In message
<4(_dot_)2(_dot_)2(_dot_)20000425090404(_dot_)00a3c3f0(_at_)pop(_dot_)dial(_dot_)pipex(_dot_)com>,
Graham Klyne wri
tes:
At 11:06 PM 4/23/00 -0500, Richard Shockey wrote:
With "always on" IP and IP on anything this is closer to reality than we
might think. In order to permit a reasonable allocation of addresses with
room for growth the idea of 25 IP address per household and 10 person
actually seems conservative.
Following this line of thought, I'd suggest taking the number of electrical
outlets and multiplying by some suitable constant (say, 10, or 1000).
And what about all those little wireless-connected gadgets; an IP address
for each TV remote-control (where everyone has their own, of course, for
personalized access and prioritizing control conflicts...)
I've been in rooms where people have walked through exactly calculation.
Let's throw a few numbers around.
Assume that the average person in the world has 1000 outlets. That's
preposterously large for even Bill Gates' house, I suspect, and it doesn't
even account for dividing by the number of people per house. But let's stick
with 1000. Assume that there are 25*10^9 people in the world -- 4x the
current population. And allocate 10 IP addresses for each of those outlets.
That means that we need a minimum of 25*10^9 IP addresses, plus allowances for
delegation on TLA boundaries, smaller provider chunks, homes, etc.
So -- when I divide 2^128 by 25*10^9, I get ~2^80. That's right -- 80 bits
worth of address space for allocation inefficiencies. If, at each of three
levels, we really use just one address out of every 2^16, we *still* have
32 bits left over.
Conclusion: if we don't do things differently, we have more than enough
addresses for any conceivable size of the Internet. (In fact, if we assume
that there are 300*10^9 stars in the Milky Way, we're not very far from being
able to handle interstellar networking, at least in terms of the number of
hosts. But we may have to redefine the TTL field to let packets last for 50K
years or so...)
OK -- if 2^128 is that large, why did I (among others) conclude that we needed
some headroom? There's a lot of history showing that address space *always*
runs out, and that the consumption often comes from changing your paradigm. A
routing/addressing scheme like 8+8 is an example of what I mean. In another
vein, I have a project going that requires 2^16 addresses per host, and maybe
2^32. There are undoubtedly other new ideas we can play with *if* we have
enough address space. But we have to engineer this in some fashion that
permits efficient use of these addresses by hosts and (especially) routers.
(An earlier poster wrote that you "just have to write the code". That's the
wrong idea -- big iron routers don't use "code" to do forwarding, they use
silicon, and very fast silicon at that. There are routers in production use
today that are handling OC-192 -- 10 Gbps -- links. You can't do that in
software.)
--Steve Bellovin