[ih] mainframe architecture. was Interprocess Communication

[John Levine]

It appears that Jack Haverty via Internet-history <jack at 3kitty.org> said:
>I have a vague memory that the IBM 360 architecture circa later 60s 
>involved the mainframe CPU and processors in peripheral equipment like 
>disk controllers, communicating over some kind of bus.  So even a single 
>mainframe was a multiprocessor.

Well, sort of.  The 360 used channels for I/O.  Each channel was in principle
a simple computer that could execute a sequence of instructions known as
Channel Command Words.  On the larger models the channels were separate
physical devices, while on the smaller ones the CPU microcode also ran
the channel.  There were standard cables known as bus and tag that connected
the channel to the device controllers.

>Also my recollection is that some form of message-oriented interaction 
>protocol was used to manage disk I/O, since I/O operations often didn't 
>complete in the same order as they were initiated. 

After a CCW started an I/O operation there was "channel end" when the
data transfer was done and "device end" when the physical operation
was done. If you did a separate seek, channel end was immediate so the
channel could do something else while the device moved the arm, and
then sent device end when it was ready for another command.

The bus+tag were sort of publish/subscribe.  The cables were daisy chained
from the channel to the controllers, and each controller had a fixed set
of I/O addresses it responded to.  So the subscriptions were set by jumpers.

> Maybe that was QTAM.

QTAM was several levels up, trying to abstract away the differences among
terminals that used differing data rates, character sets, and I/O interfaces.

https://en.wikipedia.org/wiki/Queued_Telecommunications_Access_Method

R's,
John