Estimation of correlated flows from link measurements

The real network traffic characteristics makes it difficult to estimate origin-destination flows only from statistics for routers´ interfaces. The presented approach, based on passively measured link traffic statistical properties, estimates the flows, allowing for flow interdependence and seasonality. Rescaled profile of the total daily traffic in the network is used to compensate deterministic component of flows, before the proper estimation takes place. For adequate identification of the residual flows, it is assumed that they are mixtures of a number of independent random processes, whose parameters are to be identified along with the mixing coefficients. The results obtained for real traffic in Abilene network are promising, and indicate that data preprocessing and application of correct optimization routines are of importance for further research.