Re: AS prepend, MED or local_pref question.

"John W. Stewart III" <jstewart@isi.edu> Fri, 31 January 1997 20:54 UTC

Received: from cnri by ietf.org id aa01808; 31 Jan 97 15:54 EST
Received: from merit.edu by CNRI.Reston.VA.US id aa20247; 31 Jan 97 15:54 EST
Received: (from daemon@localhost) by merit.edu (8.8.5/merit-2.0) id PAA20057 for idr-outgoing; Fri, 31 Jan 1997 15:12:54 -0500 (EST)
Received: from interlock.ans.net (interlock.ans.net [147.225.5.5]) by merit.edu (8.8.5/merit-2.0) with SMTP id PAA20047 for <bgp@merit.edu>; Fri, 31 Jan 1997 15:12:50 -0500 (EST)
Received: by interlock.ans.net id AA27184 (InterLock SMTP Gateway 3.0 for bgp@ans.net); Fri, 31 Jan 1997 15:12:48 -0500
Received: by interlock.ans.net (Internal Mail Agent-1); Fri, 31 Jan 1997 15:12:48 -0500
Message-Id: <199701312012.AA18151@metro.isi.edu>
To: Steve Francis <com2srf@ucsbvm.ucsb.edu>
Cc: bgp@ans.net
Subject: Re: AS prepend, MED or local_pref question.
In-Reply-To: Your message of "Fri, 31 Jan 1997 11:35:41 PST." <1.5.4.16.19970131113317.27173390@commsvcs.commserv.ucsb.edu>
X-Phone: +1 703 812 3704
From: "John W. Stewart III" <jstewart@isi.edu>
Date: Fri, 31 Jan 1997 15:12:38 -0500
Sender: owner-idr@merit.edu
Precedence: bulk

my personal suggestion would be to statically set MEDs
to force the traffic the way you want it to go.  this
isn't the most scaleable way to do it, but it sounds
like (1) you're not talking about a huge number of
prefixes and (2) you've got to do prefix-based stuff
anyway, and that's kind of inherently not scaleable.
keep in mind that nothing within the protocol ensures
symmetry -- if you care about it, then you have to
manage the configs such that the static-MED-setting
results in symmetry

as for why there are multiple ways, some ways make
more sense in certain contexts.  for example, in
your situation you're gonna have to have a fair
amount of manual config on all 4 routers, so you
can do it any way.  but large service providers
might not want to put so much config in place, and
instead will take the community-based approach
because it's more scaleable for them (i.e., all of
the manual config stuff is in the customer router).
as for manipulating the AS-paths, while it is
sometimes the only tool that people have and so
therefore is sometimes legitimately used, i myself
consider it little more than a hack

/jws

 > Hi.  
 > Apologies if this is not the correct forum for this question.
 > 
 > I wish to connect two separate routers in AS1 to two separate routers in
 > AS2.  I want half the networks in AS1 to use router1 going to AS2, and half
 > the networks to router 2 to go to AS2. And I want the return traffic to
 > follow the same paths, symmetrically.
 > 
 > Now, I can think of three ways to do this:
 > - at router 1, prepend a bunch of AS numbers to the routes advertisements
 > for the networks I want to come back via router2, and at router 2 prepend
 > for teh networks that should come back via router1.
 > - assign different communities to the two sets of routes, which are
 > announced both places, and apply a local_pref at the routers in AS2, so at
 > the router 1 peer in AS2, one set of routes would have a lower local_pref
 > applied, and a higher local_pref at router 2 peer in AS2.
 > - Use the MED attribute for the two sets of routes.
 > 
 > Is there a preferred way to do this?  Any advantage/disadvantage to any
 > specific approach?
 > 
 > Is it necessary for the protocol to provide so many ways to achieve the same
 > thing?  It seems to me that MED is redundant.
 > 
 > TIA
 > Steve Francis
 > Network Analyst
 > Communications Services
 > UCSB	
 > ph (805) 893 7775	fax (805) 893 7272
 >