Addressing the Internal BGP Structure
Because certain routers within the TransitNet network will no longer be required to hold BGP routes after the migration to MPLS, it is necessary to build the relevant infrastructure to support the new BGP design. This new structure can be seen in Figure 6-2.
Figure 6-2. TransitNet MPLS BGP Peering Structure
Figure 6-2 shows the desired topology, where BGP is enabled only on edge routers. These routers run iBGP sessions with core route reflectors so that all external routes can be distributed successfully between edge routers. All transit routers are now BGP-free and purely label switch packets across the backbone.
To be able to achieve this topology, and to provide a smooth migration to the MPLS solution, it is necessary as a migration step to run multiple BGP sessions from the edge routers. This can be seen in Figure 6-3, which shows the necessary old and new iBGP sessions for both of the edge routers within the London POP.
Figure 6-3. TransitNet iBGP Session Requirements
This type of migration approach has some implications?most notably, the increase in the memory requirements of BGP routers because they will need to house multiple BGP sessions and learn several copies of the same routes. However, if a smooth transition is required and memory is not an issue for a short-term migration phase, then this type of migration approach provides a seamless transition from one BGP topology to another.
Note
Our sample topology uses separate route reflectors for the migration. This is not strictly necessary because the existing route reflectors could be used as long as separate addresses are used for the iBGP sessions. (It is not possible to have multiple iBGP sessions between the same set of addresses.)
The addition of further BGP peering sessions on the edge routers does not present a problem for the BGP protocol. This is because the attributes of the routes are exactly the same; the only difference is that they have been reflected to the edge router from a different route reflector. This means that when the backbone routers are capable of label switching packets, the current BGP sessions can be removed without the danger of losing traffic.
