Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!husc6!cmcl2!brl-adm!adm!bzs@bu-cs.bu.EDU
From: bzs@bu-cs.bu.EDU (Barry Shein)
Newsgroups: comp.unix.wizards
Subject: pdp-11/55
Message-ID: <9943@brl-adm.ARPA>
Date: Fri, 23-Oct-87 19:30:04 EST
Article-I.D.: brl-adm.9943
Posted: Fri Oct 23 19:30:04 1987
Date-Received: Sun, 25-Oct-87 17:52:54 EST
Sender: news@brl-adm.ARPA
Lines: 47


From: Henry Spencer <henry@utzoo.uucp>
>> In my experience, most PDP-11s in the mid-70's made
>> their way into labs, or onto factory floors. There, I am quite sure,
>> Unix V5 would have shined not at all.
>
>Depends on what one wanted to do with it, but there were people who did
>real-time work using such early-vintage Unixes.

Indeed, at the Harvard School of Public Health my colleagues and I
were doing real-time work on PDP11/10s running V6 (mini-unix) mostly
measuring people's lungs (spirometry.) We collaborated with other
groups in the Medical School area who were doing their own real-time
projects. This would be the mid-late '70s.

It hadn't occurred to any of us that you couldn't do real-time with
Unix until that "truism" started going around (started, no doubt, by
our former RT-11 salesthing :-) We seemed to be doing just fine.

I wonder if people are as sick of these "can't do real-time" stories
as they are of these "gee, but we were doing it" stories. Granted
there have been "real-time" operating systems (so-called because they
even used the term in their names) but I don't remember a whole lot of
support in those systems which ever helped me do my work other than
perhaps ready to run device drivers for real-time clocks. Then again,
the drivers were always for the wrong flavor clock board anyhow, or
the A/D you were using was different or you wanted to handle it
differently etc etc.

I think all it ever came down to was that a lot of these systems could
set real-time interrupt priorities way up so they got serviced in a
predictable time-frame. But that was the *easy* part to add to the
kernel (it was doing threshold sampling, smoothing and conversion into
the frequency domain between interrupts in less than 20 instructions
which usually got your head scratching for a few minutes, in real-time
stuff it's reality that's always the problem.) And yes, we did
time-sharing while we sampled (eg. a technician could fill out a form
on the screen to set up the next subject as a sample was being
collected, worked well enough.)

Anyhow, it's all moot, if I were to do it today I'd plug the real-time
boards into an IBM/PC or similar and give it an ethernet interface to
Unix. Micros have made most real-time-in-a-time-sharing-system
obsolete, you can just dedicate a system to the task these days (which
is actually what we did with LSI-11's.)

	-Barry Shein, Boston University