[ntp:questions] Fwd: NTP and PPS calibration interval (convergence speed)

unruh unruh at invalid.ca
Wed May 29 22:35:28 UTC 2013


On 2013-05-29, Miguel Gon?alves <miguel.barbosa.goncalves at gmail.com> wrote:
> On Tue, May 28, 2013 at 1:46 PM, unruh <unruh at invalid.ca> wrote:
>> Not really. What would be more helpful would be graphs of the behaviour,
>> of the offsets over time, including times when the machines are heavily
>> used and times when they are not. (See for example
>> www.theory.phsyics.ubc.ca/chrony/chrony.html and past graphs as well)
>> Unfortunately the gps attached to my one ntp machine has gone nuts
>> (Garmin 18-- both my Garmins failed after about about 2 years. Not a
>> very good piece of gear-- but you can look at older graphs from 3 or 4
>> years ago to see how the ntp run machine behaved.
>
> Hi Unruh!
>
> I've tried to include the graphs but I forgot that this mailing list
> is a gateway for a newsgroup that doesn't accept binaries.
>
> If someone wants to look at them just look here
>
> http://www.flickr.com/photos/miguelbarbosagoncalves/8883017818/in/photostream/
> and
>
> http://www.flickr.com/photos/miguelbarbosagoncalves/8882397117/in/photostream/
>
> They are equal machines even though one has a Garmin 18 LVC and the
> other one a Sure.

I see nothing in those graphs about your system going crazy. It is
disciplining the computer to an offset of less than 5 microseconds. and
with the drift accuracy to about .05PPM. That is perfectly fine and is
what I would expect from a PPS interrupt trigged clock source. Not at
all sure what you are complaining about. There is the oscillation in the
offset with the Garmin. I have seen that as well, and have no idea what
it is about ( some sort of instability in ntpd?) It might be due to the
poll interval ( your offset oscillates with about a 20min period) the
PPS should not have any trouble with the clock filter (all items have
zero delay, so no selection due to delay occurs). Could you pls repeat
what you think the problem is?
>
> Cheers,
> Miguel



More information about the questions mailing list