[c-nsp] Internet routing in a vendoe MPLS environment

[Rick Martin]

This may not be a Cisco specific question but today our network is virtually 100% Cisco so I am throwing this out to this group for input.

Our statewide network with 1,800+ CE routers is currently a legacy ATM backbone with primarily ATM IMA or Frame Relay for last mile connections with a small percentage of DSL and Ethernet thrown in for good measure. Today we manage the entire L3 infrastructure with service providers delivering the L2 transport. We have evolved into a distributed hub and spoke architecture with 5 "hub" locations (ASR's and 7600's) on the network. Three of those hubs have Internet connections ranging from the smallest, a single OC3, to Gig transport utilizing BGP to manage our outbound prefix advertisements. We accept only default route and have things very well balanced and failover routing in place via BGP. We advertise 4 full class B networks as /16's and one as two /17's. Our IP assignments are geographically distributed based on the Internet connection the customer will be using.

The above scenario has worked very well for us since we implemented the second Internet connection in 2002 and have grown since then.

We are in the bidding process for migrating to a vendor owned MPLS core network, we expect to have some areas where the customer connection will terminate directly into the MPLS core and we will also have some geographic areas where we will have to aggregate customer connections into an ASR and pass those routes into the vendor's MPLS network. Aggregation will be a requirement due to the multi vendor footprint of our state. All connections into the MPLS core be it last mile or ASR aggregation router will talk BGP with the MPLS service provider.   I am very concerned about how we will carry on our Internet routing thru the vendor's MPLS infrastructure. I have no issue with what we advertise out of our network and how to distribute our prefixis, I have concern with managing our outbound traffic, we want to assure that if we advertise network A via Internet connection 1 that users on network A will egress the state network via Internet connection 1, network B exits the network via Internet connection 2 and so forth.

I have discussed this with my Cisco team and they have come up with a couple of options the most compelling at this point is GRE tunnels from CE to primary and secondary Internet connections, run our own routing protocol over the tunnels to manage the default route.

Is anybody else running this type of scenario? If so, how is it working for you?

What are you doing to accomplish diverse Internet routing in a Vendor managed MPLS environment that is different?

Thanks in advance for any info you can share.

Rick Martin
Network Architect
State of Arkansas, Department of Information Systems