MPLS Design question

[billsayegh]

We have a need to dual connect several sites (CE routers) to two data centres (PE routers) in each data centre with some CE sites dual connected by GIG or ATM PVC to the PE routers. The PE routers are connected over the MPLS area 0.0.0.0 via a GIG connection.

1. If we summarise several remote sites as a single address range for the VRF on the PE routers to the route reflectors then we loss visibility for individual site link failures and traffic for a link failure will still be sent to the PE which will be unable to route the packet.

2. Some solutions would be to create sham link over the MPLS network to link the OSPF area so that the individual subnets are passed between the PE routers.

3. Create GRE tunnels between the PE routers to pass the subnets between the VRF

4. Create a logical or physical link between the PE routers to link the VRF together to pass the subnets.

5. Would it be possible to MPi-BGP peer the two PE routers together and send the specific routes via BGP between the PE routers and send the aggregate summary MPi-BGP to the route reflectors. I tested this via normal BGP but was unable to send an aggregate route via BGP to a specific neighbor instead the aggregate was sent to the route reflector and the partner PE router.

Any ideas on how other customers are designing dual connected CE routers

[mheusinger]

Hi,

I would say, it works as designed.

In 1. you describe, that you do not want to have single CE site visibility to your data centers and the problem caused by this. True.

Why is it mandatory in your environment to use aggregation/summarization in the first place?

CE redundancy usually is simply accomplished by "normal" IP routing, i.e. let MBGP show you, where a CE is connected to the MPLS VPN.

What is not acceptable with this solution in your case, what are your requirements?

Kind regards

Martin

[billsayegh]

Thanks for your feedback.

The requirement is to have determanistic routing to a specific PE (Local Pref) and in the event that one of the dual connected remote site links fail that connect to the prefered PE then traffic will be sent to the alternate PE for this site only.

Summarisation was simply to reduce any route changes that would be propogated due to topology changes in the IGP.

Yes I agree and understand that simply sending all remote site IGP routes over MPiBGP will work fine.

Thanks

[mheusinger]

So your requirement is "CE site visibility" and reduced load on your BGP processes. Well, one suggestion would be to summarize at the CE level and not at the VRF level, if the IP address structure allows to do it.

regards