Comment 16 for bug 114505

Thomas Sprinkmeier wrote:
> @feisty,
>
> If your DHCP server isn't providing NTP servers then DHCP should have nothing to do with NTP.
> "/etc/rcS.d/S40networking" should have the NIC's up and configured long before "/etc/rc[2345].d/S23ntp" tries to start NTP.
I see the bug on feisty with a fast local network and local dhcp server.

>
> Does you modem do 'lazy connects'?
> I'm wondering if the DNS queries from NTP are the first thing the modem hears, causing it to connect.
> If the modem takes too long to connect then the DNS queries will fail and NTP won't start.
> Could you add a few IP addresses (rather than FQDN's) to your ntp.conf file?
> What does "ntpq -p" say?
>
> @Phillipp
> "Would it hurt to restart ntp whenever a network interface comes up?"
> Yes, but not much.
> NTP builds up information on remote servers and local clock drift. This is lost on restart, meaning NTP won't work as well as it could (i.e. it'll take longer to sync, and even longer to tweak the local clock frequency).
> Arguably this is not a big problem for most situations where all you want is accuracy within a few seconds (in which case "ntpdate" may be a better choice anyway).

I don't think this is a problem: the local clock drift information is
stored in /var/lib/ntp/ntp.drift, so it is not lost upon restart. The
information about servers builds up fast, and in the meantime, ntp's
application of the stored drift info to the local clock keeps it
accurate to a fraction of a second.

As of 2005, the advice was that ntp does need to be restarted:
http://lists.ntp.isc.org/pipermail/questions/2005-August/006333.html

As of a week ago, it seems that the underlying ntp problem may have been
fixed: https://support.ntp.org/bugs/show_bug.cgi?id=51
(see the bottom of that bug report).

Eric

>
> The polling interval NTP uses tends to increase as time goes on (it starts at 2^6 seconds (roughly once a minute), I've seen it as high as 2^16 (roughly every 18 hours)).
> Re-starting NTP will mean a shorter polling interval and more load on the servers (especially if you use "iburst").
>
> I think the thinking behind the "exit 0" in "/etc/network/if-up.d/ntp"
> is that, if your network address changes, it was probably DHCP that
> changed it and DHCP should be responsible for re-starting NTP.
>