[ntp:questions] Sine wave offset patterns

David Taylor david-taylor at blueyonder.co.uk.invalid
Sun Sep 8 16:12:44 UTC 2013


On 07/09/2013 18:51, Jared Watkins wrote:
> Hello list...
>
> I'm setting up several ntp servers in data centers located in many different places around the world peering off sometimes "local" sometimes distant stratum1 internet servers. (GPS backed atomic oscillators are in the works but not yet available). I've tried different configurations including setting up the more distant systems as stratum3 to our other stratum2 boxes to keep the traffic on the private wan. In pretty much all cases though when graphing the offsets I'm seeing sine wave patterns.  The period is usually about 10 hours and the amplitude can vary by +-25ms for the 'closer' ones and as much as +-100ms for the more distant ones.
>
> I also noticed that when doing a restart on one of the worst boxes.. it initially returned to a near zero offset and then fell right back into the wave pattern at the same point as if uninterrupted which I thought was very strange.  These are normal off the shelf (read cheap) servers and are reporting drift file times of between 1-8... which I believe means the local clocks are ok.
>
> I've searched the docs and the archives but I've not seen any discussion of this.  Any ideas on what would cause it?
>
> For some specifics.. the most distant boxes that are now configured as stratum3 have delay times of between 200 and 300ms and corresponding jitter of <2 and <3ms   The 'better' ones acting as stratum2 have delays of <10 to 35ms for the external servers and jitter all <1ms.
>
> Thanks,
> Jared

Jared,

I've seen oscillation effects as well - in my case with GPS/PPS sources 
fed to Raspberry Pi Linux servers.

   http://www.satsignal.eu/mrtg/performance_ntp.php

The effect is shown sampled too coarsely there (MRTG uses 5 minute 
samples), closer examination of the offset statistics shows a frequency 
of 37 cycles/hour for Raspberry Pi #1, and 45 cycles/hours for Raspberry 
Pi #2.  The magnitude of the oscillation is about 3.5 microseconds 
peak-to-peak, much worse than the offset when the oscillation is not 
present.  Sometimes restarting NTP may fix the problem, sometimes a 
reboot is required, and at other times a power-down reboot, but 
sometimes none of these steps fixes the issue.

To me, it seems like some uninitialised variable, but I would be 
surprised to find that in the NTP code.  This has happened with many of 
the development versions since I've had the Raspberry Pi cards, so 
perhaps since 4.2.7p314.  I've not tested earlier versions.

Whether this is related to what you are seeing I don't know.
-- 
Cheers,
David
Web: http://www.satsignal.eu



More information about the questions mailing list