[TUHS] update pdp11 system five?

[Clem Cole via TUHS]

below

On Wed, Dec 17, 2025 at 2:27 PM Larry McVoy <lm at mcvoy.com> wrote:

>
> Is anyone aware of a STREAMS TCP/IP besides the Lachman/Convergent and
> the Mentat ones?
>
Those were the two most popular/. IICR Mentat was also used by HP.   Also,
I was under the impression that Doug Comer's stack was once stuffed into
the TLI, and that BBN may have also developed/ported their stack to streams.

>
> And in defense of Dennis' streams that USG morphed into STREAMS, he
> intended it for tty interfaces, not for networking.  Ttys were slow
> enough that the queuing was fine.
>
Yep, it was a nice trick to replace the early idea of "line disciplines."
 As with the monokernel vs. microkernel debate, using streams as a
networking API might have solved some interesting issues.  I once had a
nice discussion with Denis about it, which was a classic theory vs.
practice style discussion.

That said, in practice, because the System V streams code performed so
poorly, especially when a BSD socket stack was the comparison. Since almost
all of the user code that did networking expected the sockets API, nobody
felt that taking the time to make a TLI-based TCP/IP stack that was as
performant as the traditional sockets was of any value.  Remember that
Metcalfe's law never talks about >>how<< you get connected, only that you
are connected.

SVR4, of course, offered both APIs, and I never saw any commercial code
base that used the TLI.