Efficiency and fairness improvement for elastic optical networks using reinforcement learning-based traffic prediction

Abstract

The characteristics of traffic in core networks have changed in recent years, with the data flows among network nodes now showing large volumes but also large fluctuations during the day. In such a dynamic environment, the static allocation of network resources by making reservations at the peak rates is wasteful, both in terms of energy consumption and resource efficiency. Periodic reconfiguration of network resources is necessary for the full exploitation of an elastic optical network’s potential. In this paper, we present a resource reconfiguration scheme for an IP over elastic optical core network. The cornerstone of the proposed scheme is a novel traffic prediction mechanism based on reinforcement learning. Using the forecasting mechanism, a heuristic routing and spectrum assignment algorithm was designed to efficiently and fairly allocate network resources periodically. Simulation results show both a reduction of the bandwidth blocking probability and an increase of the fairness index regarding the network resources allocation over corresponding algorithms in the literature.