Fast MPLS Network Optimisation using Machine Learning

This paper presents the application of a machine learning technique (neural networks) to the problem of optimising the traffic flows (label switched path placement) within telecommunications networks using multi-protocol label switching (MPLS). It is shown here that the machine learning technique achieves very fast computation of accurate solutions to the problem of placing label switched paths in order to maximise the performance of the network. Results are presented which compare the accuracy of the neural network solutions with the optimal solutions produced by the standard mixed integer linear programming technique, and the time taken to produce the solutions. As well, network simulation results are presented which show the improvement in network performance, in terms of packet loss, that can be achieved when using such a technique.