Re: Architectural approaches to multi6

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

    > From: RJ Atkinson <rja@extremenetworks.com>

    > The only architectural approach that I've seen on this list in recent
    > months is the notion of deprecating the concept of an "address", in
    > favour of using "locator" and "identifier" for the separate purposes of
    > routing and identity. Maybe I've overlooked something ?

Ahem. Separating "location" and "identification" is an architectural
approach to dealing with the fact that the address of the entity at the
other end may change 'frequently'. It's *not* an architectural approach to
handling multi-homing in a scalable way.


The only architectural approach to scalable multi-homing that seems to have
much interest is using multiple addresses (i.e. overlapping topology naming
abstractions), and switching among them as needed.

I have in the past mentioned a number of other approaches, e.g.:

- DDC's "route stubs"
- the equivalent in MD-systems (let's call them "map stubs")
- highly dynamic abstraction boundaries with a deep hierarchy

but they all involve i) big changes to the routing architecture, and ii) for
the third, at least, frequent changes in address (in other words, you need the
I/L split there too).

Overlapping naming abstractions - which we can do with the existing routing
architecture - seems to be the only choice anyone has expressed interest in.

	Noel