[rescue] Re: CDDI/FDDI from Cisco (old stuff--any good?)

[Greg A. Woods]

[ On Sunday, August 19, 2001 at 23:56:17 (-0400), Jonathan Katz wrote: ]
> Subject: Re: [rescue] Re: CDDI/FDDI from Cisco (old stuff--any good?)
>
> unswitched FE, full-duplex collides
> unswitched 10Mbit, full-duplex collides

There's only one way to have a full-duplex un-switched connection and
that's to connect two stations back-to-back with a cross-over cable.

full-duplex unswitched networks cannot suffer collisions, by definition
of being "full duplex"!  :-)

> switched 10Mbit, full-duplex collides (I think?)

no there can't be any real collisions -- both the switch and the station
can transmit simultaneously if the connection is full-duplex.

A proper full-duplex NIC or switch port will disable the loopback
circuit and turn off collision detection and jam generators when it's
put in full-duplex mode, no matter whether it's 10baseT or 100baseTX (or
1000baseT).

however unless *all* stations on the switch are full-duplex.....

> Switched FE, full-duplex doesn't have collisions.

What you might see are runts, and sometimes even FCS errors, being
counted as collisions, depending on your hardware and drivers....

Note though that there are two types of switches:  store&forward and
cut-through.  (old fashioned ethernet bridges were effectively
cut-through switches, though I doubt any ever had FDX ports! :-)

Note also that if you've got any half-duplex connections on a
cut-through switch you could also end up with runt packets on the
full-duplex connections too since the switch starts transmitting a
packet as soon as it's examined the header and thus won't be able to
check for collisions until it's already on the outgoing wire.  At that
point it stops transmitting and all that's gone out is either a runt or
broken packet (the missing data will cause an FCS error)

A store&forward switch is optionally allowed to generate PAUSE frames if
its buffers reach a high-water mark, but they may also just drop packets
too, and perhaps even generate runts, though I've not seen the latter
happen on any switch I've used (at least so long as the protocol
negotiation worked right).

If two or more stations transmit packets to a single target station
simultaneously a cut-through switch will probably drop all but one
packet (some cut-through switches can buffer one packet per port if
necessary, but you can imagine how quickly that'll fail).


If I'm not mistaken many of the low-end unmanaged el-cheapo 100baseTX
switches are cut-through switches.

-- 
							Greg A. Woods

+1 416 218-0098      VE3TCP      <gwoods at acm.org>     <woods at robohack.ca>
Planix, Inc. <woods at planix.com>;   Secrets of the Weird <woods at weird.com>