Rate-based active queue management with token buckets

Abstract

Active queue management (AQM) is applied in IP routers to protect routing resources, improve link utilization, and limit end-to-end delay. This paper proposes a rate-based AQM scheme based on adaptive token buckets per aggregated traffic flows. The approach is different from queue-occupancy-based AQM schemes because drop probabilities are not directly linked to the average queue size. In the proposed system, probabilistic dropping at routers is controlled by token buckets with committed information rates (CIR) adapted to the current router congestion status. The paper shows that this approach is better suited to the multi-modal traffic mix of short-lived responsive and long-lived non-responsive flows prevalent in the Internet. © Springer-Verlag Berlin Heidelberg 2003.