Re: slow ethernet

[Nick Holland <nick@holland-consulting.net>]

Henning Brauer wrote:
> 
> * Steven S. <ssurdock@engineered-net.com> [2004-02-16 09:18]:
> > In my opinion if you've never experienced problems with auto negotiation
> > your either very lucky or not looking for it.
> 
> no, you are probably doing something else very wrong and autoneg
> failing is an effect, not a cause.
> 
> yet, the "autoneg is problematic" myth keeps spreading, yet it is utter
> bullshit.

No.
Take an iMac 333MHz running OpenBSD.
Plug it into an autonegotiating switch.
It will pick up 100Mbps, Full Duplex.  Why? Because the HW thinks it
is best.
It will choke and cough painfully on the data, however -- sucking data
at a few kB/s.  
Bump it down to 10Mbps, and it works fine.

Why?  Because the bm(4) driver that runs the NIC on this iMac is
imperfect and can't handle 100Mbps.  You can argue that
autonegotiation worked fine, but the problem is, the system AS A WHOLE
doesn't work, and that's what people think is important: does it WORK
AS DESIRED.

The problem with extreme statements is they can be disproven with one
counterexample.  (Henning will deny it was a counter example 8-). 
Autonegotiation CAN BE problematic for a number of reasons.  You must
keep it as an option in your trouble shooting toolkit.

I have been watching something with some 3c905(B rev, I do believe,
possibly some of the older cards, don't think I've tested the C rev
yet) cards recently: they screw up autonegotiation nearly 100% of the
time, at least as they report the speed during install, but so far, I
have found this problem to be asymptomatic -- they work just fine on a
100Mbps FD link even though they just said they were at 10Mbps.  Could
it cause someone problem under some set of conditions?  No idea, but I
am NOT going to say everything is working perfectly all the time.

> i don't have a single box with autoneg problems here, and that is all
> kind of architectures with a _lot_ of different NICs to several
> different cisco switches.

But hardly every NIC, every OS, and every switch around...
A finite number of examples is not how to prove a negative...

I don't think autonegotiation is an evil.  I've had pretty darned good
luck with it, in spite of some pretty cheap switches and stepped-on
cables and second-hand NICs.  HOWEVER, I can't believe that
autonegotiation is the one flaw-free technology we have ever developed
in the computer industry.  I also have to believe the manufacturers of
a lot of equipment agree with me that it shouldn't be trusted blindly,
otherwise, why would they provide the option to manually override on
high-end stuff?

So, can we just agree that none of us has the benefit of perfect
knowledge and experience and let this silly argument die?

Nick.
-- 
http://www.holland-consulting.net