[ntp:hackers] Weird initial frequency calculations.

Brian Utterback Brian.Utterback at Sun.COM
Sat Aug 2 11:41:19 UTC 2008


David L. Mills wrote:
> Brian,
>
> I gather from your note that you stopped ntpd once each hour and:
> 1. Removed the frequency file.
> 2. Started the daemon.
> 3. Recorded the frequency between 3 and 26 minutes afger that.
>   

I think you have it. Just to be clear, on step 3 above, I recorded it 
when ntpd made the ntp_adjtime call, which turns out to be between 3 and 
26 minutes.
> Your printout apparently was in the direct_freq() routine in ntp_loopfilter.
> You should note from the code that the direct_freq() routine is called 
> from two places only and that only when actually calculating the 
> frequency after the inital measurement interval.
>
> So, how is it you show a a call when less than 15 minutes remain before 
> the only call?
>   

I am not sure what you mean by "5 minutes remain before the only call.  
But the whole point here is that if this isn't the way you think it 
should be working, then there is a problem, because this is the way it 
is working.

> I'm not getting any traction here. Take it step by step. Record 
> loopstats with the kernel disabled and then another run with it enabled. 
> COmpare the frequency and then go from there. Remember, a loopstats 
> entry is produced exactly and only when the clock offset and freqeuency 
> are updated.
>
>   

Well, up here in New England, when we don't get traction, we throw a 
little sand at the problem 8-)

Ok, I'll re-run it the tests again, alternating with and without the 
kernel discipline and also capturing  the loopstats files.
> Dave
>
>   



More information about the hackers mailing list