The MPLS WG Archive

[Eric Gray <eric.gray@sandburst.com>]

Jeremy,

    An observation: the ATM model of bi-directionality is being
assumed (at least apparently) in you comment.  It is not impossible
to establish link-by-link bi-directionality - though it is probably
more difficult than it is worth.

Consider this scenario:

   --- R1 --- R2 ---

    All that would be required would be for R1 and R2 to associate
a given pair of labels with a pair of FECs in a label negotiation
exchange - i.e. (R1, Label-1:FEC-A); (R2, Label-2:FEC-B)
where Label-1 is associated with FEC-A at R1 and distributed
to R2 while Label-2 is associated with FEC-B at R2 and the
association is distributed to R1.  Label-1 and Label-2 could be
numerically identical and FEC-A unrelated to FEC-B.

    In fact, the way that LDP is right now, this can be done in
2 PDUs (if not 2 messages): the first one (say from R1) has
two mesaages in it: a label request message for FEC-B and
an unsolicited label mapping message (with Label-1) for
FEC-A.  R2 then returns an LDP PDU with a label mapping
message (with Label-2 - possibly the same as Label-1) for
FEC-A.  If the intention is to associate the same label bi-
directionally with different FECs, it's almost a shame that
two messages are required in the first LDP PDU.

    But this is not the meaning of bi-directional that you are
discussing, is it? :-)

--
Eric Gray

You wrote:

> Dmitry is quite correct, but a comment on applicability:
> Bi-directional ATM LSPs might be useful for traffic engineering
> (for example). However scaling ATM VC-based MPLS to any-to-any
> connectivity amongst more than a couple of hundred or so eLSRs
> will typically require VC-merge capability. VC merging is an
> inherently unidirectional concept, so the applicability of
> bi-directional VCs in ATM VC-based MPLS is somewhat limited.
>
> Regards,
>
> Jeremy Lawrence
>
> At 09:12 02/26/2001 -0500, Dmitry Ryumkin wrote:
> >Paul,
> >
> >Generalized MPLS defines procedures for bi-directional LSP setup.
> >
> >Regards,
> >Dmitry Ryumkin
> >
> > -----Original Message-----
> >From:   djiang1@lucent.com [mailto:djiang1@lucent.com]
> >Sent:   Monday, February 26, 2001 8:52 AM
> >To:     Paul Moore
> >Cc:     mpls@UU.NET
> >Subject:        Re: MPLS LSPs and ATM PVCs
> >
> >
> >HI Paul,
> >
> >I believe RSVP is never intended to set up two LSPs using one single
> >operation.  We will have to use two RSVP signalling operations to set up
> >the two LSPs, one for each direction, even if the PVC may be used
> >bidirectionally.
> >
> >Regards,
> >Jane D. Jiang
> >
> >
> >
> >
> >Paul Moore <paul-patrick.moore@alcatel.be>@UU.NET on 02/26/2001 08:11:29 AM
> >
> >Sent by:  owner-mpls@UU.NET
> >
> >
> >To:   mpls@UU.NET
> >cc:
> >Subject:  MPLS LSPs and ATM PVCs
> >
> >
> >Hello,
> >
> >I have a question about MPLS and ATM. ATM point-to-point PVCs can be
> >either uni-directional or bi-directional. MPLS LSPs are only
> >uni-directional. Is there any work being done concerning bi-directional
> >LSPs, for example, to allow two LSPs to be established by a single RSVP
> >signalling operation?
> >
> >I've just joined the list, so apologies if this has already been
> >discussed.
> >
> >Thanks,
> >Paul
> >
> >--
> >Paul Moore
> >Alcatel Bell
> >+32 3 450 3536
> >paul-patrick.moore@alcatel.be