This paper starts off lamenting about the fact that network-layer multicast solutions have not been adopted even after a decade of their proposals and hence the push for application-level solutions. Application-level multicast solutions are obviously less efficient than network-level solutions and these can be quantified in terms of stress (duplicating packets on a link) and stretch (sub-optimal path to reach a node).
NICE pushes for a tree hierarchy. Each layer in the tree is divided into clusters and a member of each cluster belongs to a higher-level cluster. The cluster leader is chosen such that it has the minimum distance to all the other hosts in teh cluster. This choice is important in ensuring that new members are quickly able to find their aposition in the hierarchy.
One immediate question that pops up is why a tree and why not something else? While the authors can claim tree-based topologies to be the common thing in multicast literature, it would still have been useful to argue about this. Overall, this paper had a feel of being way too complicated unnecessarily.
Their results are against Narada and probably Narada was the new kid on the block then...but then is it really the gold standard?
I personally would vote against keeping this paper and favour the Narada or Overcast paper (the Overcast paper is extremely lucid and comes across as highly practical).