Improving lookahead in parallel discrete event simulations of large-scale applications using compiler analysis

This paper addresses the issue of efficient and accurate performance prediction of large-scale message-passing applications on high performance architectures using simulation. Such simulators are often based on parallel discrete event simulation, typically using the conservative protocol to synchronize the simulation threads. The paper considers how a compiler can be used to automatically extract information about the lookahead present in the application and how this can be used to improve the performance of the null protocol used for synchronization. These techniques are implemented in the MPI-Sim simulator and dHPF compiler which had previous been extended to work together for optimizing the simulation of local computational components of an application. The results show that the availability of lookahead ranging improves the runtime of the simulator by factors ranging front 9% up to two orders of magnitude, with 30-60% improvements being typical for the real-world codes. The experiments also show that these improvements are directly correlated with reductions by the number of null messages required by the simulations.