28 Nov
2003
28 Nov
'03
6:16 a.m.
On Thu, 2003-11-27 at 19:09, Johnny Billquist wrote:
On Fri, 28 Nov 2003, Warren Toomey wrote:
C.MMP was 12 11/40's and 4 11/20's. Each processor had 4KW of local
memory + the 4KW I/O page. The rest of the memory (1.2MW) was
accessible through a "cross-point" switch (ie it wasn't a common memory
bus...think of it as 16 port memory -- there was no memory contention
unless 2 processors wanted to access the same page (4KW) of memory).
In addition to the cross-point switch there was special IPC
(Inter-Processor Communication) hardware to allow the processors to
interrupt and communicate with each other.
The O/S that was run was Hydra a very radical capability based system
(ie everything was represented as a capability -- files, programs, I/O,
etc). If you didn't have a capability for something you didn't even
know it existed. It was very cool!
Somewhere I still have my "Hydra Songbook" which contains a bunch of
details + kernel calls about Hydra.
There was a predecessor (prototype) that supported either 2 or 4
11/40s. I remember seeing it in the same machine room as C. but don't
remember what it being used for at the time as it was "discarded" from
the C. project.
Hi all,
I stumbled across this reference to a 1975 Masters thesis:
de Brito Meyer. W., and Hawley, J.A.. III. Munix. a multiprocessor version
of UNIX. Master's thesis, Naval Postgraduate School, Monterey, Calif.. 1975.
Description of dual processor Unix.
Can anybody tell me what PDP11 platforms around 1975 had multi-CPU
capability? Also, if anybody has further information about Munix,
please let me know!
Thanks in advance for any help. I've trawled thru the Unix Archive
with no results.
I remember that CMU built a MP system out of 11/40 systems...
Search for C.mmp (if my memory is correct).
I think they built some special hardware for this. And since these
machines don't have a cache, it makes life easier...
--
TTFN - Guy