Re: [chrony-dev] Issue with version 1.25 and GPRS/GSM connections

[Miroslav Lichvar <mlichvar@xxxxxxxxxx>]

On Wed, Aug 10, 2011 at 12:31:16PM +0200, Harald Krammer wrote:
> Hello,
> I am using Chrony 1.25 and got a strange problem in the field. 
> So it can happen that the time synchronisations stops after few 
> hours. A restart of Chrony helps to get the synchronisations working 
> again, but the reason for that behaviour isn't clear to me. I can not 
> reproduce it, but I'm assuming this issue exists only on  
> GPRS/GSM connections (->high jitter).
> 
> Wireshark-Capture in case of the problem:
> http://hkr.at/ntp.pcap

The dump shows that the network jitter is in tenths of second and the
total roundtrip time can be over second. The reference ID changes
after the third iburst sample, but there are 8 iburst requests total,
which means at least samples number 4-7 were ignored. (it would be
useful to see the measurements and tracking log to confirm that)

I think I'm able to reproduce the problem in clknetsim. What happens
is that skew after the third sample is so high that the NTP test4
(peer dispersion < 16 seconds) fails even for the minimum delay on
your network and the source is stuck with the three samples. In the
simulations I see skews over 100 (not ppm!), so the roundtrip time
would have to be at most ~15 ms to pass the dispersion test.

Limiting the skew to some reasonable maximum value should fix this
problem. A workaround is to not use the iburst flag.

But you said it can happen after few hours, that would probably be a
different problem. Is it possible that you only noticed it after few
hours?

Thanks,

-- 
Miroslav Lichvar

---
To unsubscribe email chrony-dev-request@xxxxxxxxxxxxxxxxxxxx with "unsubscribe" in the subject.
For help email chrony-dev-request@xxxxxxxxxxxxxxxxxxxx with "help" in the subject.
Trouble?  Email listmaster@xxxxxxxxxxxxxxxxxxxx.