Quadibloc
2007-12-31 17:45:51 UTC
I was astounded, as I recently noted, to learn from Gordon Bell's
"Computer Engineering" that the PDP-9 was a microprogrammed machine.
The PDP-4/7/9/15 architecture involved a very simple instruction set.
If anything, the PDP-15 gave the appearance of being an enlarged
version of the PDP-8.
Thus, back in the 1970s, I had wondered - since the PDP-8 was so
cramped that it didn't have even a proper load and store instruction,
but made do with TAD and DCA, why DEC didn't make machines with the
PDP-15 architecture, priced at half again as much as a PDP-8, and make
some effort to migrate users to that, because, clearly, a
straightforwards machine would offer more for the money than one that
was cramped.
But since then, I've learned a few things.
One of them is that a bare-bones PDP-15 cost $15,500, which was close
to half again the price of a PDP-8/e, despite coming in a bigger box.
Another is that DEC had made an effort to bring that architecture to
people at a lower price; the PDP-9/L shows how hard they tried.
But even more importantly, before the PDP-8/e, there were only two
PDP-8 machines that came in a small box; the PDP-8/L and the PDP-8/S.
A PDP-8/I may have had a front panel that looked like it belonged to a
small box machine, but the CPU occupied space behind some blank panel
slots as well. So until the PDP-8/e, the technology for a small-box
PDP-15 didn't exist.
Very shortly after the PDP-8 came out - not *so* shortly after that
DEC didn't have time to introduce the LINC-8 and PDP-8/S as well -
Hewlett-Packard entered the minicomputer fray with the HP-2116. And
then there was the Honeywell 316.
Like the PDP-9, they shared the same basic instruction format as the
PDP-8, except they had the 16-bit word one normally associates with a
typical minicomputer.
So, making a PDP-9/I, a PDP-9 without the extra features of the
PDP-15, but implemented with integrated circuits, would have produced
a "me-too" machine,
Another thing I didn't realize back in the 1970s was that the PDP-4
and its successors had instructions for both two's complenent and
one's complement versions of the basic arithmetic operations. This was
wasteful and outmoded.
The original PDP-11, the PDP-11/20, first shipped in June, 1970, which
was before the PDP-8/e first shipped in March 1971. So it wasn't as if
DEC could have shipped a smaller version of the PDP-9 shortly after
March, 1971, while the radical new PDP-11 delayed DEC's move into more
powerful minicomputers, as opposed to bulky medium-range systems, by
several years.
As is well known, the PDP-11 had an orthogonal two-address
architecture. The program counter and stack pointer were included
along with six general registers in the set of eight registers it
could utilize with a number of addressing modes. Minis from Interdata
and Data General and even HP, with their HP 3000 series, would
eventually offer fancier addressing modes than those of old-style
minis, but the PDP-11 led the way in offering a modern, powerful
architecture to minicomputer users.
And so it was wildly successful and immensely influential.
Thus, DEC made the right decision to come out with the PDP-11, rather
than to attempt to groom the venerable PDP-4/7/9/15 architecture for
the same role. The PDP-15, despite having a bigger number, first
shipped in February 1970, which is why it had a somewhat older-style
front panel, harking back to the PDP-9/L (as I but recently
discovered, thanks to Al Kossow).
Had the PDP-15 come out after the PDP-11/20, it would have been clear
that it was simply there to provide existing users with an upgrade
path, instead of being a new and exciting machine in its own right.
The importance of the PDP-11 in the early development of UNIX, and in
much other innovative work with minicomputers, can hardly be
overstated.
The PDP-1 was a one's complement machine, but the PDP-4 mixed two's
complement and one's complement. The PDP-5 was too small to do
anything like that, and it needed multiple precision arithmetic badly,
and so it was a pure two's complement machine. So was the PDP-6.
The PDP-6 was a 36-bit machine with 16 general registers (like the
Univac 1107!) and included the innovative feature of being able to
handle characters of arbitrary size within its words. Except for the
choice of a 36-bit word length, which eventually became unfashionable,
there wasn't really anything wrong with the PDP-6 architecture.
In fact, there was a lot right with it. It was very successful at
timesharing.
