[c-nsp] full duplex mismatch speed - dynamips

[John Neiberger]

>>>> Adam, you are my new best friend. I've been saying this for the past
>>>> few years and people still think I'm crazy. I flat out refuse to
>>>> manually configure speed and duplex for someone unless it is
>>>> demonstrated (or I can verify) that a duplex mismatch is actually
>>>> happening or there is some other extenuating circumstance that
>>>> requires it.
>>>
>>> JFTR, count me in that camp as well (and that has been discussed here
>>> on c-nsp just a few months ago as well).
>>>
>>> The PA-FE-TX is really the only thing left in our network that needs
>>> force-duplex.
>>>
>>
>>
>> I, believe it or not, still have a 3640 floating around with a NM-4E set
>> to 10/full on all interfaces. It hasn't died yet and does its intended
>> job perfectly.
>>
>> I'd tried to turn up a fast ethernet with Verizon a while back and their
>> policy was to set the ports on the overture to 100/full, no auto.
>>
>> But these are both fringe cases these days. Other than that it's autoneg
>> all the way. It's not the 90's anymore. There's absolutely no reason to
>> not use it.
>
> I think the nailing issue is still more of a problem in "telcos" as opposed
> to "isps". Telcos being in general larger, slower and more fond of process
> and procedure which once instituted are impossible to remove :)
>
> I've worked for a telco which insisted on forcing ports both internally and
> customer facing. They went so far as to force 100/full on ports facing
> phones and printers.
>
> It took a while to remove the practice. I found challenging people to tell
> me when the last time they saw an issue caused by an autonegotiating port
> where the other side wasn't forced helped quite a lot, as no one could think
> of any, it was just something they'd always done, for fear of the autoneg
> boogeyman!
>
> adam.

Even more insidious is the fact that hard-setting both sides to
100/full can actually cause a duplex mismatch. Many NIC drivers still
expect to see an autonegotiating link partner even when hard set. If
they don't detect a partner participating in Nway, they
will--according to spec--assume they are connected to a hub and fall
back to half duplex even when configured for full duplex. This is
important because most Cisco switches made in the last eight years or
so completely disable Nway if you hard set them. If you connect a
PC/server that expects an Nway partner, you'll get a duplex mismatch.
I've personally corrected this on a few hundred occasions.