[ntp:questions] NTPD can take 10 hours to achieve stability

Harlan Stenn stenn at ntp.org
Mon Apr 18 19:37:28 UTC 2011

Bill wrote:

> However, do the following test. While ntpd is running, use adjtimex to
> change the system clock rate by say 20PPM. Watch the errors and see
> how long it takes to settle down to the original accuracy.

I could be wrong but I think you are describing a documented "don't do
it - this is a bad idea" case.

Please note I am freshly awake, and almost functional.  I have not dug
thru any documents on this point.

What are the documented operational parameters of ntp's model?  Is a
spontaneous frequency change of at least 20ppm one of them?

If so, what is the documented "recovery" time for that?  Does ntpd meet
that specification?

If a spontaneous 20ppm frequency change is not within the stated
operational parameters, why should it be?  As an aside, in reality (ie,
the "natural" reality, not the one where humans mess with it just to see
what happens) what sort of spontaneous frequency changes are actually

What happens if you *stop* ntpd, then shift the freq by 20ppm, and then
restart ntpd?


More information about the questions mailing list