-- thus allowing a partitioning of the IPv6 address
space in a way that is simultaneously both
topologically aggregatable _and_ policy-based
That would be good.
Actually I'm of the opinion that IP addresses exist to name attachment
points (and sets of attachment points) in the network, and this governs
how they are structured. They are used to name other things also, and
this is often a useful optimization. But any attempt to change the
structure of IP addresses to do something besides name attachment points
is likely to compromise this structure.
128 bits is probably enough for naming network attachment points for
quite a long time, especially with due care in how they are assigned.
But if you try to add policy to this structure, 128 bits becomes
quite a small number.