> It doesn't make sense to attempt nanosecond synchronisation using
> these techniques

Actually, since to a very good approximation the systems are at rest
with respect to one another, it does.  Relativity does say that by
choosing an appropriate inertial frame, two events separated by a
spacelike distance can appear to occur in either order.  But for
computers on this planet, we have a distinguished interial frame, the
one with respect to which both machines are at rest to an excellent
approximation.  And it *does* make sense to (try to) synchronize their
clocks *with respect to that frame*.  (Yes, if you're driving - or
flying! - around with your machine, these assumptions may be violated.
In that case you may *not* get nanosecond accuracy - but that's not a
reason to give up on getting it for stationary machines.)

Yes, the packets are long with respect to the accuracy being attempted.
This was true when NTP was first deployed, too, but it didn't stop it
from achieving amazing things given what it ahd to work with.  I expect
it to be capable of achieving subnanosecond synchronization given a
network link with, proportionally, comparably stable behaviour,
hardware that can do timestamps to that sort of precision, and, yes,
machines close enough to at rest with respect to one another.

What's more, Moore's Law is showing no signs of giving out.  It's not
going to be all that long before networking hardware can fit a whole
packet inside a nanosecond.

/~\ The ASCII				der Mouse
\ / Ribbon Campaign
 X  Against HTML	       mouse@rodents.montreal.qc.ca
/ \ Email!	     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B