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

Spoon devnull at localhost.com
Mon Apr 2 12:30:37 UTC 2007


Hal Murray wrote:

> Spoon wrote:
> 
>> In my situation, I don't care what time it is, but my clock needs
>> to tick at the correct rate.
> 
> How accurate do you need it?
> 
> You can get "pretty close" if you figure out what the drift
> should be and use that.
> 
> You can get that two ways.
> 
> One is to let ntpd run while connected to the outside world.
> It will leave the answer in log files and /var/ntp/drift or
> wherever.

The problem with that solution is that the frequency offset of the 
system clock changes by a huge amount every time the system reboots.

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

> The other is to do it by hand.  Start your system with no ntpd.
> Compare it's time with a known good clock.  Wait a day/week/whatever.
> Compare the clock again.  Now compute the drift.

Same problem as above.

> If you can measure time within 1 second, you will have to wait
> ~2 weeks to get within 1 ppm.  The drift on some of my boxes is
> ~100 ppm, so that would correct most of the error.
> 
> Note that temperature has a significant influence on drift.
> Your correction factor may change over the seasons or as
> you rearrange things in your machine room.

What are the other parameters that influence frequency offset?

> You may get the wrong answer if you "connect to the outside
> world" by moving the system to another location with a different
> temperature.

I cannot set the frequency offset once and for all and hope for the 
best. There has to be some form of synchronization.

Regards.




More information about the questions mailing list