As is well known, researchers at Xerox PARC once built their own clone
of a PDP-10 for their computing research rather than try to use one of
XDS' own Sigma machines. (The Sigma computers were mainframes with
datatypes similar to an IBM 360, but with fixed-length 32-bit
instructions: if the Univac 1107 and the PDP-6 were modernized
imitations of the IBM 704, the Sigma was a retro imitation of the
System/360, allowing it to be cheaper even though it was built from
discrete transistors, not ICs.)
I remember when the first advertisements for the DECsystem-20 came
out. The days of the affordable mainframe seemed to be approaching!
(Well, they were, although I would have to wait until after the 386 SX
came out... and after upgrading from 2 MB to 4 MB of RAM, I remember
installing Yggdrasil Linux in a partition... )
And I remember reading in the trade press about DEC's decision to
abandon that platform cold in favor of the VAX.
Well, it *was* always VAXes, and never DECsystem-20s, that Russian
spies kept trying to smuggle out of the Free World. They may have
known something.
Given the success of the PDP-11 versus DEC's older machines, the
PDP-15 and even the PDP-8, it shouldn't have been surprising that they
would favor the VAX-11/780 and its descendants as representing
modernity, versus the (apparently) dated PDP-6/10/ [DECsystem-] 20
lineage.
However, they did bring out the DECsystem-20 (that system *was* based
on 2900 bit slice technology, IIRC) with great fanfare, and there
really wasn't anything terribly wrong with the PDP-10 architecture.
Without having to rely on hindsight, of course it was wrong to
unexpectedly abandon loyal customers completely. Whatever assistance
DEC might have offered to sites migrating from DECsystem-20 hardware
to VAX hardware at the time of their next upgrade, there would likely
be third-party software which now would have to be replaced at the
very least.
Could DEC have copied IBM, which allowed some models of the System/360
line to emulate the 7094-II? The VAX was microcoded, after all.
While DEC was a much smaller company than IBM, and so couldn't bear
the expense of supporting multiple lines of product indefinitely
(assuming that the multiple lines of product only divided up the slice
of the pie DEC would have without them)... there was a Jupiter project
that had reached an advanced stage of development that was cancelled
at this time as well.
The VAX-11/780 was announced on October 25, 1977.
The DECsystem-20 name was around before then; the KS10 processors, the
ones using AM2900 bit slices and offering the impressive price
reductions, were from 1978. They brought new customers to DEC,
attracted by mainframe power at mini prices.
Had the DECsystem-20 side of the business been losing money badly for
DEC, abandoning it miight have been the only choice. Even then, one
ought to do everything possible to treat one's customers right. If
that wasn't true, then phasing it out gradually and migrating
customers to the VAX would have meant going ahead with Jupiter, aiming
it primarily at existing accounts, and treating the DECsystem-20 side
as a "cash cow".
Why didn't that happen?
I can think of a couple of reasons.
On the day that Jupiter was released, being the newest system from
DEC, it would have been the one with the best price/performance. So,
to keep new accounts from going to Jupiter instead of to VAX systems,
it would have been necessary to make price cuts on the VAX line on
that date - and that would send the wrong message, making it look as
though it was the VAX side that was declining!
If the DECsystem-20 side of the business was small compared to the VAX
side, then it wouldn't have been much of a "cash cow" to, say,
subsidize VAX development.
Internal corporate politics seems like the cause of the hasty
abandomnent of the 20 even from a complete ousider's perspective. The
PDP-11 was the future, the PDP-10 was the past. So it likely was
argued that the DECsystem-20 was diluting the company's focus,
consuming an inordinate share of resources that could be better put to
use on developing the VAX further (was there an internal manpower
shortage within DEC at the time?). The (clearly) specious argument
that selling an old-fashioned 36-bit machine would hurt the company's
image more than abandoning loyal customers may also have been
advanced.
I'm not trying to make excuses for a wrong decision, but while
favoring the VAX can be explained by the desire to be modern and up-to-
date, one doesn't expect an insane decision from a big company. If DEC
needed a lot of additional programmers and engineers in a hurry to
take the VAX where they wanted it to go, the quickest way to obtain
them would be internally. (Even then, a way to finish Jupiter on a
shoestring could have been found, of course.) I expect, therefore,
that it must have seemed rational at the time for some reason.
