A structure based decomposition approach for GSPN

We present a decomposition approach for the solution of large generalised stochastic Petri nets using p-invariants to identify submodels. Due to the structure of interfaces between submodels, types of interactions are defined. An interaction graph is derived, in which the information flow among submodels is represented by the direction of arcs. According to solution quality and efficiency, the information graph is refined to get a suitable partition of the model. The submodels of this partition are aggregated in a special way to preserve the interface structure and its throughput. Combination and solution of the aggregates results in a second step to include the interaction influence into interface substitutions. The isolated solution of the expanded submodels with interface substitution results in approximations of the marking probabilities. The solution process may be iterative, depending on the interaction types among submodels in the solution partition. As all steps of the evaluation are based on model structure, the derivation of the reachability set and the corresponding Markov process is avoided