Performance Petri net analysis of communications protocol software by delay-equivalent aggregation

Performance analysis of Petri net models is limited by state explosion in the underlying Markovian model. To overcome this problem, an iterative approximate technique is obtained, using a number of auxiliary models, each of much lower state complexity. It is demonstrated on a substantial model which represents a parallel implementation of two layers of protocols for data communications. The model represents ten separate software tasks and their interactions via rendezvous, and is based on a testbed implementation in the laboratory. Submodels can be constructed in various ways, and this is illustrated with four different decompositions. Their state space complexity, solution time and solution accuracy are evaluated