John Savard
"Computer Engineering" that the PDP-9 was a microprogrammed machine.
The PDP-4/7/9/15 architecture involved a very simple instruction set.
If anything, the PDP-15 gave the appearance of being an enlarged
version of the PDP-8.
Thus, back in the 1970s, I had wondered - since the PDP-8 was so
cramped that it didn't have even a proper load and store instruction,
but made do with TAD and DCA, why DEC didn't make machines with the
PDP-15 architecture, priced at half again as much as a PDP-8, and make
some effort to migrate users to that, because, clearly, a
straightforwards machine would offer more for the money than one that
was cramped.
But since then, I've learned a few things.
One of them is that a bare-bones PDP-15 cost $15,500, which was close
to half again the price of a PDP-8/e, despite coming in a bigger box.
Another is that DEC had made an effort to bring that architecture to
people at a lower price; the PDP-9/L shows how hard they tried.
But even more importantly, before the PDP-8/e, there were only two
PDP-8 machines that came in a small box; the PDP-8/L and the PDP-8/S.
A PDP-8/I may have had a front panel that looked like it belonged to a
small box machine, but the CPU occupied space behind some blank panel
slots as well. So until the PDP-8/e, the technology for a small-box
PDP-15 didn't exist.
Very shortly after the PDP-8 came out - not *so* shortly after that
DEC didn't have time to introduce the LINC-8 and PDP-8/S as well -
Hewlett-Packard entered the minicomputer fray with the HP-2116. And
then there was the Honeywell 316.
Like the PDP-9, they shared the same basic instruction format as the
PDP-8, except they had the 16-bit word one normally associates with a
typical minicomputer.
So, making a PDP-9/I, a PDP-9 without the extra features of the
PDP-15, but implemented with integrated circuits, would have produced
a "me-too" machine,
Another thing I didn't realize back in the 1970s was that the PDP-4
and its successors had instructions for both two's complenent and
one's complement versions of the basic arithmetic operations. This was
wasteful and outmoded.
The original PDP-11, the PDP-11/20, first shipped in June, 1970, which
was before the PDP-8/e first shipped in March 1971. So it wasn't as if
DEC could have shipped a smaller version of the PDP-9 shortly after
March, 1971, while the radical new PDP-11 delayed DEC's move into more
powerful minicomputers, as opposed to bulky medium-range systems, by
several years.
As is well known, the PDP-11 had an orthogonal two-address
architecture. The program counter and stack pointer were included
along with six general registers in the set of eight registers it
could utilize with a number of addressing modes. Minis from Interdata
and Data General and even HP, with their HP 3000 series, would
eventually offer fancier addressing modes than those of old-style
minis, but the PDP-11 led the way in offering a modern, powerful
architecture to minicomputer users.
And so it was wildly successful and immensely influential.
Thus, DEC made the right decision to come out with the PDP-11, rather
than to attempt to groom the venerable PDP-4/7/9/15 architecture for
the same role. The PDP-15, despite having a bigger number, first
shipped in February 1970, which is why it had a somewhat older-style
front panel, harking back to the PDP-9/L (as I but recently
discovered, thanks to Al Kossow).
Had the PDP-15 come out after the PDP-11/20, it would have been clear
that it was simply there to provide existing users with an upgrade
path, instead of being a new and exciting machine in its own right.
The importance of the PDP-11 in the early development of UNIX, and in
much other innovative work with minicomputers, can hardly be
overstated.
The PDP-1 was a one's complement machine, but the PDP-4 mixed two's
complement and one's complement. The PDP-5 was too small to do
anything like that, and it needed multiple precision arithmetic badly,
and so it was a pure two's complement machine. So was the PDP-6.
The PDP-6 was a 36-bit machine with 16 general registers (like the
Univac 1107!) and included the innovative feature of being able to
handle characters of arbitrary size within its words. Except for the
choice of a 36-bit word length, which eventually became unfashionable,
there wasn't really anything wrong with the PDP-6 architecture.
In fact, there was a lot right with it. It was very successful at
timesharing.
