[ppml] [address-policy-wg] Those pesky ULAs again

[David Williamson]

On Tue, May 29, 2007 at 02:58:36PM +0000, Paul_Vixie at isc.org wrote:
> according to <http://www.isc.org/index.pl?/sw/dhcp/dhcp4_0.php>, there's
> now code in "alpha test release" status to handle DHCPv6.

That's good news!
 
> imho, the days of EUI64 are numbered.  at home i'll probably use a /120 for
> each LAN.  at work, we might splurge and use /96's.  not that a /56 isn't
> enough for my house or anything, i just want the sparseful wastitude of the
> new address bits in IPV6 to all be at the top end.  i'm using a /124 for my
> T1, mostly to make the PTR's easy to write and read.

This is the most sane thing I've heard in this conversation.  Frankly, I'm
baffled by the hard architectural boundary at /64.  Perhaps I'm limited in
vision, but it's difficult to imagine a new layer-2 technology that wouldn't
be entirely overwhelmed by a /64 subnet that wasn't extremely sparse.  At
some point, that subnet has to get to a router, and it's unusual for a router
interface to be rated at more than 10x the bandwidth of the devices attached
to that interface.  That typically implies that there are less than 100 devices
attached to that router port.  Even if some new technology comes along that
makes those (very loose) numbers off by a factor of 5 orders of magnitude,
we're still looking at a very sparse /64.

It's almost as if we're back in the early 80s, and trying to find the
same potholes to step in.  Didn't we learn anything from IPv4?
Apparently not.  We still have the EIG/RID split problem, with no
solution apparent on the horizon, and we've simply made a new network
protocol with more address space and incompatible headers.

> >   I predict with the current allocation procedures IPv6 will be
> >   "used up" in my lifetime.  I also predict the groups today getting
> >   /32's (and larger) will look like the legacy class A holders in
> >   20 years time.  When your doorknob automatically requests a ULA-C
> >   /64 when you bring it home, and your house has 2,000 of them as every
> >   individual system talks to each other we'll be looking at this quite
> >   differently.
> 
> i include this only so that i can say, i nearly agree.  unless we have an
> IP architecture that splits EID/RID, those doorknobs will not be globally
> reachable.  (not that this is a problem for doorknobs but it might be for
> microwave ovens or something.)

Yup.  Giving someone address space in no way makes that space
routable.  At some point, the routing system with either break, or it
will become entirely monitized and you'll need to pay to get your
routes where you need them.  If you are a global content provider,
that's going to be a big surprise.  (As a side note, I wonder if the
economics of this will work out such that shorter prefixes cost less
than longer ones.  That might make a big incentive for aggregation.
Hmm.)

-David