In: 5th International Workshop on Petri Nets and Performance Models, Toulouse (F) 19.-22. October 1993, pages 238-247. 1993.
Abstract: In fault-tolerant systems and in communication networks rare events, like system crashes or buffer overflows, are a great importance to predict system behaviour. In principle stochastic Petri nets are well-suited to model such systems using Markovian and simulation techniques. But unfortunately both Markovian as well as simulation techniques cause crucial problems for modeling rare events. The simulation of rare events gets infeasible because rare events are also rare in simulation experiments, and therefore conventional simulation methods yield no reasonable statistical results. In this paper it is shown how variance reduction techniques can be used simulating rare events in stochastic Petri nets. The application of fast simulation methods are illustrated for several classes of stochastic Petri nets: GSPN's, deterministic and stochastic Petri nets, and Petri nets with generally distributed firing times are studied. For the latter case different state change policies as resampling and age memory are distinguished. Some experiments show the usefulness of this approach