As is well known, researchers at Xerox PARC once built their own clone
of a PDP-10 for their computing research rather than try to use one of
XDS' own Sigma machines. (The Sigma computers were mainframes with
datatypes similar to an IBM 360, but with fixed-length 32-bit
instructions: if the Univac 1107 and the PDP-6 were modernized
imitations of the IBM 704, the Sigma was a retro imitation of the
System/360, allowing it to be cheaper even though it was built from
discrete transistors, not ICs.)
I remember when the first advertisements for the DECsystem-20 came
out. The days of the affordable mainframe seemed to be approaching!
(Well, they were, although I would have to wait until after the 386 SX
came out... and after upgrading from 2 MB to 4 MB of RAM, I remember
installing Yggdrasil Linux in a partition... )
And I remember reading in the trade press about DEC's decision to
abandon that platform cold in favor of the VAX.
Well, it *was* always VAXes, and never DECsystem-20s, that Russian
spies kept trying to smuggle out of the Free World. They may have
known something.
Given the success of the PDP-11 versus DEC's older machines, the
PDP-15 and even the PDP-8, it shouldn't have been surprising that they
would favor the VAX-11/780 and its descendants as representing
modernity, versus the (apparently) dated PDP-6/10/ [DECsystem-] 20
lineage.
However, they did bring out the DECsystem-20 (that system *was* based
on 2900 bit slice technology, IIRC) with great fanfare, and there
really wasn't anything terribly wrong with the PDP-10 architecture.
Without having to rely on hindsight, of course it was wrong to
unexpectedly abandon loyal customers completely. Whatever assistance
DEC might have offered to sites migrating from DECsystem-20 hardware
to VAX hardware at the time of their next upgrade, there would likely
be third-party software which now would have to be replaced at the
very least.
Could DEC have copied IBM, which allowed some models of the System/360
line to emulate the 7094-II? The VAX was microcoded, after all.
While DEC was a much smaller company than IBM, and so couldn't bear
the expense of supporting multiple lines of product indefinitely
(assuming that the multiple lines of product only divided up the slice
of the pie DEC would have without them)... there was a Jupiter project
that had reached an advanced stage of development that was cancelled
at this time as well.
The VAX-11/780 was announced on October 25, 1977.
The DECsystem-20 name was around before then; the KS10 processors, the
ones using AM2900 bit slices and offering the impressive price
reductions, were from 1978. They brought new customers to DEC,
attracted by mainframe power at mini prices.
Had the DECsystem-20 side of the business been losing money badly for
DEC, abandoning it miight have been the only choice. Even then, one
ought to do everything possible to treat one's customers right. If
that wasn't true, then phasing it out gradually and migrating
customers to the VAX would have meant going ahead with Jupiter, aiming
it primarily at existing accounts, and treating the DECsystem-20 side
as a "cash cow".
Why didn't that happen?
I can think of a couple of reasons.
On the day that Jupiter was released, being the newest system from
DEC, it would have been the one with the best price/performance. So,
to keep new accounts from going to Jupiter instead of to VAX systems,
it would have been necessary to make price cuts on the VAX line on
that date - and that would send the wrong message, making it look as
though it was the VAX side that was declining!
If the DECsystem-20 side of the business was small compared to the VAX
side, then it wouldn't have been much of a "cash cow" to, say,
subsidize VAX development.
Internal corporate politics seems like the cause of the hasty
abandomnent of the 20 even from a complete ousider's perspective. The
PDP-11 was the future, the PDP-10 was the past. So it likely was
argued that the DECsystem-20 was diluting the company's focus,
consuming an inordinate share of resources that could be better put to
use on developing the VAX further (was there an internal manpower
shortage within DEC at the time?). The (clearly) specious argument
that selling an old-fashioned 36-bit machine would hurt the company's
image more than abandoning loyal customers may also have been
advanced.
I'm not trying to make excuses for a wrong decision, but while
favoring the VAX can be explained by the desire to be modern and up-to-
date, one doesn't expect an insane decision from a big company. If DEC
needed a lot of additional programmers and engineers in a hurry to
take the VAX where they wanted it to go, the quickest way to obtain
them would be internally. (Even then, a way to finish Jupiter on a
shoestring could have been found, of course.) I expect, therefore,
that it must have seemed rational at the time for some reason.
John Savard