Re: slow ethernet

[Brian Keefer <chort@amaunetsgothique.com>]

On Sun, 2004-02-15 at 10:00, Dom De Vitto wrote:
> Our Cisco 7900s, 7200s, 6500s, 5500s, 2900s, are not garbage 
> (well ok, the 2924/48s are crap, but the 2950 is ok), neither is the
> Cisco, Sun, Dell and HP kit I've hung of them.  
> 
> I've had autoneg problems with all of these, which show up as 
> a mix of errors on the links (going in/out of full duplex causes
> madness).  This is with autoneg on both ends or autoneg on just one
> end. Fixing both ends solves the problems and stops the errors.

and

> On Sun, Feb 15, 2004 at 04:34:43PM -0000, Dom De Vitto wrote:
> > Generically, whatever the OS, and whatever the devices:
> > 
> > Autonegotiation does _not_ work reliably.
> > 
> > It may do at first, but not the next, or fifth, or thirtieth reboot.
> > It may even just change after a little while, and then change back a 
> > while later - even if the cables were plugged in the whole time.

I have to agree here.  I install network appliances for a living, and
while it doesn't happen all the time, it has happened often enough for
our support department to have it in their "ass-biter" list. 
Particularly we've seen a problem with 2900s and old firmware not
autoneging with fxps correctly.

Forcing both sides to 100/full usually fixes the problem, but
occasionaly a firmware update was necessary on the switch before it
would work at anything over 10/half.  Yes, that's right... we've had to
hardcode some devices to 10/half after switching out everything, cables,
switch ports, etc in order to get them to work.  Then magically,
everything was great after upgrading the Cisco firmware.

Autoneg might be simple in theory, but experience tells me that
different hardware manufacturers implement it differently, and also
firmware has the potential to be buggy.

-- 
Brian Keefer, CISSP
Systems Engineer
CipherTrust Inc, www.CipherTrust.com