MAC address confusion

Tue Mar 3 01:31:19 UTC 2009

> Date: Sun, 1 Mar 2009 00:40:00 +0200
> From: Saku Ytti <saku+nanog at ytti.fi>
> 
> On (2009-02-28 22:38 +0100), JAKO Andras wrote:
> 
> Hey,
> 
> > > http://standards.ieee.org/regauth/oui/oui.txt
> > > 02-07-01   (hex)		RACAL-DATACOM
> > 
> > After enabling DECnet routing, the interface MAC address turns to 
> > something like this:
> >   Hardware is BCM1250 Internal MAC, address is aa00.0400.0a04 (bia 000b.bffd.fc1a)
> > AA-00-04   (hex)		DIGITAL EQUIPMENT CORPORATION
> 
> > in the list. I don't know what 02-07-01 is, but I guess that could be 
> > something similar: The OUI belongs to a company, but they don't use the 
> > addresses to burn them into interface cards.
> 
> I guess you shouldn't be able to assign 02 (or AA) to a company for ethernet
> number, much in the same way you shouldn't be able to assign RFC1918.
> However you are right, it seems that these are unexplained exceptions to rules:
> 
> http://www.iana.org/assignments/ethernet-numbers
> 'some of the known addresses do not follow the scheme (e.g., AA0003; 02xxxx)'
> 
> Would be interesting to see what are the historical reasons.Perhaps they simply
> predate the scheme or some might not even co-exist in ethernet network to begin
> with, in which case they might be better documented elsewhere.
>   In any case, to avoid collision with history, better start with 06 which
> seems cruft free, instead of 02, when choosing local MAC address prefix.
> 
> As a side note, the 40 prefix used as local MAC in IXP here, seems to be
> just mistake in assuming ethernet follows tokenring in numbering scheme.

OK. Time for a history lesson.

In the olden days, Xerox, the company that made the first Ethernet and,
in a consortium with Digital and Intel, developed the first public
standard for Ethernet. (This was referred to as the Blue Book or the DIX
Ethernet.) 

At the time, the Ethernet networking schemes all assumed that the MAC
address would be changes to embed the network address. This would allow
hosts to know the MAC address of any other system running the same
protocol. XNS and DECnet both did this and registered the locally
assigned prefix that they would be using. This seemed quite reasonable
at the time. Several companies registered the OID they were going to use
for their networking systems as well as the hardware ones. This assured
that no two hosts would have locally assigned address.

Then along came IP over Ethernet and ARP. IP won the protocol battle
(pretty easily) and the concept of dong MAC to host mapping via a
protocol like ARP or NDP became the norm. But the OIDs for several
protocols were already in the registry when it was turned over to the
IEEE after 802.3 was ratified. IEEE agreed to retain existing
registrations and they have remained there.

The scheme was actually not a bad one, just one that never really caught
on for modern networking.
-- 
R. Kevin Oberman, Network Engineer
Energy Sciences Network (ESnet)
Ernest O. Lawrence Berkeley National Laboratory (Berkeley Lab)
E-mail: oberman at es.net			Phone: +1 510 486-8634
Key fingerprint:059B 2DDF 031C 9BA3 14A4  EADA 927D EBB3 987B 3751