Daniel Stutzbach wrote: > On Mon, Mar 20, 2006 at 11:42:40PM +0100, Oskar Sandberg wrote: >>Structured DHT networks are brittle. > > > Kad and OverNet seem to be doing okay. They have around 1 million and > half a million nodes, respectively. I imagine the BitTorrent DHT is > pretty big by now, too, but I haven't played with it yet.
Is there any more detailed information about these networks somewhere? I have played with the former a few times, and found that while it seems to work for its purpose, it does seem to give pretty strange and not always consistant results. I was under the impression that it used only a subset of stable users in the actual grid, though, so a million and a half users seems large. (Haven't most users had been scared off these networks?) What surprises me most with these networks, actually, is that they haven't been taken down by DoS attacks. Saying this isn't an attack on any particular system, since every system I have seen is vulnerable to the point where it ought to take only one sufficiently motivated script kiddie (/ national organization) with a few hundred purposely broken nodes. >>If a Chord node, for instance, for some reason thinks that a node on >>the other side of the ring is actually its closest neighbor, bad >>things will happen, and in theory it could cause the whole network >>to degenerate. Similar scenarios are possible for other structured >>DHTs. > > > That wouldn't be problem in Kademlia-based DHTs (nor in other DHTs > that use parallel queries). If there's a problem, it routes around > it. No, you can have similar inconsistancies that would cause hypercube based systems like Kademlia to degenerate. Parallel queries can only offset such problems a little, unless the network updates itself based on which queries work well. You can make a hypercube based mesh more flexible if you start being very lax about having the "correct" neighbors and just ask nodes to try to match the levels as well as possible. But the result is essentially the same thing as Kleinberg's model (though since you are using bitwise distance it will remind more of [1] than the geographic papers). // oskar [1] http://privacy.cs.cmu.edu/dataprivacy/papers/socialnetworks/40.pdf _______________________________________________ p2p-hackers mailing list p2p-hackers@zgp.org http://zgp.org/mailman/listinfo/p2p-hackers _______________________________________________ Here is a web page listing P2P Conferences: http://www.neurogrid.net/twiki/bin/view/Main/PeerToPeerConferences