Performance evaluation of conservative algorithms in parallel simulation languages

Parallel discrete event simulation with conservative synchronization algorithms has been used as a high performance alternative to sequential simulation. In this paper, we examine the performance of a set of parallel conservative algorithms that have been implemented in the Maisie parallel simulation language. The algorithms include the asynchronous null message algorithm, the synchronous conditional event algorithm, and a new hybrid algorithm called Accelerated Null Message that combines features from the preceding algorithms. The performance of the algorithms is compared using the Ideal Simulation Protocol. This protocol provides a tight lower bound on the execution time of a simulation model on a given architecture and serves as a useful base to compare the synchronization overheads of the different algorithms. The performance of the algorithms is compared as a function of various model characteristics that include model connectivity, computation granularity, load balance, and lookahead.