Potentially useful things to know about the Debian NTP packages...

 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

A list of Public NTP Time Servers is available on the web at

http://www.eecis.udel.edu/~mills/ntp/servers.htm

Please make sure to read *and follow* the Rules of Engagement there, please!

 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

The ntp-simple, ntp-refclock, and ntpdate packages all share a common config
file for listing servers at /etc/default/ntp-servers.  There is debconf 
support in those packages for setting a list of servers.  In the case of
ntpdate, any changes made to that file manually or through the debconf
interface will take effect the next time /etc/init.d/ntpdate runs.  

If you want to manually set up /etc/default/ntp-servers, it needs to have a
single line like 

NTPSERVERS="server_one server_two etc" 

listing one or more space-separated servers.  Any other content in the file
will currently be ignored.

For the daemon packages, it's more complicated, because the /etc/ntp.conf file
does not support an include file syntax.  There is, therefore, a debconf
variable to indicate whether the package maintainer scripts should re-create
/etc/ntp.conf each time the data changes.  I'm not really happy about how that
works right now, and there's at least one bug open against the packages about
it, so I'm investigating Manoj's ucf package as a potential way of doing a
better job handling changes in /etc/ntp.conf.  Stay tuned.

 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

Several people have reported that ntpd fails on SMP boxes unless the "Enhanced
Real-Time Clock" support is enabled in the kernel.  This is known to be 
essential on SMP Alpha systems, and is believed to also be necessary on SMP
Intel systems.

 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

By default, ntpd will log via syslog().  Since Debian does not have a LOG_NTP
facility defined, the daemon will use LOG_DAEMON, leading to ntpd log entries
going to /var/log/daemon.log.  If you define a logfile location in 
/etc/ntp.conf, the daemon will do direct filesystem writes to the specified
file, avoiding syslog().  Previous Debian packages did this, with the side
effect that they had to ship a weekly cron job that stopped the daemon, 
rotated the log, then restarted the daemon.  This is moderately evil for 
high-stratum NTP servers, where ntpd should be allowed to run more or less
forever.  To handle both modes of logging, the cron.weekly script is still 
provided to rotate /var/log/ntpd if present, but it will not cycle the daemon 
if syslog is in use.  By default, new installations of this package will use 
syslog.

 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

NTP and hwclock issues:  hwclock (from the util-linux package) is normally
called on startup and shutdown. You must ensure that hwclock --adjust is never
called (make sure it is disabled in /etc/init.d/hwclock.sh) in a host which
has ntpd active.  You should allow hwclock --systohc to be called on shutdown
unless you are running the NANO kernel patch, because the kernel does not 
fully update the RTC time, and it could be off by a multiple of 30 minutes in
the next boot if hwclock --systohc is never called by the shutdown sequence.
See the hwclock README files in the util-linux documentation for more 
information.

 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -

If you want to configure a machine to use its own processor clock as a
reference, so that for example you can get a set of other machines to sync to
it even if the cluster of machines isn't synced to anything else, then you 
need to make sure to use a less-attractive stratum... just in case the cluster
someday mingles time with the real world.  An example of how to do this is
to configure the "authoritative" machine's ntp.conf like so:

server 127.127.1.1
fudge 127.127.1.1 stratum 10

Then, have the other machines peer with each other to some extent, and peer
(maybe using 'server' instead of 'peer') with the authoritative server, and 
they'll all go to stratum 11 sync'ed to this machine eventually... and if you
ever get a better source of time, everything will latch onto it instead.

 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -