On Saturday 04 May 2013 07:09:17 Oskar Sandberg wrote: > I looked over both papers.
Thanks! I'm going to reply separately re each paper. > > Second paper: > > The reason I never let the nodes independently draw and change IDs is that > it seems to me that, and this was borne out in experiments, this would > cause all the IDs to converge to a single point. This happens with churn anyway, though hopefully randomisation resolves this. Randomisation and churn mean that we already don't have trivially provable uniformity though? > The math that motivates > the formula c(u) in their paper actually only works when the IDs are in > fixed grid - if let the space be continuous the distribution degenerates at > as the distance between points approaches zero, so it isn't necessary that > the Markov chain should converge to any meaningful distribution at all. > > If it works, than it works of course. But I would approach with some > caution. Their simulations say it works. I guess you can prove anything by tweaking your simulations... The results they give only apply to 10k nodes, though a later presentation suggests they've tried bigger networks. Ẃe're a long way away from having 10k darknet nodes right now (we have about that on opennet). But the published attacks are hideous, and they add some more, and your/Michael's Pitch Black fix is rather tentative and may not fix their other attacks... > > (Note that if a node is running the Markov chain all on its own it should > be possible to analytically calculate, or at least estimate, its stationary > distribution from a particular set of neighbors, making actually running > the chain unnecessary.) You still need some randomness to escape false minima though, right? I don't suppose you have any ideas on how often we should "swap" etc? I'm inclined to deploy this as soon as we can validate our implementation. Even if it does eventually collapse, our current system is also a "toy", because the attacks are so devastating. What do we need to do to be sure?
signature.asc
Description: This is a digitally signed message part.
_______________________________________________ Devl mailing list [email protected] https://emu.freenetproject.org/cgi-bin/mailman/listinfo/devl
