🚀 go-pugleaf

On Fri, 27 Jul 2001 08:44:00, mikeki...@cix.co.uk (Michael J Kingston) 
wrote:

> I recollect reading that the 9020 was, or at least started as a 
> multiprocessor version of the S/360 Model 40.

50 and 65.

> No doubt there were changes 
> and additions to the instruction set.

Principally to manage the multiprocessor configuration.  Test Config, Set 
Config, that sort of thing.  The 9020E, which was used as a display processor,
had some additional instructions but I forget what they were.  Line intercept 
sort of things, to determine what was outside the view of the display, I 
think.  Maybe some special block moves.  The E also had display refresh 
memory.

There were also "Peripheral Adapter Modules" to interface to teletypes, radar 
digitizers (12-bit bytes), and like that.

A typical configuration was three 65s, three 50s (serving as I/O processors), 
and three PAMs.  Plus a display system, either Raytheon or a 9020E, with 
Raytheon consoles.

> It would be interesting to learn what changes other than the mechanics of 
> shared memory were made.
  
The System Journal article ought to say.  The multiprocessor was hardware 
configured into online and offline systems.  The online system did air traffic
control and the offline system did offline kinds of things.  Offline hardware 
could not interfere with nor reconfigure the online.  When the online system 
detected a failure, it configured out the failed unit and brought in a unit 
from the offline system.

All that hardware has been replaced, but the software, which first went 
operational in the early 70s, is still in use but now in duplexed 370s instead
of multiprocessor 360s.  The multiprocessor was always a rotten idea, since 
there wasn't enough hardware to configure two complete systems.  This was done
supposedly for efficiency, but the consequence was that there was no way to 
test a new software delivery except to take the system down.  We told them 
that but the guy in charge wouldn't listen.

-- 
    John Varela