Characterization of queue fluctuations in probabilistic AQM mechanisms

Abstract

We develop a framework for studying the interaction of a probabilistic active queue management (AQM) algorithm with a generic end-user congestion-control mechanism. We show that as the number of flows in the network increases, the queue dynamics can be accurately approximated by a simple deterministic process. In addition, we investigate the sources of queue fluctuations in this setup. We characterize two distinct sources of queue fluctuations; one is the deterministic oscillations which can be captured through the aforementioned deterministic process. The other source is the random fluctuations introduced by the probabilistic nature of the marking schemes. We discuss the relationship between these two types of fluctuations and provide insights into how to control them. Concrete examples in this framework are given for several popular algorithms such as Random Early Detection, Random Early Marking and Transmission Control Protocol.