MPLS LDP transport address should it be always loopback IP

Question

Can the transport address for the LDP be different from the loopback IP? LSR ID would be loopback while the transport address would be interface IP. For example, if the interface IP is the transport address, is there a drawback of not using it as a loopback.

Answer 1

Since the LDP is running using TCP, it needs to establish session with every LDP speaking neighbor. Sometimes those neighbors are one hop away and you can use the point-to-point addresses configured on interfaces, but sometimes those might be a few hops away (e.g. when using L2 vpns/EoMPLS) and then both routers must be able to discover each other first before they're able to establish this communication. Similar to iBGP using loopback0 for update-source (think about it this way). Therefore, if you're to establish a TCP session with a neighbor that is 2-3 hops away, you'd want to use your loopback address to avoid LDP adjacency failures during interface failure.

Consider the following topology.

(A) et1 --- et1 (B)
   et2       et2
     \       /
      \     x <--- failed link
     et1   et2
        (C)

Router C's et2 towards Router B fails. In this case, if you don't use your loopback address to establish adjacency, router C will use it's p2p address used for the link between C<->B to peer with B. When the link goes down, Router C will try to establish adjacency with (B)'s router-id (which is the IP of the failed interface et2) because B is not using loopback address either, so C will try to use it's et1 address to talk to (B)'s failed et2 (again, this is only in case they have indirect (i think it was called targeted) LDP session (EoMPLS/L2 circuit) and this won't work, because the link will be down.

Using Loopback addresses solves these problems and as long as you have route between peers - the LDP adjacencies will be intact.