Network tomography is a problem of estimating network properties such as the packet loss rates of links using available packets. There are two kinds of methods to measure packets: active and passive. An active measurement specifies link information (paths) of packets a priori while a passive measurement gets only the origins and destinations of packets. The conventional methods for estimating the packet loss rate of each link, one of the network tomography problems, utilize only active measurements because passive measurements have no link information. We propose a method to utilize passive measurements also. The method regards the link information in the passive measurements as latent variables and estimates the variables and the loss rates of links simultaneously in the framework of Bayesian inference. We show through numerical experiments that our method outperforms the conventional algorithm with only active measurements in the estimation accuracy.
展开▼