[c-nsp] Ethernet Interfaces Speed and Duplex - Force or Auto

[Kaegler, Mike]

FWIW, the following is an old email I sent to our internal IT list last
year reinforcing our current practice of auto/auto on all ports except
where autoneg is proven to fail.
It's a bit of historical review of reasons to hard-set.

~ Begin ~

>> Why did we continue to use 100/full?"

Social inertia.

Back in the day (1995) when FastEthernet was new, not all the
manufacturers had their... stuff... together. 802.3u (autoneg) had been
written, and everyone started working to build hardware on it. Most
manufacturers interpreted the standard one way... predictably cisco (and
a few others) interpreted it another. Cue surprised face. Disparate
hardware would not autonegotiate, or if the hardware was smart enough to
handle the different autonegotiation methods, it would be stupid enough
to do so unreliably.
Network admins, burned, disabled autonegotiation.

The vendor disagreements were settled in 1998. 802.3 was released.
In 1999, the current 802.3ab was released, to support gigabit and some
extra code pages.

Its 2009, a decade later. Well, for some of the network equipment we
run, its 4 years later. Anyway. The reasons to disable autoneg have been
dead for a very long time. In my entire career, I have only seen it fail
spectacularly once, oddly enough on a packeteer when it failed and
closed the relay to connect its in and out ports.

So the advantage is never having this problem.
The disadvantage is the theoretical possibility of having a suboptimum
link speed if the ethernet firmware on the server side doesn't strobe
correctly to signal capabilities and noone notices.

No vendors currently recommend disabling autoneg.
-porkchop


~ End ~