Optimal Buffer Management Algorithm with Auto-Tuning Reference Queue Length

Abstract

Most Active Queue Management (AQM) algorithms keep the queue length of the router at a reference value. While when the network traffic is heavy, these algorithms tend to cause high packet loss ratio, large queue oscillations, etc. To address these issues, a new method called Auto-tuning Reference Queue (ARQ) is proposed in this paper, which can be incorporated in any existing AQM algorithms with fixed reference queue length, such as RED and PI. According to the average packet drop probability, AQM algorithms with ARQ adjust the reference queue length. They obtain low delay by keeping low reference queue length when the network traffic is light. On the other hand, they obtain low packet loss ratio by increasing the reference queue length in the case of heavy traffic. The simulation results show that AQM algorithms with ARQ can make a good tradeoff between delay, packet loss ratio and throughput.