clock synchronization

[R. Hirschfeld]

Thanks Bill, Julian, and Rick for your responses concerning my clock
problem.

Perhaps I didn't make clear that the problem is not with the Hardware
Clock (RTC).  That is remarkably stable and accurate and in fact my
system does not lose time under MacOS (pre-OSX), only under Linux.

The problem is rather with the interrupt-driven system timer, which
runs about .03% slow if I don't adjust it.  So if it is a hardware
issue, I suspect it is not with the battery that powers the CMOS clock
but rather with the oscillator that regulates the processor, which is
more difficult to replace.

I did ask Apple to check the battery when my machine was recently in
for repair and they didn't replace it, so either it was OK or they
neglected to check.

There are no man pages for "ntp" or "ntpd" on my system, but I just
discovered some html docs in /usr/share/doc.  I notice that the ntp
package includes /sbin/tickadj with functionality similar to adjtimex.
The "hwclock" man page is interesting reading but is really more
applicable to keeping a drifting RTC in sync with an accurate system
timer rather than the other way around.

I wasn't sure whether ntpd had to recalibrate and settle in each time,
which is why I thought it might be useful to pre-adjust the kernel
variables.  But I gather that /etc/ntp/drift keeps track of this (as
/etc/adjtime does for the RTC).

Ray