On 09/16/2015 12:23 PM, Sean Caron wrote:
And I actually got to play with NOS ... many years after the fact ...
never thought I'd see that! What the cray-cyber.org guys are doing is
remarkable.
Sad that they don't have any early software. In the beginning there was
COS (Chippewa Operating System), followed by SCOPE (which doesn't really
describe a specific OS, but more on that later). On the 6000, both were
essentially PP-oriented, leaving the CP to do the real work.
MACE was essentially a bootleg product mostly done a night on the Arden
Hills QA floor by (Dr.) Dave Callender (he of the bats) and Greg
Mansfield (whom I had the privilege of introducing to the delights of
gelato--but that's another issue. Greg wanted a slimmed-down operating
system for the CE community--hence, the name "Mansfield's Answer to
Customer Engineering). Greg, unfortunately, was your basic geek, not
good with people, but very talented. Without Dave's promotion, I doubt
that it would have gotten anywhere.
MACE did make its way into the academic community. I believe that
Purdue added and extended it quite a bit. One aspect is that the OS was
considerably more CP-involved than SCOPE. So, for the interactive stuff
like PLATO, MACE was re-named KRONOS and launched as a separate official
product. At the user level, the two were pretty much compatible, but
internally, they were very different. For example, SCOPE had a rather
elaborate disk driver, called 1SP (for "Stack Processor"), which sorted
and prioritized disk requests based *partly* on the distance that a seek
to cylinder involved. KRONOS on the other hand, used a much simpler
"first-come, first-served" principle.
Mostly, it was a battle of cultures. SCOPE was maintained out of Palo
Alto (Porter Drive) and later, Sunnyvale (Moffet Park Drive) and KRONOS
out of Arden Hills. Sometime around the advent of the Cyber 70 series,
management made a non-decision (with 128 vice presidents, how could it
be otherwise?) So, KRONOS was re-christened NOS (network operating
system) and SCOPE became NOS/BE (batch environment). Eventually, enough
of the BE aspect was rolled into NOS that only one--NOS survived.
Besides, CDC was doing their best to get rid of as much staff as they
could at CDC Sunnyvale. Certain key people, however, refused to
transfer, so it took well into the 80s before the last light went out in
Sunnyvale.
Now, that's about Cyber 70/6000 NOS/SCOPE. The 7600 was a very
different beast. For one, the 7000 PPs didn't run as slots in a
barrel--they were fully independent. Secondly, the PPs were
hard-assigned a buffer region each in SCM--they did not have unfettered
access to memory. This meant that the PP-resident aspect of 6000 SCOPE
(and KRONOS) was unworkable on the 7600. So, 7000 SCOPE was
born--mostly, what I saw was SCOPE 2 and SCOPE 2.1. With the PPs
dedicated to pretty much nothing but simple I/O, the whole business of
job control and user interface fell to the CP. Note that the 7600 had
very fast SCM, but not much of it, so storage of transient programs was
left to slower LCM. The scheme adopted was to have various functions of
the OS be nested, matryoshka-doll style with the user program in the
inside. So outside of the user program, you could have the job
supervisor, then the record manager, then the buffer manager and so on..
So 7600 SCOPE was nothing like 6000 SCOPE (or NOS).
There were other 6000-series operating systems, not very well known to
this day. For example, the ROVER people had their own version of what I
assume was an early SCOPE--I'm not certain if this has been declassified
even today, so the less said the better. There was TCM, Time-Critical
Monitor which claimed to have a maxiumum average event latency of 100
nsec ZODIAC was built on the TCM OS, used lots of ECS in
multi-mainframe configurations to host "chains" of real-time
transactions (a bit complex to explain) that essentially operated with
its own I/O and filesystem.
Life was interesting back then...
By the time you hit the 80s and the CYBER 180, not so much.
--Chuck
