Merging parallel simulations
MetadataShow full item record
I n earlier work cloning is proposed as a means for efficiently splitting a running simulation midway through its execution into multiple parallel simulations. In simulation cloning, clones usually are able to share computations that occur early in the simulation, but as their states diverge individual logical processes (LP’s) are replicated as necessary so that their computations proceed independently. Over time the state of the clones (or their constituent LPs) may converge. Traditionally, these converged LPs would continue to execute identical events. We address this inefficiency by merging of previously cloned LPs. We show that such merging can further increase efficiency beyond that obtained through cloning only. We discuss our implementation of merging, and illustrate its effectiveness in several example simulation scenarios.