QRP06-6: Estimation of Flow Lengths from Sampled Traffic

In this paper, we consider the problem of nonparametric estimation of the original length of traffic flows, based on sampled flow data. The proposed approach is a two-step one. In the first stage, the flow length distribution is estimated by an expectation-maximization (EM) algorithm that in addition provides an estimate for the number of active flows in the link. In the second stage, two estimators are derived for the original flow length, using information from the posterior distribution previously obtained. The proposed approach is illustrated on a number of synthetic and real data sets.