[119241] in North American Network Operators' Group
Re: BGP Traffic Engineering question
daemon@ATHENA.MIT.EDU (Aaron Hopkins)
Tue Nov 10 15:05:01 2009
Date: Tue, 10 Nov 2009 12:04:11 -0800 (PST)
From: Aaron Hopkins <lists@die.net>
To: Drew Weaver <drew.weaver@thenap.com>
In-Reply-To: <F3318834F1F89D46857972DD4B411D70017D1E99CE@EXCHANGE.thenap.com>
Cc: nanog@nanog.org
Errors-To: nanog-bounces+nanog.discuss=bloom-picayune.mit.edu@nanog.org
On Tue, 10 Nov 2009, Drew Weaver wrote:
> If you have several transit providers connected to your network and much
> of your traffic is generally directed by the "BGP tiebreaker" (i.e. lowest
> IP address) is there a way, without specifying on a per-prefix basis to
> prefer the "tie breaker winner" slightly less often?
Assuming Cisco, set "bgp always-compare-med", "bgp deterministic-med", and
in your route-map in, "set origin igp" and "set metric X". You can then
vary X as you see fit as an alternate tie-breaker. As long as you never set
the metric the same on two different paths for the same prefix, it'll never
fall back to router-id.
Depending on the transit provider, you can often match bgp communities to
determine which are customer routes or the region where the announcement was
heard, which you can then use as a tie-breaker when setting the metric.
Barring that, as-path access-lists matching specific path fragments can do
the same thing, but seems to take more work to maintain as relationships
change over time.
-- Aaron
