[c-nsp] CEF-based per-packet load-sharing under MPLS VPN

[Everton da Silva Marques]

Hi Oliver,

Oliver Boehmer (oboehmer) wrote:
> 
> Trying it in a fresh environment, I fail
> because of the same reasons (0.0.0.0 as
> next-hop)..

We have solved the load-sharing problem
under MPLS VPN. Using eBGP across the
the parallel PE-CE to enable recursive
load-sharing does work, as you and others
have recommended.

A colleague of ours seems to have found
another possible way to perform per-packet
load-sharing for the general MPLS VPN
case, using static routing instead of
eBGP between PE-CE.

The answer is, in the PE with parallel
links, to attach a distinct VRF for each
link. One will remember that in this case,
load-sharing works evenly for packets
coming from other VRFs in the same PE.
But packets coming from VRFs in remote
PEs remain unbalanced since BGP, due to
the single next-hop, installs only one
of the routes. The trick to correct the
same next-hop problem seems to lie on
the "bgp next-hop" VRF parameter, like
in the example below. The first VRF uses
Loopback0 as the default BGP next-hop.

ip vrf balanced1
 rd 1:1
 route-target both 1:1
!
ip vrf balanced2
 rd 1:2
 route-target both 1:1
 bgp next-hop Loopback1
!

We have performed some testing and it
seems to work.

Regards,
Everton