Re: PI/metro/geo [Re: The state of IPv6 multihoming development]

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

    > From: Brian E Carpenter <brian@hursley.ibm.com>

    >>> E.g. if a (large enough) bunch of customers band together and create
    >>> an exchange, which buys service from multiple ISP's, and which is
    >>> given an address which is visible over the same scope as those ISP's,
    >>> and the customers are addressed as part of that exchange, you meet
    >>> the policy goal (being able to change providers) *without* so-called
    >>> "PI" addresses. But the addresses are still connectivity-based.

    > The problem with this (and the reason I have serious doubts about
    > Tony's PI proposal, and all geographical proposals going back at least
    > to Bill Simpson's metro proposal years ago)

Well, you usefully point out that all geographical proposals basically amount
to the proposal to create exchanges! :-) At least, if the routing is to remain
functional!! :-) :-)


    > is
    >  There is no economic reason to create such an exchange.
    > If we could get past that, of course it works.

Ah. You misunderstood the point of my comment, which was not to advocate the
creation of exchanges, but solely to point out that the terminology of
"provider-independent" is not technical terminology, but rather
economic/policy terminology.

In other words, there are addressing scheme which are "provider-independent"
but which differ in extremely important technical ways from the PI addresses
proposed by various people.


The important division in addressing schemes is between "connectivity-based"
and "non-connectivity-based", and it occupies the same place in networking
discussions as the division between "second-law-machines" and
"perpetual-motion-machines" does in engine discussion.

There are many useful finer subdivisions of the "connectivity-based" class
(e.g. hierarchical, provider-dependent, exchange-dependent, etc - some
orthagonal to each other), just as there are many useful finer divisions of
the "second-law-machines" class. However, at least in discussion of engine
designs one normally doesn't have to spend a lot of time discussing
perpetual-motion machines; everyone involves understands that although that
is the most fundamental division, the second class isn't worth discussing,
and not a lot of time gets spent discussing things in that division.

One day the networking world will get there too, and the only discussion of
routing-names (i.e. those names/identifiers/whatever which the routing
calculations use - addresses, in the IPv4/6 architecture) which people will
bother with are the "connectivity-based". I'm not holding my breath.

If people don't like some of the properties that being connectivity-based
induces in routing-names, then the answer is simple: define another
namespace which has characteristics you like better, and map back and forth
between the two.

If the price of that is too high, then you just have to deal with the
hassles of connectivity-based names, just like we have do live with the
hassles of friction, entropy, etc.

	Noel