Robust adaptive capacity allocation algorithm

We consider the capacity allocation problem arising, e.g. in the capacity management of a VP in an ATM network, where the occupancy of the VP is periodically observed and based on this information new capacity allocation is made for the beginning period. We show that with a simple stochastic allocation policy it is possible, without any knowledge of the intensity of the Poissonian call arrival process, to obtain a long term blocking probability which for any practical purposes has exactly a prescribed constant value. Thus, the algorithm is inherently robust. Moreover, since the algorithm is adaptive in the sense that the achieved blocking probability is practically independent on the arrival intensity, it is plausible that the same blocking probability prevails even when the intensity is a slowly varying function of time.