[Beowulf] raw ethernet

[Tim Mattox]

On Thu, 22 Jul 2004 10:14:43 -0400 (EDT), Robert G. Brown
<rgb at phy.duke.edu> wrote:
> On Thu, 22 Jul 2004, Daniel Ridge wrote:
> > It's easy to take hardware too seriously. If you are building a cluster
> > with
> > a private network, you don't need to worry about IANA and what they
> > think
> > your Ethernet MAC addresses should be. Just staple your node number into
> > your hardware address* and be done with it. Essentially all network
> > adapters
> > provide some way to load a user-supplied MAC address into the card and
> > nearly all linux network device drivers expose this facility.
> >
> > * and make sure that the resulting ethernet address is a legit unicast
> > address
> >
> > Regards,
> >       Dan Ridge
[sniped great anecdote from rgb on how bad things can happen if you
just wing it]
>   c) It isn't that hard to do correctly.
[snip]
Exactly... the standard includes room for locally administered MAC addresses.

The one thing to make sure when modifying the MAC address of any
device, is to make sure you set the Locally Administered bit in the MAC.
See section 3, Figure 1 in:
http://standards.ieee.org/regauth/groupmac/tutorial.html

If your script or tool forces the U/L bit to be set (and the Group bit
to be cleared),
then you would never conflict with a MAC address that's "from the factory".
However, you do need to worry about conflicting with other locally administered
MACs.  As long as you have a router (or firewall preferably) between your
cluster and the "campus" network, then even that is not a concern.

As far as the beginning of this thread, as Joachim Worringen mentioned,
the GAMMA project is something really worth looking into as far as raw
ethernet protocols for clusters.  It's open source, and has a long
development life, including an MPI port.
-- 
Tim Mattox - tmattox at gmail.com - http://homepage.mac.com/tmattox/