Abstract
ATM has been designed to support various classes of multimedia traffic with different bit rates and QoS requirements. Recently, the demand for higher transmission capacity and higher bandwidth to the users by multiservice and multimedia applications has put ATM networks at the forefront of several competing switch technologies. In ATM networks, due to unpredictable fluctuations and burstiness of traffic flow, congestion can occur frequently. An important issue is the ''efficient coexistence'' of Constant Bit Rate (CBR) services, Variable Bit Rate (VBR) services and ''best effort'' services (ABR). In order to avoid congestion, it is necessary to design appropriate congestion control mechanisms to regulate the ABR input traffic rate of the network such that all entering cells can be completely delivered using the existing network resources (i.e. queues, link transmission capacity). In the last decade, control has been developed for robust stabilization in control system design against the modeling error, external disturbance and system uncertainties. In this paper, a simple controller is introduced for rate-based congestion control of ATM networks from minimax perspective. A PID controller, plus a smith predictor is developed to overcome instabilities due to large propagation delay and to eliminate the effect of disturbance and the system uncertainties, as well as to avoid cells loss. In the algorithm, ABR source rates are adjusted according to virtual channel (VC) queue lengths at intermediate nodes along the path. The goal is to fully and promptly utilize the available bandwidth left by quality of service (QoS) constrained traffic (CBR+VBR) for transmitting the less QoS stringent ABR traffic. With the aid of Smith predictor we can overcome the effect of propagation delays (delay will be out of the feedback loop, and therefore does not affect stability), a robust PID controller is proposed to guarantee the stability, to reject the influence of variational output rate of cells leaving the queue and to avoid cell loss. Since no closed form solution can be found for this PID optimal control problem, an effective implementation algorithm (Genetic Algorithm) to find the PID control parameters is also proposed. Since the PID controller has only three parameters to be specified, conventional optimization techniques cannot be employed to obtain a closed-form control solution. In this study, a design procedure is proposed to the PID control algorithm for ATM congestion control. In the first step, a Smith predictor is specified to overcome the effect of propagation delay. In the second stage, based on Routh-Hurwitz criterion, the stability domain of three PID parameter space, which guarantee the stability of the closed loop system is specified. In the third step, from the subset of the stability domain in PID parameter space in step 2, three PID control parameters are chosen to achieve optimal control. This is generally considered to be a highly nonlinear minimization problem, in which many local minima may exist. A local minimum can be reached via conventional algorithm. Genetic algorithms (GAs) have recently found extensive applications in solving global optimization searching problems. They are useful when the closed-form optimization technique cannot be applied. Genetic algorithm are parallel, global search techniques that emulate natural genetic operators. Because a GA simultaneously evaluates many points in the parameter space, it is more likely to converge toward the global solution. Based on genetic algorithm, the three PID parameters will be tuned in the stability domain to achieve optimal control. The main advantages of the proposed congestion algorithm consist in: (1) the use of simple time-delay dynamic model for queue levels and simple but efficient control algorithm, (2) the queue occupancy is maintained in a desired level with small cell loss, despite system uncertainties and disturbance in network, (3) the queue occupancy dynamic is always stable. From the simulations in network, the proposed PID control scheme can achieve a desired queue level in bottleneck queues under propagation delay and variation of bandwidth, which gives a guarantee for the fairness and high utilization rate of queues per VC.