This project explores methods for improving the performance of parallel discrete-event simulation by combining features of both conservative and optimistic synchronization strategies. Rather than attempting to avoid rollback, which is accomplished by slowing down the fastest components of the simulation, the strategy is to avoid state checkpointing, which should speed up the slowest components. The project is experimental in nature. A single implementation will allow one to make fair comparisons between the performance of pure conservative synchronization, and a hybrid method. Furthermore, by adding a mechanism for limiting the degree of optimism in both the pure optimistic and the hybrid cases, it will be possible to verify or refute the hypothesis that the overhead of optimistic rollback is a second-order effect compared to the overhead of state checkpointing. Subject to the successful verification of this hypothesis, this research will continue with heuristics for dynamically adapting the degree of optimism in a simulation during its execution.
