydrodynamic (MHD) finite difference code used to simulate plasma flow switches and nuclear radiation. A taxonomy relating functional levels of a code to levels of parallelism is presented and used as a model for analyzing existing large codes. It is shown that although parallelizing lower level code segments (e.g. algorithms and loops) on shared memory systems is generally easier to accomplish, in some cases an entire large code is most easily parallelized at a high level; via domain and functional decomposition. Also a multi-decomposition scheme is introduced in which acceptable load balances can be achieved for functional decompositions and heterogeneous data partitionings.