Two big debates I recall:
Datagram (packets) versus virtual circuits (telephone calls).
Berkeley (BSD) sockets versus Bell Labs (dmr) "streams" APIs.
https://cseweb.ucsd.edu/classes/fa01/cse221/papers/ritchie-stream-io-bellla…
It is important to note that the STREAMS stuff that eventually was foisted on
unfortunates from USG in System V was - despite being derived from Ritchie's
work - yet another ... thing.
https://en.wikipedia.org/wiki/STREAMS
dmr was trying to replace the abomination that the tty drivers had become (and
I still wish we'd adopted his APIs and work for that purpose to this day),
but trying to force everything into a connection-oriented framework for a
datagram/packet network has conceptual limitations, and to AT&T's chagrin,
the Internet is not an X.25 or Datakit network.
Plan 9 also had some interesting work in it that I wish we'd adopted.
Erik Fair