Using graphical model for network tomography

It is essential to have link-level performance data, such as loss ratio and delay on each link, for our understanding of the dynamic features of a network. One way to achieve this is based on end-to-end measurement to assess the performance feature. Instead of using classical statistics to do the inference, we use the graphical model which has advantages of both efficiency and accuracy. Simulations, based on the network simulator 2 (ns2) were conducted, and data collected were inferred by the expectation-maximization (EM) algorithm, the result is almost identical to the result produced by the maximum likelihood estimator previously proposed.