Search Results

1. V. Timonen, Monte Carlo Simulation of a Banyan Net, Special Assignment, Networking Laboratory, Helsinki University of Technology, 2001 (bib)

   Abstract: Blocking probabilities characterize the grade of service in networks. Often it is computationally ineffective to calculate these probabilities exactly and probabilities are estimated via simulation instead. In this paper we compare estimating of the blocking probabilities in a Banyan net via static Monte Carlo method and via inverse convolution method. The blocking probabilities are rather small in actual networks and thus to achieve a sufficient accuracy, a great number of samples is required with the static Monte Carlo method. Inverse convolution method generates samples directly into the blocking region and thus less samples are required to achieve the same accuracy. However, to operate effectively it necessitates that the load of the system is low and it has a higher memory requirement. The numerical results show that in terms of variance, the inverse convolution method can be even $6700$ times more effective than the static Monte Carlo method.