22 Nov
1999
22 Nov
'99
11:39 a.m.
Eric wrote...
I'm not talking about I/O cables that are many
feet long between a card
and
an external device, nor cables under two inches
between just a few boards
in the same backplane.
I'm talking about using over-the-edge cabling for bussing signals between
many cards in the same backplane, and that's what I was referring to the
book about. This was done in the PDP-6, IIRC, and they decided that it
was a maintenance nightmare. If HP somehow made this scheme work reliably
and somehow kept it from being a headache to troubleshoot, I'd be
interested
in hearing how they did it. Obviously HP must not
have thought it was
a particularly great technique, since they generally aren't using it
today.
As a matter of fact, HP did virtually the same thing as you mention above as
a matter of course on the 21MX line. Take a look at the memory extender
units and the I/O extender units. Each was a separate rackmount card cage to
extend either the memory or I/O backplane and they ran at exactly the same
speed as the main bus (ex: memory access time was identical in either main
memory cage or extender cage). I believe the cable length limit was 15
feet. I doubt this was problematic because it was introduced sometime around
the early 70's and followed through to the end of the F series at least
which was approximately the mid 80's. And if the memory or I/O extender
isn't quite the situation you're talking about, then the memory section of
any HP 21MX M,E, or F certainly is. What about that daisy chain cable in the
memory section that connects all the memory cards (and if you're using FCS
you have up to 3 ribbon cables, not just one).
HP 21MX's a DEC me-too copy? That's just ridiculous. I have yet to see a DEC
that split the backplane the way HP did on the MX line. Without that
fundamental change you would be stuck in the "gee I've got to remove some
memory cards to fit in my extra I/O cards" scenario. Not to mention the
optimizations you can do by having a separate bus for memory vs. I/O. Also,
I'm not familiar with mid-time period PDP's (read: Unibus, but hey, we're
talking 8's anyway), but I don't recall seeing one that had several loader
roms that you could choose from via the front panel, and the loaders would
be patched for I/O addresses via the switch register at the press of a
single button. Sure, on an 8 you could put in a hard wired rom board, but
there goes a precious slot and a hefty amount of current draw. The 21MX's
also have an interesting technique of varying the microinstruction cycle
time to speed up certain control section operations. HP made it clear that
users were encouraged to microprogram the system due to the wealth of
documentation and software they provided for doing so. Does the PDP-8
support remote IPL? And these are just a few trivial issues, not fundamental
design ones (of which there are many). Then talk about flexibility, look at
the 21MX K series - basically built to customer spec!
And by the way on a different thread, WCS and UCS was a very common option
on HP 21MX's, and on at least one model it was built in.
I'm not going to pursue the discussion about HP's being "cool" or not
any
further - I feel very strongly about the subject and might lose a few
friends if I took up the argument seriously :). Dont get me wrong - I love
my DECs too. They have their own things I like better than HP. But don't say
the old HP line is just a PDP-8 wannabe/knockoff.
Jay West