5 Nov
2005
5 Nov
'05
3:37 p.m.
"Dwight K. Elvey" <dwight at ca2h0430.amd.com> Wrote:
Hi
My thought is to make a variable word length processor using
a single bit ALU. Of course, I've been thinking in terms of a relay
based machine.
Dwight
Hi Dwight,
Having helped design a variable word length computer (B1000 series machines
(B1955, B1965), I'm of the opinion that it isn't a useful construct.
Burroughs justified the idea by saying they ran "microcode" on the hardware.
It was really a basic machine language just like any other computer. The
only real difference was that all the compilers compiled down to intermediate
languages which were run on interpreters. It didn't yield very good
performance even though the hardware wasn't that slow for it's era.
There are a couple of interesting side effects of the choices they made. They
needed to have a variable word length rotate capability, and also chose to
address memory on bit boundaries. If you had a 13 bit word length, then
memory delivered 13 bit elements to you (wasting alot of bandwidth in the
process since it was organized by 32...)
You also need the ability to pull off the carry bit from any particular point
in the ALU.
If I were the Burroughs architect, I wouldn't have gone to those extremes
knowing what I know now.
Steve Wilson
5 Nov
5 Nov
5:17 p.m.
On 11/5/2005 at 12:37 PM stevew wrote:
"Dwight K. Elvey" <dwight at
ca2h0430.amd.com> Wrote:
>Hi
>My thought is to make a variable word length processor using
>a single bit ALU. Of course, I've been thinking in terms of a relay
>based machine.
You might as well make it decimal, too, Dwight. That way you can use the
8-4 and 8-2 combinations to mark the end of a word... :)
(I guess it's been done!)
Cheers,
Chuck
5:28 p.m.
...
You might as well make it decimal, too, Dwight. That
way you
can use the
8-4 and 8-2 combinations to mark the end of a word... :)
(I guess it's been done!)
Kinda. The IBM 1401, 1440, 1410, but they used an extra bit in each (6 bit)
called a word mark rather than blow off a digit. It was designed to hold
alphabetic info as well as doing decimal math.
Gil
Cheers,
Chuck
5:59 p.m.
Gil Carrick wrote:
...
Umm what about the sign bit?
Bytw this 2000 not 1950 so we can use ascii encoding. :)
You might as well make it decimal, too, Dwight.
That way you
can use the
8-4 and 8-2 combinations to mark the end of a word... :)
(I guess it's been done!)
Kinda. The IBM 1401, 1440, 1410, but they used an extra bit in each (6 bit)
called a word mark rather than blow off a digit. It was designed to hold
alphabetic info as well as doing decimal math.
Gil
6 Nov
6 Nov
11:15 a.m.
...
Kinda. The IBM 1401, 1440, 1410, but they used an
extra bit
in each (6 bit) called a word mark rather than blow off a digit. It was
designed to hold alphabetic info as well as doing decimal math.
That should have said "in each byte (6 bits)..."
Umm what about the sign bit?
Since the decimal digits only took 4 bits and the word mark was a 7th bit
there were two bits to carry the sign. This machine was actually designed to
be an off-line card reader/ punch/ print processor using tapes to offload
the biger 7094 type machines. So it was very card oriented. The other two
bits represented the "zone" punches off the card. I can't remember which
combination(s) was considered a negative number, but it was the same for any
machine reading cards.
The instruction to read a card was "R" and it read into a fixed location (0,
IIRC). The instructions were also variable length, varying from 1 to 8 bytes
long. An instruction could leave off the addresses (indicated by a word mark
on the next byte) and the addresses would be inferred to continue from the
previous instruction. The 1401 machine had 16KB MAX. memory, so saving a 3
byte address was a big deal.
Gil
12:55 p.m.
On 11/6/2005 at 10:15 AM Gil Carrick wrote:
Since the decimal digits only took 4 bits and the word
mark was a 7th bit
there were two bits to carry the sign. This machine was actually designed
to be an off-line card reader/ punch/ print processor using tapes to
offload
the biger 7094 type machines. So it was very card
oriented. The other two
bits represented the "zone" punches off the card. I can't remember which
combination(s) was considered a negative number, but it was the same for
any machine reading cards.
