[SUMMARY] OBP setting for full duplex NIC?

From: Neil Schellenberger <nschelle_at_crosskeys.com>
Date: Wed May 15 2002 - 12:30:07 EDT

Briefly, the original question was:

  Is there any way to set the hme link mode (duplex) at the OBP level?

There were only three responses, one of which was a clipping of the
FAQ material on what to set in /etc/system, which isn't directly
germane to this problem.

Darren Dunham helpfully pointed out a twist that I hadn't considered:

    > I can't simply change the boot image /etc/system because a) that won't
    > help with the tftp portion of the boot, and b) I need to be able to
    > jumpstart both half and full-duplex machines.  I suppose I could
    > maintain two boot images, but that seems like overkill (as well as
    > being error prone).

    You will need to anyway.  While an OBP command would probably help the
    TFTP portion, once the kernel loads the hme driver, it will probably
    completely reset the interface, and redo negotiation (unless set).

    If at all possible, I recommend staying away from /etc/system mods and
    instead put in an explicit rc script that ndds the interfaces.  Such a
    script could have logic in it to do its work only on a set of machines
    that you need it to (perhaps identified by hostname or IP).

Matthew Stier kindly pointed me at:


Unfortunately, this material doesn't really help in this particular
situation.  During further discussion with him, he opined:

    As to your problem, I'd forget using full-duplex.  Before upgrading our
    network to 4000's, we used to have 5000's with these cards, and just
    locked them at half-duplex.

    Although Cisco support full-duplex on it's 10Base-T ports, 98% of all
    NICs do not.  It is a hardware limitation. (To support full-duplex,
    CSMA/CD must be disabled; and old 10Base-T designs did not permit this.)

    Full Duplexing did not become popular, or even a part of the standard
    until the ratification of the 100Base-T standards, and it was hapazardly
    backfitted onto 10Base-T cards. (Actually, I believe that those cards are
    really 10/100 cards with the 100Base-T feature permanently disabled.)

The only other piece of infomation I've managed to find is how to set
the link speed at the OBP via nvedit:

    probe-all install-console banner
    apply transfer-speed=10 /path/to/hme/device

Unfortunately, there doesn't seem to be an equivalent setting for link
mode.  Sigh.

Basically, it seems that you're SOL before /etc/system or /etc/rc?.d
are read.  Even then, in order to have a single jumpstart miniroot,
you'd need to have some sort of external logic for an rc script to
decide whether a given hosts should be full or half so that it can ndd
appropriately.  All of this seems predjudicial to having a simple,
self-maintaining, automatic jumpstart system in this environment.  It
seems that the sysadmin will need to manually change the Cisco port
configuration while machines are jumpstarting and then switch it back
again after.


Neil Schellenberger          | Voice : (613) 599-2300 ext. 8445
Orchestream Americas Corp.   | Fax   : (613) 599-2330
350 Terry Fox Drive          | E-Mail: nschellenberger@orchestream.com
Kanata ON, Canada, K2K 2W5   | URL   : http://www.orchestream.com/
sunmanagers mailing list
Received on Wed May 15 12:51:55 2002

This archive was generated by hypermail 2.1.8 : Thu Mar 03 2016 - 06:42:43 EST