On Tue, Nov 04, 2003 at 09:59:12AM +0000, Ian Clarke wrote: > Toad wrote: > >This strongly suggests that routing is not doing terribly great. The > >suspicion is that there is some form of network fragmentation going > >on. Ian refuses to admit the possibility of network fragmentation, so I > >am wondering how to produce extraordinary evidence to support my > >extraordinary claim :). > > I don't refuse to admit the possibility, I just think that it is > unlikely and that there are much more likely explanations (such as > routing just not working). It simply seems statistically very > improbable that the network could split into two independent segments > with no interconnections. > > I would like to verify the claim that those splitfiles were in-fact > inserted recently. I have noticed very poor splitfile performance, but > I have only tried splitfiles that are months old. > > Looking at the Estimators many of them appear quite chaotic, and I am > fearful that they simply aren't doing a good job, note that NGR's > routing ability is entirely dependant on the effectiveness of the > ResponseTimeEstimator implementation. The statistic I have been > suggesting we implement for ages now, the mean difference between > estimated and actual response times, would tell us whether these > estimators are really doing their jobs.
That stat is absolutely useless and impossible to implement properly. I explained why on another thread, please reply to it. > > The problem could be that many of them simply aren't receiving enough > individual data to generalize effectively. This could be addressed in > at least two ways: > > * Reduce the number of peers from 50 to, say, 15 or even 10 Then we have to be very careful with adding new peers to the RT. > * Embed Estimators in DataSources and in the seednodes file so > nodes share experience of other node's performance > > Of course there is always the chance that there is just a bug somewhere, > perhaps even in the splitfile code. > > Ian. -- Matthew J Toseland - [EMAIL PROTECTED] Freenet Project Official Codemonkey - http://freenetproject.org/ ICTHUS - Nothing is impossible. Our Boss says so.
signature.asc
Description: Digital signature
_______________________________________________ Devl mailing list [EMAIL PROTECTED] http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl