Protecting Scalable Video Flows from Congestion Loss

To minimize packet loss caused by bandwidth probing, a video congestion control algorithm may use forward error correction (FEC) to reproduce dropped packets. Unfortunately, if a dominant part of the competing flows adds redundancy, the extra overhead may cause increased packet loss ratios in the network, and in turn do more harm than good. In this paper however, we demonstrate both analytically and with simulations, that by using a standard DiffServ dropping mechanism and assigning a lower precedence level to FEC packets, the negative effect of the added overhead can be strongly pacified. The analysis also show that the FEC benefit is clearly maximized when router buffers are small, favoring multimedia flows with delay and jitter requirements.