We review the development of the concept of effective bandwidth from its origin in the planning and management of ATM networks. We start with the extension of the telephone network to multi-rate circuit-switched networks and then to bufferless networks with variable rate traffic. We also show how the concept can be extended to IP-type networks with buffered queues.

We use two simple traffic models that are amenable to exact computation to show the accuracy of two definitions of effective bandwidth, one based on a single-queue model and the other on the Chernoff bound of the distribution of a variable. We use this to illustrate the notion of admission region and its relation to effective bandwidth. We discuss the accuracy of the two definitions and the scope of applications.