Classification-Model Applied to Routing Problem in Flexible-Grid Optical Networks

In recent years, machine learning algorithms have been widely used in optical networks to solve complex problems such as routing, resource allocation, among others. In routing, modulation and spectrum allocation (RMSA) problems, machine learning algorithms can be used to learn patterns in historical data and find good solutions without having to explore all existing solutions. In this paper, we propose an algorithm based on a classification model to solve the routing problem in elastic optical networks. This algorithm predicts the route according to the call request information and the state of the network links. The dataset used to train the proposal is obtained through a dynamic routing algorithm. With this dataset, two versions of the proposal are evaluated with different sets of routes according to the frequency distribution of these routes. Three network topologies are used to evaluate the routing algorithms: six-node, NSFNET and European optical network. The results are compared with two other routing algorithms: Yen’s algorithm (k shortest routes) and the spectrum continuity based shortest path (SCSP) algorithm. This last algorithm is used to train our proposal. Our proposal outperformed the Yen’s algorithm in the three network topologies in terms of blocking probability. When compared to the SCSP algorithm, our proposal obtained an average performance gain of 15% and 25% in the six-node and NSFNET network topologies, respectively. In the European network topology, our proposal achieved an average performance gain at the lowest network loads of 23.19%. In all network topologies considered, our proposal reduced the time spent to find the RMSA solution compared to the SCSP algorithm.