| Quality of Service (QoS) guarantees on a per node, per aggregate/class basis have been proposed to achieve scalability. However, traffic at an aggregate level is unpredictable and sometimes has unknown statistical characteristics. Thus, a conventional approach for quantitative QoS guarantees such as static bandwidth allocation is not only ineffective but also has serious limitations because it assumes some specific traffic model and needs traffic parameterization.; This dissertation considers the use of Adaptive Bandwidth Control (ABC) algorithms for a quantitative packet loss guarantee to aggregate traffic. ABC starts with some initial amount of bandwidth allocation to the queue and adjusts it over time, based on on-line measurements of system states to ensure that the allocated bandwidth is just enough to attain the specified QoS requirements. Consequently, no a priori traffic information is required, making ABC more suitable for efficient aggregate QoS provisioning. We present three ABC algorithms. The first algorithm is based on integral control using loss feedback. The second algorithm is based on fuzzy control which maintains the average queue length at an appropriate target to achieve the desired loss rate. The last algorithm is an augmentation of the above fuzzy control, called the A-Fuzzy control, which uses packet loss information in addition to the average queue length feedback to improve the performance. An extensive simulation study on these ABC algorithms under different traffic types, non-stationary traffic conditions, and various system parameter configurations clearly shows that ABC, especially the A-Fuzzy control, is indeed a viable alternative and improvement to static allocation. Additionally, we consider the admission control problem for ABC algorithms under a limited link capacity, which is complicated by the time-varying allocated bandwidth. We show that practical admission controls for the A-Fuzzy controller can be constructed and the controller can operate without noticeable performance degradation. |
