作者: Kevin J Bowers , Ron O Dror , David E Shaw , None
DOI: 10.1063/1.2191489
关键词:
摘要: The evaluation of interactions between nearby particles constitutes the majority computational workload involved in classical molecular dynamics (MD) simulations. In this paper, we introduce a new method for parallelization range-limited particle that proves particularly suitable to MD applications. Because it applies not only pairwise but also involving three or more particles, can be used both nonbonded and bonded forces simulation. It requires less interprocessor data transfer than traditional spatial decomposition methods at all lowest levels parallelism. gains an additional practical advantage certain commonly communication networks by distributing burden evenly across network links decreasing associated latency. When parallelize MD, further reduces requirements allowing computations with short-range interactions, long-range electrostatics, migration use much same communicated data. We variants significantly improve balance load processors.