> one of the issues with the descriptor approach is bandwidth. This very much depends. Yes - it is overhead (which may be small when well designed) - it may need superfluous TCP opens (when not too well designed) - but it also opens up a whole new set of optimizations, of efficient caching, of trust management, of (partial) mirrors and more ways of avoiding fetching the wrong file as entropy does its usual (and unavoidable) think.
So I think that in the long run this trade off may actually be on the bandwidth (or at least perceived latency) side of the coin. Dw
