[ntp:questions] NTPd is not sync for some poll interval values
Richard B. Gilbert
rgilbert88 at comcast.net
Thu Feb 19 20:43:19 UTC 2009
Gaurav Narain Mathur wrote:
>> From: unruh-spam at physics.ubc.ca
>> To: questions at lists.ntp.org
>> Date: Fri, 6 Feb 2009 14:18:09 +0000
>> Subject: Re: [ntp:questions] NTPd is not sync for some poll interval values
>> narainmg at hotmail.com (Gaurav Narain Mathur) writes:
>>> I have a problem where the NTPd would not sync the time with the
>>> timeserver (10.0.0.1) if minpoll value is 7 and maxpoll is 8. It works
>>> if the minpoll is 4 maxpoll is 5. Can anybody please help me out in
>>> identifying what the problem could be? The "flash" never becomes
>>> "0" I think. Also another host on the same subnet sync with the same
>>> poll values and with the same server (10.0.0.1).
>> Your offset its too large. Use ntpd -g to have it get rid of the large
>> offset initially.
> Thanks,. I'll try that. What puzzled me most was that a machine on the
> same subnet, syncing with the same time server and with the same initial
> time as the one I am working with syncs ok. The machine I am interested
> in does not.
> Digging in a little deeper, I have found that this might have something
> to do with the routine precision values calculated on the two machines.
> In the one where the minpoll and maxpoll of 7 and 8 work, the precision
> is almost always calculated between 1-5 usec. The one I am working with
> calculates the precision between 70-80 usec. I think this skews the
> peer_unfit calculation to always be above the threshold value and the
> peer is never deemed fit in my machine.
I think you are confusing precision with accuracy! Precision is the
smallest possible time change in two consecutive readings of the clock.
It's usually expressed as a negative power of two. On the Sun SPARC
platform running the Solaris O/S, precision is -21.
Accuracy is the difference between the time on your machine and the
atomic clock at the root of the NTP tree.
More information about the questions