HP Labs Technical Reports

Modeling A Fibre Channel Switch with Stochastic Petri Nets

Kotov, Vadim; Cherkasova, Ludmila; Rokicki, Tomas; Ciardo, Gianfranco

HPL-94-107

Keyword(s):

Abstract: Discrete-event simulation is the most common method of performance analysis because it allows the modeling of arbitrarily complex systems with a minimum of assumptions. Unfortunately, programming errors, improper statistical methods, and runs that are too short can yield inaccurate results. The numerical analysis of Markovian models can supplement discrete-event simulation, but only if the system can be simplified to reduce its state space to a manageable level. The approximations introduced through such simplification can also cause a significant amount of error. Recent advances in stochastic Petri nets and Markov chain analysis, and the availability of fast workstations with large amounts of memory, allow the numerical solution of much larger and more complex Markov models. In this report we discuss the analysis of a complex packet switch through both discrete-event simulation and fixed-point approximate numerical solution of Markovian stochastic Petri net models. We compare the results obtained from them and identify the error in the various approximations. We conclude that stochastic Petri net analysis can be a useful technique even for complex systems.

Back to Index