🚀 go-pugleaf

Greg Pfister <pfi...@us.ibm.com> writes:
> 
> Related -- Once upon a time that same friend told me that you
> actually have to be able to do three levels of virtualization of
> memory to be general -- the top-level VM system just doing it is
> not enough; you have to be able to nest a VM system virtualizing
> what the top-level system is also virtualizing, and a third level
> below that. I never did understand why the third level was
> necessary, but this guy was very, very good and knew that code
> intimately, so he was probably right.

multi-level reference
http://www.garlic.com/~lynn/99.html#7

three levels of CP/67 under which ran CMS. standard cp/67 running on
the real hardware in an "open" environment (university students &
others on effectively a public access machine), 

then a modified CP/67 running in "360m67" mode ... but providing 370
virtual machines (primarily relocation tables different structure
between 360m67 & 370), and then a further modified CP/67 running in
"370" mode ... which then provided 370 virtual machines (running CMS)
... aka

     real 360m67
-      cp/67
        standard 360 & 360/67 virtual machines
-         CP/67 modified to provide 370 virtual machine
            370 "R" virtual machines
-             CP/67 modified to run on 370 w/virtual memory
                370 "R" virtual machines
-                 CMS

This was about two years before virtual memory was available on 370
(and possibly a year before 370 engineering models with virtual memory
were available). Issue was to try and obscure the fact that
development for unannounced (or even unavailable) products was going
on.

while there was quite a bit of use of CP/67 & VM/370 inside the
corporation for the testing of new operating systems ... in terms of
CPU cycles, probably the biggest use was straight interactive
computing (edits, compiles, non-operating system testing, email,
documents, GML, etc). Part of this give rise to the largest network
in the world (larger than arpanet until around '85)


'83 ... had nearly 1000 mainframe nodes at point when arpanet was about
255 total nodes. random refs:
http://www.garlic.com/~99.html#112

there was another way of getting machine time. the disk engineering
labs were doing all their development in a stand-alone environment
(MVS with a single "test-cell" MTBF was on the order of 15 minutes).
I rewrote i/o supervisor so it wouldn't fail & they could operate 6-12
test-cells concurrently. The priority list on processors ... was that
typicall as soon as the CPU/processors engineers had the 2nd one
running; it was made available to the disk engineering labs for their
testing.

random refs:
http://www.garlic.com/~lynn/2000d.html#0
http://www.garlic.com/~lynn/subtopic.html#disk

-- 
Anne & Lynn Wheeler   | ly...@garlic.com -  http://www.garlic.com/~lynn/