On Monday, March 24, 2003, at 06:04 PM, der Mouse wrote:
>>> [...two-KA630 speculation...]
>> The question, though, is how useful would it be.  The KA630 was
>> designed to support a multiprocessor system, ***EXCEPT*** that the
>> multiple processors do NOT share main memory.
>
> They do if they want...with some restrictions.
>
> It's not fully symmetric sharing; the processor that owns the memory
> has to explicitly export it to the Qbus, and other processors pay a
> speed penalty (because they access it via Qbus cycles rather than
> through their own memory interconnect).
>
> It's also limited by the address space of the Qbus, which as I recall
> imposes a 22-bit (= 4MB) limit, which limit applies to the total amount
> of memory exported by all processors.  And in practice the limit is
> even less than that because you have to allow space for devices in that
> 4MB.

   This brings up an interesting possibility.  Processes that have 
multiple threads (i.e. that would take advantage of multiple 
processors) could have their virtual addresses mapped by the VM system 
to physical addresses in the "shareable" space.

   But wait...why would this have anything to do with Qbus (and its 
22bit address space) at all?  Wouldn't such 
other-processors-accessing-my-memory transactions happen over the PMI 
bus?

        -Dave

--
Dave McGuire             "I've grown hair again, just
St. Petersburg, FL           for the occasion."       -Doc Shipley