The ARPA network made a transition to TCP/IP gradually sometime
around 1981-82.
Actually, it was sort of sudden. I remember a lot of fuss during
the NCP-to-TCP cutover. (The switch was required, though, because
NCP had an 8-bit "host number" and thus was limited to 256 hosts
on the entire worldwide network.
) There was a lot of prep
work first, of course -- and as I recall, a separate network
(or perhaps several) continued to use NCP for some time.
(The University of Maryland got onto what was becoming the Internet
in the mid-1980s, via a connection by ECUs to an IMP that was
actually on MILNET. The remote end of that connection was located
at the NSA, somewhere in or near Fort Meade, which made for a
remarkable snippet of phone conversation one time. ECUs had some
reliability problems, and too-often needed a manual reset. This
had to be done at their end as well, which meant we had to call
someone there to do it. One time when we called -- I was not the
one on the phone, but was present -- I heard Jim, the caller, say
something like: "uh, it's green", followed by an aside to me: "he
wants to know, `what color is my phone'.")
(I will also note here that we were on Usenet by late 1982 or so,
certainly by 1983 anyway. The net.lang.c group existed by the time
I first read Usenet news, though.)
Ah, yes, Bolt-Beranek-Newman. Without one you could not connect.
To bring things on topic for a moment
we should also mention
the BBN "C machines", so called because they were designed to run
compiled C code. These were probably the first machines so designed,
since earlier architectures generally predated the language.
The "C machines" had 10-bit bytes.
Yes, the first machines specifically designed to work with C had
bigger-than-8-bit bytes.
(See <
http://www.isi.edu/in-notes/ien/ien164.txt>. The last
paragraph on page 2, before the footnotes, describes the memory
resources on the C/50 and C/70.)