Subject: Re: revised ethernet/802.xx input header processing.
To: Bill Sommerfeld <sommerfeld@orchard.arlington.ma.us>
From: Bill Studenmund <wrstuden@nas.nasa.gov>
List: tech-net
Date: 10/30/1999 17:14:18
On Sat, 30 Oct 1999, Bill Sommerfeld wrote:

> there's also the more complex 802.x framing scheme (which can be used
> on ethernet, but isn't, normally, at least for IP), which replaces the
> ethernet type with a 2-byte length field (the possible values of which
> are fortunately not valid ethertypes), and then follows it with an LLC
> header, which includes a source and destination protocol id ("LSAP")
> (why anyone would want to send a packet from X.25 to IP is beyond me,
> but I digress..).

Ethertalk phase 2 also uses LLC_SNAP_LSAP for its packets..

> To make life more interesting, one possible LLC type is a "SNAP"
> header, which can contain an ethernet protocol type..
> 
> Other 802.x media (including token ring, fddi, 802.11, and apparently
> some ATM encapsulations) always start off their packets with LLC
> headers.

[rearrangement snipped]

> We can then fix up if_fddisubr.c and if_tokensubr.c and a forthcoming
> if_80211subr.c to call ieee802_input() instead of having
> near-identical copies of code in them.

What would the extra soubroutine overhead add to performance & packet
throughput? (probably mattered a lot more when folks used vaxen for
routers..)

Take care,

Bill