Fair and efficient call routing and admission control algorithms

We consider the problem of call routing and admission control in general topology networks. Given a network, a call request consists of an origin-destination pair, and a bandwidth requirement. For each request, the routing algorithm must find a path in the network satisfying the bandwidth requirement, and the admission control algorithm must decide whether or not to accept the call. If the call is to be accepted, the required bandwidth needs to be allocated on the path selected throughout the duration of the call. The goal of the admission control algorithm is to decide online which calls to accept, without prior knowledge of future calls, so as to maximize the network throughput over time. An additional desired quality is fairness, which means that calls in different classes have the same probability of being accepted. In this paper, we evaluate the performance of online admission control algorithms based on those proposed by several researchers in the context of competitive worst-case analysis. We first show that the behavior of EXP is much more unfair than the behavior of the greedy algorithm along several dimensions. We propose small variations to the admission control algorithm which each greatly reduce the unfairness of EXP on a dense commercial network topology, without affecting the throughput. Additionally, we conducted experiments on the recently proposed KPP algorithm. In our results, a specific variant of this algorithm performs almost as well as EXP, and always achieves better fairness.