Load balancing in cellular networks using first policy iteration

This paper discusses load balancing in cellular networks. We compare the performance of both static and dynamic allocation policies for a simple model of two base stations with overlapping cells. In particular, we investigate the performance of the policy obtained by the first step of the policy iteration algorithm (FPI policy). When starting with a static policy, the two base stations can be modeled independently as Erlang loss systems, for which we can easily determine the relative costs in each state. This makes the first step of the policy iteration algorithm of low complexity, and therefore applicable for large instances. The idea is to approximate the optimal policy with the FPI policy. Numerical experiments for small instances provide a deeper understanding of the FPI policy. As it turns out, the optimal static policy not necessarily leads to the best performance of the FPI policy. When choosing an appropriate static policy, we show that the FPI policy is extremely close to the optimal policy, and performs much better than the other considered policies. This strengthens our belief in the practical relevance of the FPI policy.