>>>> [bridges, repeaters, and switches, oh my!]
>>> [a repeater is...and introduces no delay...]
>> [What, *no* delay?]
> Epsilon, unless it's a switching hub (i.e. a learning bridge);

For suitable values of "epsilon".  Yes, repeaters introduce a delay,
but that delay is on the order of the signal propagation time from one
end of the wire to the other.  (This is the reason for the rule that no
more than four repeaters may appear in the path between any two hosts:
otherwise, a collision may fail to be noticed before the collision time
has passed, due to the delay the repeaters introduce.  You violate this
rule only at risk of late collisions and/or corrupted packets.)

> they [ie, switches] actually have to process addresses (at minimum,
> for a device that offers what has come to be called "cut through
> switching"), which takes a lot longer than simply regenerating a
> signal.

And not only that, because it mustn't start transmitting until it's
received the destination MAC address, it must be prepared to buffer the
packet in case the destination segment is busy when it decides to start
forwarding.  (I suppose it could simply drop it, but as it's then too
late to reflect the collision back to the originating station this
seems - to me - like a very bad idea.)

Of course, the problem with cut-through switching is that if the packet
is a runt or jabber, you end up forwarding a runt or jabber to the
destination segment.  But you do delay the packet by only the collision
listen interval plus 48 bit times plus processing delay (which, one
hopes, is overlapped with the 48 bit times), versus delaying by the
entire packet size for a store-and-forward switch.

					der Mouse

			       mouse@rodents.montreal.qc.ca
		     7D C8 61 52 5D E7 2D 39  4E F1 31 3E E8 B3 27 4B