[mpls] Comments on draft-ietf-mpls-ldp-multi-topology-01

[IJsbrand Wijnands <[e-mail]>]

Dear Quintin,

During the MPLS WG presentation you asked for feedback on the different options to signal the MT-ID. See below;

I've worked on this with Kamran Raza and we came to the conclusion there is really only one good option to signal the MT-ID, and that is within the FEC. This is true for LDP, and very much true for mLDP. We documented the preferred solution for mLDP, which would also apply to LDP. See; 

draft-iwijnand-mpls-mldp-multi-topology-00.txt

Below is a summary of the reasons;

- The LDP spec allows for multiple FEC's to be signaled in a single label mapping. If you don't add the MT-ID into the FEC its becomes difficult to match the MT-ID the right FEC, so you can't mix and match FEC elements, for example for IPv4 and IPv6. For the same reason Address Family is part of the FEC.

- The MT-ID is something that is directly related to the Prefix/Root address encoded in the FEC, so it makes sense to combine them.

- LDP messages and specifications that use FEC (e.g. Typed Wildcard,
End-of-LIB) will automatically be extended for MT. There is no need to specify for which messages (map, withdraw, release etc..) the MT-ID applies.

- If you are doing mLDP Live-Live using MTR (or MRT) and the 2 LSPs merge/overlap for some reason, you potentially duplicate the traffic and prevent live-live from working. For that reason the MT-ID MUST be in the FEC to make sure the 2 LSPs are unique and will never accidentally merge.

- With mLDP a router may receive label mappings from multiple downstream routers, each of them including a different MT-ID. At that point you would need come up with some selection logic on which MT-ID you need to select and signal upstream. If the MT-ID is part of the FEC, you don't have that problem.


Thx,

Ice & Kamran