Not sure if spooling was the designed purpose of the 1401, but there was an
official IBM acronym for the 1401-7090 type combination. But there were
plenty of standalone 1401's.
The 11 punch would be the negative sign indicator, changing 1, 2, 3, 4,
etc. to J, K, L, M. On the earlier decimal machines, ABCDEFGHI is followed
immediately by JKLMNOPQ. When EBCIDIC was introduced, the card code
mapping was preserved, leading to annoying gaps in the alphabetic collating
sequence (e.g. A=C1, B=C2, but J=D1, K= D2, with CA-D0 unassigned). IMOHO,
this is one of the strong points of ASCII--alphabetics are contiguous in
the code.
The 12 punch (the next row toward the top of the card) was used for
ABCDEFGHI in combination with 1-9 numeric punches. The 0 punch (the next
row toward the bottom of the card from the 11 row) was used for STUVWXYZ
(0-2, 0-3, etc).
I suspect that card code (12 positions per column) was one of the reasons
behind the proliferation of binary machines with words that were a
multiples of 12 bits in the 1950's and 1960's.
Cheers,
Chuck
4:04 p.m.
...
Not sure if spooling was the designed purpose of the
1401,
but there was an
official IBM acronym for the 1401-7090 type combination.
But there were plenty of standalone 1401's.
SPOOL actually means Simultaneous Peripheral Operations On-Line, so that is
not the correct term to use in this case. That refers to using a disk drive
as a buffer for record oriented I/O devices so that program execution is
separated from the slower I/O devices. The 1401 was an off-line system for
card to tape input & tape to punch/print output.
But you are definitely correct that the 1401 sold way beyond IBM's
expectations. Once a few people got a look at it and saw how easy it was to
program it took off like a rocket. IIRC they sold in the thousands, most
with only card & print. There was no official OS that I ever saw, but
several I/O packages.
...
Gil
4:24 p.m.
On 11/6/2005 at 3:04 PM Gil Carrick wrote:
SPOOL actually means Simultaneous Peripheral Operations
On-Line, so that
is
not the correct term to use in this case. That refers
to using a disk
drive
as a buffer for record oriented I/O devices so that
program execution is
separated from the slower I/O devices. The 1401 was an off-line system for
card to tape input & tape to punch/print output.
Mostly a matter of sneakers and one's concept of time. :) But what I can't
recall was IBM's term for the 1401-7090 setup--there was one, I'm sure.
And you're correct that it was quite prevalent--I used such a setup at
Purdue. IIRC, the 7090 (actually 2 of them) were turned into frontends for
the CDC 6500 that was later acquired. PUFFT, anyone?
Cheers,
Chuck
5:23 p.m.
Not sure if spooling was the designed purpose of the
1401, but there was an
official IBM acronym for the 1401-7090 type combination. But there were
plenty of standalone 1401's.
The 1400 series was specifically designed to clear out a bunch of the
older, hard to maintain, unit record equipment. Use as an effective
sidekick for some of the 7000 series was a happy surprise for IBM.
William Donzelli
aw288 at osfn.org
6:19 p.m.
On 11/6/2005 at 5:23 PM William Donzelli wrote:
The 1400 series was specifically designed to clear out
a bunch of the
older, hard to maintain, unit record equipment. Use as an effective
sidekick for some of the 7000 series was a happy surprise for IBM.
Ironic, in the sense that I knew of several trade schools who were still
teaching how to program a 407 accounting machine and 514 reproducing punch
plugboards long after most 1401's had hit the scrap pile. I think a lot of
places kept the 407's around just to do 80-80 listings.
The 407 was a pretty neat box for a piece of unit record equipment.
Cheers,
Chuck
11:45 p.m.
...
The 1400 series was specifically designed to clear out
a
bunch of the older, hard to maintain, unit record equipment.
Use as an effective sidekick for some of the 7000 series was
a happy surprise for IBM.
