| The market opportunity is tremendous for cost-effective and flexible MPLS VPN services overall. QoS control and guarantees for provided VPN services is a critical issue to be addressed such that service providers can maximize their profits and keep their competitive edge in the market. The focus of this thesis is to solve the quantitative QoS control problem in the provisioning of QoS enabled VPN services across the core IP/MPLS networks, especially in the packet loss case.;A QoS control framework is proposed in a control-theoretic perspective, where the open-loop and closed-loop control techniques are combined together. As one of the key components of the solution devised, a traffic monitor is first designed and implemented in Java. The packet loss estimators calculate the packet loss probability based on traffic measurements, where two estimation formulae are derived by applying the large deviation theory. A reactive estimator is then constructed on-line, which surpasses the original two estimators with the capacity of adapting itself to model imprecisions. Furthermore, a Kalman filter is employed to optimally estimate the packet rate mean and variance. The resulting reactive estimator is applied to the service admission control for admitting an appropriate amount of services and traffic flows into the core network.;The focus of the thesis is on the dynamic packet loss probability control for VPN services. Preliminary controllers are first designed and evaluated for the analytic loss model. System identification technology is then applied to identify the dynamic loss model, where both of the offline and online prediction error methods are explored. Controlling the identified model is examined by both classical and modern controller design technologies. Since the identified linear model contains time-varying and norm-bounded parameters, a Linear Matrix Inequality (LMI) approach for designing robust and optimal controllers is also studied. The performance of the packet loss control system is evaluated with real measured traffic from the live NCIT*net2 network. |
