In message <20020822220546.D8CC18B4@ab.use.net>Sean Doran writes
>
>| 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).
>
>Teensy nit: variance and *symmetry* of delay.

Just so. However, painful (best forgotten) experience with the early
internet link to NZ (a 9600bit/sec satellite link, horrendously
overloaded in the FIX-West->NZ direction) showed that even with
grossly asymmetric delays, NTP doesn't do too shabbily.

Perhaps Mark Davies recalls the numbers better than I?

>I think Jonathan knows alot of this but...


... many of these conservative people have reason to worry: they do
have user bases which care about "accurate to nanosecond" vs. the more
pedestrian ``time to nanosecond resolution, accurate to within a
microsecond or so'' which NTP-heads hope for.  These users include, if
memory serves, VLBI and outfits like SLAC.

Dave Mills was kind enough to introduce Jeff Mogul, Ulrich W, PHK and
I (rfc2783) to some of the US NIST staff who care about this stuff.
They were interested in the PPS-API: not necesarily for NTP,
but they can definitely use it.