After investigating some strange estimates, it appears that tSearchFailed can take some absurdly high values, because of the different things it is used for: QueryRejected's, timeouts in place of a QueryRejected, timeouts after Accepted, timeouts after QueryRestarted's.
Options: 1. Add separate estimators for each option. If it works, it may partially explain routing breakage, and may be less alchemical than exponential backoff. 2. Remove pSearchFailed from the estimators altogether. Leave load balancing entirely to exponential backoff.
Anyone in favour of option 1) ? :)
Me. At this time, I think option 2 would be throwing the baby out with the bathwater. If we can fix the estimator() so it's more accurate, then let's do it. Presumably, that's why tSearchFailed gets absurdly high values.
NGR gets better when estimator() is more accurate. Option 1 makes it more accurate. Option 2 makes it less accurate.
-Martin
_______________________________________________ Devl mailing list [EMAIL PROTECTED] http://dodo.freenetproject.org/cgi-bin/mailman/listinfo/devl
