A node-to-node communication architecture for congestion avoidance of live video over the internet

• Authors: A. Srinivasa Reddy;S. V. Raghavan
• Venue: ICCC '02 Proceedings of the 15th international conference on Computer communication
• Year: 2002

To transmit VBR video over the Internet, neither single rate reservation techniques (such as CBR) nor its extensions of renegotiation techniques (such as RCBR) provide deterministic guarantees for packet loss. This is because it is difficult to have a service curve that mimics the traffic generation rate due to its multiple time scale bit-rate variability [1,4]. Also, in reservation techniques, the maximum share of frames that are having size greater than that of the reserved rate are intra-coded frames. The losses in these frames are carried to the consecutive inter-coded frames that are encoded depending on them. In such techniques, it is difficult to minimize the packet losses with moderate reservations because, it is difficult for the source node to predict the exact congestion at the intermediate nodes and/or the congestion state information may not be available in time. One possible solution is to transcode the video at the intermediate nodes. Transcoding improves the quality, as the quantization loss is less perceptual than the packet loss. In this paper, we propose an intermediate node-to-node (n2n) communication architecture that transcodes the video at any given node (on the end-to-end path) depending upon the anticipated congestion at its successor node. Simulations show that transcoding improves the video quality up to 20% over the conventional packet-drop based mechanism for moderate bandwidth reservations.