On 2012 May 24 (Thu) at 01:21:53 +0200 (+0200), Kurt Roeckx wrote: :On Sun, May 13, 2012 at 03:36:07AM +0200, Peter Hessler wrote: :> On 2012 May 13 (Sun) at 01:48:51 +0200 (+0200), Kurt Roeckx wrote: :> :I think a lot of the problematic servers run openntpd. :> :> Please do not spread FUD. : :The last released portable version of openntpd is 6 years old, and :doesn't call ntp_adjtime() / adjtimex(), but just calls adjtime(). :As a result the kernel does not compensate for the drift of the :clock. : :If you have a clock with a drift of 100 ppm, which is very normal, :and a poll interval of 1024 seconds, you're basicly off 102.4 ms :every 1024 seconds. : :I understand that a patch has been available for a very long time, :at least since 2005 from what I can find, but clearly wasn't used :everywhere, and that the upstream portable branch now has a :simular patch applied. But it's still not released. : : :Kurt :
As far as I know, the portable version of openntpd is un-maintained. HOWEVER, openntpd is written by, and is the default daemon in, OpenBSD. Instead of ntp_adjtime/adjtimex, use adjfreq which does adjust the clock frequency as you expect. http://www.pool.ntp.org/scores/81.209.183.113 is my server running openntpd/openbsd on commodity hardware. As you can tell, it is plenty accurate for pool usage. If you see *real* problems in production, please let us know. -- Don't look back, the lemmings are gaining on you. _______________________________________________ pool mailing list [email protected] http://lists.ntp.org/listinfo/pool
