Active queue management for congestion avoidance in multimedia streaming

Reducing the queue length and queuing delay in the buffer which in turn reduces congestion and latency in the network are of great importance for real-time video services. This paper proposes a congestion avoidance active queue management, tailored for scalable video coding (SVC). This distortion-constrained (DC) queue management mechanism maximally decreases the queue length and queuing delay of video traffic for a given maximum allowable distortion in each video. DC prioritizes video layers in accordance with their hierarchical layer structure into multiple virtual queues inside the buffer. Losses in different layers of the hierarchy introduce different amounts of reduction in PSNR.

Therefore, we determine the optimal drop rate in each virtual queue which causes minimal distortion to the stream while achieving a maximal drop rate. This in turn produces the highest reduction in the average buffer queue length and queuing delay for a given distortion. Compared with the widely deployed tail drop mechanism, our proposed scheme would increase the fraction of QoS-satisfied users and maintain a considerably shorter queue length in the buffer which would combat congestion and decrease queuing delay.