On Fri, Aug 07, 2009 at 10:26:29AM +0100, Alison M. Wheeler wrote: > I would like to suggest, therefore, that in the case where there is no > response (for two/three attempts) then instead of reducing the score > without limit that it is immediately dropped to the 'new server' score > of -20, but does not decrease any further during the period of the > inaccessibility.
There's been quite a bit of discussion about this in the past and the general consensus has always been that the current algorithm isn't perfect but anything else just moves the problem cases around and doesn't fix anything fundamental. > Going by the restart curve I see on my own server after a network fail > (yes - personal interest in this!) this would bring the server back > online and available to the pool within 6-8 hours, rather than day(s). > > Alison Wheeler http://www.pool.ntp.org/scores/78.33.97.18 You graphs show it varying by upto 200ms, this is a lot of variation and is why your box never really reaches a score of 20. This is a lot of variance and I'd be tempted to fix this first. I see an order of magnitude less variance and I'm worried about it as there are people with getting on for an order of magnitude less than me. As a quick hack, I chucked the following together: http://samason.me.uk/~sam/tmp/awheeler.pdf (you) http://samason.me.uk/~sam/tmp/smason.pdf (me) http://samason.me.uk/~sam/tmp/ksandler.pdf (the third link on the pool's "how do I join" page) -- Sam http://samason.me.uk/ _______________________________________________ timekeepers mailing list [email protected] https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
