In this paper we presented a novel protocol called SCUBA that complements existing rate-adaptation algorithms with a mechanism for discovering and exploiting receiver interest. As with RTCP and SRM, SCUBA embraces the announce/listen metaphor and uses periodic and robust receiver reports to explicitly rank active sources. When combined with layered transmission, SCUBA enables a source to infer an arrangement of signal layers onto network channels that best suits a heterogeneous set of layered receivers. We formalized a model for mapping signal layers onto network channels and presented an algorithm that satisfies a number of important and natural constraints.
By exploiting sampling, we overcame the linear growth in the convergence time of a naive version of the distributed SCUBA algorithm. Each source uses sampling to determine its share of the session bandwidth in a reasonable time with a high level of confidence independent of the session size. We pointed out the shortcomings of this approach and are currently developing new versions of the protocol that we believe will surmount these problems.
Finally, we outlined how SCUBA could be integrated into and exploited by several important applications such as traditional video conferencing, media gateways, and floor control.
We believe that reflecting receiver interest back to the sources in a multimedia conference is a key mechanism for future collaboration technologies. SCUBA achieves this goal in a light-weight, robust, and scalable manner. By dynamically accounting for receiver interest in the allocation policy, sources can share the available session bandwidth in a way that best reflects the natural model for human-to-human communication.