Importance sampling, jump distributions and event-time distributions

Abstract

Two different methods have been proposed for performing asymptotically optimal simulation to obtain the statistics of buffer overflows in queuing networks, with both using large deviations and importance sampling. In the first, based on heuristic arguments, the distributions of interarrival and virtual service times are analyzed to find the simulation system. In the second, it is the distribution of jumps occurring in a Markov chain that is examined. In the present work, the authors show that the approaches will produce identical fast simulation systems for an arbitrary GI/GI/1 queue. Show more