[c-nsp] Traffic engineering in mixed OSPF/iBGP environment

[Garry]

Hi,

I've been trying to figure this out for a while now, but can't get a
grasp on how to get this to work the way I want to ...

In essence, here's the relevant part as a small drawing:

        CPR1------CPR2
         |         |
        PER1     VSwitch
         |         |\
        PR1        | \
         :         |  CPR3..CPR6
        PRn        |
           \       /\___DSLGW-----CPR7..CPR12
            \     /
             CoreR1---CustomerFW--Internet


We have a customer network with a redundant connectivity, running as a
distinct MPLS VRF in our backbone. Both links are terminated with a
separate router (CPR1 & 2). Link 1 at CPR2 goes to a virtual WAN switch
with additional sites as well as a link to our backbone (CoreR1). We
have OSPF running in that broadcast domain, as well as between the two
CPR1/2 routers, distributing the different subnets that are reachable.
Link 2 between CPR1 and PER1 is also running OSPF, though in the MPLS
backbone, routes from the VRFs are learned through a route reflector;
PER1 redistributes the RR routes to CPR1 via OSPF.
Additionally, there are some DSL locations which are also terminated
through another router at the CoreR1 location. Again, locally the routes
are learned through OSPF and redistributed to the BGP RR.

At this point, all routing and redundancy is working fine. If links go
down, routing is recalculated and converges in satisfactory time.

As for the problem I have: the customer is running applications at a
service provider that is connected to the network via CPR3. Traffic from
CPR1/CPR2 as well as the other customer locations correctly take the
route through the VSwitch. Now, traffic to the Internet (in essence
everything that is not a connected destination inside the VRF but the
default route) from CPR1/CPR2 is supposed to be routed via the PER1...
connection, utilizing the otherwise unused backup link. This is where
I'm having trouble. If the whole network were running on OSPF, putting a
higher OSPF cost on the CoreR1 towards the VSwitch link would more or
less ensure that the link would only be used as a backup, otherwise CPR1
has PER1 as lower cost towards the CustomerFW, while the traffic to CPR3
is still cheaper towards the VSwitch. But with the BGP-redistributed
routes at the PER1-CPR1 link, the OSPF cost metrics do not help. I've
already tried messing around with the redistribute statements a bit, but
even with altering the BGP-learned routes to be imported to OSPF as E2
routes, the routing still kept the vswitch routing up. I'm somewhat out
of ideas on how to implement this while still keeping redundancies
operational (with either traffic able to be routed via either link) and
not using error-prone hacks ... any pointers?

Thanks, -garry