-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 (copying Oskar - I think you will want to read this)
On 17 Aug 2006, at 14:15, Matthew Toseland wrote: > On Thu, Aug 17, 2006 at 12:16:34PM -0700, Ian Clarke wrote: >> Perhaps, in which case the solution is for someone inside the >> firewall to connect to a darknet node outside the firewall, they can >> then connect to opennet nodes. In this case the user in the hostile >> regime is still just 2 hops from the opennet. > > There is a limited supply of friendly westerners, and there is also a > limited intersection of content between the two networks. If the > network > is to work well for the chinese then it will have to scale > *internally*, > so that people can add their friends without rapidly slowing down > their > own access. What you suggest is analogous to me running a proxy for a > few of my chinese friends; if they connect their friends to that > proxy, > and their friends connect their friends, pretty soon it is intolerably > slow. You need a large network with lots of internal nodes > connected to > each other, and relatively few external connections. I agree that if we end up in a situation where we have large parts of the network only connected to each other through a very small number of links that this will be problematic as those links will quickly be overloaded. I'm not yet convinced that this situation will occur, but I agree that it is a possibility. I think the fundamental reason for this problem is the migration towards a more simplistic notion of node specialization in 0.7. The more flexible approach of 0.5 where nodes can have more than one specialization, and varying degrees of specialization in response to demand, I believe, would be able to deal with this situation. 0.7's simpler approach may not. I don't think the solution is to have some different routing behavior depending on whether it is a darknet or an opennet node, because this doesn't solve the problem that the information you want is still very likely to be outside your isolated corner of Freenet. Perhaps if nodes maintained two specializations, one for "local darknet" and another for "global opennet", that could solve the problem, but that strikes me as being rather ugly.os For now I suggest that we wait and see, if we do start to see a network topology that essentially consists of multiple small world networks that are poorly connected to each-other, then we may need to consider moving back to something closer to the 0.5 approach to node specialization. Ian. -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.5 (Darwin) iD8DBQFE5OEcQtgxRWSmsqwRAq04AJ9eJopVTpgKg8FofnukjGIow5PzKQCdGTb1 ePKFPPg9tWcqHhg3LYT2Ncg= =roc7 -----END PGP SIGNATURE-----
