> I believe that 802.2 addressing is in fact what is used for 802.3,  
> so it is "the same" for all IEEE media layers (for some definition of  
> the term).  Is there something specific that is lacking in this  
> support?
> 
> Ethernet V2 and 802.2 do indeed coexist on the same wire.  The two  
> bytes following the source and destination are treated as a packet  
> length field for 802.2, if smaller than the Ethernet MTU (1514), and  
> as a protocol id for Ethernet V2 if larger than the MTU.

As of today, the Ethernet V2 frame format *is* officially part of the
IEEE 802.x standards. Read recent postings in comp.dcom.lans.ethernet
from Rich Seifert about this.

> Probably, today, it's anyone's guess.  IP is almost exclusively V2  
> on ethernet (except for some legacy gear from HP and compatible  
> implementations).  AppleTalk uses 802.2 exclusively.  Then there's  
> IPX, which, as I understand it, can use any of a handful of  
> mechanisms from V2, 802.2 (with and without SNAP headers, ...).  I'd  
> expect that IP is generally the lion's share, but it really depends  
> on the installation.

I made some measurements a few years ago in a reasonably large university
environment. At the time, we routed IP, IPX, Appletalk and DECnet. On the
university backbone, something like 95% of the packets had the Ethernet
type field, while less than 5% had a length field. (Doesn't mean that the
backbone was 95% IP, there was also a good bit IPX with 0x8137 type field.)
Of the total IP traffic, less than 0.1% used a length field.

Steinar Haug, Nethelp consulting, sthaug@nethelp.no