One of the key issues in the design of the future ATM-based broadband integrated services digital network (B-ISDN) is the guarantee of quality of service (QoS) requested by users. Specifically, the QoS guarantee includes cell delay variation (CDV), maximum cell transfer delay (max CTD) and cell loss ratio (CLR) guarantees. The CDV and max CTD can be guaranteed by limiting the buffer size in ATM multiplexers. To guarantee the CLR, two major approaches have been adopted by researchers in the past. The first approach focuses on the measurement of the CLR in the network in real time. Since CLR can be of the order of l0^-6 to l0^-9, direct measurement is unlikely to be accurate. The second approach is based on the characterization of the input traffic streams using some parameters and assumed...[ Read more ]

One of the key issues in the design of the future ATM-based broadband integrated services digital network (B-ISDN) is the guarantee of quality of service (QoS) requested by users. Specifically, the QoS guarantee includes cell delay variation (CDV), maximum cell transfer delay (max CTD) and cell loss ratio (CLR) guarantees. The CDV and max CTD can be guaranteed by limiting the buffer size in ATM multiplexers. To guarantee the CLR, two major approaches have been adopted by researchers in the past. The first approach focuses on the measurement of the CLR in the network in real time. Since CLR can be of the order of l0^-6 to l0^-9, direct measurement is unlikely to be accurate. The second approach is based on the characterization of the input traffic streams using some parameters and assumed traffic distributions. This model-based approach fails to give a practical tool to be used in network management since it is based on the knowledge of some unpredictable traffic parameters such as burst length distribution, silence length distribution, and more generally the auto-correlations of the different traffic streams. Moreover, the computation times and the complexity of the analytical models increase exponentially with the number of traffic streams.

In this thesis, we investigate two kinds of problems arising in ATM traffic controls. The first part of this work focuses on the fast estimation of the CLR in ATM multi-plexers when traffic characteristics such as burst length distributions are not known. We use an adaptive fuzzy system to predict the CLR for a large-sized system (e.g., large buffer size or large link capacity) based on the values of CLR obtained from small-sized systems and the asymptotic behavior for very large-sized systems. The CLR in a small-sized system is relatively large and can easily be obtained either by quick simulation or by real-time measurement. Extensive experiments have shown that the fuzzy approximation can accurately estimate the CLR and does not depend on any queueing model. In addition, the computation times have been shown to beii negligibly small. The second part of the thesis is dedicated to the problem of QoS guarantee for real-time Variable Bit Rate (VBR) video in ATM networks. VBR video coders are known to produce variable bit-rate output with unpredictable statistical properties. It is very difficult to characterize real time VBR video traffic. Cells will be discarded/delayed by the usage parameter controller if the actual video output does not stay within their negotiated traffic parameters. This cell loss/delay will degrade the QoS. By introducing a sufficiently small delay, a fuzzy-based rate control scheme is proposed to ensure that the video coder output stream always conforms to the pre-declared traffic parameters. The admission control algorithm can simply be based on these parameters to determine if the connection can be handled without violating the QoS constraints. Furthermore, the control scheme can reduce the fluctuations of both the traffic bit rate and the image quality.