[Simh] terminal multiplexers

[Clem Cole]

​below..​

On Thu, Nov 12, 2015 at 2:08 PM, Rich Alderson <
simh at alderson.users.panix.com> wrote:

> Hmm.  Come to think of it, the purpose of these was to convert serial lines
> to telnet.  My first encounter with "milking machine mode" (telnet to
> serial
> lines) was a Cisco ASM connected to an IBM 4994 (headless Series/1) to
> allow
> telnet into the IBM 4381s at LOTS, which was around 1989 and hardly early.
> I don't know whether earlier Cisco terminal servers or Stanford EtherTIPs
> (as they were called) had that capability in the standard software load.
>

​Great point.  When do we start differentiate them?  A few interesting
factoids/ pieces if history that I lived to follow:

First if we exclude the front end idea (more in a minute) and think purely
of serial to network converter like 3Com, Bridge, UB, etc, the first one I
ran into was a box at Tek Labs (Tektronix) called the  Network Interface
Black Box (aka NIBB) by the same folks that developed the 68K UNIX Magnolia
Workstation (a couple of years before Sun and Apollo).

The NIBB was a Z80 with some custom SW and MSI TTL to talk to originally an
ethernet-like stream we had developed and were using to show off the idea
of an "instrument controller".  But we used a lot of them to connect glass
TTYs to our 11/70 and 11/60 UNIX systems.    IIRC, there were a couple of
flavors of NIBBs but the serial version supported 2 serial ports since that
was what Zilog support with their "SCC" dual USART chip.​  Again, memory is
fading here, I think when we got the first 3Com boards (Tek Labs was 3Com's
first customer - a different but fun story), I think we upgraded the NIBBs
to run a true 10 M ethernet.  (But an issue at the time was the cost of the
ethernet transceivers which were about $500 ea plus $100 transceiver cable
in 1979 dollars - which I recall was about the same price as a NIBB itself).

That said, Rich made me remember something else.  In the early/mid 1970s we
used to use mini-computers to connect to terminals and connect the minis to
larger/more expensive systems - aka a terminal concentrator.   At CMU we
called this the "Front End."   This was done for a couple of purposes: it
allowed the terminal driver to pulled out of the PDP-10 and OS work like
echo, canonicalization, etc was moved upstream so on the 10s saw "cooked"
data; and it also meant that you did not need a big peg board (like what I
would see at UCB a few years later).  All terminals were connected that FE
and then it connected you to which ever system you desired (at CMU CS in
the 70s - this was the 3 PDP-10's and C.mmp).

I'm pretty sure MIT & Stanford had something similar, in fact I believe
that the old MIT SUPDUP protocol (and alternative to Telnet) was created
for just this sort of use.

BTW, CMU/MIT et al were not unusual in this type of Front End configuration
at the time.  Commercial folks did it too.  I remember that there was
timesharing service in Pittsburgh's North side that sold time on a few
PDP-10's that  a number of my friends worked.  All the terminals came into
two 11/45's which then switched up back to the 10s.  Even the IBM types
like Mellon Bank (and I believe the airline systems) used to use PDP11's to
do that type of trick because it was just cheaper to connect to a
mini-computer than directly to the mainframe.

In fact, a few years later a former boss of mine and one-time head of East
Coast regional sale for IBM, once told me he believed that the primary
reason IBM developed the Series 1 Mini was because IBM was losing so many
sites to PDP-11s for front end work.  i.e. they sold it as a terminal
concentrator.

Also, by the late 1970s, when the Vaxen started to show up at CMU and there
were more than just the 4 big systems; CMU started to develop the
"distributed Front End" - which originally was being done on LSI-11s and
3Meg Xerox ethernet.  I played with/helped hack one some of this before I
left for Tek Labs; as we were trying to get all the main systems across
campus on a big network, so you get to a EE or CS system from Mellon
Institute which was about a mile away.    I know that the Distributed Front
End was eventually moved to 8085's on Multibus boxes from the LSI-11s.  A
project that Phil Karn (aka KA9Q or sometimes just lovingly known as "the
nerd") and I did was build a tape system for one these for the graduate RT
time compute class.   Phil and I cons's up a C compiler for the 8080 and I
remember the distributed front end folks asking for it (which I have no
idea where the sources are today).

My memory was that the multibus ethernet board needed the space of a full
multibus board, but it was still cheaper than the LSI-11 solution.   If he
did not directly worked on it, Belchteshiem certainly knew about that
scheme as he was knocking around the CMU HW lab in those days (i.e. before
he went to Stanford a few years later and did the Sun-1 and their
distributed front end).

The point of all this being and to answer the question of how far back it
all goes --> using some other processor and a network to make a serial
connection this was a pretty natural progression and not surprising.
Putting terminals on a larger computer was always "expensive" in both HW
and what it takes to service them in the OS.    There were numerous ways
people tried to reduce that cost over the years, so by the time of LAT or
like; it was really a bit of old hat.

Clem
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.trailing-edge.com/pipermail/simh/attachments/20151112/9e57d340/attachment.html>