A Recurrent Neural Network approach to traffic matrix tracking using partial measurements

Traffic matrix allows network engineers and managers to solve problems in design, routing, configuration debugging, monitoring and pricing. Direct measurement of traffic matrix is usually not implemented because it is too expensive. Instead, we can easily measure the loads on every link and inference traffic matrix by using network tomography technology. In this paper, we develop a novel network tomography approach using recurrent neural network (RNN) that track origin-destination traffic matrix based on partial measurements without any prior information. Our RNN approach not only allows us to estimate traffic matrix and can also be used to predict traffic. Using real data collected from a Ailebant network, we illustrate that our proposed approach can achieve lower errors than general Gravity model prior.