End-to-end quality of service seen by applications: A statistical learning approach

The focus of this work is on the estimation of quality of service (QoS) parameters seen by an application. Our proposal is based on end-to-end active measurements and statistical learning tools. We propose a methodology where the system is trained during short periods with application flows and probe packets bursts. We learn the relation between QoS parameters seen by the application and the state of the network path, which is inferred from the interarrival times of the probe packets bursts. We obtain a continuous non intrusive QoS monitoring methodology. We propose two different estimators of the network state and analyze them using Nadaraya–Watson estimator and Support Vector Machines (SVM) for regression. We compare these approaches and we show results obtained by simulations and by measures in operational networks.