[ppml] IPv6>>32

[Stephen Sprunk]

Thus spake <Michael.Dillon at radianz.com>
> > I think this /48 per room stuff is just a little over the top.
> > Is this "just because we can"??  Let's try to show a little sense
> > and some version of reality rather than extreme future fantasy...
> > are there really going to be 64K *subnets* in each dorm room???
>
> I think people are confused about whether dorm rooms are
> like private apartments or like hotel rooms. If they are
> like private apartments, then they should get a /48 but if
> they are like hotel rooms, then a /64 should be able to
> service a whole floor if not more. I suspect that the majority
> of dorms are more like student hotels than they are like
> private apartments. In any case, I don't think it is fruitful
> to try to count the number of subnets or hosts in any particular
> room.

I would classify dorms as "student hotels" and assign a /64 per floor (or
per building even) unless a student specifically asked for their own /64 or
/48.  Unfortunately, I'm not sure that's permitted under existing policy if
the student is paying the university -- or if a guest is paying a hotel -- 
for Internet access.

Extending this to the absurd, if the university (or other ISP) were offering
service via 802.11, would we still require them to hand out a /48 per
user/customer?  Are there even any consumer-grade devices that work in such
a scenario?

> > which brings me another complaint about this thread...  people
> > are talking like networks should assign a /64 per host.  these
> > are subnets, people!!  they might even have 10s of hosts using
> > one prefix!
>
> Again, what are we talking about? Are hosts single machines with
> a single port providing an IPv6 network attachment point? Or are
> they devices containing several internal IPv6 networks, each
> communicating with different external IPv6 networks doing jobs
> like service monitoring, control plane, data collection...
> Again, I don't think it is fruitful to start getting into the
> internal architecture of devices to decide whether or not they
> are hosts.

The reality is that today and for the conceivable future a single human will
only have a few to a few dozen devices needing a handful of addresses
each -- and that's allowing for a lot of growth from today's norm of one IP
per user -- which is well within the realm of simple, existing, cheap L2
technology.

We can throw around ideas of fruit crates (or even fruits) that need their
own subnets for internal communication, but until such devices actually
appear (and IMHO, it will not be within my lifetime), we shouldn't be
wasting 80 bits of addressing on each human when only one to five bits will
be used in the vast majority of cases.

> If the person in charge of said device says that
> a /64 is required, then I'm happy to give a /64.

Agreed.  But policy shouldn't _force_ people to deal with having their own
/64 or /48 when they're happy getting one or more /128s out of a shared
pool.

> Simple rules of thumb can be applied to resolve these kinds of
> issues.

ARIN doesn't operate on simple rules of thumb; it operates on policies.

S

Stephen Sprunk      "Those people who think they know everything
CCIE #3723         are a great annoyance to those of us who do."
K5SSS                                             --Isaac Asimov