An Approximate Approach to End-to-End Traffic in Communication Networks

All end-to-end traffic in a network constructs Traffic matrix (TM) which reveals all traffic traversing the whole networks. In this paper, we investigate TM estimation problem in large-scale backbone networks. We propose an accurate approach to estimate it, based on the Recurrent multilayer perceptron (RMLP) which has a powerful ability of modeling. According to constraint relations between link loads and TM, we introduce their temporal and spatial correlation to modify the traditional RMLP and establish our models. And the outputs of our models take into account the constraints that TM itself is satisfied with. Trained with input-output data pairs, our models can learn and grasp all kinds of characteristics of TM and all weight parameters are determined. Finally, we use the real data to validate our method. Simulation results show that our method can perform the accurate and fast estimation of TM very well.