Flux Research Group / School of Computing

Auto-tuning active queue management

Joe Novak and Sneha Kumar Kasera

International Conference on Communication Systems and Networks (COMSNETS) 2017.



Auto-Tuning Active Queue Management

Active queue management (AQM) algorithms preemptively drop packets to prevent unnecessary delays through a network while keeping utilization high. Many AQM ideas have been proposed, but none have been widely adopted because these rely on pre-specification or pre-tuning of parameters and thresholds that do not necessarily adapt to dynamic network conditions. We develop an AQM algorithm that relies only on network runtime measurements and a natural threshold, the knee on the delay-utilization curve. We call our AQM algorithm Delay Utilization Knee (DUK) based on its key characteristic of keeping the system operating at the knee of the delay-utilization curve. We implement and evaluate DUK in the Linux kernel in a testbed, that we build, and in the ns-3 network simulator. We find that DUK can attain reduced queuing delay and reduced flow completion times compared to other algorithms with virtually no reduction in link utilization under varying network conditions.