[c-nsp] eBGP multihop, CE default route, using PBR instead of dynamic routing?

[Roger Wiklund]

Hi

We have an MPLS customer who is running IS-IS on their LAN, and then
redistributing that into BGP to our core.

This was the original standard setup:
PE----ebgp-----CE----ebgp-----CUSOMER----ISIS

So that worked just fine, but the customer wanted the IS-IS metric to be
injected into BGP MED. This can be done, but with the setup above, MED is
only sent to the CE router, after that its removed.

So what we did was to setup eBGP multihop from the PE directly to the
customers router. We then used BGP on the CE to the customers router, and
from the CE to PE we used a default route.

Now, this site is the customers HUB site so somewhere in their LAN, they
have an Internet breakout. So the customer is injecting a default route from
their router, into the MPLS.

So what happened now is when another stanard site in the MPLS tried to reach
the internet, we had a loop between the PE and CE. Cause the PE will send it
to the CE, and the CE will have a static default route back to the PE.

So to fix this, I skipped the default static route on the CE, and enabled
eBGP between the PE and CE. That way the CE have full knowledge about each
sides.
However, this is not an optimal soultion, I dont want to have 2 BGP peerings
on the PE.

So, what I came up with, and this is where I would like your input on.

In my lab, I have the same setup, so I removed all the static routes and
dynamic routing on the CE. So basically everyting is broken, because the CE
doesnt know where to send the traffic to.
I then configured policy based routing, and created an ACL permit all
traffic, and created 2 route-maps, that matches on the ACL, and sets the
next hop. I then applied the route-maps to each interface on the CE.

So, when traffic coming into the CE from the PE, I match on everything, and
set the next hop to the customers router. And vice versa in the other
direction. I tested it and it worked, and it has no dynamic routing what so
ever.

But this is just in the Lab, I really cant say what will happen in the live
network.

Have anyone done anything similar? Will PBR eat up all the CPU process? Any
other problems that may occur? I mean, all I want to do on the CE is shuffle
the traffic from one interface to another.

Thanks

Regards
Roger