Event-driven hard-particle molecular dynamics using bulk-synchronous parallelism Original Research Article

We describe and analyse a new parallel algorithm for event-driven simulation of hard-particle systems that is based on the ideas of the bulk-synchronous parallel (BSP) model. This model provides a unifying approach for general purpose parallel computing which in addition to efficient computation ensures portability across different parallel architectures. The hard-particle system is divided into regions that are owned by a unique processor. Events involving particles located in the region boundaries are called border zone events and are used to synchronize the parallel operation of the processors. On a P-processor BSP computer the algorithm works doing iterations composed of two phases: the parallel phase where each processor is simultaneously allowed to simulate serially and asynchronously its own region, and the synchronization phase where the occurrence of at most P border zone events is scheduled for the next iteration considering the state of neighboring regions. We show that this algorithm is efficient in practice provided that a sufficiently large hard-particle system is to be simulated.