On Mon, Nov 09, 2009 at 08:44:41AM -0800, Ask Bjjjrn Hansen wrote: > On Nov 9, 2009, at 7:00, Ryan Tucker wrote: > >Of course, in the "long run" the bad server will be automatically > >removed from the pool, but that takes a relatively long time [...] > > Yeah - up to an hour(!) currently. It'd be nice to cut it down, but I > don't know any other way than just making it easier to fall out of the > pool and harder to come in.
Speeding up polling when servers start to look bad (only enough to determine how transient the error is) and slowing down when they're good (to counter the increased bandwidth requirements from above) would be possible. > I have all the measurements the pool has done since September last > year (about 43 million) if anyone is good with statistics and applied > math and want to help come up with a better scoring algorithm or > parameters it'd be most welcome. What are you trying to optimize for? You don't have data on the DNS requests as well do you? It would be interesting to try and partition the network up into ntp and sntp clients. Each want a different sort of server, ntp doesn't really care about transient errors sntp is less tolerant. -- Sam http://samason.me.uk/ _______________________________________________ timekeepers mailing list [email protected] https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
