[CentOS] Strange NTP problem

[Paul Heinlein]

On Tue, 20 May 2008, Alfred von Campe wrote:

> I have 30 identical Lenovo desktop systems running CentOS 5.1.  On 
> one of those systems the clock is running slow (5+ minutes from 
> yesterday to this morning and another minute since this morning) 
> despite the fact that NTP is running on all of them and they all 
> have the exact same /etc/ntp.conf file (I compared the MD5 sums of 
> that file on all the systems).  Here is the output of "grep ntp 
> /var/log messages" on the system with the problem since I restarted 
> the NTP daemon earlier today:

A slew of 5 min/24 hrs should be in the range of fixable.

> May 20 11:35:38 hepdsw03 ntpd[31792]: frequency initialized 0.000 
> PPM from /var/lib/ntp/drift

This is very suspect. Are there any SELinux or other log messages 
suggesting that ntpd isn't able to write to its drift file? Your local 
clock is definitely drifting, so a 0.000 value is bogus. It may 
indicate that there's a disconnect between ntpd and the filesystem.

I'd be interested in the output of "ntpdc -c kerninfo"; on most 
systems the 'pll frequency' value is a close match to the figure in 
the drift file.

> May 20 11:38:55 hepdsw03 ntpd[31792]: synchronized to LOCAL(0), 
> stratum 10
> May 20 11:38:55 hepdsw03 ntpd[31792]: kernel time sync disabled 0001
> May 20 11:39:59 hepdsw03 ntpd[31792]: synchronized to 10.101.32.104, 
> stratum 3

This is ungood. Sync-ing to local before your network time server 
means that your machine doesn't want to believe your server -- and you 
should see a "kernel time sync enabled" message once the machine has 
sync-ed with the time server.

You said the machines are identical. Could there be any variation in 
the BIOS revision level or its settings? Sometimes ACPI stuff can mess 
up ntp.

Also -- the log messages you provide have no "step time server" 
reference. Do you have a valid /etc/ntp/step-tickers file?

-- 
Paul Heinlein <> heinlein at madboa.com <> http://www.madboa.com/