Optimized out-of-order Parallel Discrete Event Simulation using predictions

Parallel Discrete Event Simulation (PDES) enables efficient validation of ESL models on multi-core simulation hosts. Out-of-order PDES is an advanced scheduling technique which allows multiple threads to run in parallel even in different simulation cycles. To maintain simulation semantics and timing accuracy, the compiler performs complex static conflict analysis so that the scheduler can make quick and safe decisions at run time and issue threads early. Often, however, out-of-order scheduling is prevented because of the unknown future behavior of the threads. In this paper, we extend the analysis in order to predict the future of candidate threads. Looking ahead of the current simulation state allows the scheduler to issue more threads in parallel, resulting in significantly reduced simulator run time. Our experimental results show simulation speedup up to 1.92x with only negligible increase in compile time.