The MPLS WG Archive

[David Charlap <David.Charlap@marconi.com>]

Nic Neate wrote:
> 
> It's good to see a proper technical issue on the mailing list rather
> than just a virus for a change.

:-)

> I think the key point to bear in mind here is the possibility of 
> per-interface labels.  The main advantages of facility over
> one-to-one backup are when global labels are in use.  However, I
> believe per-interface labels are still catered for.
> 
> In this case, the facility backups have to be signaled prior to local
> repair (as they are in the detour case).  The MP can then expect to
> receive path messages from multiple upstream PLRs at once.  We
> therefore have to view the backup Paths as signaling new LSPs rather
> than just refreshing the protected LSP, and the MP needs to be able
> to distinguish between them in order to manage the different labels
> correctly.

Right.  I forgot about that.

> Finally, following local repair, the protected Path state can time
> out as you say, and the MP won't forward a PathTear downstream until
> all merging backup Path state is gone too.

I seem to remember reading this.

What I'm still not clear on, however, is how a merge point can identify 
which protected LSP (arriving on a normal interface) corresponds to what 
protected LSP (arriving through a bypass).

It would seem that all LSPs in the same session with the same LSP ID are 
assumed to be be backups of one another.  But it seems that there can be 
situations where this will match when the LSPs don't have a primary/ 
backup relationship (for example, if extended tunnel ID of zero is used 
to signal a many-to-one multicast LSP - which is perfectly valid.)

Or is this simply a category of LSP that is effectively deprecated in 
the eyes of the working group?

> That's how I understand it anyway.  Hope that make sense and I've not
> missed anything critical.

-- David