The MPLS WG Archive

[Bob Thomas <rhthomas@cisco.com>]

Jim,

> Forgive me if this has already been discussed, just point me to the thread in
> the archive.
> 
> I have a question about the loop from LMp.17 to LMp.31
> Here is my interpretation of what should happen:
> 
> When processing a incoming label map message the LSR needs to figure out
> what other peers this mapping needs to be propagated to.  While processing
> the list of all peers:
> 
> -if the LSR has previously sent a mapping, and the attributes of the new
>  mapping are different then the attributes of the old mapping then send an
>  updated mapping (LMp.22-LMp.27).
> -if a peer is operating in DU then the LSR should propagate the mapping
>  (LMp.20,LMp.21)
> -if a peer is operating in DOD AND the LSR has a record of an outstanding label
>  request from that peer, propagate the mapping and delete the record of the
>  outstanding request (LMp.29-DOD).
> 
> If this is a reasonable interpretation then what is LMp.29-DU for?
> Didn't the LSR already do that work in LMp.20,LMp.21?

LMp.20+LMp.21 handles (only) the situation where a label mapping
has not previously been sent to the upstream peer AND the mode is
DU Ordered.

LMp.29-DU handles the situation where there are outstanding label
requests from the peer (which won't be the case if LMp.20+LMp,21
is executed since in the unlikely event that the peer had
requested a label in DU Ordered mode LMp.20+LMp.21 would have satisfied
the request).  In addition, LMp.29-DU handles the situation where
DU Independent is being used, which LMp.20+LMp.21 does not.

This procedure is complex because the number of combinations of modes
that must be handled (DU/DoD advertisment, liberal/conservative retention,
ordered/independent control).  There are no doubt ways to re-arrange
it to achieve the desired effect.


Bob