This site says otherwise.
http://simh.trailing-edge.com/i1401.html
I am sure I have read it elsewhere as well.
The IBM archive just says it was used for both, but gives no design
objective.
I was on the user side, so I have no special information. It would be
interesting to hear from an insider.
Gil
William Donzelli
aw288 at osfn.org
7 Nov
7 Nov
9:41 a.m.
This site says otherwise.
http://simh.trailing-edge.com/i1401.html
I am sure I have read it elsewhere as well.
The IBM archive just says it was used for both, but gives no design
objective.
Check out the (more or less) official history books on IBM put out by MIT
press.
William Donzelli
aw288 at osfn.org
1:33 p.m.
On 11/6/2005 at 5:23 PM William Donzelli wrote:
The 1400 series was specifically designed to clear out
a bunch of the
older, hard to maintain, unit record equipment. Use as an effective
sidekick for some of the 7000 series was a happy surprise for IBM.
I think the reality here is somewhere between "yes they did" and "no they
didn't".
There was never a 1401 setup was even remotely competitive with a small DP
operation that used unit-record equipment. A quick perusal of the numbers
shows that the lease cost of a 1401 with tape, printer and punch was in no
way competitive with the lease cost for a simple set of unit record
equipment (407, 519, 082). A single 1401 could be much faster and do more
complex operations than the unit record equipment, but the old stuff was
very robust and economical.
However, for the larger DP customers, a 1401, owing to its use of tape
(handling fewer trays of punched cards) and faster I/O (the 1403 printer
was a comparative speed demon to the 407 printer) was a very attractive
alternative for a big roomful of unit record equipment. What is more
significant, it was extremely attractive as a replacement for earlier
business computers, such as the vacuum-tube 650. (The 1401 could simulate
650 code). On the other hand, it meant a lot of retraining--many shops
kept some of the UR equipment, such as the 082 for sorting, since it didn't
necessitate re-punching cards after sorting (there are some operations that
UR equipment does better).
IBM was actively trying to upgrade their customer base from UR to computer
for a long time. I recall that a friend in the late 60's was excited that
his employer, a small Chicago area manufacturer, had just upgraded from UR
equipment to a 360/20 on the advice of an IBM SE. Lots of interesting new
stuff and new tricks to perform. He thought it was wonderful.
Every morning, the DP head would deliver a thick bundle of paper to the
president, summarizing the state of the corporate accounts. Of course, the
president simply filed it without reading--he already knew, through
interaction with the sales department, how things stood.
But the new equipment required taking on extra employees and was a lot more
expensive. After awhile, the push to add more equipment started and the
president began to wonder why, since business hadn't increased all that
much, why more of the corporate overhead was dedicated to DP operations.
IBM was holding a seminar for prospective customers in the area, so the
president (without telling the DP manager) attended. He laid out the
extent of his operation as if he'd had no EDP at al to one of the SE's and
received the reply that some unit record equipment should do the job just
fine. He asked if perhaps a computer may not be a better choice and the
sales guy replied, no, your operation isn't large enough to justify it.
The 360/20 was out of the door two months later. The extra employees
followed shortly thereafter.
The point of this is that even during the 60's, IBM was still actively
recommending and leasing UR equipment.
Cheers,
Chuck
3:56 p.m.
I think the reality here is somewhere between
"yes they did" and "no they
didn't".
Pages 465 thru 479 of "IBMs Early Computers" holds the story. Required
reading.
The roots of the 1401 project, called SPACE, goes back to the RAMAC days,
apparently.
A single 1401 could be much faster and do more
complex operations than the unit record equipment, but the old stuff was
very robust and economical.
Remember, however, that the unit record equipment was starting to show it
age. Being electromechanical in nature, failures went up over time, just
like an old car. Fixing the things was mostly on IBMs dime.
IBM was actively trying to upgrade their customer base
from UR to computer
for a long time. I recall that a friend in the late 60's was excited that
his employer, a small Chicago area manufacturer, had just upgraded from UR
equipment to a 360/20 on the advice of an IBM SE. Lots of interesting new
stuff and new tricks to perform. He thought it was wonderful.
