

To determine whether this estimate was accurate for real multicast distribution trees, we measured the metric, delta, for a number of real groups. Furthermore, we explored the metric's dependence on temporal and spatial factors such as receiver duration and interarrival times as well as receiver distribution.
The first graph to the left is for a real group where the temporal domain has been stretched by randomizing receiver duration and interarrival times. The second graph stretches the spatial domain by using random receiver distributions. To collect this "synthesized" dataset, we traced paths from a local source to a random collection of receivers over the current multicast infrastructure of the Internet. So, although the receiver distribution was random, the underlying topology was not.
So, the efficiency estimate holds over a wide range of group dynamics and distributions within a small range of values of epsilon, from .30 to .34. The estimate tells us how efficient an interdomain multicast should be. With even a small number of receivers, multicast outperforms unicast by a wide margin. For 20 to 40 receivers, we can expect a 6070% increase in efficiency, reaching 80% for 150 receivers.
