Congestion control in a constrained-worst case framework

We present an approach to worst-case reactive congestion control in ATM-based broad band networks. The effective capacity of a virtual circuit in such a network can vary dynamically because of traffic fluctuations, call rerouting or equipment failures; we pro pose a single mechanism by which to ensure that the traffic offered to the circuit remains matched to capacity, whatever the source of the variation. The approach is non-statistical. It applies to circuits subject to Leaky Bucket regulation at the network interface, and to networks implementing a CPS-like service schedule at the nodal output buffers. The key element is an algorithm which modulates the parameters of the Leaky Bucket on the basis of measured values of instantaneous capacity, and in such a way as to avoid cell loss, to the extent possible, inside the network.