IPv4 address length technical design

[bjorn at leffler.org (Bjorn Leffler)]

On Wed, Oct 3, 2012 at 12:13 PM, Chris Campbell <chris at ctcampbell.com> wrote:
>
> Is anyone aware of any historical documentation relating to the choice of 32 bits for an IPv4 address?

I've heard Vint Cerf say this himself, but here's a written reference
for you. They had just finished building arpanet, which was expensive
to build. Hence why they estimated two networks per country.

http://www.domainpulse.com/2012/06/06/world-ipv6-day/

When developing IPv4, Cerf said that he and Bob Kahn ?estimated that
there might be two national-scale packet networks per country and
perhaps 128 countries able to build them, so 8 bits sufficed for 256
network identifiers. Twenty-four bits allowed for up to 16 million
hosts. At that time, hosts were big, expensive time-sharing systems,
so 16 million seemed like a lot. We did consider variable length and
128-bit addressing in 1977 but decided that this would be too much
overhead for the relatively low-speed lines (50 kilobits per second).
I thought this was still an experiment and that if it worked we would
then design a production version.