The following is part of the transcript of Bob Cringley's interview of Bob Kahn in which he discusses the number of networks they thought the Internet would have to interconnect:
Bob Kahn: The ARPANET had 16-bit addresses. The way you would address something was you spelled out in the 16 bits which machine you wanted to go to. Well, guess what? They were all one net. Right?
So you need to say what net, but you had to say what node of the net to go to and then you had to say which wire out of the net. They were four wires so you can connect four machines basically. And so that's what it was. It was sort of like two bits to say which of the four wires and then another I guess six bits to say which of the 64 nodes that you could handle, and so those were using 8 of the 16 bits. And maybe there were a few more used for other minor purposes, but to first order that's what happened in the ARPANET. We say, "Well, if we're going to go to a larger set of networks, we need more bits," so we went to 32 bits. And our thought was -
Bob Cringely: That's enough.
Bob Kahn: - this is going to be good for the indefinite future. I mean, what we did was we took the 32 bits and said we'll take the first 8 and use that to designate which net. Well, if you figure out how many combinations of 8 bits you can have, there are 256 combinations, and we figured, well, AT&T would probably have a net of their own and maybe the defense department. That was two. Maybe there'd be one in Europe. That's three. We figured, well, maybe there'd be one in Japan or Asia Pacific or something. That'd be four. We double it. That's eight. Let's double it again. That's 16. Double it again. That's 32. And no matter how we could calculate it, 256 seemed to be more than we'd ever need. But what did we not think about? Well, we hadn't contemplated the personal computer being generated. We hadn't been thinking about local area nets that connect lots of computers in one building, for example, but within a very short time after that, all of that hit us. Xerox part came out with their altos and the Ethernet came right on the hills of it, so very quickly we realized that 8 bits wasn't going to do it. And so we had to kind of rethink how we were going to deal with that. We still kept the 32 bits. In fact, there are still 32 bits today. It's called version four of the IP protocol. IPV4. But people are now trying to move toward a 128-bit version, which is called IPV6, and it's hard to make those transitions. Anytime you move from one regime in which all the pieces sort of work together and you try and change some pieces then you've always got the issues of backward compatibility and how do you make the transition happen in a smooth fashion. So that was one of the things that I think we didn't get quite right initially, but we had enough flexibility in the system to cope with it.
Bob Cringely: Sure. Oh, yeah. The fact that we're still -
Bob Kahn: We're still using it.
Bob Cringely: - functioning with 32-bit addressing.
Bob Kahn: But we knew within six months to a year after that that 8 bits wasn't going to hack it.
Bob Cringely: Really?
Bob Kahn: But we thought 8 bits was perfectly fine when we did the original work.