On the Speedup/Delay Trade-Off in Distributed Simulations

Assume a local simulator (LS) of a given system is available and we wish to turn it into a distributed simulator (DS). In the DS case, the LS is partitioned into segments called federates, each federate being run by a separate host. Before implementing the DS (i.e., at design-time) we wonder: will the DS execution time be shorter than LS one? In some cases the DS may run slower than the equivalent LS. To answer this question we are to consider that the execution time of a distributed simulation system depends on 3 interacting factors: 1) the speedup (or runtime gain) resulting from the partitioning of the local simulator into federates. 2) The network delays in the federate synchronization messages exchange. 3) The network delays inthe federatedata messages exchange. The combination of such factors makes very hard predicting the benefits of the LS-to-DS transformation. In this paper, a LS/DS decision procedure to support the LS/DS decision process at design-time. The procedure is guided by a performance model of the DS. The use of the High Level Architecture (HLA) distributed simulation standard is assumed.