[ntp:questions] Using different timebase for ntpd
mills at udel.edu
Mon Mar 23 23:46:48 UTC 2009
An external timebase has been implemented in two ways. The most
desireable is via the kernel and an appropriate shared memory device
driver. An alternative way is to used a driver such as the KSI/Odetics
TPRO refclock driver. There might even be a refclock driver for the old
TrueTime TT-560, although Symmetricom gutted the TT product line when
they ate them.
There are numberous gremlins in your project. If all you want to do is
stabilize the motherboard timer oscillator, poke the timer pin with a
stable stignal and let the kernel/NTP feedback loop do the discipline
and interpolation. If you want to do frequency discipline, you will need
to engineer some critical loop parameters. See Chapter 4 in my book.
Patrick Loschmidt wrote:
>I'm trying to replace the linux/unix system clock by a special clock
>maintained on a PCI card. The clock can be set, rate adjusted, etc., so
>I have the equivalent for gettimeofday(), settimeofday(), adjtime(), etc.
>Unfortunately, I'm confused by the number of different abstraction
>levels within the NTP package. I first started replacing the functions
>in libntp/systime.c (get_systime, adj_systime, set_systime). Then I
>found out, that there are direct calls to adjtime() from within ntpd.
>Further I see some ntp_adjtime() calls.
>Can anybody enlighten me, where would be the best place to use a
>different timebase for ntpd, or do I have to go trough all the functions?
>questions mailing list
>questions at lists.ntp.org
More information about the questions