Marking dependency in non-Markovian stochastic Petri nets

Abstract

Marking dependency is a powerful tool that allows different firing time distributions to be associated with a stochastic Petri net transition, depending on the marking. Through this feature, the modeler can easily and compactly represent advanced properties and behaviors of the system. While a semantics and specific solution techniques have been provided for generalized stochastic Petri nets thus covering homogeneous Markovian aspects, in the non-homogeneous/non-Markovian case marking dependency still needs to be investigated. To fill this gap, this paper provides a formalization of marking dependent semantics in non-Markovian stochastic Petri nets (NMSPNs) and a solution technique, based on phase type distributions and Kronecker algebra, able to deal with such a feature allowing both transient and steady-state analyses. To motivate the actual need of marking dependency in NMSPN modeling and to demonstrate the potential of such a feature as well as the validity of the proposed solution technique a case study on a multi-core CPU system with power management facilities is explored.