The MPLS WG Archive

[Jeremy Lawrence <jlawrenc@cisco.com>]

Another example (thanks to Rob Raszuk) is with the multi-provider
application of BGP+MPLS VPNs: section 10 of
http://www.ietf.org/internet-drafts/draft-rosen-rfc2547bis-01.txt

As described earlier, there are usually no *top-level* LSPs
established across the two (or more) provider ASes involved, so
it can be argued that
1. the two ASes are separate administrative domains.
However there are some LSPs established across the two ASes, at
a lower level in the label stack. So, it can be argued that
2. the two ASes are the same administrative domain, in so far
    as the two providers agree to allow lower-level LSPs to
    be established across the two ASes

I think that (1) and (2) are both true, which implies that
different definitions of the boundary of the administrative
domains can exist with respect to different levels in the label
stack. It is also (in hindsight) obvious that different MPLS
domain boundaries can exist with respect to different levels of the
label stack. 

Jeremy

At 17:21 05/30/2000 -0700, Lane Patterson wrote:
>The proposed application of MPLS across public exchange point switches/LSRs
>(for example http://www.nanog.org/mtg-0002/feldman.html) is an example that
>stretches the definition of "administrative domain".  In this case, the
>administrative domain is not the AS, it is the common practices and
>coordination necessary by exchange point participants.
>
>The primary motivation for this is to provide higher-speed and layer-2
>agnostic interfaces as a means of public peering, although there is little
>if any operational experience in this area.
>
>The implementation is likely two-hop [IXP LSR, peer LER/LSR] explicit route
>LSPs using signalling and static routes, no IGP.
>
>Regards,
>-Lane
>