Date: Fri, 7 Jun 96 16:26:13 EDT
From: Milan Merhar <milan@xyplex.com>
To: P8021@hepnrc.hep.net
Subject: Priority Bits Discussion


Steve said (apologies in advance for yanking out of context):
   Using 3 bits of the VLAN tag:

   Reduces VLAN tag space to 13 bits.  Personally, I thought 16 was too small.  I 
   believe that we will kill the long term prospects of VLANs if we continue to 
   steal "name" space from the field.

   Another way of saying this is that the argument for 3 bits of priority
   (when 4 or 5 would do for all applications imagined today) is for architectural
   growth. However, the same argument can be made for keeping the VLAN ID 16 bits wide.

I suppose that if we REALLY need to eat our cake and have it too, we
could define a subset of priorities (1 or 2 bits) that can be
signalled with the 802.1Q single-level encapsulation scheme, and
define a full 16-bit flags field in the two-level encapsulation
scheme. This expanded field would have room for three bits of
priority as well as the RIF flag for source routing, and all of the 
other flags that we suspect might be needed in a big network.

Disadvantages include two different flag field formats to parse
(identified by different protocol types, and possibly implemented in
different products, however) and additional length in the two-level
header.

A meta-disadvantage might be that we _haven't_ limited the VLAN
qidentification space to a "small" number such as 4K or 8K VLANs, so that
simple linear lookup schemes become more costly. Let's defer the "how
small is small enough to be cheap" vs. "how big is big enough to be
useful" discussion to another time, though....

- milan

----------
Milan J. Merhar, Consulting Engineer 
Whittaker Xyplex   295 Foster St. Littleton, MA 01460  USA
Internet: milan@xyplex.com    Phone:(508)952-4713   Fax:(508)952-4887