Research On Self-Adaptive Differentiated Services For IP Network

With the development of Internet and communication technology, various new network services become available over the past decade. The "best-effort" service cannot satisfy the requirements of new network applications, and the research of control mechanisms and algorithms for quality of service (QoS) are the focus. As for its characteristic of simplification and high scalability, the differentiated services (DiffServ) model becomes one of the primary architectures for the QoS.Motivated by improving the self-adaptive ability of DiffServ, the control mechanisms of DiffServ is studied in this dissertation. The main research works and contributions are listed as following:The control mechanisms on the edges of DiffServ, including marking algorithms, the shaper, meter algorithms and active queue management (AQM) algorithms, are studied in this dissertation. Firstly, in order to solving the fairness problem of bandwidth sharing, a new marking algorithm called adaptive fair bandwidth traffic conditioner marker (afbTCM) is proposed. By allocating the available bandwidth among different service classes proportionally, afbTCM can archive high performance in the fairness of bandwidth sharing. Secondly, aiming to describe the performance of the finite storage shaper (FSS) by algebra, based on network calculus, a generalize model of FSS build by min-plus algebra and the performance parameters are presented. Thirdly, according to the weakness of the computational complexity of "long-range dependent" traffic prediction models, two novel traffic prediction models namely minimums mean square error predictor (MMSEP) and normalize minimums mean square error predictor (NMSEP), which are based minimum mean square error, are put forward. These two models can achieve high estimated accuracy while keeping simplification in computation, and can be used both as online internet traffic prediction models and as algorithms for the meter on the edge of DiffServ. Finally, the weakness of the PI algorithm for the slow response to the changing of traffic is noticed, and a novel AQM algorithm named adaptive parameters control (APC) is proposed by adding adaptive mechanism to proportional controller for the sake of making up the weakness of PI algorithm. Furthermore, APC and NMSEP are combined together to propose an AQM algorithm namely rate prediction based AQM (RPB), which offers more precise proportional lossrate, faster response, higher link utilization and less loss ratio when compared with APC.How to extend end-to-end AQM algorithm to support QoS is also studied. Based on APC, this dissertation proposes a new AQM mechanism called weighted APC (WAPC) to support proportional loss rate differentiation. Compared with RIO and WRED, WAPC is easier in configuring and is more scalable while keeping more precise loss rate.Arrival rate guarantee mechanism in the end-to-end DiffServ is discussed. A novel arrival rate guarantee mechanism called rate ensure mechanism (REM) is presented. By labeling packet's header on target rate, REM can provide approximated arrival rate guarantee in complexity network environment.Based on the works mentioned above, the deployment of edges in DiffServ is discussed. A new edged configure architecture is brought forward. By reconfiguring the location of shaper and by introducing REM, this architecture can improve the "self-adaptive" ability, robustness and scalability of DiffServ. In the end, how to deploy the algorithms and mechanism presented in this dissertation, whose intent is to support numbers of service classes and to ensure the DififServ's performance in various environment, are also discussed.