The MPLS WG Archive

[Anouk Rocher <anouk@ureach.com>]

Eric,

I am sorry I mis-wrote. You are absolutely right (and you did
answer my question except for the follow on below).

I meant that the "outer" label is established by RSVP TE and the
inner by LDP. According to the martini drafts, the inner label
is usually associate with a layer 2 entity like frame-relay DLCI.

I see your point about problems with independent label spaces
operating at different levels of the stack if PHP is used. 
However, is it a MUST (in ietf terminology) to support PHP?
Ie if I have network which does not rely on PHP, is it otherwise
reasonable to allow independent label spaces - in the above example
one for the inside of a network (outer label, using rsvp-te) and
one for the edges (inner-label, using LDP).

Thanks!
Anouk.


________________________________________________
Get your own "800" number
Voicemail, fax, email, and a lot more
http://www.ureach.com/reg/tag


---- On Tue, 10 Jul 2001, Eric Gray (eric.gray@sandburst.com) wrote:

> Anouk,
> 
>     I assume that David can answer for himself, however I would like to
> get some things clarified.  Please see comments/questions below.
> 
> Anouk Rocher wrote:
> 
> > David,
> >
> > Let me clarify with a more specific example. Lets say we want
> > to deploy one of the l2 vpn mechanisms - say the one in
> draft-maritini-*
> > The inner label there is negotiated between the ingress and egress
> > LSRs via LDP (sort of multi-hop). The inner label is negotiated (say)
> > via RSVP-TE.
> 
> I believe one of these should be 'outer label'.  I assume you mean
> higher
> level, when you say 'outer' and lower level when you say 'inner' but,
> even
> so, I am not sure which should be which in your description above (just
> that they cannot both be the same).
> 
> Would you recast your question somehow distinguishing which is which?
> 
> >
> >
> > I don't see a reason why LDP and RSVP-TE be forced to use the same
> > label space in this case. (I am sure there are similar cases while
> > using wavelengths as labels.)
> >
> > Am I missing something obvious?
> 
> So, one reason would be if PHP is used.  In this case, the egress for
> the
> lower level label will not have received the higher level label and will
> not
> be able to distinguish it from other labels it receives from the
> penultimate
> hop.  Give that this is the case, then the egress LSR must not
> distribute
> a label to be used by the remote LSR at the other end of the LSP tunnel
> with a different meaning then an identical label distributed to the
> local
> LSR that happens to be the punultimate hop for the LSP tunnel.
> 
> This is an example of when the label space MUST be the same.  Does this
> answer your question?