Re: wireless lans with multiple accesspoints

> Date: Wed, 19 Jun 2002 10:56:17 -0400
> From: Peter Radcliffe <[email protected]>
> Sender: [email protected]
> 
> Kevin Oberman <[email protected]> probably said:
> > The use of that term goes way back and it really is no longer
> > appropriate in the era of full-duplex where collision simply don't
> > exist. But it matters here.
> 
> It is still appropriate in the era of full-duplex, and collisions do
> exist even with full-duplex switches.
> 
> Switches are still the same broadcast domain, you can still get
> collisions for any packet that is not unicast or is broadcast to all
> ports by the switch for another reason. If the per port buffering
> is small or there is just too much traffic ... it's much, much
> less likely to happen, but can still happen.
> 
> Some switches produce collisions to signal the kit on that port to
> back off if the egress port is highly congested, for example.

Simply stated, no. Per 802.3, collisions are a local event. There is
no way for a switch to "send a collision" and the section of 802.3 on
full-duplex is specific that collision detection MUST be disabled. A
switch does not divide a broadcast domain, but does divide a collision
domain.

There is a push-back signaling system in 802.3, but it is not a
collision. Collisions do back-off and retransmit at the MAC
level. Switches are "store and forward" devices, even if they do
cut-through forwarding, and they handle deferrals due to busy ports in
firmware and not the MAC. There simply are no collisions here.

R. Kevin Oberman, Network Engineer
Energy Sciences Network (ESnet)
Ernest O. Lawrence Berkeley National Laboratory (Berkeley Lab)
E-mail: [email protected]			Phone: +1 510 486-8634

To Unsubscribe: send mail to [email protected]
with "unsubscribe freebsd-mobile" in the body of the message