[ntp:questions] Re: synchronization
david at djwhome.demon.co.uk
Fri Mar 10 08:07:13 UTC 2006
In article <slrne121b1.j8c.kostecke at stasis.kostecke.net>,
teve Kostecke <kostecke at ntp.isc.org> wrote:
> On 2006-03-09, Richard B. Gilbert <rgilbert88 at comcast.net> wrote:
> > Ron Croonenberg wrote:
> The frequent time steps show in the log extract included in Ron's
> original article suggest that his clock drift rate exceeds ntpd's
> correction limits. This is the problem that must be solved first.
The magnitude and frequency of the steps would indicate that the crystal
frequency error was way outside even the lax tolerances expected of PC
crystals, and therefore that it is possible that the crystal is either
dying or is completely dead and the frequency is being determined by
components other than the crystal. As such, the only reasonable fix for
a frequency error is a new motherboard.
Another reason for frequent and consistent time steps is that there is
some other software trying to discipline the clock, although this is not
normally the case for Linux distributions, unless someone uses "hwclock -a"
in a cron job. The sheer magnitude of the error makes me favour this
Monontonic forward steps can be due to lost timer interrupts, which are
quite common on some Linux systems because the kernel, on fast machines,
has been built for 1ms clock ticks (HZ=1000) when the device drivers,
mainly the IDE driver, disable interrupts for more than this at times.
However, they tend to less consistent than this.
> > You should be using four servers.
>From what I've seen there seems to be some dispute as to whether or
not this includes the client's clock, i.e. you might only need three
> This is true, but it is hardly germane to the problem.
There does seem to be a tendency to reply to all problems with a request
to make the configuration match the responder's own preferences before
even looking at the evidence provided.
More information about the questions