Re: BGP-4+ [QOS et al]

"John G. Scudder" <jgs@ieng.com> Tue, 24 December 1996 19:56 UTC

Received: from cnri by ietf.org id aa06790; 24 Dec 96 14:56 EST
Received: from merit.edu by CNRI.Reston.VA.US id aa15035; 24 Dec 96 14:56 EST
Received: (from daemon@localhost) by merit.edu (8.8.4/merit-2.0) id OAA27806 for idr-outgoing; Tue, 24 Dec 1996 14:31:06 -0500 (EST)
Received: from interlock.ans.net (interlock.ans.net [147.225.5.5]) by merit.edu (8.8.4/merit-2.0) with SMTP id OAA27801 for <bgp@merit.edu>; Tue, 24 Dec 1996 14:31:03 -0500 (EST)
Received: by interlock.ans.net id AA14727 (InterLock SMTP Gateway 3.0 for bgp@ans.net); Tue, 24 Dec 1996 14:30:59 -0500
Received: by interlock.ans.net (Internal Mail Agent-1); Tue, 24 Dec 1996 14:30:59 -0500
Message-Id: <v0300780baee1da54b53c@[194.154.144.50]>
In-Reply-To: <199612182011.MAA23562@puli.cisco.com>
References: Your message of "Wed, 18 Dec 96 14:41:12 EST." <199612181941.OAA26301@idrp.merit.net>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Date: Tue, 24 Dec 1996 21:26:29 +0200
To: Yakov Rekhter <yakov@cisco.com>
From: "John G. Scudder" <jgs@ieng.com>
Subject: Re: BGP-4+ [QOS et al]
Cc: 6bone@isi.edu, dkatz@cisco.com, yakov@cisco.com, bgp@ans.net
Sender: owner-idr@merit.edu
Precedence: bulk

One thing: being able to associate NLRI with (only) an address family seems
like an unnecessary restriction.  Instead, how about permitting the
association of NLRI with an arbitrary RIB, as qualified by <whatever>?  The
additional parameter that springs to mind is QOS, so a RIB would be
identified by <AFI, QOS>.  There isn't any particular need to restrict
things so that only AFI and QOS are usable to qualify the RIB, though.

One way to do this would be to use a more general RIB ID of some sort, and
associate a RIB with <AFI, other stuff> in a separate step.  One would want
to make the RIB establishment step sufficiently extensible that one could
(at least) associate a QOS with a RIB.  I see no particular reason to
restrict the <other stuff> part to support only QOS, though.

Alternately, I suppose you could just stick the QOS right into the
attribute each time you send it.  This would be less complicated, at the
cost of each PDU being more piggy.

For bit-counters, using a RIB ID could be useful if you assume that 256
RIBs are sufficient for a particular peer-peer association -- you can save
8 bits per PDU (compared to the current proposal) in that case by making
RIB ID a single octet instead of the two octet AFI.

Note that you can't properly support QOS by making it yet another path
attribute, for the same reason you can't support AFI by making it just a
path attribute in the obvious way -- one may want to advertise (and
withdraw) the same NLRI multiple times with different QOS.

By the way, one conceivable benefit of doing something like this is that it
moves yet another piece of IDRP functionality into BGP.

If this is considered to be a good idea, the question would be when and how
to establish the RIB ID<->RIB attribute binding.

--John

--
John Scudder                        email:  jgs@ieng.com
Internet Engineering Group, LLC     phone:  (313) 669-8800
122 S. Main, Suite 280              fax:    (313) 669-8661
Ann Arbor, MI  48104                www:    http://www.ieng.com