RE: (multi6) requirements draft comments

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

    > From: "Tony Hain" <alh-ietf@tndh.net>

    > So there is a prefix for every combination of possible providers a
    > customer can subscribe from

No - at least not the way you seem to be meaning when you say those words).
I.e. you seem to think everyone with service from Verizon DSL and Cox Cable
would share the same prefix - which is distinctly not what I meant.

What I meant was that all the people who are connected to both:

- i) the same segment of the York County Cox CATV system here in Virginia,
	and
- ii) the same Verizon DSL office

would share the same one-level-up topological naming entity - and that would
be a very different one-level-up topological naming entity from the people
who have Cox CATV service in, say, Maryland, and who are also getting Verizon
DSL service from an office there.


Anyone thinking about anything related to routing needs to take a sharp
implement, and engrave on their forehead "REAL TOPOLOGY IS KING". For routing,
the *only* thing that matters is *actual* connectivity; i.e. the actual wires
and routers. I often hear people *claim* they are paying attention to this -
but then they go off and say things which show clearly that they haven't
*really* understood it (and taken it to heart).

Just because two wires are owned by the same company doesn't mean a damn
thing. The only thing that matters is the real network topology - i.e. the
graph of the real, physical assets that make up the network.


The other thing that people need to keep firmly in mind is the K-K lesson,
which is that the way you get the routing to scale, even in mindbogglingly
large networks, is to use LOTS OF LAYERS OF ABSTRACTION. Three or four real
layers of abstraction for a network of 10^8 nodes is *not* going to cut it.


    > How would the providers know where the overlaps are to route correctly,
    > ... How would routing work if those overlap areas were discontiguous?

Some of these questions seem to be based on your incorrect understanding of
what I was proposing. Both of these questions make no sense if each area of
overlap has a (globally) unique Sij name (but one which may itself be subsumed
in the name of a one-level-higher topology aggregate).

    > This approach moves the scaling problem from the technical space to the
    > administrative space, where the complexity would create prohibitive
    > costs.

No, I was speaking of a system that basically automically assigned
routing-names, and did so basically *purely* on the basis of connectivity;
i.e. it would deduce for itself where to put abstraction naming boundaries.
The *last* thing I want is bureacrats trying to do it.

No doubt this would not sit well with the providers, but I don't mind waiting
for the tide to cover their thrones.

	Noel