Re: [mpls] Solicit Opinions on Definition of MPLS Global Label

[Stewart Bryant <[e-mail]>]

Robin

Others have made many of the points that I would have made
in this discussion of global labels, and I have yet to see a
compelling case emerge.

I am responding to the following from your slide deck:

BGP VPN LSP is another type of multi-point to point
LSP which faces the challenge of the measurement of
packet loss proposed by [I-D.bryant-mpls-synonymous-
flow-labels].

Firstly SL can be made to work in this environment. One of the
points that I have been making is that the SL label exchange can
be thought of as part of the measurement session establishment
protocol and thus works underneath any existing label
advertizement/exchange protocol without disruption to the
existing protocol.

If your concern is that you want to remain within the BGP VPN
paradigm and have an advertizement only method of operation,
with no session, then the advertizing PE could advertise either
a set of labels it wanted to use to map between either a
source/VPN label pair or as a source identifier. Alternatively
you could take a leaf from the SR book use a virtual global
label set and advertise the PE offset in the VPN advertizement.
This latter approach would mean that a single identity was
distributed, but the labels in the stack were scoped to the
destination in conformance with RFC3031.

I will make one other point, and that is *IF* there is a compelling
case for a tagged network layer with truly global labels, then there are
other stack based designs that could be considered, which are possibly
more powerful than the existing MPLS design.  However it is difficult
to imagine a sufficiently compelling commercial case for their
deployment in the near future.

- Stewart