Performance prediction of a parallel simulator

There are at least three major obstacles thwarting widespread adoption of parallel discrete-event simulation: lack of need; lack of tools; and lack of predictability in behavior and performance. The plain truth is that most simulation studies can be adequately done on ordinary serial computers. Parallel simulation tools are products of research efforts, and simply do not stand up to the demands of modern software engineering. The results of 20 years of research in parallel simulation reveal it to be a highly complex endeavour, with performance results very much dependent on implementation details and model characteristics. The Scalable Simulation Framework (SSF) is an effort to address some these concerns. It addresses lack of need in two ways; it provides a modeling API that is attractive both for serial and parallel simulation, with parallel execution requiring no change to the model, and it targets large-scale telecommunication system modeling, an application area that requires the computational capabilities of parallelism. We address the concern over unpredictable behavior. We show how we measured the internal overheads of the Dartmouth implementation of the SSF API (DaSSF), and how those measurements can be used to predict the performance of a given model, using given features of the simulator, without having to run, or even build, the model