Re: errors on my interfaces

From: Mikael Abrahamsson (
Date: Sat Jan 05 2002 - 10:02:47 EST

On Sat, 5 Jan 2002, Donald Becker wrote:

> I write this many times a day:
> Don't use forced-full-duplex!
> Anytime the speed or duplex is forced, autonegotation is disabled. Read

That is actually incorred. It should be "Anytime the speed AND duplex is
forced". It is a very common problem and there are way too many people
that miss this. The above is true for cisco for instance. Speed 10 and
duplex auto will still autoneg but only at 10 megabit. With Extreme
Networks you can either "auto on" (both speed/duplex) or "auto off" (and
then you have to set both). As far as I know switches should be able to
still auto but only announce certain capabilities (Sun boxen does this for
instance, you can turn off capabilities (10/half, 10/full and 100/half)
and still use auto, and the machine will only announce 100/full over NWAY
and only use that, but any switch connected to it will auto to 100/full)

Otherwise you are right of course. Rule of thumb:

If you force a speed/duplex at one end, you have to force it at the other
as well. Use auto at both ends and it works 99% of the time.

Only time I force full duplex is over fiber links when I use media
converters from copper to fiber and back again, and then I make damn sure
I force it at both ends.

If you really REALLY have to force it and you have problems, always force
half duplex and see if it works with ok performance. If you force half
duplex at one end, the other party will either autoneg to half duplex, or
not autoneg at all and default to half duplex, in any case you'll be ok.
The exception here again is the fiber links, especially over long
distances. You do not want to use half duplex over 500 meters (or even

Signs of duplex problems:

Late collisions, CRC errors and other errors, and few megabits of thruput
and packet loss.

Mikael Abrahamsson    email:

- To unsubscribe from this list: send the line "unsubscribe linux-net" in the body of a message to More majordomo info at

This archive was generated by hypermail 2b29 : Mon Jan 07 2002 - 21:00:40 EST