Parallel sorting algorithms for optimizing particle simulations

Real world particle simulation codes have to handle a huge number of particles and their interactions. Thus, parallel implementations are required to get suitable production codes. Parallel sorting is often used to organize the set of particles or to redistribute data for locality and load balancing concerns. In this article, the use and design of parallel sorting algorithms for parallel particle simulation codes are discussed. As a typical example, the particle simulation code PEPC is considered and a specific parallel sorting algorithm for this application is presented. The resulting parallel simulation code was implemented on an IBM Blue Gene/P system and corresponding performance results are shown.