[c-nsp] does duplex mismatch affect UDP throughput?

[Dantzig, Brian]

On Tue, Aug 02, 2011 Scott Granadose wrote:

> Nowadays, more vendors have problems with hard settings "not quite
working"
> (because that code doesn't get tested so well, I'd assume) than in the
last century.

> The notable exception being the Cisco 7200 (single-port) FastEthernet
modules 
> (PA and IO-board).  Those can not do autoneg at all, and need their
counterpart 
> to be hard set.


> Vendor problems aside, the problems with hard setting is not so much
> "things not working as set up" (that usually works) but "things get
> replaced".  So, for example, a device breaks, gets replaced by a 
> new one, and the person doing the replacement forgets to set the
> ethernet port to "hard set".  Been there, seen that, and *these*
> problems are much more frequent these days than "just set all ends to
> autoneg".

Carriers probably stick with fixed duplex as a legacy issue. Auto
negotiation used to be somewhat iffy. Sun in particular had problems
with it in the past. While I've not had problems with Sun for about 8-10
years. Once this gets baked into your network, it's hard to get rid of.
It also eliminates the possability of a negtiation issue. If both sides
are auto, there is a chance it won't work right. If both are full, it
works. You might call this determinalistic provisioning.

A good thing to remember is that if you are auto-negotiating, and your
side comes up half-duplex, the other side is probably full-duplex no
auto-negotiate. Yes, you could be connected to some odd equipment that
is actualy running half but, 9 out of 10 times it's configured
full-no-auto. 


Brian Dantzig