Asymptotic Analysis of Rate Adaptive Multimedia Streams

Abstract

We investigate dynamic adaptation policies for rate adaptive multimedia streams in a network where each route traverses at most one bottleneck link. Dynamic adaptation allows clients to dynamically adapt the stream subscription level, i.e., time-average stream rate, in response to changes in available link capacity, and allows the system to maintain a lower blocking probability than is possible with non-adaptive streams. We define the quality of service for rate adaptive streams using the metrics of time-average subscription level, rate of adaptation, i.e., change in subscription level, and blocking probability. We investigate two baseline policies, namely, fair share adaptation and two rate randomized adaptation, and show that each suffers from significant implementation drawbacks. We then show that the adaptation policy which maximizes the mean subscription level overcomes these drawbacks, although streams with a duration near a critical threshold may experience unacceptably high rates of adaptation. This motivates the investigation of admission policies for rate adaptive streams where a stream is given a static subscription level at the time of admission which it maintains throughout its lifetime. We identify the asymptotically optimal admission policy for rate adaptive streams and show that it achieves an expected subscription level equal to that under the optimal dynamic adaptation policy. We also show that it maintains the asymptotic zero blocking property achievable using dynamic rate adaptation but does not incur the implementation overhead and QoS drawbacks of dynamic rate adaptation. The conclusion is that near optimal QoS can be obtained using a simple admission policy which gives the maximum subscription level to short duration streams and the minimum subscription level to long duration streams.