Resource optimization in realistic mobile backhaul networks

This paper proposes a resource management system that allows maximizing the traffic which can be served by a given mobile backhaul network. It does so by means of a novel path optimization algorithm that consists of an offline component (based on predicted traffic demand information) and an online component (for excess traffic). As opposed to previous traffic engineering approaches, which usually assume a fluid traffic model, our algorithm explicitly takes flow classification and thus splitting granularity into account. To improve network utilization, we introduce t time slots and allow m traffic matrices, where the number of m depends on the QoS classes to be differentiated. The algorithm is able to compute a highly efficient resource allocation for each QoS class (also taking availability requirements into account) and avoids re-routing of QoS-sensitive traffic. We present a novel Genetic Algorithm for solving the offline problem. Evaluation is done using synthetic as well as real mobile backhaul data from European operators.