[ntp:questions] What prevents continuous time within an operating system ?

MAYER Hans mayer at iiasa.ac.at
Wed Jan 20 19:37:06 UTC 2016


Many thanks for your answer. 

> If those are peerstats offsets, plot those two against loopstats offset and ... 

Yes, it is the peerstats offset. 
Ok, I will look for the loopstats.
What I have done I made a plot with the difference of both peerstats ( some small awk scripting ) 
and this shows more or less a straight line. 
so both offsets go at the same time into the same direction. 

> Post the plot and ntpq output here for more help

I will do. Currently I am testing with the system something different. 

> Set min/maxpoll for local references to 4. 
This I will give a try. This was also mentioned by Charles. Thanks. 

> Set a preferred source to label the PPS ticks e.g. server ... prefer.
> Add good LAN or nearby internet sources.

Here I can't follow you. 
I have two PPS sources at the same server. The measured offset for each is  typically in a range of +- 5 us ( see image ) 
Any time source over the network is much worse.  Typically by a factor of 10. 
And I need this with a poll interval of 64 time seconds or less. 
Finally I don't want to poll a foreign time server in minute intervals. 

> Have you set affinity for ntpd or is it pinned to a non-zero CPU
> to reduce contention? 

What do you mean ?  Never heard "non-zero CPU"
I googled a little bit but results didn't make sense for me. 

> Have or can you disable any power management features?

There is nothing like this. Always running, always idle. 

// Hans 





________________________________________
From: questions [questions-bounces+mayer=iiasa.ac.at at lists.ntp.org] On Behalf Of Brian Inglis [Brian.Inglis at SystematicSw.ab.ca]
Sent: Monday, January 18, 2016 12:42 AM
To: questions at lists.ntp.org
Subject: Re: [ntp:questions] What prevents continuous time within an operating system ?

On 2016-01-16 08:26, MAYER Hans wrote:
> I connected a GPS based 1PPS signal to a Banana Pi as stratum 0 source.
> I watched the offset in the peerstats file and it goes up and down sometimes.
> So I assumed this happens due to the fading of the signal from the GPS satellites.
> Later on I installed a second 1PPS signal based on a rubidium frequency standard.
> So I realized that both time-stamps have the same issue. See here
> https://farm2.staticflickr.com/1628/24296211776_a7e56c71b4_b.jpg
> https://farm2.staticflickr.com/1642/24214179542_89e2b28d21_b.jpg
>
> 127.127.22.1 is rubidium, 127.127.22.0 is GPS.
> 127.127.22.1 is the time source, 127.127.22.0 is configured "noselect".
> As you can see 127.127.22.1 offset is typically between +/- 5 micro seconds
> but currently at around 18:00 h the offset is 40 micro seconds.
> the same behavior with the GPS signal.
> Both fit very well. So it's not the time source, it must be the operating system.
> And of course this doesn't happen always at the same time, it's different.
>
> I am not sure if this is related to the Banana Pi and it's operating system.
> OS is Bananian OS which is actually a Debian OS with kernel  3.4.108+
> It has 1GB of RAM where almost all is free. Nothing else as NTPD and GPSD
> is running.

If those are peerstats offsets, plot those two against loopstats offset and
frequency drift, and you may see a correlation between system clock frequency
drift and PPS timing.

Watch ntpq -p -c rv -c cv values and check what states the daemon, references,
and other sources show.

Post the plot and ntpq output here for more help.

> I also wonder why NTP needs almost 2 hours to find back to the "normal" offset
> if it does read the time stamp each 32 seconds ?

Other suggestions to improve stability:

Set min/maxpoll for local references to 4.
Set a preferred source to label the PPS ticks e.g. server ... prefer.
Add good LAN or nearby internet sources.

Post server entries in ntp.conf here for more help.

Have you set affinity for ntpd or is it pinned to a non-zero CPU
to reduce contention?
Have or can you disable any power management features?

--
Take care. Thanks, Brian Inglis, Calgary, Alberta, Canada
_______________________________________________
questions mailing list
questions at lists.ntp.org
http://lists.ntp.org/listinfo/questions


More information about the questions mailing list