Hello David,
do the route-distinguishers on different PE routers (each servicing the same customer at different locations) need to have the same route distinguishers on the VRFs of the different PE routers?
The values of route distinguishers (RDs) can be largely arbitrary. The must be unique for different VRFs on the same PE, but for two corresponding VRFs on two different PEs, they may or may not be the same, it does not really matter. In simpler deployments, they can be the same, but in larger or more complex VPNs, it is more probable that they differ.
Secondly, are route-distinguishers carried in MP-BGP VPNv4 advertisments?
Yes, they are, as part of al network addresses that are in a particular VRF and are advertised to another PEs.
Allow me to explain the process of advertising the routes from one PE to another, and let me start first by talking about the other attribute configured in VRF - the Route Target (RT).
The RT is a BGP attribute of a route - technically, it is an extended community attribute. Being an attribute means that this value is a property of the route that further specifies how it shall be processed but does not form a part of the route's network address or mask. It is included along with the advertisement of that network to a BGP peer but just as with other BGP attributes (next hop, local preference, metric, AS_PATH, etc.), it is just a property of the advertised network, not an address part. Certain BGP attributes do not need always to be present, or can be present multiple times in an update for a route. The RT in particular is used to say into which VRFs can the particular route be imported. Saying that a route has an export RT of 1:1 and 1:2 means that the route can be imported to any VRF that imports routes with either 1:1 or 1:2.
The presence of the RT and the possibility of having multiple RTs for a single update allows for great flexibility in designing MPLS VPNs with several locations and partial visibilities between them. However, having only RTs would not be sufficient.
Consider two VRFs on a single router, V1 and V2, each of them containing a single network 10.0.0.0/8. For now, assume that there is no concept of RD. If BGP sends the routes from this router to a peer, it always sends them in some sequence. Let's assume that the first routes sent are the ones from one VRF, say V1, then routes from another VRF, say V2, are advertised. V1 uses RT of 1:1, V2 uses RT of 1:2.
Now imagine how the BGP updates would look on the wire: first, an update comes, saying that there is a network 10.0.0.0/8 with the RT set to 1:1. What would the neighboring router do? It would place the route into the corresponding VRF V1. Now, a second update comes in (or a second entry from the single update is processed), and guess what - it says that there is the same network 10.0.0.0/8, just with the RT set to 1:2. So what would the neighboring router do? It would consider this second update to be a replacement of the previous update - the same network but different attributes! So it would remove the network 10.0.0.0/8 from the VRF V1 and add it to VRF V2!
This would happen because for BGP, the network and its netmask are so-called keys in its database. An update to a network is performed simply by sending the update about this network again, with new attributes explicitly specified. There is no need to first withdraw the route. And because BGP does not see the difference between the network 10.0.0.0/8 from V1 and 10.0.0.0/8 from V2, it gets confused because it thinks that the information about the same network simply got updated.
This is where RD comes into play - by extending the formal network address for all routes in a particular VRF using a unique value. If V1 has the RD also set to 1:1 (it may very well be different from RTs and usually it is) and V2 has RD set to 1:2, then the route from V1 will be advertised as 1:1:10.0.0.0/8 while the same route from V2 will be advertised as 1:2:10.0.0.0/8. For BGP, it is now immediately obvious that these two networks are not the same. The trick here, therefore, is to use the RD to temporarily extend the IP addresses of networks in a VRF with unique values to work around the BGP's attitude to identical-looking networks. However, the RD is not used to sort routes into different VRFs after receiving them through BGP, and exists purely within the BGP communication.
So the RT says into which VRF a route can and possibly will be imported. The RD helps BGP understand that network X from one VRF is not the same as the network X from another VRF. Both RT and RD have the same format but that is their only similarity. RD must be present exactly once in each VRF on a single PE and needs to be unique. RT must be present at least once in each VRF and does not need to be unique (if VRF route leaking is desired). RD becomes a part of the network address within BGP updates, RTs are carried as attributes of these networks. RDs are never used to sort routes among VRFs - that is the purpose of RTs. That is also the reason why the RD may or may not be the same in two corresponding VRFs on two different PEs - it actually does not matter.
Please feel welcome to ask further!
Best regards,
Peter
