[c-nsp] BGP weight

[Jon Lewis]

On Wed, 19 Apr 2006, Michael K. Smith wrote:

>> Most of the routes I see in my BGP table are three AS's away.  So with a
>> tie (my understanding) the first peer up get's the traffic.
>
> The last step in the algorithm relies on lowest peer IP address, so that is
> the one that will be preferred if it's making it that far.

Long time ago it was that way.  Then cisco changed it to be the oldest 
route wins.  Those of us who don't like such surprises use "bgp bestpath 
compare-routerid".

> In general, I would say you should come up with a community strategy that
> makes interesting netblocks stand out from the rest, and then apply
> route-maps with higher local prefs to those communities.  You could also
> match using as-path access lists and relevant regular expressions, or some
> combination of both.

Like Pete, I prefer to be more gentle and use prepending such that I'm not 
forcing traffic to take a certain provider regardless of what happens to 
the path length via that provider.

----------------------------------------------------------------------
  Jon Lewis                   |  I route
  Senior Network Engineer     |  therefore you are
  Atlantic Net                |
_________ http://www.lewis.org/~jlewis/pgp for PGP public key_________