Asymptotic Analysis of Optimal Adaptation of Rate Adaptive Multimedia Streams

We first discuss the Quality of Service (QoS) of a rate adaptive stream and argue that the time-average mean rate is a reasonable first-order proxy for user perceived performance. Next, we discuss general system level tradeoffs associated with rate adaptive streams. We then identify optimal adaptation policies under different assumptions of available information. The optimal policies make clear that it is beneficial to discriminate among streams based on stream duration, i.e., the system level benefits are maximized when shorter duration streams are granted a priority over longer streams. Another insight obtained from the optimal adaptation policies is that two rate adaptation is near-optimal under some fairly mild assumptions. We next investigate the question of which entity, the sender, network, or receiver, should be responsible for implementing the adaptation. We argue that the receiver is ideally suited to handle adaptation, especially because receiver adaptation generalizes easily to a multicast scenario. Finally, we identify an appropriate load/capacity scaling regime for rate adaptive streams. Scaling our equations for QoS allows an asymptotic analysis of the tradeoffs involved in rate adaptation, including the load capacity scaling parameter, the application adaptivity parameter, the stream duration distribution, and the adaptation policy. Our results suggest that implementation of optimal adaptation policies can significantly ameliorate the QoS cost of rate adaptation. These equations can be thought of as the rate adaptive asymptotic analogues of the Erlang blocking probability formula.