Applying Wavelet De-noising to Improve TCP Throughput in AQM queues with Existence of Unresponsive Traffic

In the current Internet, congestion control is performed jointly by the end systems running the TCP protocol and by routers running active queue management (AQM) algorithms. Due to the TCP protocol´s AIMD congestion control algorithm and its round trip time delay to react to packet losses, it is very hard to maintain high TCP throughput with a low packet loss rate in routers. In addition, unresponsive traffic, such as short HTTP sessions, do not react to AQM packet loss/marks. Hence, these unresponsive traffic may cause high packet loss in the AQM queues due to its high bursts. In this paper, we provide a solution to identify unresponsive traffic in AQM queues that does not need packet header examination. In our solution, periodic features of TCP traffic due to its AIMD behavior is first estimated in the mixed incoming traffic. We then design a wavelet de-noising filter to separate the high bursts in unresponsive traffic from the TCP traffic and allow those bursts to bypass the AQM queue. By allowing bursts of unresponsive traffic to bypass AQM queues, we not only avoid the impact of unresponsive traffic to AQM queues, but also avoid dropping extra packets in those unresponsive flows. Our proposed de-noising scheme is suitable for high speed networks, where per packet header examination is expensive. Our simulations show that the proposed de-noising scheme is effective on heavily congested links. An analysis of the computational complexity of the proposed scheme is also provided.