The "fake" S/360 was the second attempt to kill off the old stuff. S/3 was
the third and final stab, and it pretty much worked (although 083s were in
use until the very end).
The point of this is that even during the 60's,
IBM was still actively
recommending and leasing UR equipment.
I would think that was the exception rather than the rule. Certainly they
would have wanted to lease or sell more 1401s than the older systems.
William Donzelli
aw288 at osfn.org
4:49 p.m.
On 11/7/2005 at 3:56 PM William Donzelli wrote:
Remember, however, that the unit record equipment was
starting to show it
age. Being electromechanical in nature, failures went up over time, just
like an old car. Fixing the things was mostly on IBMs dime.
Which is why the customer resistance to change was such a tough nut to
crack--and why IBM met with only mediocre success with the 1401 in
replacing UR equipment.
I don't deny that IBM's plan all along was to replace the UR with the 1401;
just that the reality wasn't achieved for a number of good reasons, cost
being one of them.
The timing of the introduction of the S/3 was perfect--we'd just put a man
on the moon and the old 1940's vintage UR equipment did seem kind of
klunky. Smaller cards, cartridge drive, fast printer, very clean
appearance, it was the right product for the right time.
Cheers,
Chuck
8:08 p.m.
Which is why the customer resistance to change was
such a tough nut to
crack--and why IBM met with only mediocre success with the 1401 in
replacing UR equipment.
That is true. Mainframers, as you know, are still hard to change in their
ways.
According to the book, the 7070 team were the first ones to grab hold of a
1401 as a sidekick, and the 7090 team shortly thereafter. This was while
the 1401 was still in the prototype stage.
The timing of the introduction of the S/3 was
perfect--we'd just put a man
on the moon and the old 1940's vintage UR equipment did seem kind of
klunky. Smaller cards, cartridge drive, fast printer, very clean
appearance, it was the right product for the right time
And with it, a return to the good old days of completely incompatible
lines of machines (S/3 to S/32 to S/34 and on and on).
I think the little cards, while cool, are actually sort of dumb. They
should have killed the cards off at the start.
William Donzelli
aw288 at osfn.org
8:40 p.m.
On 11/7/2005 at 8:08 PM William Donzelli wrote:
And with it, a return to the good old days of
completely incompatible
lines of machines (S/3 to S/32 to S/34 and on and on).
I think the little cards, while cool, are actually sort of dumb. They
should have killed the cards off at the start.
They never did make a big dent in the punched card world, did they?
Darned silly encoding too--BA8421 for the basic character set and the
completely separate C and D rows if you needed 8 bit encoding. I wondered
if the US Treasury would keep up with the times by shrinking our dollar
bill by half (I know that Hollerith's dollar was bigger). It seems that
typical governmental intelligence came into play however, and they got it
backwards--around 1980, they kept the physical size the same, but halved
its value. :)
While the S/3 instruction set bears certain similarities to the S/360, I've
wondered what IBM's logic was for making it so different from the S/360. I
doubt that it was technical--the 360/20 was far more brain-dead than the
S/3 and customers still used them, in spite of the crippled instruction
set.
Cheers,
Chuck
10:06 p.m.
They never did make a big dent in the punched card
world, did they?
And to add to that, later S/3s could be fitted with 80 column card
devices!
While the S/3 instruction set bears certain
similarities to the S/360, I've
wondered what IBM's logic was for making it so different from the S/360. I
doubt that it was technical--the 360/20 was far more brain-dead than the
S/3 and customers still used them, in spite of the crippled instruction
set.
The S/3 line was...special.
I think the bizzare instruction set of the S/3 stems from the the need for
the machine to deal with characters - RPG and such - almost
exclusively. FORTRAN was offered but it must have really sucked.
There are only 30 or so instructions in the whole set, yet a number of
them are for complex string handling (kind of inpressive for a
non-microcode machine to do string inserts, actually). There are no simple
byte level instructions - everything that goes thru the ALUs are strings,
basically. Everything is a string. Thw world is a string to an S/3.
