[ntp:hackers] [gsoc]student applying for idea "Implement adjfreq()"

Harlan Stenn stenn at ntp.org
Mon Mar 19 23:10:39 UTC 2012


Hal wrote:

> The problem is that the startup code that computes the TSC frequency doesn't 
> get a repeatable answer.  It's OK if the answer isn't super-accurate, the 
> drift calculations will correct for small errors.  The problem is that it's 
> not repeatable so if you reboot your system the old drift calculation isn't 
> valid any more.

The same problem happens when folks use tickadj to change the value of
"tick" away from the default, and then restart ntpd using a drift file
value that was based on the "other" version of "tick".

Years ago I tried a few times to get the value of tick written to the
driftfile, but that didn't happen.

I gather "more recent:" versions of the linux kernel do a better job of
coming up with the same value of the frequency.

Even so, I still think there is benefit to writing this value (when
available) and the value of tick (when available) somewhere stable (like
the drift file) so we can at least notice when these values change.

H


More information about the hackers mailing list