Improved Techniques for Optimistic Modeling

Running a simulation on a parallel or distributed computing platform can decrease the execution time and improve real-time responsiveness. Optimistic Parallel Discrete Event Simulation (OPDES) model performance improvements derive from partitioning portions of the overall model to distributed computational nodes and executing them concurrently. Executing concurrently yet independently may cause some model events to execute out of order (causality errors). Optimistic techniques allow these causality errors to occur but detect and correct the errors through the use of rollbacks. During a rollback, computations are restored to a previous model state in which events can execute in proper non-decreasing time order. Optimistic modeling uses optimistic synchronization techniques when developing a parallel or distributed simulation model. This typically involves writing code to handle some aspects of rollback processes. While writing models with rollback code can be straightforward for experienced developers, any errors in the code can be difficult to identify and correct, resulting in longer development cycles. Our solution to these problems, named PROMOTE, has focused on diagnosing, fixing, and reducing rollback and repeatability errors in optimistic models and simulations. Optimistic modeling has the potential to reduce simulation times and increase responsiveness for large system models. Approved for Public Release | 18-MDA-9710 (6 Jul 18)