[c-nsp] BGP weight

[Paul Cairney]

As mentioned, weight is not redistributed however if you only have a single router that is not an issues. What is more commonly used is local preference, which is redistributed to iBGP neighbors. 

You can assign localpref using route maps for each peer however assigning any sort of metric to each provider, if one is higher than the other it will always be used unless a more specifc route exists from another peer.

Please see http://www.cisco.com/warp/public/459/25.shtml for further information on the BGP bestpath selection proccess.

In order to attempt to balance traffic across both links, you need to find some way to differentiate between routes from each provider. One of the simplest ways of doing this would be based on prefix lenght, however for more fine grained control you should ask your upstreams to provide you with bgp communities.

If your upstreams support them, you could use communities to prefer routes received from one of your upstreams transits or to assign a lower preference to prefixes received from a congested peering exchange or a 'budget' transit porvider.

Anyway YMMV depending on your upstreams willingness to provide communities.



On Tue, Apr 18, 2006 at 08:22:56PM +0200, Kristian Larsson wrote:
> On Tue, Apr 18, 2006 at 10:57:37AM -0700, Voll, Scott wrote:
> > Forgive me if this is a newbie question, but I have setup my BGP with
> > two ISPs and everything seems to be running well.  But a large
> > percentage of my outbound traffic is going out one ISP and I would like
> > it to go out the other.  The AS_Path to most routes are pretty equal (3
> > AS's away)  I have setup a weight to each Neighbor.  One 110 and the
> > other 90.  but there are routes under a show IP BGP that have 0 weight.
> > Shouldn't all routes have one weight or the other?
> First off be aware that weight does not distribute
> the traffic so that one gets more than the other.
> If all prefixes learned from one neighbor has a
> weight of 110 and prefixes from the other has 90
> all with 110 will be chosen. Thus you effectively
> chose one peer or the other.
> What you must do is come up with some "sneaky"
> system of marking a few more routes with 110 on
> one of the peers to balance it out.
> This is not as easy as one might think.
> 
> Now, why your routes are not marked at all is
> probably due to some misconfiguration.
> Could we please have a look at the configuration?
> 
> Regards,
>   Kristian