[c-nsp] Negotiation problem with Catalyst 2950 and Cisco 2821

[John Neiberger]

On 12/4/05, Dave Temkin <dave at ordinaryworld.com> wrote:
> I've also seen numerous instances where both Compaq and Intel NICs had
> driver issues where they lied about being hard-set to 100/full.  The
> driver would report 100/full, however it was obvious from the switch
> counters that it was 100/half, and if you set the switch to auto it would
> auto to 100/full.

That's expected behavior, actually. Sort of. Some NICs or switchports
still participate in autonegotiation when they are hard-set, and some
don't. The Intel NICs do, so they don't play well when you connect
them to a switch port that doesn't, like just about any Cisco switch
made in the last four years.  The NIC thinks that it's link partner is
not capable of full duplex and will try to be helpful by falling back
to half duplex.

This became a big issue when we began replacing 2924XL switches with
2950s. The 2924XLs still participate in autonegotiation if the ports
are hard-set, but the 2950s don't. We immediately began having gobs
and gobs of duplex mismatches when installing the 2950s. We also have
the same problem with our 6500s. The solution was to begin setting
everything to auto unless we knew that it wasn't going to work for
whatever reason. We noticed that a lot of our Sun servers had NICs (or
NIC drivers) that didn't autonegotiate correctly, but everything else
works great on auto.

John