Linux: concerns over systemd adoption and Debian's decision to switch

Stephen Satchell list at satchell.net
Thu Oct 23 06:51:18 UTC 2014


On 10/22/2014 08:20 PM, Simon Lyall wrote:
> On Wed, 22 Oct 2014, Miles Fidelman wrote:
>> And maybe, you should check out some of the upstream bug reports re.
>> systemd interactions with NTP.
> 
> If you think the current situation is all good then maybe you should
> look at other bugs for ntp. eg this one I that affected me with Ubuntu
> Disktop. They only run time syncing when the network is bounced so if
> you have a stable network then your machine will never sync:
> 
> https://bugs.launchpad.net/ubuntu/+source/ntp/+bug/1178933
> 
> Note that the importance of this has been set to "wishlist".
> 

Looking at the ticket you submitted, I see this statement:

> On Ubuntu 12.04 desktop, in System Settings > Time & Date, when the
> time is set to update Automatically from Internet it syncs once but
> then drifts out over a period of days.
> 
> In Syslog I can see that ntpdate is called on boot but is never called again.

I'm a long-time user of NTP, and what you are asking for is a no-good
way of doing things.  What you are supposed to do is use the ntpdate(8)
utility *ONCE* on boot to initially set the system clock, then you have
ntpd(8) running to do two things for you:  sync up to one or more time
sources, and discipline the local clock.

What is "discipline the local clock?"  This is the process of
determining the *exact* frequency of the crystal clock in your local
computer, and tuning your local clock hardware so that local real time
is in sync with the world.  This is because the accuracy of the crystal
is specified to be rather loose (hundreds of parts per million), but the
relative accuracy of a crystal is actually within tens of parts or even
single-digit parts per million.  So if you can measure the drift, you
can in software compensate for it so you get a true-chimer.  That's what
the ntpd daemon does.

Now, what is supposed to start the ntpd daemon in Ubuntu?

DHCP has no significant effects on the filters used in ntpd(8).

I didn't have any problem getting this to work "the right way" on
Slackware, Red Hat, and Debian.  It works "out of the box" with Fedora
and CentOS, to the point that I don't even think much about it other
than in the GUI to point to my local time source on system install.

It even works with the Ubuntu 8.04.4 LTS server that was forced down my
throat a few years back, which I'm in the process of trying to retire in
favor of a CentOS 6.5 server implementation.  (At least THAT
Ubuntu-loving sysadmin is gone now.)

What happens on your box when you do "/etc/init.d/ntp start"?  And how
frequently do you reboot your box?  It sounds like it's up all the time,
which means the local clock can be trimmed very accurately indeed.

That's the SERVER way of running a time synchronization.  So it would
appear that you have a quarrel with GUI support, not with NTP itself.



More information about the NANOG mailing list