Re: Label Set Object

[Anca Zamfir <ancaz@cisco.com>]

At 03:53 PM 5/9/2002 -0400, Adrian Farrel wrote:
>Suppose (just suppose) that I want a unidirectional sonet LSP and that I 
>want to
>constrain the label choice to a specific value.  Note that this is 
>distinct from
>the question of what labels to use in each direction for bidirectional.
>It is clear how I signal it.  No issues. Label Request, Label Set with one
>member...

You can also allow the downstream node to pick one (maybe complex) label 
from a set with more than one member by including multiple labels in the 
Label Set.

>Now suppose I also want a bidirectional sonet LSP.
>How do I indicate that it is bidirectional?
>I  believe the only method is to include an Upstream Label.

Correct.

>Could we say that bidirectional uses Upstream Label only, and unidirectional
>uses Label Set only?
>Yes.

This is probably true for SONET LSPs. For other types this restriction may 
be undesirable.

>Would that be a good idea?
>I don't see think that having multiple choices of presence of objects is 
>useful.

IMO, assuming most GMPLS implementations are meant to cover more than one 
type of LSP domain, for the reason of maximizing the code re-use, it would 
be nice to have as few domain specific restrictions as possible.

>Could an implementation choose to ignore one of the objects and act only 
>on the
>other?
>Of course.
>
>Adrian
>
>----- Original Message -----
>From: "John Drake" <jdrake@calient.net>
>To: "'Anca Zamfir'" <ancaz@cisco.com>; "Zafar Ali" <zali@cisco.com>
>Cc: "'Juneja, Manoj'" <m_juneja@trillium.com>; "Vinay Vernekar"
><vinay.vernekar@wipro.com>; "Manoj Agiwal" <ManojA@netbrahma.com>; "'Ccamp
>(E-mail)" <ccamp@ops.ietf.org>; <skatukam@cisco.com>; "mpls@UU. NET (E-mail)"
><mpls@UU.NET>
>Sent: Wednesday, May 08, 2002 10:42 PM
>Subject: RE: Label Set Object
>
>
> > Lou and I just reviewed section 5.1.1 of
> > draft-ietf-mpls-generalized-rsvp-te-07.txt (rather than
> > draft-ietf-mpls-generalized-signaling-08.txt, which is what I looked at 
> this
> > morning), and it's pretty clear.
> >
> > The upstream node processing the ERO for the outgoing link checks whether
> > there are explicit label control subobjects for upstream or downstream.  An
> > explicit label control for upstream is replaced in the ERO with an upstream
> > label and an explicit label control for downstream is replaced with a label
> > set with the same label.  This is so done so that there is only one
> > mechanism for an upstream node to specify the downstream label; explicit
> > label control is only used by a remote node.
> >
> > So the only question is whether for SONET/SDH LSPs, this mechanism is
> > required.
> >