I agree, but I actually use TWO local, internal time servers. also,
I try to have all the other machines use each other as peers.

I found before with only one internal server, my other internal machines
could sometimes drift as much as a full second off, and when using nfs,
this sometimes caused make to complain that timestamps were in the future.

so now, I have two ntp servers, each one using as many stratum 2
internet time servers as possible (I think I usually get 6 or so for
each server). call my internal servers s1 and s2, and say I also have
other machines on the network called m1, m2, and m3

then m1's ntp.conf looks like this:

driftfile /etc/ntp/drift
server s1
server s2
peer m2
peer m3

it's cool. s1 and s2 end up being stratum 3 servers, and m1, m2, and m3
are all stratum 4 servers.

ntpq -p

will list where the machine is getting its time from.




On Tue, Nov 23, 2004 at 02:32:38PM -0700, Robert Ambrose wrote:
> 
> My personal $.02:
> 
> When setting up a network to use NTP I designate a box to be the local NTP
> server.  All the other local devices get their time off the local NTP
> server.  This box is the only system which gets it time from the Internet.
> 
> Whenever passable get your NTP source for the NTP server from your local
> ISP. All ISPs will have (or should have) an internal NTP server.  
> 
> You can try to call the ISP and (usually after a certain amount of pain)  
> get the name or address of their own NTP server.  Good luck if the ISP is
> one of the major national ones.
> 
> Failing that you can try to find a NTP server by doing a traceroute to an
> arbitrary host on the Internet and then guess, based on domain names, a
> host name for a NTP server.  time.ISP-DOMAIN is often available.
> 
> NTP is meant to be hierarchal.  The idea is to keep most of the NTP
> traffic local and to keep the load off the stratum 1 servers.
> 
> The ntpq utility is your main debugging tool.  Some (perhaps) useful tips:
> 
> 1. ntpd will not sync when the time is way out of sync.  Use the ntpdate
>    utility before starting ntpd.
> 
> 2. It can take a while for ntpd to sync.
> 
> 3. To see what's going on:
> 	a) From the ntpq utility, use the 'as' command.  The fields of 
> 	   interest are condition and assID.
> 	b) If the condition is 'sys.peer', time is being maintained.
> 	c) If the condition is not 'sys.peer', use 'rv <assID>', where
> 	   <assID> is the value from the assID column.  The main thing I
> 	   look at are the time samples.  ntpd needs at least 4 samples 
> 	   before it will start keeping time.
> 
> OTH, rna
---------------------------------------------------
PLUG-discuss mailing list - PLUG-discuss@lists.plug.phoenix.az.us
To subscribe, unsubscribe, or to change  you mail settings:
http://lists.PLUG.phoenix.az.us/mailman/listinfo/plug-discuss