The MPLS WG Archive

[Eric Rosen <erosen@cisco.com>]

The issue of tunneling MPLS  packets through an IP-but-not-MPLS backbone has
started to attract some  interest, judging from draft-worster-mpls-in-ip and
draft-rekhter-mpls-over-gre. 

I believe that  tunneling MPLS through GRE is in fact  legal and standard as
things stand right  now.  Draft-rekhter-mpls-over-gre provides some explica-
tion  of how  to do  it,  but doesn't  really propose  anything which  isn't
already standard.

Draft-worster-mpls-in-ip proposes a way to  tunnel MPLS through IP with less
overhead. 

In  some cases  of  tunneling, MPLS  will  really need  to  go through  GRE.
Suppose, for  example, two routers are  "connected" via a  GRE tunnel, which
they treat as an interface.  Many  different kinds of packets may go through
that interface.   If MPLS  packets need to  go through that  interface, then
they will  go inside GRE  packets, just like  all the other packets  that go
through that  interface.  Having to  use a different encapsulation  for MPLS
than  for all  the  other tunneled  packets  would just  be a  complication.

There may be cases though where  there is no particular reason to choose GRE
over anything  else, and where  minimizing overhead is deemed  important.  I
don't have any really strong  objection to the use of draft-worster-mpls-in-
ip for such  cases.  (I thought about raising the ritual  "do we really need
another tunneling  protocol?" issue,  but this would  only give rise  to the
ritual  "no other  tunneling  protocol  is good  enough  for the  particular
application we're focused on now" response.)

I do  think though  if draft-worster-mpls-in-ip is  to advance, it  needs to
make it clear that what it is describing is just one of a number of standard
tunneling techniques.