The MPLS WG Archive

[Kireeti Kompella <kireeti@juniper.net>]

Hi Ron,

> Have we been discussing two distinct questions, as if they were the same?
> These are:

> 	1) Overlay model vs. Peer model
> 	2) UNI vs MPLS signaling between client IP network and optical transport
> network

> The first issue addresses the degree to which a client IP network can access
> the optical transport network's signaling. The second issue addresses the
> syntax of the interface between the client IP network and optical transport
> network.

Good distinction.

> These two issues are orthogonal to one another.

In principle, yes.

In practice, though, the connotation of "UNI" is pretty much
"overlay" -- the distinction between U and N means that there is
a restriction in information exchange.  And in any case, the
current UNI spec has a very restricted flow of information across
the U/N border -- it's primarily a signalling protocol, and the
information exchange is "end point discovery" and "LSP status".

However, in GMPLS, both models are supported equally.  For the
peer model, one puts all boxes in one IGP area.  One easy way to
accomplish the overlay model is by the notion of IGP areas; another
is by means of IGP instances.  The finishing touch is to define a
good means of doing inter-area TE which includes GMPLS TE (watch
this space).

> Whether we choose a UNI or MPLS signaling, flexible access policy is
> required. That way, an optical carrier could decide exactly how much
> privilege to grant each client, on a case by case basis.

Absolutely.  A lot of folks don't seem to get this point: for
example, one could have OXCs and *some* routers in the backbone
area; other routers in non-backbone areas, and yet other clients
(routers, ADMs, ...) accessing the same OXCs via the UNI.  And
one could run NMS systems as well for provisioning legacy boxes.

> Please re-orient me if I am missing the point, entirely.

You aren't.  Note however that two other points have been raised:
a) is IP the "right" protocol for the control plane; and
b) is GMPLS the "right" protocol for the control plane.
However, I haven't seen convincing arguments that IP and GMPLS
aren't suitable; nor have I seen alternate proposals.

> P.S. I hope that I have not just opened Pandora's box.

No, it was already wide open :-)

One thing that has come out of this debate is that carriers are
thinking more deeply about requirements, which is a welcome step
forward.  At the same time, several deep-rooted biases are being
exposed, and I'll be the first to admit my bias towards using
MPLS/GMPLS for the control plane.  Another step forward, but this
one is more painful.

Kireeti.