On the solution of GSPN reward models

Abstract

We extend the basic GSPN (generalized stochastic Petri net) model to the GSPN-reward model. This allows the concise specification of both the underlying stochastic process and the rewards attached to the states and the transitions of the stochastic process. The classical method for the steady-state solution of GSPN models, based on the correspondence between GSPNs and continuous-time Markov chains (CTMCs), is compared with a method based on discrete-time Markov chains (DTMCs) previously judged poor. We show that there are GSPNs where the DTMC-based method performs better than the classical method (and others where it performs worse). Finally, we discuss how to perform parametric sensitivity analysis of the measures computed from a GSPN using either solution method.