| Virtual Private Services (VPN) are an important service offering for any service provider. It can be justified that in a metro environment some networks will carry only such traffic. In addition, metro networks are characterized by small link spans, which make the propagation delays negligible. In this environment, this dissertation proposes and investigates a service aware network control platform for packet switched networks that is technology agnostic.; The basic premises of our work are that VPN are provider based, have well defined QoS behavior specified by the long term loss, and have a degree of user-controllability.; The dissertation proposes a number of aggregate traffic loss estimators, based on the theory of large deviations. Our assumptions are that the buffers at each multiplexor are finite, and therefore the loss ratios are computed. It has been noted that large deviations based estimators can be estimated in two main ways: (1) assuming a traffic model and by computing the loss using that model (2) estimating the moment generating function directly.; The work compares, through a large number of simulations, the performance of the proposed estimators in both cases. The results show that under the Gaussian model the model-based estimation is very accurate. This conclusion is important because the model-based estimator has better computational complexity when compared with the modeless estimator. Therefore, it is better to be used in practice.; The dissertation proposes a new method for atomically admitting VPN service instances in the network. Dubbed Service Admission Control, this generalizes the classical admission control methods by using an atomic operation over a set of class of service instances. Furthermore, we propose and investigate a method for dynamic loss control that is of use when the traffic injected in the network has long timescale changes. |
