Skip to Main Content
Markovian networks of finite capacity queues are widely used models for performance evaluation of systems and networks. Unfortunately, excepted in some specific situations, these models are not tractable analytically. Perfect simulation provides a new technique to sample steady-state and avoids the burn-in time period. When the simulation algorithm stops, the returned state value is in steady-state. We applied this technique first to Markov chain with sparse transition matrix, and to queueing networks with finite capacities and complex routing strategies. The sofware have been developed to validate this simulation approach and applied to in the context of low probability events estimation. The design of the software architecture is presented.