[prev in list] [next in list] [prev in thread] [next in thread] 

List:       cisco-nsp
Subject:    RE: set as-path prepend
From:       "Martin, Christian" <cmartin () gnilink ! net>
Date:       2001-04-26 4:55:40
[Download RAW message or body]

Neil,

No problem!

Given two paths from two different ASs, same AS Path length, route A with
MED 1 and IGP cost 2, route B with MED 2 and IGP cost 1.  The router chooses
route B because it has a lower IGP cost and MEDs aren't compared between
ASs.  Then along comes route C with same AS path as A but with MED 3 and it
is external.  Well, now it chooses route C, even though it has a higher MED
than A because A was dropped out of the process in the previous step.  If
the PATHs ae sorted in groups based on AS, then route A would have beaten
route C, and then route B would win (and stay installed).  By configuring
bgp deterministic-med, you can achieve this behavior.  To take it a step
further, configuring bgp always-compare-med will cause route A to win,
because it has the best MED, even though the IGP cost is different.

So to summarize:

Deterministic MED causes BGP to sort prefixes in the RIB based on AS, then
MED.
Always-compare-med causes BGP to sort prefixes in the RIB based on MED
alone.
If non-deterministic is applied, the sorting is done on based on ASPATH.

Note that Local Pref, weight, and local origination still come first.  Also
note that always-compare-med must be used with caution, as it can also cause
loops if not deployed throughout an AS.  Besides, since MEDs are relevant
only to the AS announcing them, doing a comparison based on them isn't very
useful, unless you overwrite them inbound.  Then they can make for some good
load-balancing between ASNs.

chris

> -----Original Message-----
> From: Neil J. McRae [mailto:neil@COLT.NET]
> Sent: Wednesday, April 25, 2001 3:21 AM
> To: Martin, Christian; Desmarais, Jonathan; cisco-nsp@puck.nether.net
> Subject: RE: set as-path prepend
> 
> 
> > See http://www.cisco.com/warp/public/770/fn12942.html for 
> details on this
> > issue.
> 
> Chris,
> Thanks for this, although we saw this problem before this 
> field notice went
> out, TAC didn't come up with a solution so we scaled down our route
> reflecting deployment - What does BGP deterministic MED actually do?
> 
> Cheers,
> Neil.
> 

[prev in list] [next in list] [prev in thread] [next in thread]