Available bit rate services in ATM networks

This thesis is concerned with designing a comprehensive strategy for supporting the ABR traffic class in ATM networks. This class has been defined for reliable data service support in future high speed networks. Since data traffic is relatively delay tolerant, the use of feedback control schemes is appropriate. Among other things, it is necessary that the proposed algorithms provide fair resource (band-width) distribution among competing connections and exhibit good scalability to many different network scenarios.; Although a number of ABR flow control proposals have appeared in the literature, some important and crucial issues still require further investigation. Some of these include operation in dynamic environments, containment of transient congestion effects, and provision of MCR guarantees. Hence there is still a need for designing improved algorithms which address the above issues effectively.; The proposed EDERA (enhanced distributed explicit rate control) flow control algorithm is designed to address the needs for ABR data users in ATM networks. The algorithm complies with the ATM Forum guidelines and implements the MCR-plus-equal-share bandwidth fairness criterion. Effective capacity tracking algorithms coupled with improved congestion indicators allow the scheme to function in dynamic network environments. A simple, robust rate allocation algorithm is used to achieve fair bandwidth distribution. Simulation results show that the scheme performs very well in many scenarios and outperforms the draft EPRCA (enhanced proportional rate control algorithm) proposal in the ATM Forum. Scalability is good and queue sizes are also properly controlled.; To further improve ABR services integration with other traffic types, a complete cell-level scheduling strategy is also proposed. Namely, a hierarchical approach called hierarchical fair queueing (HFQ) is used to ensure that the throughput guarantees for data traffic are met along with the delay guarantees of real-time traffic. Numerical results show that the scheme gives improved delay performance for real-time connections over a non-hierarchical strategy.; The convergence behaviour of the algorithm (under some fixed network conditions) is analyzed, and the conditions for non-oscillatory steady-state rate behaviour are also formulated. Furthermore, bounds on the worst case transient queue buildups are also derived in order to help in the buffer dimensioning problem. However, the results indicate that the complex behaviour of the scheme makes it difficult to derive tight bounds in the generic network case.