The MPLS WG Archive

["J. Noel Chiappa" <jnc@ginger.lcs.mit.edu>]

    > From: Bala Rajagopalan <braja@tellium.com>

    > I'm not sure what you mean by "endpoints" in the dynamic overlay
    > case.

I think he means "border router" (generally exit border router), where the
BR is also a participant of some sort in the lower-level network (even if
it's purely as a client).


    > the addressing is the same in IP and optical domains.

As is I hope obvious from my earlier (long, sorry) note about abstractions,
which names you use is far less important (indeed, unimportant, really)
than what representation of the topology you're passing across.

    > In the "open" model scenario, there could be routing exchange between
    > the domains (e.g., BGP), so there is really no need to know about
    > optical endpoints corresponding to destinations in the IP domain (but
    > as in the dynamic overlay case, the optical network will take care of
    > routing within).

I'm not sure what you meant by this (in part since you used the term
"endpoint", of which you just finished saying, above, you didn't know what
it meant :-). Even after staring at this for a long time, I can't single
out your likely meaning - can you restate it in a bit more detail?

(Also, can you please be careful to distinguish between i) "domains" -
which in this particular discussion I am thinking of as an area of the
network where a particular technology [e.g. optical] is being used - and
ii) "layers"/<other-term> - which to me implies something about the
functional relationship between two different systems, e.g. optical routing
and IP routing. I think you're tending to use the same word for both, and
it's a bit confusing...)

(Can I also be a bit of a nit-picker and point out that no matter what
model you're using, there is *some* sort of "routing exchange" between the
layers - even if it's a static configuration - since *some* representation
of connectivity has to be propogated into the IP layer before it can send
traffic. I would assume what you really meant was "a dynamic routing
exchange".)


    > a distinction has to be made between control and data planes. Even if
    > you use the peer model with full exchange of topology etc in the
    > control plane, the connectivity between routers in the data plane is
    > over a pipe in the optical domain. That is, the interior nodes in the
    > optical domain cannot see the IP data or LSPs

True. What's important is i) who is in control of deciding paths, and ii)
to the extent that there are two separate entities doing so (e.g. long-haul
IP routing, and a local optical path-selection mechanism), what
representation of the topology the lower-level entity is propogating to the
higher-level one.

Note that sometimes the "propogating a represntation" may be implicit; e.g.
if the optical nodes are peer participants at the control plane for routing
(i.e. operating a common IGP with some non-optical nodes, and it's all one
area), that counts as "full topology advertisement" - even though there's
no interface where anyone is handing across a full map.

	Noel