Title of Thesis
ADAPTIVE FEEDBACK SCHEME FOR ATM TRAFFIC CONTROL WITH EMPHASIS ON ABR TRAFFIC CONTROL

Summary:
Future high speed networks are expected to use the Asynchronous Transfer Mode (ATM) in which the information is transmitted using short fixed-size cells consisting of 48 bytes of payload and 5 bytes of header. It is also expected that the future ATM networks will be dominated by bursty traffic. Unpredictable statistical fluctuations and burstiness of traffic sources make it difficult to guarantee quality of service and high utilization of the network resources using source traffic parameters declared by the users, such as average rate and burst duration. Gains due to statistical resource utilization come at risk of potential congestion when many applications compete for network resources. The problem is especially difficult during periods of heavy load, particularly if the traffic demands can not be predicted in advance as in the case of data traffic. Hence congestion control is critical in ATM networks and although only a part of the traffic management, is its most essential aspect. It helps ensure efficient and fair operation of networks in spite of constantly varying demand. The thesis presents an effort to design a scheme able to remarkably improve, in respect to existing schemes for traffic management/congestion control in ATM networks, quality of service for the delay sensitive Variable Bit Rate traffic by reducing its queuing delay and queuing delay variance. For this purpose an explicit rate-based closed loop congestion control scheme has been developed. The scheme does not dictate a particular switch architecture. The existence of lower priority (controllable) and higher priority (uncontrollable) traffic groups in the network is assumed. The controllable sources’ allowed cell rates (ACR) have been dynamically shaped by the explicit feedback messages based on the instantaneous queue length in the ATM switch output buffer. As the dynamics and parameters of the traffic flow process are poorly known, the adaptive feedback gain depends not only on the parameters estimates, but is also modulated by the parameters current uncertainty. Therefore, the control algorithm implemented here expresses its dual nature by simultaneously performing two functions: perpetual parameters estimation i.e. “learning” and the process control. The requirements tested include: utilization, queuing delay, delay variance and queue size The results of a simulation study are presented which suggest that the scheme can provide for higher priority traffic considerably shorter queuing delay, delay variance and average queue size, particularly on shorter links, while retaining utilization unchanged.