Global State Monitoring in Optimization of Parallel Event–Driven Simulation

Abstract

The paper presents results of experimental work in the field of optimization of parallel, event-driven simulation via application of global state monitoring. Discrete event simulation is a well known technique used for modelling and simulating complex parallel systems. Parallel simulation employs multiple simulated event queues processed in parallel. Absence of proper synchronization between parallel queues can cause massive simulation rollbacks, which slow down the simulation process. We propose a new method for parallel simulation control with monitoring of global program states, which prevent excessive number of rollbacks. Every queue process reports its local progress to a global synchronizer which monitors the global simulation state as timestamps of recently processed events in distributed queues. Based on this state the synchronizer checks the progress of simulation and sends signals limiting progress in too advanced queues. This control is done asynchronously, and thus it has small time overheads in case of correct simulation order. The paper describes the proposed approach and the experimental results of its basic program implementation.