multihoming

[Michel Py]

> Bill Manning wrote:
> How many /64s does each lightbulb get?

I would say each light bulb get its own unique 64 bit MAC address. Given
the typical lifespan of a light bulb GE will be able to recycle MAC
addresses for light bulbs after some years. In my house, each light bulb
is a /128 address present on three /64 subnets. I'm sure that everybody
recognizes the vital role of managing an IPv-6 enabled light bulb to the
point that each light bulb must be multihomed to at least three ISPs.
It is none of the manufacturer's business neither the utility to know
when my individual light bulbs are on.

So, if I get three /48s for my home, I can have each room being a
separate subnet, where light bulbs can be multihomed to three different
ISPs
And I still have addresses to spare.

> Or are you making the tacit assumption that everyone 
> gets enough space to address all the things that are of
> interest to them, with their OWN block of v6 space?

Generally speaking, yes. I would say that any home owner _really_
interested in multihomed light bulbs will get a /48 block of
portable-within-the-area, provider-independent addresses on top of few
number of /48 PA blocks). However, I feel that a very large part of the
world's population will be happy to manage their IPv6 light bulbs on a
single-homed single-subnet with a /64 block of addresses provided by
their ISPs.

> IMHO, the whole point of CIDR in v4 was to address TWO
> problems, first, address exaustion. Delegation policies
> were increasingly finetuned to only delegate as much space
> as was really needed.  second, a constraint on routing table
> size.  

Correct. Since address exhaustion is not a problem in v6, the idea is to
allocate more addresses than one would ever need, because it will allow
aggregation and therefore a reasonably manageable routing table size.


> It does not mean that we should hammer
> IPv6 into the IPv4 mold nor should you restrict your thinking 
> to using v4 routing protocols for v6 address space.

I fully agree.
	

> Neither the delegation problems nor the routing problems
> are tractable with current thinking.  Hierarchical
> delegation/routing, while known to work, do not meet the
> wants of the user populace. They will find a way around
> what they perceive as impediments.

I beg to differ. It is possible that you have not been following the
current multihoming developments.

Michel.