> shouldn't only one cpu take clock interrupts?  or at least, only one
> cpu should update the time.  the others should not.

That's what I would expect too.  However, running an MP kernel with
clockinterrupt as its time source can show results like:

      remote           refid      st t when poll reach   delay   offset  jitter
 ==============================================================================
 +orthanc         83.138.191.59    3 u   80 1024  377    7.812   -7.839   7.812
 +janus           212.13.198.71    3 u   71 1024  377    7.812   -7.148   7.812
 *213.205.157.156 129.69.1.153     2 u  104 1024  377   60.486   -6.891   7.812

which isn't too bad (notice the value of 7.812, which appears to be the
minimum value for delay, jitter and noise on an MP kernel) but can also
show:

      remote           refid      st t when poll reach   delay   offset  jitter
 ==============================================================================
 +orthanc         83.138.191.59    3 u   20   64  377    7.812  481.824 262.121
 +janus           212.13.198.71    3 u   49   64  377    7.812  482.246 275.808
 *213.205.157.156 130.149.17.21    2 u    2   64  377   49.745  504.329 275.861

and syslog entries:

 Jan 31 16:45:27 sparky ntpd[19367]: time reset +0.513950 s
 Jan 31 17:12:28 sparky ntpd[19367]: time reset +0.525693 s
 Jan 31 17:40:32 sparky ntpd[19367]: time reset +0.736815 s
 Jan 31 18:06:53 sparky ntpd[19367]: time reset +0.424785 s

so all is not right with either the time updating or the time reading.

I haven't spotted what causes it to be unable to sync time.  I initially
thought it was load related, but this box was idle (apart from the daily
cron jobs) for 22 hours and then started resetting the time a couple of
times an hour (while still idle).

Thanks,

J

-- 
  My other computer also runs NetBSD    /        Sailing at Newbiggin
        http://www.netbsd.org/        /   http://www.newbigginsailingclub.org/