[ntp:questions] Remaining synced on an unsynchronised peer?

Richard B. Gilbert rgilbert88 at comcast.net
Mon Nov 30 20:24:03 UTC 2009

Michael Butow wrote:
> Hello,
> I am seeing behaviour with RHEL5.3 (ntp-4.2.2p1-9.el5) which I have a 
> question about, and am wondering whether someone knows an answer (or can 
> tell me that it's a bug):
> I have 2 machines, call them host1 and host2, which both get their time 
> from the same external NTP servers ntp1 and ntp2.
> Host1 and host2 are also peered. In short, each of their ntp.conf looks 
> like (with the partner host as "peer"):
>   server ntp1 version 4 iburst prefer
>   server ntp2 version 4 iburst 5 
>   peer <partner> version 4 iburst
> When I disconnect the LAN that connects host1/host2 to the external 
> servers ntp1 and ntp2, I see something that I do not understand:
> After both host1 and host2 lose reachability to their external servers, 
> one of them will become synchronised on its peer, ie. ntpq -p will show a 
> '*' on the peer, despite the peer not having any '*' (no external clocks).
> The reachability of the peers to each other will fluctuate between 376 
> and 377.
> There will be a message in /var/log/messages on the host that has no '*' 
> that it has lost sychronisation, however, on the other one there is no 
> such message (this conforms to what is visible using ntpq -p).
> I waited more than 77 minutes, but the machine with the '*' did not 
> become unsynchronised, even though it's peer had no more reliable time 
> source.
> I was expecting both host1 & host2 to become unsynchronised after a short 
> time - our application depends on it and it was a test that was 
> successful on a prior version (running on RHEL3.8, ntp-4.1.2-4.EL3.1).
> Am I seeing a bug, or do I have a wrong expectation/understanding?
> Any insights would be appreciated.
> Regards,
> Michael
> P.S. Detail that I omitted: we use minpoll=maxpoll=5 (for historic 
> reasons in our system).

"Historical Reasons" is a pretty feeble excuse.  NTPD will, if permitted 
to do so, adjust that polling interval as needed to get the most 
accurate time.  By pegging minpoll and maxpoll you are crippling the 
ability of NTPD to get the greatest possible accuracy.

More information about the questions mailing list