Stability and Fairness Issues in Layered Multicast

Abstract

Layered multicast is a promising technique for broadcasting adaptive-quality video to heterogeneous receivers. Past evaluation of layered multicast protocols has focused on the effectiveness of determining the maximal number of layers that can be delivered to each receiver. Unfortunately, this is only a partial metric and does not capture all aspects of the “viewing experience”. This paper extends past analysis by investigating stability of layered multicast schemes and the related issues of fairness and scalability. Our study is motivated by the desire to transmit small-scale TV broadcasts (hundreds or thousands of viewers) over IP-based networks with heterogeneous receivers. Although users are aware of their physical bandwidth limitations, they will still expect TV-like characteristics of their video such as consistent quality (i.e., stability). Our results shows that Receiver-Driven Layered Multicast (RLM) exhibits significant and persistent instability. Previous work also raises fairness as a potential problem with RLM. We show that with CBR traffic RLM can be arbitrarily unfair, but with VBR traffic, RLM provided better fairness than we anticipated. Our analysis of stability in RLM prompted us to question RLM’s ability to select the optimum number layers for a non-shared access link. We show that RLM is very conservative in its choice resulting in low link utilization. 1