Measurement based quality of service control for communications networks

This thesis presents a purely empirical approach for the estimation of effective bandwidth of aggregated traffic flows, suitably accurate for use within traffic engineering based network performance optimisation strategies, such as QoS-aware network planning and admission control. The effective bandwidth estimation algorithm proposed, is an off-line approach, which uses a packet trace collected from the network and processes it through a FIFO queue. Based on queue buffer analysis, an appropriate queue service rate for the FIFO queue that ensures stated QoS targets on packet delay are maintained is chosen to represent the effective bandwidth estimation of the packet trace. The algorithm is evaluated in two traffic engineering based scenarios. The first is a cost efficient process for establishing a network-wide demand matrix from available network accounting data. We use effective bandwidth coefficients to enhance such a demand matrix for QoS-aware network planning. This process is demonstrated to be reasonably accurate for the scenarios we consider, and importantly at a fraction of the operational and capital expenditure that would be incurred by the deployment and operation of a direct measurement approach. Secondly, we propose two empirical effective bandwidth estimation based IPTV focussed admission control algorithms suitable for optimising utilisation of bandwidth while maintaining QoS targets on packet delay of admitted traffic. The first algorithm employs a simple evaluation of whether there is sufficient bandwidth available to ensure, with an appropriate degree of confidence, that QoS targets will not be violated if a requested ow is admitted. The second algorithm utilises information within the admission control process relating to the cost, duration and request frequency of specific IPTV content to prioritise flows that maximise revenue