> You would be better off running xntpd on both hosts, with one
> peering with the other.
>
> We *really* should remove timed from NetBSD.

Hm, why?  In many settings, sub-millisecond (or even sub-second)
accuracy is not a requirement, and timed has the advantage over
xntpd that it's "configure-less", you just need to turn it on if
there's a local timed master available.

As for size, I get a smaller physical memory footprint from timed
than from xntpd, even though the code size isn't smaller than with a
stripped-down xntpd (the uninitialized global array of "hosttbl"
structs in timed seems to be mostly responsible for this):

rype% ps axuw | egrep PID\|xntpd\|timed
USER    PID %CPU %MEM   VSZ  RSS TT  STAT STARTED       TIME COMMAND
root     97  0.0  1.6   268  392 v0- S<   Sat04PM    2:12.39 /local/sbin/xntpd 
root    100  0.0  0.7   376  172 ??  Is   Sat04PM    0:15.27 timed -M 
he    15736  0.0  0.0   536    0 p3  RV   11:34AM    0:00.00 egrep PID|xntpd|timed (tcsh)
rype% size /local/sbin/xntpd /usr/sbin/timed
text    data    bss     dec     hex
106496  16384   75488   198368  306e0   /local/sbin/xntpd
36864   4096    306984  347944  54f28   /usr/sbin/timed
rype%

Oh, BTW, I've never seen xntpd give up synchronizing the clock on
this host, an i386 still running 1.2.

- H=E5vard