The MPLS WG Archive

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

In message <3.0.5.32.20010410211927.008a14c0@10.1.1.249>, Mark Duffy writes:
> >
> >Mark> (I supposed  you [Curtis] will tell me  that the whole use  of LDP for
> >Mark> topology driven  set up should be  freely ignored :-) and  I would not
> >Mark> necessarily disagree with you.)
> >
> >Just don't  expect to  interoperate with others,  unless you are  using MPLS
> >only for traffic engineering. 
> 
> Interoperability is certainly an important goal for me.  2547bis specifies
> LDP as a common point of interoperability, however I have heard some
> dissenting voices on this list speaking in favor of using TE tunnels.
> Outside of 2547, I'm also interested in the use of hierarchical LSPs to
> interconnect virtual routers, where pretty much the same considerations
> apply.  For all these reasons therefore I'm trying to probe the sentiment
> of this list.

Let me be clear about what I said.  Interoperability is goodness.  We
do LDP for RFC2547 and it works with both Cisco and Juniper
(interoperability testing has occurred in out lab, UNH, and GMU).  We
also allow the LDP tunnel to ride inside a RSVP/TE tunnel.  This
offers benefits (which I won't enumerate again).  If this feature is
used, we still interoperate with others running LDP and others running
RSVP/TE in the same network, although afaik we have not checked an
Avici router on one end of the TE tunnel and another router on the
other end (but we should get there).

If the TE tunnels go all the way to the edge and if this is done in a
way that scales well, then the need for LDP disappears.  My
understanding of the real reason why we are using LDP for VPN is that
it had more to do with the scalability of a particular RSVP/TE
implementation rather than an issue of protocol scalability (though
authors would never admit this).  If RSVP/TE does prove to be very
scalable, particularly with multi-area support and hierarchical
tunnels, then the need for LDP in some networks (the "in some
networks" is an important qualifier) may go away.  As long as there
are routers in a network which have a scaling problem with RSVP/TE are
in a network (older routers tend to migrate toward the edge) there
will be a need to support LDP (possibly just near the edge).

Curtis