[ntp:hackers] difference between kernel and daemon mode

Miroslav Lichvar mlichvar at redhat.com
Thu Apr 23 15:02:36 UTC 2009


I was doing some tests with latest Linux kernels and I noticed a
difference between ntpd running in kernel and daemon mode. I know they
are not exactly the same, but I was expecting similar responses,
especially in low poll intervals where FLL doesn't kick in.

What I'm seeing is this: after ntpd running in daemon mode has
stabilized and the clock is stepped by 0.1 sec, the clock drift
subtracted by real hw drift will peak at only about third of the
peak in kernel mode. Is this ok? I wrote a simple simulator of the
daemon mode and I'm getting results closer to the kernel mode.

Here is a table with my results, also with the time it takes to cross
zero offset after the step:

mode		poll	zero(s) drift peak(ppm)
daemon		4	900	8
daemon sim	4	700	22
kernel		4	700	20
daemon		6	4000	1.8
daemon sim	6	3000	5.5
kernel		6	3000	7

Can someone confirm this is the correct behavior, or what results
should I be getting for kernel and daemon mode?


Miroslav Lichvar

More information about the hackers mailing list