| To generate profit and support high volumes of traffic, IP backbone networks are required to provide efficient scalable end-to-end Quality of Service (QoS) for a variety of telecommunication services. Scalability is achieved by using class-based QoS provisioning approaches such as Differentiated Services (DiffServ), while efficiency requires the appropriate use of traffic engineering mechanisms such as those provided by Multi-Protocol Label Switching (MPLS). This thesis presents a delay-margin based Traffic Engineering (TE) approach to provide end-to-end QoS in MPLS networks using DiffServ at the link level. The TE, combines mapping flows to routes, mapping routes to QoS classes, and the definition of each class delay. The thesis formulates traffic engineering as a nonlinear optimization problem that reflects the inter-class and inter-link dependency introduced by end-to-end QoS requirements of flows, and their aggregation into DiffServ classes. Three algorithms are used to provide a solution to the problem: The first two, centralized offline route configuration, and link-class delay assignment, operate in the convex areas of the feasible region to consecutively reduce the objective function using a per-link per-class decomposition of the objective function gradient. The third one is a heuristic that promotes/demotes connections at different links in order to deal with concave areas that may be produced by a trunk route usage of more than one class on a given link. Approximations of the three algorithms suitable for online distributed TE operation are derived. Simulation is used to show that the proposed approach can increase the number of users while maintaining end-to-end QoS requirements.To estimate the queue length and delay distributions versus changes in the traffic characteristic, the available capacity variation, the loading, and the change in queue weights that are introduced by the proposed TE approach, a multi-scale queue performance analysis technique is proposed. At each scale, the queue weights or priority dependencies are exploited to convert the multi-queue problem into a set of single-queue problems. The core of the analysis is in using Variable Service rate Multi-Scale Queuing (VS-MSQ) to estimate the multi-scale capacity available to each queue. The thesis shows the hierarchy of this estimation and the dependency of the each queue variable capacity on the unused capacity of the other queues and their weights or priority. Simulation and analytical results on the queue length and delay survivor functions are in a good agreement. |
