This reminds me: I remember that back when Amdahl was trying to get their Unix system 
working on the 580 dual processor they had problems with the MVCL.  Actually I think 
this is why I knew that MVCL was a problem in multiprocessor systems.

It may be that Amdahl didn't implement MVCL properly.  There were some variances 
between IBM hardware and Amdahl hardware.  The UTS group also had problems with the 
implementation of channels on the 580 and the asynchronous terminal controller (not 
the 3270 controller).  I don't remember much detail, but it had to do with when 
interrupts occurred.  When testing UTS under VM it worked, but when running native it 
failed.  Sometimes VM can mask these kinds of problems.

-----Original Message-----
From: Phil Payne [mailto:[EMAIL PROTECTED]
Sent: Tuesday, December 16, 2003 5:15 AM
To: [EMAIL PROTECTED]
Subject: Re: Technical Specs


> I wonder what the significance of being interruptible is if not garbling
results.

MVCL came along with System/370 - at that time uniprocessors were more common and a 
lot of IBM
hardware had strict timing considerations - 1419 interrupts HAD to be handled within 
so many
milliseconds or you "lost" the document - sometimes literally.

So an instruction with a potentially long duration HAD to be interruptible for I/O 
interrupts
to be handled in time.  I'm not sure what the current situation actually is - the
time-critical peripherals have gone and most systems are multi-CPU with the SAP 
offloading a
lot of I/O processing.  I doubt that MCVLs get interrupted much these days.

> I imagined that MVC was "atomic" in nature meaning that the second processor couldn't
"garble" things until the MVC had run it's course.

I think that's "implementation-dependent", as IBM would say.  Unwise to rely on it.

--
  Phil Payne
  http://www.isham-research.com
  +44 7785 302 803

Reply via email to