>Is this paper anywhere around, online?

The relevant details are in xntp-kernel.tar.Z on louie.udel.edu.
I forget the exact path.


>Where's the "free-running counter" live?

It's an undocumented turbochannel bus-cycle counter on the IO asic.


>if it's a reasonably univeral register then the thing to do would be
>to use it for microtime() only -- at each clock interrupt, get its
>value, then when microtime is called, use the offset from that value,
>etc.

3100s and 5k/200s don't have this register.  I think other decstations
do. In any case one wants to discipline the time sources to avoid
introducing cumulative roundoff errors.  Dave Mills makes a living
doing research in this kind of thing; if possible, I'd take his code
and use it.

--Jonathan