A Kalman filtering approach to traffic flow estimation in computer networks

In this paper, data-based estimation of traffic flows in computer networks is dealt with. The knowledge of the traffic matrix, containing the values of flows among the nodes in a computer network, could be very useful in many aspects of network management, like capacity planning, traffic load distribution and anomaly detection. The proposed technique exploits a new observer based on Kalman filtering, suitably modified to better fit into the specific application framework. Experimental results on a Metropolitan Area Network in Milan (Italy) show that the proposed method may obtain very low error rates even in presence of severe technological constraints.