[Info-vax] VMS 5.0

[glen herrmannsfeldt]

Richard B. Gilbert <rgilbert88 at comcast.net> wrote:

(snip, someone wrote)
>> Whoa... how long did they run a 360/91.  I was last in front of a 360
>> at Monmouth Medical  when I worked for DEC in 85 or 86.

(snip)
>> My slow obsolete android 2.x tablet's probably got a 360 worth of power
>> in it. 8-).

> Yeah!  And my "smart phone" too!  Folks, you have to understand that 
> we are talking about the technology of the late 1960's and early 1970's. 
> "Core" was really an array of ferrite cores that were magnetized 
> "clockwise" for a zero and counter-clockwise for a one.

And on many S/360 models, the processor registers were in a
high-speed (smaller cores) core array.  Most models of S/360
were microcoded to allow a smaller machine to run a big machine
instruction set at relatively low cost.  Also, I/O devices had
a common interface so that you could move up to a larger, faster
model without replacing all the I/O devices.

> I think of the core "cycle" as something like firing a revolver over 
> your shoulder to determine if there is someone behind you.  If you hear 
> a body fall, you knew that there was somebody behind  you.  So you read 
> a core by setting it to zero, if you got a pulse on the "sense" line you 
> knew the core *was* in a "one" state.  No pulse meant the core was in a 
> "zero" state.  You then put it back like you found it.  

Note that the DRAM we use now has to write back just like core.
The data is stored by the charge on a little capacitor, with the
charge being drained off by a read, and then restored.

In the core days, there were some machines that could do a
read-modify-write cycle, where the old value isn't restored,
but the new one is.  (Consider an increment or decremement.)
In the early DRAM days, some might have done that, but now
the restore is done internally as part of the memory cycle.

> It was a kludge and, by our standards, incredibly slow!  
> It also generated heat.  

But done well it was pretty reliable, which wasn't true of
many of the older systems.
 
> The IBM/360 Model 91 had sixteen megabytes of RAM in sixteen 
> one megabyte water cooled "bombs".

Sixteen way interleaved, but not that big.  There were 2MB 
and 4MB boxes (each containing 16 core arrays designed for
a slower machine), which you could have one, the other,
or both, so up to 6MB, and 780ns cycle time.  The processor
cycle was 60ns, so with 16 way interleave and some luck you
can get one fetch per machine cycle.

But the low end S/360 were much slower.  The 360/30 uses
a byte wide core array with a 1.5us cycle.

-- glen