Efficient calculation of rare event probabilities in Markovian queueing networks

We address the computation of rare event probabilities in Markovian queueing networks with huge or possibly even infinite state spaces. For this purpose, we incorporate ideas from importance sampling simulations into a non-simulative numerical method that approximates transient probabilities based on a dynamical truncation of the state space. A change of measure technique is applied in order to accomplish a guided state space exploration. Numerical results for three different example networks demonstrate the efficiency and accuracy of our method.