Cluster identification on a distributed memory multiprocessor

The cluster identification step is often the bottleneck in multiprocessor simulations of spin models for statistical mechanics. We have applied a connected component labeling algorithm originally developed for VLSI circuit extraction to the cluster identification problem. The algorithm is extended to more than two dimensions, abstracting away unnecessary spatial information to simplify implementation in higher dimensions. We identify two types of spatial locality in cluster configurations, and present optimizations to exploit each type of locality. Performance results are presented from two and three-dimensional Ising model simulations.<
>