Alternative methods for incorporating non-exponential distributions into stochastic timed Petri nets

Abstract

A natural and compact way to incorporate nonexponential distributions into stochastic Petri nets is described. It allows users to directly specify the nonexponential transitions at the next level without providing the detailed construction; for example, to specify an Erlang distribution, the user only needs to provide the number of stages and the mean of the distribution. The refinement of the transition with a general distribution is performed automatically with a net-independent mechanism. The resulting net is a GSPN that can be solved with standard techniques. The authors also show how to expand conflicting transitions under the race-enabling policy (without the interconnection of places and transitions internal to the expansion of the different transitions), and have identified the different semantics introduced by nonexponential distributions, when a model does or does not use a control place.<>