EIGRP with Dual CEs to Dual PEs and Asymmetrical Routing

rcolter · ‎03-21-2007

I have MPLS with a carrier. I'm running EIGRP with 2 CE routers at my host connected to 2 different PE routers at the carrier. My goal is to have one CE as the primary and the second CE as the backup. When I bring up the backup circuit I get asymmetrical paths. Data goes out the primary CE, but some remote CEs reply through the backup CE. I tried modifying the bandwidth statement on the backup CE with no success. Is this a carrier route reflector issue or can I solve this through my configurations. The carrier will not modify the PE configurations.

mheusinger · ‎03-21-2007

Hi,

make sure the backup CE announces a worse metric. This alone should help, but the underlying problem is the BGP decision process in the SP network.

If the SP prefers the path through one PE over another PE, there is little you can do. If he does not modify BGP default behaviour, route selection should be based on MED values in your case. They are directly copied from the EIGRP metric during redistribution.

Thus, announcing a higher metric from the backup CE should lead to the desired result.

You can try to lower bandwidth or increase delay:

interface FastEthernet0/0

bandwidth <1-10000000> Bandwidth in kilobits

delay <1-16777215> Throughput delay (tens of microseconds)

This should be the link to the backbone, i.e. where routes are received, before they are passed to the PE.

Check the metric in both CEs with "show ip eigrp topology" and make sure the Feasible distance on the backup CE is larger. You might need to play with parameters to achieve the desired result.

Hope this helps! Please use the rating system.

Regards, Martin

royalblues · ‎03-23-2007

Well i have a similar situation in my network except that i run BGP with the service provider.

I have asked our SP to always prefer the routes through a particular CE which he is doing via local preference.

As martin suggested this could also be achieved by sending MED values to the ISP but should be recogonised by ISP.

Another way to achieve this is by using communities which most service providers prefer.

HTH

Narayan

IVAN PEPELNJAK · ‎03-28-2007

This is what the SP can do:

===========================

If they use route reflectors in the MPLS VPN cloud, make sure the two CE routers are connected to different VRFs and use different RDs (so the VPNv4 route in the BGP process will be different and not acted upon by the route reflectors).

Also, the PE router on the primary PE-CE link could set local preference to a high value to ensure the route is always preferred over the backup BGP route (redistributed on backup PE router).

This is what you can do:

========================

If you change bandwidth on outbound EIGRP interface (CE -> PE link), it will not affect the metric on the PE router. You should use the offset-list out router configuration command to increase the delay of the routes sent from your CE router over the backup link.

ilya.varlashkin · ‎03-29-2007

We use only BGP and RIP as PE-CE routing protocols, but approach could be easily applied to EIGRP as well. I've created route-maps that for BGP match community and and for RIP match metric then set local-pref and MED for prefix within our (provider) backbone. If your provider is using Cisco as PE, they also need to reset weight of routes redistributed from EIGRP to BGP, otherwise locally redistributed routes are always preferred by that PE.

You say that carrier is not willing to modify PE configs, but it's absolutely necessary that PE-to-PE routing is properly implemented and that's beyond your control. Talk to your provider about how they can resolve problem.