[ntp:questions] Re: Taming the pinball machine

David L. Mills mills at udel.edu
Tue Nov 18 17:15:45 UTC 2003


Tom,

If you started ntpd with no frequency file, it will take one or two
15-minute measurements to quickly calculate the intrinsic frequency
offset. A considerable time offset might accumulate during this process,
especially if the intrinsic offset is large, like hundreds of PPM.
However, the process should converge without frequency overshoot as you
report. Even if started with a frequency file with large frequency error
it may take some hours to converge, but not overshoot the frequency. The
time overshoot should be less than ten percent in all cases, as
confirmed in both theory, simulation and practice.

I have to conclude from your report that something external to ntpd is
tweaking the frequency. Certainly, with both xntpd and ntpd in Solaris I
have nothing like the problems you report.

Dave

Tom Schulz wrote:
> 
> In article <3FA9CC42.58299EA5 at udel.edu>, David L. Mills <mills at udel.edu> wrote:
> >Guys,
> >
>  ------------------ text removed --------------
> >
> >I recommend that only folks that fit the profile above try the test
> >version at ftp.udel.edu:pub/ntp/software/ntp-4.2.0.gz. This is only a
> >temporary spot under an assumed name until the crew rolls a suitably
> >versioned model to the web.
> >
> >Dave
> 
> Well I don't fit the profile, but the new code fixed a problem for me.
> I tried the original 4.2.0 on our HP-UX systems and found that if the
> clock was not set almost perfectly before starting ntpd, ntpd would
> jump between large negative and positive offsets and would never
> stabilize.  Ntpd was using the default values for all the polling
> parameters.  With the test version, ntpd quickly gets the clock
> to the right time without using a huge frequency offset and quickly
> settles down to a stable frequency offset with no jumping around.
> I also noticed that the Solaris version of ntpd stabilizes in less
> time after a restart using the test version of ntpd.
> --
> Tom Schulz
> schulz at adi.com



More information about the questions mailing list