Closure?

[Thomas Narten]

Shane,

> Actually, I am worried by both the overconfidence and the apparent
> classfulness and inflexibility of IPv6.  We've started with 128 bits,
> threw away half, split the rest up on 16-bit boundaries and now we're
> left with 13 bits to play with all of a sudden.

Just to clarify the above point. I believe it is a misunderstanding of
the intent of the various IPv6 addressing documents to compare them
with classful addressing.  The only firm boundaries within addresses
that have been defined by the architecture are the 64/64 split
(routing part and interface identifier). This was done as an explicit
engineering tradeoff decision in which the benefits of stateless
address autoconfiguration were believed to be large, and the belief
that leaving 64 bits for routing left more than ample room for being
able to address the number of networks as called for in the original
IPng requirements.

>From the architecture's perspective (and from what implementations
know about), addresses are just 128 bits long. From a routing
perspective, longest match is done. The internal address boundaries as
specified in RFC 2374 (for example) are really there for address
assignment and management, not because implementations need to be
aware of them. Indeed, no IPv6 document that I am aware of calls for
(or even suggests) that an implementation might want to take those
boundaries into consideration.

Having said that, the concern that implementations (hardware or
software) might hardcode such boundaries has been heard from several
sources, so I suspect that more forceful wording on this point would
be useful. But I think that wording is probably more important to have
in other IPv6 documents than in the addressing recommendation under
discussion here.

> The analysis that
> "well, it's only 1/8 of the available space" is fundamentally
> flawed,

I don't think folks are making this argument to justify giving /48
everywhere. Or if they are, I certainly don't agree with this line of
reasoning. What I believe folks *are* saying though is they believe
that 45 bits of heirarchy is really a lot, and *is* sufficient to
allow the giving out of /48s. However, do note that if the analysis is
turns out to be flawed (not that we expect it to be) we do still quite
a lot of remaining space that we can use more conservatively.

Thomas