The MPLS WG Archive

[Curtis Villamizar <curtis@workhorse.fictitious.org>]

In message <3CBB30E5.F76B0159@marconi.com>, David Charlap writes:
> Sudheer Dharanikota wrote:
> > David Charlap wrote:
> >>
> >> What's the point?
> >>
> >> In an optical network, you are signaling wavelengths or time
> >> slices.  How could you conceivably have two time-slices share
> >> resources?  Or two wavelengths?
> > 
> > Only to imply that a time-slice can be allocated to one of the
> > *backup* paths when a primary path fails.
> 
> That's not the definition of SE style.  SE style may be used for many
> things other than signaling backup paths.
> 
> To assume that SE style can only be used in the way you want is
> unwarranted.  You may end up with undefined behavior if you do.
> 
> If you need a mechanism for explicitly indicating which LSPs serve as
> backups for others, then you should use a mechanism that is meant to do
> this.  If none exists, then you should define one and use this working
> group to make it into a standard.
> 
> Unilaterally assigning a new and undocumented meaning to an existing
> construct is never a good idea.
> 
> -- David


Delayed response...

What SE is used for in MPLS/TE is always to indicate that resources
are shared.  It doesn't signal any kind of reroute but the sharing is
among the primary and backup in local protection beyond the merge
point (not applicable to fully disjoint paths from ingress to egress)
or among the primary and another path that will be used in a
make-before-break fashion.

It makes perfect sense to reuse the same lambdas on some legs to
overlap in make-before-break fashion and turn off one laser at the
ingress (for lambdas) and turn on another and have a (very near) no
impact reroute that does not require a complete teardown before the
setup.  It also makes sense to do a local repair this way.  It also
may be applicable to time slices if there was a way to send nothing on
time slice on the one path not in use at any given time in the
transition.  This would be like sending empty sonet frames on one of
two OCx and having an intermediate pick the non-empty frames from
either or two ingress OCx and send them out another OCx.  This is very
APS like if used for repair in the TDM context.  This clearly involves
too much inspection for burst switching and is more like what a router
or ADM can do and less like what optical burst switching can do.

IMHO, the UNI is useless so direct discussion to OIF.  Or devnull or
anywhere but the IETF.

Curtis