An ABR congestion control algorithm feeding back available bandwidth and queue level

Classical control theory is proposed to give a general framework for designing congestion control algorithms for ABR service in ATM networks. The algorithm is entirely implemented at the source whereas the switches in the network are in charge of supplying the feedback. Switches feed back to the ABR sources both queue level and available bandwidth. The queue level is fed back to track a queue threshold, which guarantees that the bottleneck queue is never empty. The available bandwidth is fed back to allow the source input rate to track the bottleneck bandwidth so that, in the steady state condition, zero queuing is necessary to ensure full link utilization. Considering that RM cells supply feedback in a sampled form, and that available bandwidth is bursty, the Nyquist sampling theorem implies that feedback of available bandwidth cannot give any contribution to reducing the buffer capacities that are required to ensure full link utilization in the presence of bursty traffic. However, this feedback reduces to zero the buffer capacities which are required to ensure full link utilization in the steady state condition. The properties of the algorithm, such as stability and full utilization of network links, are shown via mathematical analysis. Therefore, they hold for a general network topology and traffic scenario.