(2006-08-24 08:53 +1000), David J. Hughes wrote:

From operational experience, what are peoples thoughts on B
vs C? I can image that multi-hopping the C solution
could get ugly unless you have tight geographical / topological
grouping of VPN endpoints in the peer AS.

If it would be implemented as RFC requires, that is B would
only allow labeled packets in that contain label, that it has
previously advertised out. Then B would be clear choice when
doing VPN interconnects with out organizations or with boxes
that are not in controlled space (such as PE in customer
premices).
But as today both cisco and jnpr do not do this check, B
is not very useful. You either tend to use A in eg. situations
mentioned above, or if no such problem exists straight to C.

I personally want to see label checking in B as soon
as possible. A is really cumbersome.

(2006-08-25 09:09 +0300), Saku Ytti wrote:
previously advertised out. Then B would be clear choice when
doing VPN interconnects with out organizations or with boxes
with /other/
That is, with some other company or with customer etc.