[[ Sean Doran says much...]

All of this is very hard to object with.

However, dont forget that:
  * NTP is very carefully crafted so that what matters ot it
    is not aboslute delay, but *variance* in delay (aka jitter).
    Dave Mills has this down to where he can pull the minimum
    rtt out of the filters on his local-campus machines
    and give very good estimates of the raw,
    guaranteed-not-to-exceed forwarding latency of local switches.

* In internets (small i, aka intranets) the end-systems often are the
  bottleneck.  I dunno about you, but i have more gigabit-connected
  machines than I can keep track of, and I'm waiting for prices  on
  10-gig blades to start falling.


* the coarseness of NetBSD (and lack of Mill's "nanokernel") is
  already a drawback of NetBSD. Even I am looking serious at FreeBSD
  for my GPS radio.  This thread has also kicked off some discussion
  of  PHK's timecounter abstraction   (the idea, not necessarily the extant
  code or kernel  API).