Present MPLS backbone based on 3750ME and 6504/Sup32 to all main locations.
Rigid (and stable) structure at each site with C - CE - PE, eBGP pr. VRF between CE and PE, static/HSRP between C and CE.
C typically 6500/Sup720, CE either 3560(G) or 3750(G).
Only backside to present design is that connectivity between to CEs at a site is via L2 in Cs, that is CE - C - C - CE.
That has provided some interesting failure scenarios :(.
Static configuration with many configuration steps for new networks also provides possible errors.
We're looking at removing the CE and combine the C/CE functionality into the 6500/Sup720 on each campus.
Then we can either run traditional eBGP pr. VRF between C/CE and PE or we could move to a solution with PE-CE eBGP VPNv4 prefix exchange, such as would be used when service providers exchange label info across AS boundaries.
The latter solution means all PEs will accept all VPNv4 routes rather than just the ones for which they have locally defined VRFs.
But it also provides flexibility combined with ease of the configuration, as the PEs are basically reduced to P functionality, and at the same time the usage of eBGP between C/CE and PE means we don't have to include the C/CEs in the IGP (IS-IS) of the backbone and convert them to full PEs, compromising the stability of the MPLS core, while maintaining all the control handles of BGP.
The C/CEs would need to BGP peer with the PEs in the global routing table for exchange of VPNv4 routes and next hop information.
Does this sound like a feasible solution? Caveats?
Next issue at hand is mVPN.
We'd like to support pr. VRF multicast across the MPLS backbone, and mVPN seems like the answer to that. Unfortunately it's not supported now nor will it apparently be so on the 3750ME.
The obvious answer is to move to e.g. 7604 as PE platform, which is a long term strategic option, but it's not gonna happen overnight.
Assuming we implement the PE-CE eBGP VPNv4 route exchange, would we be able to switch the mVPN functionality into the C/CE routers, this taking away the restraints of the 3750MEs?
We would run pr. VRF PIM on the C/CE and define default- and data-MDTs for each VRF on the C/CE.
Obviously we would need multicast routing in the global routing table, both on the PEs (based on PIM-SM, Bidir or SSM), and between the C/CEs and the PEs (based on PIM or perhaps MP-BGP?).
Would this work? Caveats?