Umm, I think the example being given was this:
You are piloting your avater locally on you host. The motion of the
windmill is deterministic. Your browser performs collision detection
locally, with the windmill. Thus, when you walk into the windmill, the
response that you see is instantaneous -- no network lag. Your
collision detection routine then checks the deterministic equations
of motion for the windmill. These equations might include a rule=20
that says something like "if (collision) then {slow by 10%}".
After your browser has finished esarranging your avatars anatomy, it
will then send update messages to the central server. The central=20
server will then resend these updates to anyone in the vicinity who=20
might have been watching you (your avatar). Because of network =
delay,
server load, network load, other people may not see you walk into,
and be hit by the windmill, until 300ms, a second, several seconds
after you actually hit it. =20
The point here is=20
-- your responses to what you do is instantaneous, as it should be.
-- what other people see may be delayed, but so what? Did it rsally
matter that they see what you do when you do it? Wouldn't the=20
delay be acceptable?
-- Its scalable. The server ("beains") doesn't have to be a =
supercomputer,
doing collision detection for 500 avatars against 500 animated =
objects
50 times a second. It only needs to keep track of who is likely to =
be=20
looking at what, so that it limits re-narrowcast of updates to =
only
those browsers that might actually be looking at the event.
I think this is what was trying to be said.
--linas