[ntp:hackers] ntpd as pinball machine
David L. Mills
mills at udel.edu
Fri Jul 20 01:20:59 UTC 2007
As experiment I started a known good machine with a broken frequency
file (-500 PPM) and let it coast for a few days. To heighten anxiety, I
left the kernel and Autokey running. While the expected frequent steps
continued during that period, nothing unexpected happened. However,
after three days the frequency error was still in the 350 PPM range,
ssomething more heroic was needed.
The code has been modified to count the number of stepa in one hour. If
more than two, the discipline state machine is forced to the FREQ state
where it measures the frequency directly. With this change and the same
initial condition, the time was within 1 ms and the frequency within 1
PPM in about one hour. It took four steps to do it.
Truly, this is wrangling the herd.
I'm not completely happy with the code that knows when to write the
frequency file. Workin' on it.
More information about the hackers