Evaluating Computer-Communication Systems using Infinite-State Stochastic Petri Nets

For modelling and evaluating the performance of modern computer communication systems, quasi-birth-and-death models (QBDs) can often be used. For such QBDs, a variety of highly efficient numerical solution algorithms have been developed. For models of practically relevant size though, the specification of QBDs at the state level is cumbersome or practically infeasible. To overcome this problem, we advocate the use of a high-level model specification technique that hides the QBD from the modeller. To that purpose we propose a formalism based on an extended class of stochastic Petri nets (SPNs), the so-called infinite-state SPNs (iSPNs). Secondly, we present the software tool Spn2Mgm, which provides support for system-oriented modelers to specify and solve iSPN performance models. The derivation of the underlying QBD, its solution via one of the five supported methods and the computation of relevant performance measures is completely transparent for the tool user. The third part of this paper is devoted to a comparison study of the most efficient algorithms currently available for the steady-state solution of QBD models. As Spn2Mgm provides an efficient implementation of these algorithms, we can easily perform a comparison based on a very large case study. In our comparison, we both focus on the computational speed and the accuracy of the algorithms.