This paper approaches the problems of scalability and deployment of IP Multicast.
The main contribution of this paper is to examine and elaborate on End System Multicast in the context of audio and video conferencing. The authors explore various schemes for constructing overlay networks on about twenty hosts distributed around the Internet. The problem of handling audio and video multicasting on the Internet is still relevant today and will probably become moreso in the future, so this exploration of various schemes is still important today.
The paper concludes that for End System Multicast to satisfy performance requirements for audio and video, self-organizing protocols must adapt to both latency and bandwidth metrics.
I think this paper is moderately significant because it tackles a still existing problem of handling real-time reliable audio and video multicasting on the Internet.
The authors are fairly convincing in their methodology, and their evaluation of the resulting data is thorough. They conclude that overlays must be constructed with both bandwidth and latency in mind. Their data suggest that protocols that consider no network metrics perform far worse than protocols that consider only latency or only bandwidth, and protocols that consider both perform far better.
This paper concludes with a number of questions remaining to be solved. One is whether the overhead of active measurements scales well to groups much larger than twenty members. Another question is how much time self-organizing protocols must take to converge on efficient overlays given no initial network information. The authors also are unsure of performance degradation in highly dynamic environments, and leave the problem of exploring mechanisms with shorter time scale adaptation to address these environments.