2 Jul
2010
2 Jul
'10
2:08 a.m.
allison <ajp166 at verizon.net> wrote:
On 06/30/2010 11:27 AM, Johnny Billquist wrote:
> > I'll reply to this one last time, and then I'll give up.
(I can't seem to keep out, can I? :-) )
I can't add too much to this regarding what parts
and what DEC
designators applied
but here are memories of the time frame.
The first multiprocessor 11/70 was built with existing hardware and a few
wire wrap and jumper mods. Memory said there were 4 total, three inside DEC
and one at CMU that they hacked together possibly with DEC help.
CMU did multiprocessor PDP-11s before DEC did, I think. However, they
went about it differently than the 11/74 (or whatever you want to call
it). Search for C.MMP and similar stuff on the net for more information
about CMUs multiprocessor PDP-11 projects.
The 11/74 systems were designed and built inhouse, although they might
have talked with CMU to get help, experience and whatnot. Reportedly
more than three systems were built. Rumors have it that they even had
some systems out to external customers for test, but all systems were
returned at the end of the tests (even though there is a persistent
rumor about Ontario Hydro keeping their).
I think I know of/heard of three systems that were in use inside DEC
long after the system was officially cancelled. We had, of course,
CASTOR:: which was the RSX engineering system, and which was up and
running as late as 2002 (2005?) or so. This was a 4-CPU system.
Then we had POLLUX::, which I think was a 2-CPU system. Not sure, but I
think it might have been DECnet engineering who had it. The third I've
heard about is PHEANX:: which might have been POLLUX:: after a move to
field service, and possibly also using bits and pieces from other
places inside DEC.
As far as I know, all of these systems, as well as the ones gone out on
field test, were KB11-CM cpus. So, no CIS option ever made it out of
prototypes, nor any KB11-E.
The boards from the 11/74 systems that were returned were allegedly used
in plain 11/70 machines inside DEC afterwards. They were, after all,
plug compatible with the normal 11/70 systems. The KB11-E boards would
not have been that, though.
It
would evolve
to a design project to make that buildable as marketing felt they could
sell it.
However at the same time VAX/11/780 was real and also the various product
groups were feeling the effects of FCCs new class A and B limits for
RFI/EMI.
That and the high end market had been moving to more addressable memory
for bigger datasets and computationally wider data words as the tasks were
getting bigger. At that time the big calculations that were important were
atomic physics and weather models and both were associated with massive
[by that eras measure] datasets. In many respects the same pressures
repeated
themselves in the 32bit to 64bit evolution [Alpha].
Indeed. But looking at the papers on the 11/74, their aim was more
towards high availability. Thus the total redundancy in the system, as
well as the ability to bring CPUs and memory on- and offline while the
system is running, and even run diagnostics on one CPU while the others
were serving. Even going as far as being able to physically remove
hardware from a running system.
So, high performance and large memory applications were not the target
of the 11/74. In fact, a 4 CPU 11/74 had about three times the
performance of an 11/70, but that was only aggregated performance. A
single task ran no faster on an 11/74 than on an 11/70. Possibly slower.
The question I think DEC asked itself wether there would be more point
in just selling four 11/70 machines to the customer, or one 11/74. And
four 11/70 won.
I know that RFI/EMI became a problem around this time. I think that
originally DEC planned to stop the 11/70 because of this, but since it
was such a popular machine, and no real replacement existed for quite a
while (the VAX was not a good enough replacement for an 11/70 in some
applications, mostly realtime), they were eventually forced to redesign
the 11/70, and that is where the DEC Datasystem 570 came from.
So the 11/70 in the corporate cabinet was ok with regards to RFI/EMI
radiation, while the older style full height (H960?) cabinets are not.
Or at least that is my understanding.
It was my understanding that the 11/70 continued as a
grandfathered
EMI and the new multiple cpu died due to EMI issues (plethora of cables
and multiple racks) and it was a faster number cruncher than VAX-11/780.
The VAX had higher potential as the new reigning super minicomputer. It
wasn't long after that I'd seen a VAX-11/782, 785 and VAXclusters.
The long cables from multiple CPUs to the memory boxes might have been
an problem with RFI/EMI in an 11/74, I don't know. But the 11/74
machines I have seen in pictures have been in the newer corporate
cabinets, which would imply that they were designed to pass the RFI/EMI
requirements.
There were several of the PDP11 flavors that would die
or morph as a
result of manufacturing and serviceability issues.
I bet. :-)
Johnny