And yes, the S/3 is a real variable word length machine! Yes, you can add
two 569 byte integers with one instruction!
William Donzelli
aw288 at osfn.org
11:10 p.m.
On 11/7/2005 at 10:06 PM William Donzelli wrote:
...There are no simple
byte level instructions - everything that goes thru the ALUs are strings,
basically. Everything is a string. Thw world is a string to an S/3.
Well, yes, there are a couple of byte-level instructions--the immediate
ones (MVI, CLI, SBN, SBF) and the very limited register instructions (you
have to be able to manipulate the index registers somehow!) (L, ST, A, LA).
But yeah, the instruction set is nothing short of bizarre and it'd be
pretty hard to carry out much useful work using only the index registers!
Cheers,
Chuck
12 Nov
12 Nov
5:48 p.m.
The story I hear was that the System/3 was built by a CE in his home using
salvaged parts and adopted by IBM as a product. I have no references to
prove it. If so, then there was no logic to the decision. The 360 was just
way too big a design to strip down as the comments on the model 20 show. The
sys 3 was for really small offices.
IMHO the most bizarre thing was the twinax cabling. That line and some Wang
systems were the only ones I ever saw. Not being a hardware guy I have no
idea if there is any advantage for twinax over coax.
Gil
...
...
While the S/3
instruction set bears certain similarities to
the S/360,
I've wondered what IBM's logic was for
making it so
different from the
S/360. I doubt that it was technical--the 360/20
was far more
brain-dead than the
S/3 and customers still used them, in spite of the crippled
instruction set.
The S/3 line was...special. There are only 30 or so instructions in the whole set,
yet a
number of them are for complex string handling (kind of
inpressive for a non-microcode machine to do string inserts,
actually). There are no simple byte level instructions -
everything that goes thru the ALUs are strings, basically.
Everything is a string. Thw world is a string to an S/3.
And yes, the S/3 is a real variable word length machine! Yes,
you can add two 569 byte integers with one instruction!
William Donzelli
aw288 at osfn.org
5 Nov
5 Nov
6:54 p.m.
On 11/5/2005 at 4:28 PM Gil Carrick wrote:
...
bit)
You might as well make it decimal, too, Dwight.
That way you
can use the
8-4 and 8-2 combinations to mark the end of a word... :)
(I guess it's been done!)
Kinda. The IBM 1401, 1440, 1410, but they used an extra bit in each (6 called a word mark rather than blow off a digit. It was
designed to hold
alphabetic info as well as doing decimal math.
Yeah, I know (I've been there and can even remember a bit of autocoder).
7080 or 1620/1710, which is what I had in mind when I wrote the above.
1620 used 5 bits (the 6th bit was parity and not addressable by program
control, although it was displayed and mentioned in the manuals). 82 was a
record mark--84 was a numeric blank, 842 was a group mark (used for disk
operations, mostly). The 1 bit didn't matter to the hardware for those
characters.
That flag bit on the 1401 and 1620 always seemed like a great waste of
resources. It was handy for the math tables (You don't really want to do
your math by lookup tables, do you?) to denote a carry, but I suspect that
much less than 20 percent of the digits in memory had a word mark set.
So, let's see--you really have 6 undefined characters at your disposal.
82, 821, 84, 841, 842 and 8421. Forget about addressing by the high-order
position (i.e. record addressing) for the time being. Let's see what can
be done with just 4 bits:
A 5 digit number could be encoded as (82)99999 if positive; or (821)99999
if negative, addressed by LSD. You could keep numbers in 10's complement,
instead of sign-magnitude.
That still leaves you with 84, 842, 841, 8421 for other record types and
devious schemes.
As far as ASCII goes, you've got 100 possibilities, which should be enough
for upper and lower case and a healthy dose of punctuation with a few
control characters.
Cheers,
Chuck
6828
days inactive
6835
days old
20 comments
6 participants
participants (6)
-
aw288＠osfn.org -
bfranchuk＠jetnet.ab.ca -
cclist＠sydex.com -
gcarrick＠cse.uta.edu -
gilcarrick＠comcast.net -
stevew＠ka6s.com