| This thesis presents a model-based feedback control approach for network Quality-of-Service (QoS), that achieves accurate control, better resource utilization, and faster response to network and traffic changes than other prevalent techniques. It is a significant departure from the traditional approaches in providing network QoS, that either calculate and provision a priori the required network resources in a feed forward way to meet certain QoS requirements, or use ad-hoc adaptive solutions, that need significant tuning and calibration efforts. Models are used as the basis for performance control of the network problems under study. We apply our control approach at several points in the network, starting with differentiated packet marking, where we introduce a new marking algorithm for assured services based on a feedback model of TCP congestion control. Then, we develop models for per-hop QoS using statistical modeling techniques and build model-based controllers for QoS over network nodes using these models. We further use control-theoretic design techniques to build QoS-controlled FIFO-based network services. These services are reservation-less and self-controlled, so they are capable of tracking reference QoS in a way that could not be easily achieved using operator-controlled or other commonly-used techniques. The problems and applications studied in the thesis demonstrate the effectiveness and the importance of using new and versatile methods such as statistical analysis, design of experiments and control-theory in studying network QoS. We evaluate the new control approach using both simulation and experimentation on a Linux-based network testbed showing its correct functionality as well as comparing it with other available QoS schemes. We also present a two-tier QoS Gateway architecture that works at the first mile to facilitate the use of these controls for QoS-sensitive applications and end-user devices and enables better deployment and utilization of QoS frameworks in general. The overall research in this thesis will provide a better understanding of QoS modeling and control problems and enable wide QoS deployment. |
