A Latency-Hiding Algorithm for ABMS on Parallel/Distributed Computing Environment

A latency-hiding algorithm for the parallelization of large scale agent-based model simulations (ABMS) on parallel/distributed computing platform is proposed. The key idea of this algorithm is using redundant computations to hide communication latencies. An analytical model for this algorithm is presented to tell how to select R value to reach the best speedup. Compared to B+2R algorithm [1], theoretical analysis and experimental result show that the algorithm proposed in this paper has the following advantages when running on CPU cluster: 1) it can hide more communication latency, 2) a smaller R can satisfy the best communication latency hiding.