Optimal intercell coordination for multiple user classes with elastic traffic

We consider the intercell coordination problem between two neighboring cells, assuming that the traffic in the system consists of elastic downlink data flows. In this case, there is an option of completely switching off one base station at certain times, which reduces interference and enables a higher service rate in the neighboring base station. We use a flow level queueing model to describe the evolution of the system based on a symmetric capacity region. Recent results by Verloop and Núñez-Queija show that, assuming a single class of flows for each cell, the stochastically optimal dynamic policy is to have both stations switched on whenever there are users in both cells. In this paper, we consider a system where the two stations are able to provide services to two different classes of users - the near ones and the far ones. In this setting, the stochastic optimality of the Both Stations On policy does not necessarily hold, but it may still be a close-to-optimal policy, at least for minimizing the mean flow delay. We present a systematic method based on the policy improvement algorithm of the theory of the Markov Decision Processes to generate a near-optimal state-dependent resource allocation policy. Our numerical experiments with these two dynamic policies indicate that the Both Stations On policy is, indeed, close to optimal even when there are multiple user classes.