In this dissertation, new and efficient traffic management schemes based on fuzzy logic for ABR services in ATM have been suggested. End-to-end closed loop feedback control of the source rates has been adopted for ABR traffic control method. The cell-transmission-rates are regulated according to the explicit rate values calculated from the network. Though many control algorithms have been proposed so far, most of them can lead to unfairness problems unless their control parameters are set properly. In addition, large time-delay incurred in the feedback path and the statistical variation in the link capacity at the ATM node can cause buffer saturation or unfairness. The control algorithm should not interfere with delay-sensitive traffic to assure given QoS and minimal cell loss.
We propose two adaptive fuzzy explicit rate allocation algorithms which are composed of linguistic rules and fuzzy inference engine for deciding explicit rates of all sources. The parameters of the fuzzy rules are adapted to minimize the given performance index in both cases. The controllers estimate future queue length from the previous input loads while it monitors current buffer length, degree of fairness, link-utilization and available bandwidth. The level of network congestion is monitored through the occupancy q of the ABR buffer, with the control target being set at a certain threshold $q_d$.
Between two algorithms, one mechanism is based on a heuristic algorithm. The performance index is composed of difference between q and $q_d$ and fairness degree of each connection allocated the available bandwidth. The other mechanism is based on the projection algorithm. In this, the stability of the overall control system is analyzed and proved. Theses new traffic management approaches have both adaptive and learning behaviors without any assumption on the traffic pattern. The performances of the proposed control mechanisms are validated for various ABR and VBR demand patterns. It is noticeable that the proposed schemes can operate by only a few parameters and simple learning algorithms without any assumption on the traffic characteristics. The overall control system is analytically proved to be stable even in very long propagation delays. Simulation results show that in various tests (LAN and WAN) the desired characteristics (max-min fairness and maximum link-utilization) are obtained when the proposed scheme is used.
Finally, we have applied the proposed mechanism to ATM over satellite networks. In the satellite environment, we should consider some obstacles which do not matter in terrestrial environment. The first thing is the satellite link-delay of 125 $ms$ between a satellite and ground terminals. The next is the multiple access scheme for utilizing the limited network resource (power and bandwidth). Traffic control schemes for the terrestrial networks should be modified to be successful in the satellite environment. The concept of virtual source and virtual destination is implemented to reduce the effect of time-delay, and the anti-windup scheme is designed to prevent buffer saturation. In the case of satellite-ATM networks which have extremely long propagation delays and limited resources such as bandwidth and power, our scheme showed superior performance to the well-known Explicit Rate Indication for Congestion Avoidance (ERICA) algorithm and ERICA+ algorithm.
