Timed Petri nets and prediction to improve the Chandy–Misra conservative-distributed simulation

Abstract

The main contribution of this paper shows that distributed simulation of timed Petri nets (TPN) can take advantage of their structure to obtain a significant lookahead which is usually difficult to compute with other models. In this paper, we introduce a conservative-distributed simulation with a reduced number of control messages and without deadlock resolution. This approach is based on a part of optimism computed on the prediction time each logical process can determine for its advancement. Obviously this prediction time must be computed easily according to the structure of the simulated logical process. Timed Petri nets meet these requirements and we use their structure to evaluate the depth of the prediction. In conservative-distributed simulation, it is known that the deeper the prediction, the better the efficiency of the simulation. We present a method we have devised based on channel time prediction. We compare its performance to the Chandy–Misra method and to some related Petri nets approaches (Chiola). Experiments carried out on Sun stations show that there is more parallelism and a reduced number of null messages in the cases of deadlock avoidance. Moreover, considering deadlock detection and resolution technique we observe that in many cases no deadlock occurs with less control messages.