[c-nsp] The myths of autonegotiate vs forced (was: full duplex mismatch speed - dynamips)

[Mikael Abrahamsson]

On Fri, 20 Aug 2010, Peter Rathlev wrote:

> The duplex thing is about Ethernet legacy; you don't have the problem on 
> fiber links, since these can't be simplex (AFAIK, please correct me if 
> I'm wrong). But any copper port _might_ be connected to a hub from 1993 
> some day, and the standard tries to make that work.

I've used AUI-transceivers with 10BASE-FL (fiber) that by definition were 
half duplex, but this was in the mid 90:ties so it's doesn't really apply 
anymore.

It emulated collissions over the fiber layer as far as I could discern.

+1 on the "use autoneg unless you really have to force duplex".

In case of having to force, do use "speed auto / duplex full" if your 
equipment supports it. If you connect something that is "auto / auto" to 
it then autoneg still succeeds, if you connect something that is "10/full" 
or "100/full" to it, then you still get something that works. If you force 
half duplex or connect it to a hub then you'll get a duplex mismatch, but 
at least you have a much higher chance of it working even if you do 
something wrong.

Duplex seems to be a big mystery in most organizations, I've heard so many 
misconceptions about it it's scary, I'd say it's one of the biggest causes 
of bad performance in modern networking, at least in equipment which 
humans can actually configure. The simple L2 switches with no 
configuration is more likely to work.

-- 
Mikael Abrahamsson    email: swmike at swm.pp.se