On Wed, Mar 31, 2010 at 5:42 AM, xor <x...@gmx.li> wrote: > On Wednesday 31 March 2010 06:32:58 am Evan Daniel wrote: >> On Wed, Mar 31, 2010 at 12:22 AM, Ximin Luo <xl...@cam.ac.uk> wrote: >> > have you joined the freenet-dev mailing list? in future i'd like to have >> > these discussions there so that other people can see it too. >> > >> > (03:53:26) lusha: hi, can i ask a question about WOT? >> > (03:55:33) evanbd: No need to ask permission :) >> > (03:56:12) lusha: is there any document for this? >> > (03:56:32) lusha: i dont quite understand how they evaluate trust >> > >> > (i think) WoT uses a flow-based metric similar to advogato >> > (www.advogato.org) - see the source code (plugin-WoT-staging), or ask p0s >> > on IRC (xor on the mailing list) for specific details. atm the >> > implementation requires retrieving trust scores for everyone on the >> > network, which won't scale in the long run. >> >> No. The current WoT code is neither flow-based nor particularly >> related to the Advogato algorithm. It's purely alchemical, having >> neither a proper specification as to the problem being solved nor any >> sort of theoretical basis to believe it solves that unspecified >> problem. > > That is true. I should finally do this and answer to your mail w.r.t. to your > prosed alternative algorithm. I'm sorry, I'm just trying to make > everyone > happy. People want a release of FT/WoT soon so as long as I didn't have much > time/day I was trying to spend it on writing code. > > BUT we should also state that the algorithm itself fortunately is only a small > part of WoT. Most of the work which is required for a working WoT was writing > the class architecture, the captcha stuff, the FCP stuff, adding proper > synchronization and general glue code. Those are all done and they work. So > now we have a proper "nest" for embedding any proper trust/score-based > algorithm in.
Please don't misunderstand: right now I think usability and any internal changes you need to do to get WoT / FT ready for release are far higher priority. I'm greatly appreciative for the work you've been doing, and think you should keep doing it. There's time enough for WoT algorithms after that. > >> Retrieving trust lists for large numbers of nodes should scale fairly >> well, as long as the updates can be slow. IMHO the only real problem >> presented is startup for a new user (downloading a few tens of MiB of >> scores might take a little while). Specifically, I don't think the >> scaling problem is any different or worse than the scaling problem >> inherent in trying to retrieve messages from that many users. And, >> whether it's a problem or not, there are *vastly* more important >> things to worry about than what to do once we have 1M users -- like >> how to get that many users in the first place. That sort of scaling >> problem gets put in the "nice problems to have" category in my book. > > It can also easily be cut down to logarithmic complexity by only downloading > trust lists from identities which are directly trusted and from their > trustees... I'll do that as soon as Freetalk is deployed and the core features > are working. Hmm? You mean limiting it to 2 degrees of separation? Doesn't that mean a lot of the network isn't visible (especially if you're assuming lots of new users)? I already have a lot of known rank 3 identities. (Also, I don't see how that's O(log(n)) -- it sounds more like O(1) to me.) Evan Daniel _______________________________________________ Devl mailing list Devl@freenetproject.org http://osprey.vm.bytemark.co.uk/cgi-bin/mailman/listinfo/devl
