[ Getting into COFF territory, I think ]
On Thu, 30 Jan 2020, Clem Cole wrote:
BTW: Dave story is fun, but I think a tad apocryphal.
He's right that
DEC marketing was not happy about people using it, but it was well
spec'ed if you had CPU schematics. They way they tried to control it
was to license the bus interface chips (made privately by Western
Digital for them IIRC but were not available on the open market). IIRC
if you did not use DEC's chips, you could have issues if you >>similar<<
function chips from National Semi. I remember Ken O'Munhundro giving a
talk at a USENIX (while he was CEO of Able) talking about 'be careful
with foreign UNIBUS implementations.' If I recall it was the analog
characteristics that were tricky with something like the BUS acquisition
for DMA and Memory timing, but I admit I've forgotten the details.
Ah; the chips could explain it. I can't remember where I heard the story,
but it was likely in ";login:" or some place. Hey, if the DEC marketoids
didn't want 3rd-party UNIBUS implementations then why was it published?
I think you are confusing VAX's SBI with UNIBUS.
With the Vax, unlike
PDP-11, the systems did not come with complete schematics for all
boards. So to design for the SBI you had to reverse engineer the CPU
and Memory boards. DEC having successfully won the CalData suit, went
after Systems Industries who was the first to build SBI controllers.
DEC lost, but the truth was that because they had work had been reverse
engineering, SI was close but not 100% right and they had a number of
issues when the boards first hit the street, particularly with UNIX
which did a better job of overlapped I/O than VMS did. At UCB we had a
logic analyzer in one of the 780s at all times, and the phone number of
the SI engineers. We eventually helped them put out a couple ECO's
that make the original boards work in practice much better.
No; it was definitely UNIBUS (I wasn't aware of the SBI at the time).
As for overlapped seeks, when they were implemented in Unix it broke the
RK-11 controller, and DEC pointed the finger at Unix (of course) since
their own gear worked. To cut a long story short, they were forced to use
some fancy diagnostic (DECEX?) which hammered everything at the same time,
and the problem showed up. Turned out that their simpler diagnostics did
not test for overlapped seeks, because they knew that it didn't work; out
same the FE to modify the controller...
BTW: My friend Dave Cane lead the BI at DEC after
finishing up the
VAX/750 project (he had designed the SBI for 780 before that). In
fact, the BI was >>supposed<< to be 'open' like Multibus and VME
chips were supposed to be from the merchant market. But at the last
minute, DEC marketing refused and locked down the specs/stopped shipping
schematics with the new systems destined to use BI. Dave was so pissed,
he left DEC to found Masscomp and design the MC500 (using the
Yet another reason why DEC went under, I guess...