[c-nsp] bgp maximum-paths

[Peter Rathlev]

On Wed, 2010-04-14 at 16:26 -0500, Chris Gotstein wrote:
> Does this seem like a valid way to load balance?
> 
> http://ccnalab.net/bgp-routing/bgp-load-sharing-2-isp/

(For the lazy on the list: The document explains prepending your own AS
on an inbound route-map from each provider, thus making the routes
appear from same AS.)

If it works (the example tends to say it does) it would balance your
outbound traffic only, i.e. what you send to your upstreams.

Multipath balanced BGP-traffic is problematic in that the balancing is
typically some kind flow based of hashing of L3/L4 information, and that
makes it difficult to predict what upstream a given flow ends up using.
Troubleshooting can be very problematic; a problem from one host towards
ftp.example.com might not exist on another host in the same subnet.

For outbound balancing it's probably more desirable to do as Manolo says
and adjust traffic per-AS with local-pref. 

Inbound traffic balancing is another thing completely. The easy method
is asking your upstreams for a list of communities you can use for
selective propagation of your prefixes through their net. The bad (for
the size of the DFZ) is to chop up your prefixes and announce different
more specifics to each provider.

-- 
Peter