Network traffic demand patterns for video over relative differentiated services networks

Relative service differentiation mechanisms such as DiffServ Assured Forwarding are promising techniques for future delivery of video services due to their desirable scalability properties and the ability to trade-off cost and loss rate. This trade-off is important since the loss of certain parts of a video bitstream (e.g. motion information) has a catastrophic effect on the quality of the decoded sequence whereas the loss of other information (e.g. high-frequency transform coefficients) does not. We investigate the performance of these systems under realistic conditions where per-packet tariffs must remain fixed for relatively long periods of time but the network conditions may change over short timescales. We also consider how the presence of many users on a network, each with their own unique notions of utility interact in the context of a single shared network. We show that so long as the network behaves in a certain way as the overall load fluctuates the distribution of traffic among service classes remains largely unchanged. An essential consideration is not simply the mapping of video packets to service classes but also the rate of generation of video packets and thus the amount of protection that can be given to each packet. This is an important result as it demonstrates that such a network “makes sense” and will not simply regress to behave like a best-effort network.