Unicast-based inference of network link delay distributions using mixed finite mixture models

As telecommunication networks grow larger and more complex, it is important to monitor internal link characteristics for operation, monitoring, and diagnosis purposes. Since router link monitoring is not practical due to high communication overhead, there has been considerable interest in monitoring from edge (end-to-end) observations. This paper focuses on the estimation of internal link delay distributions from edge measurements. Discrete and continuous delay models are introduced and we propose a new mixed finite mixture model for link delay probability density functions (p.d.f.). When collecting end-to-end unicast packet delays from edge nodes, we are able to estimate internal link delay distributions using the EM algorithm. Simulation results are given to illustrate our method.