On Mar 3, 2008, at 1:23 PM, Henry Sinnreich wrote: > Ted, > >> The overlay ends up being a path of last resort > > Having the media go over 5-10 hops, some of which may pass via several > continents due to the hashing in the DHT makes such real time media > worthless due to the delay and cut-offs. > > Bruce: Can you comment on this? Reload-03 has no clear text on > symmetric > routing for media. >
I think that's because we assume a TURN server will be used, if necessary. I'm somewhat in Ted's camp of "it will be used as a last resort, regardless of what the draft says" although I think some of my fellow authors would like to require that implementors explicitly prevent it from happening. With that said, I would say that if you did really, really want to relay media through an overlay: - compress the path as much as possible, i.e. add an edge to the connection table that reduces the number of hops to the lowest number possible - implement route optimizations. Ordinarily I oppose routing options other than symmetric because of the complexity involved with the need to have fallbacks to symmetric for various healing/joining scenarios. But if you're going to insist on routing media across an overlay, you're going to have to pay the complexity cost of fancier routing. I think Henry's point that geography comes into play is very important here. Henning has proposed that AS number be included in selection of a TURN server. It's not clear to me that is the right rule in all cases or that we have a working example of how to really implement this that isn't far too researchy at this point, but we might have some hope of doing it for TURN service discovery. Most of the DHT proposals I think we've seen for P2PSIP have tried to avoid the complexity of using something like PASTRY. I really want to leave this out of scope for now, except for the TURN service discovery portion. Just too much complexity and too many headaches for little gain. Bruce _______________________________________________ P2PSIP mailing list [email protected] https://www.ietf.org/mailman/listinfo/p2psip
