Effective bandwidth shaping: a framework for resource dimensioning-extended version

Effective bandwidths is a statistical descriptor in the context of stochastic models for statistical sharing of resources. It is generally known that effective bandwidths assigned to independent flows being multiplexed behave additively. Furthermore it has been shown by Wischik that under the many sources limiting regime, an effective bandwidth assigned to a flow passing through a switch is not changed, and inductively a flow has the same effective bandwidth through the entire network. If all flows feeding a network under consideration have well defined effective bandwidths and QoS demands, resource dimensioning can be done based on simple large deviation principles. We address the problem of specifying and limiting effective bandwidths of arriving flows by introducing a new shaping and policing algorithm based on an on-line effective bandwidth estimator. The presented algorithm's functionality is verified by both synthetic and real data traces and the performance of algorithm variants is discussed. Finally, it is shown that the invariance of effective bandwidths (as proved for an infinite number of sources by Wischik) approximately holds already for a finite ensemble of competing connections, even in the presence of TCP traffic.