An Improved AQM Scheme with Adaptive Reference Queue Threshold

Due to exponential increases of traffic, congestion is more intractable than ever before in the Internet often leading to longer delay, high jitter and massive packet losses. Consequently, quality of service (QoS) deteriorates. Active Queue Management (AQM) is a powerful supplement of transport layer (i.e. TCP) congestion control for IP layer to regulate network load without increasing the complexity of end-user protocols. In our previous work, an AQM scheme considering traffic prediction called ATPAQM had been proposed. Indeed, it works well because its traffic predictability and packet dropping adaptivity. However, as a common flaw of most AQM schemes, it maintains a fixed reference queue (Qref) threshold. The fixed threshold results in unnecessarily additional delay in the case of light traffic load, while bring about excessive packet losses when traffic is heavy. In this paper, an improved version called ARTAQM with minor revision to ATPAQM is presented. In ARTAQM, a functional relationship between traffic load and the Qref is built. The difference between the estimated traffic rate and link capacity serves as the input of the function to adjust the Qref. Simulation results show that ARTAQM outperforms other schemes in queue stability, packet loss ratio and link utilization.