[ntp:questions] Getting NTP to correct only the clock skew

Harlan Stenn stenn at ntp.isc.org
Fri Apr 6 04:08:40 UTC 2007


>>> In article <461555bb$0$6969$426a34cc at news.free.fr>, Spoon <root at localhost> writes:

Spoon> Harlan Stenn wrote:
>> Are you using iburst?

Spoon> Yes.

>> Do you have good values in your ntp.drift files?

Spoon> No!

Spoon> cf. thread titled "Clock skew changes drastically between reboots"

Spoon> On the systems I have to work with, ntpd finds a different frequency
Spoon> offset every time. There is no "good" value for the frequency offset.

Are you running the kernel time discipline stuff?  If so, please see
http://bugs.ntp.isc.org/740 .

Spoon> I have to remove the drift file, and let ntpd spend 15 minutes (?) in
Spoon> state 3 to compute the correct frequency offset, then 15-60 minutes
Spoon> to fix the time offset.

Do you reboot the machine after removing the drift file?  Again, I'm asking
about the kernel time stuff

>> The other approach may be to get a good PPS signal distributed to all
>> your machines.

Spoon> I'm trying to get my boss to test a GPS-based solution. The problem
Spoon> is that the systems will be running in underground server rooms, and
Spoon> some clients don't like having to run a cable to a GPS antenna.

You might be able to use a single (decent) real PPS source and get the
signal propagated that way.  Also note the RIPE driver, which uses a Trimble
GPS unit where all the smarts are in the GPS antenna.  The RIPE driver can
use standard ethernet wiring to get the signal to its desired location.

And if you get cell phone coverage, there are some CDMA refclocks, too.

H




More information about the questions mailing list