[ntp:questions] offset slowly increases while ntpd is running?
Richard B. gilbert
rgilbert88 at comcast.net
Wed Feb 7 14:35:16 UTC 2007
matthew.garman at gmail.com wrote:
> We have an ntp server running on "server3", to which three other
> machines sync: "server1", "server2", and "server6".
> servers 1 and 2 sync correctly to server3. server6 is a new machine
> that we are having troubles getting to sync (and stay synced).
> Basically, from server6, we can run "ntpdate server3". Then we
> start the ntp daemon on server6. Using "ntpdate -d server3" and/or
> "ntpq -p", we can watch the offset between server6 and 3 slowly
> creep upwards. Today, within about a 10 minute window, the offset
> drifted up to about 50ms. A rapid increase in offset seemed to
> correlate with heavy CPU load.
> We are still doing testing to determine the specifics of the
> Has anyone seen anything similar to this? We're not particularly
> well versed with ntp, so we're not even sure if we're testing things
> FWIW, this new machine is a Dell PowerEdge 1950 with dual quad-core
> Thank you for any assistance!
> questions mailing list
> questions at lists.ntp.isc.org
You didn't mention the O/S you are running!
Some flavors of Linux behave badly during periods of high disk usage.
It appears that interrupts are either masked or disabled for a period
long enough to lose two or more clock ticks. If you are running Linux
and the clock is slow with respect to the server this may be your problem.
Some newer machines do fancy things with power management that result in
changing the CPU speed. This is practically guaranteed confuse ntpd.
It is also possible that server six has a really bad clock! NTP cannot
correct a clock with a frequency error greater then 500 PPM. If,
without running ntpd, this machine gains or loses more than 43 seconds
per day, it will never synch! If the machine really keeps time this
badly without running ntpd, ask Dell to fix it under warranty!
More information about the questions