> > The only alternative is to create an artificial scarcity in address space.
> No, it's certainly not the 'only' alternative. Network translation would
> allow many companies to use nonexistant blocks of addresses.
We're already doing that. The thing is, it's not a long term solution. It's
a one-time percentage cut in an exponential growth curve.
Paul mentioned that a group of people have recovered 20% of the address
space that was unused. Now let's say that you can cut the address space
by another 20%. With the number of hosts on the net doubling every two years
(I've heard estimates from 12-18 months, but let's be conservative) then
you've held off the day the address space runs out by 40% of 2 years, or
about 10 months. Let's say that 75% of the growth is behind firewalls,
so you get another factor of 4. That means you have a bit over three
years extra time before IPv4 runs out.
Meanwhile the existing (artificial) scarcity of IP address space has kept
ISPs from implementing lots of cool stuff, and increased telephone line
congestion. Pretty much the only reason that dynamically allocated IPs
for end-users work is that they're using circuit-switched connections for
the "last mile".
This ties up huge amounts of local resources in the telephone network, which
would be unnecessary if the user could drop and restart a phone call without
losing their IP connections. With ISDN it takes less than a second to set
up a dial-on-demand link. Two-way dial-on-demand should still be cheaper than
nailed (even for the duration of a session) lines in areas with time charges.