[ntp:questions] Different behavior between ntpd 4.2.0 and 4.2.4?

Johnny Ljunggren j.ljunggren at no.parkairsystems.com
Sat Oct 4 20:14:25 UTC 2008

Thanks for the input

>> LAN connections ok, and restart of ntpd. I let ntpd sync and see with
>+ ntpq -c peers that server nanna is in use. If I disconnect the lan
>+ connection to nanna eventually ntpq -c peers shows that local clock is
>+ being used. However, if I connect the lan to nanna again ntpq -c peers
>+ shows that nanna is never accepted as ntp server, and local clock is
>+ used no matter how long I wait.
>> Doing the same with ntpd 4.2.0 automatically reinstates nanna as ntp

> Interesting.  If there is such a difference, it might explain why we are 
> getting more reports of machines locking onto local clocks.

> Never configure a local clock on a pure client and only configure one on 
> a server if you understand the tradeoffs.  Even without it, the clock 
> frequency will continue to be corrected during a server outage.

I tried to keep the first post brief, but I'll fill in some more data here. This machine is also a backup server for another machine on the network. A slightly strange redundant setup. A typical setup of ours is a small number of computers (say 10) where two of the machines work as clients to a hardware NTP server (GPS-based) that has been provided by us or to us. One has got the setup in the first post, with the local clock, and the second has gps time server and server one as servers to connect to. As I understand it server two would not sync to server one if connection to the gps ntp server is lost without local clock on server one?

These computers are set up with dual everything so the reason is to have server one and server two always in sync, so if one of the servers is down the rest of the machines will still be in sync.

> Also, I hope you own nanna, as using minpoll 4 on a public server is 
> likely to result in a kiss of death. You may want to review the "Slow 
> convergence" thread to see if your reason for overriding the poll limits 
> possibly is similar to the one that Dave Mills considers to take the 
> application outside the scope of NTP.

It's ours :) so no problem there. I'm not sure why minpoll is there in the first place, since we don't really need it. I'll read the thread and do some tests without it.

questions mailing list
questions at lists.ntp.org

More information about the questions mailing list