__Search Results__

*P. Lassila and J. Virtamo*, **Using Gibbs Sampler in Simulating Multiservice Loss Networks**, COST257, TD(98)19, 1998 (bib)**Abstract:** In this article we consider the problem of calculating the blocking probabilities of calls in a multiservice network by using simulation. The traditional approaches suffer from the fact that the state space explosion inherent in the system causes their efficiency to decrease correspondingly. We develop a method that alleviates the effect of the state space explosion. The method is based on using the so called Gibbs sampler to generate a Markov chain with the desired stationary distribution. In particular, by making an additional ``virtual'' step from each state and calculating the expected contribution from this step analytically, we are able to collect information from a subset of the whole state space for each generated sample. This leads to a smaller variance of the estimate for a given computational effort.