Network tomography: an iterative bayesian analysis
Résumé
The origin-destination (OD) traffic matrix in a network is useful for network configuration and management, traffic engineering and also for pricing. This matrix cannot usually be measured directly, but less informative repeated link counts can be obtained easily. The estimation of
the traffic matrix is often based on a statistical model for the OD flows. Generally, the OD counts are supposed to be independent
from one time measurement period to another. On the contrary we suppose in the present work that the OD flows are Markov modulated and we generalize the problem of estimating the OD traffic matrix to this case. We make the most of the information of dependence between the successive counts to improve the estimation of the traffic matrix. For that purpose the estimation is performed iteratively. (i) First the expected value of each OD flow is estimated on the basis of the observed link counts. (ii) Then, for each OD flow, the successive Markov regimes of that flow are estimated from the estimated successive counts; the updated estimate of the regime is then provided to step (i) in form of a Bayesian prior over the traffic matrix.The process is iterated until convergence