Talk:ILLIAC IV

Cray I faster or slower than ILLIAC IV?

The article on Cray Inc. states that "The Cray-1 was a major success when it was released, faster than all computers at the time except for the ILLIAC IV." But this article on ILLIAC IV states that "(ILLIAC IV) was finally ready for operation in 1976, after a decade of development that was now massively late, massively over budget, and outperformed by existing commercial machines like the Cray-1." These two articles seem to contradict each other. —The preceding unsigned comment was added by 71.50.113.68 (talk • contribs) 21:18, 25 April 2006 (UTC)

Not only that, but the ILLIAC article contradicts itself. It says at the top of the page that the Cray-1 was faster, but says at the bottom of the page that the Illiac began operation in 1976, the same year the Cray-1 was released with roughly the same performance. --Blainster 08:25, 26 April 2006 (UTC)

I never had a chance to use the ILLIAC IV, but I know many people who did and had an office in the building which housed it (at the moment I am across the street). So the issue boils down to what many people call an "apples-to-apples" comparison. If you are an armchair supercomputer watcher as most are, you only look at clock speed. If that were true then any latter (more recent) machine will generaly be faster. I never directly used a Cray-1 or a Cray-1S either, but I did use the X-MP/2 (upgraded to an X-MP/4) which succeeded the 1S which replaced the IV.

The ILLIAC had a slower individual processor element (PE) or control unit (CU) than the Cray-1. The serious computational philosophy question is whether one can add parallel work in sum. Software engineers like Fred Brooks say no. Managers of budgets today have to say yes (is 64 slower elements comparable to a machine 64 times faster than an individual element? No simple answer).

Additionally, speed ignores memory: both RAM and disk in this case. The ILLIAC had woefully little memory. It was only usable with the fixed head disks holding the main part of the problem and the processing unit memories acting as cache and the programmer acting as memory manager. Serious (real, not toy) supercomputing problems take up all memory (only toy problems sat in processor memory).

To this day the ILLIAC was faster in I/O, with at best a weak comparison to Connection Machine Data Vaults (non fixed head disks, much slower), than any other machine since. Apparently the much touted optical memory as tertiary store never worked. I have seen some of the optical strips (it was not disk).

--enm 1 jun 2006

"destined to be the last" really need to be edited out, as even the page for ILLIAC notes the ILLIAC 6.

--enm 1 jun 2006

I'm not sure I see the point in your comment? I assure you as the primary author of the article, I certainly don't compare machines based on clock speed. I thought the article was fairly fair on this. If the article reads this way, and it doesn't seem to, please edit away! Maury 04:24, 28 November 2006 (UTC)

No, I am not the author of the article. Every one but the knowledgeable have contributed to this article (seriously). It appears mostly a rehash of the literature and second hand knowledge that people read or heard. Comparison is done on the basis of real codes and balanced systems. People remember CPUs and their clock cycles, they rarely remember the other parts of systems like storage. Too much in this article needs a rewrite for an old system which I never used which I don't have time for.... --enm 143.232.210.150 (talk) 16:07, 16 September 2008 (UTC)

This last comment is correct. When I recently discovered this page I found that it was almost entirely based on what can only be called rumor or a complete fiction, not even second hand information. Not only was the history wrong, but so was the computer science. As a member of the Illiac IV group at Illinois, worked on the machine in Paoli and was involved in the campus demonstrations (my office was fire-bombed, luckily it didn't go off), I have now updated it (and I conferred with other members of the project in doing it). More of course could be said, but what is there now is accurate. Jeanjour (talk) 15:04, 21 December 2012 (UTC) John Day

1 Tbit optical storage device?

I flagged the mention of a 1 Tbit optical storage device as "citation needed". That's roughly 3 orders of magnitude more storage than state of the art magnetic disk drives in 1976, i.e., it's hard to believe that the claimed capacity is anywhere near accurate. Paul Koning (talk) 18:29, 14 October 2013 (UTC)

I have added a reference to the description of the "laser memory", the paper can be found at a New Zealand university website (unable to include the URL due to some blockage). Quoting the paper, it has the following description:

1) Laser memory: The B6500 supervises a 10^12-bit write-once read-only

laser memory developed by the Precision Instrument Company. The beam

from an argon laser records binary data by burning microscopic holes

in a thin film of metal coated on a strip of polyester sheet, which is

carried by a rotating drum. Each data strip can store some 2.9 billion

bits. A "strip file" provides storage for 400 data strips containing

more than a trillion bits. The time to locate data stored on any one

of the 400 strips is 5 s. Within the same strip data can be located in

200 ms. The read and record rate is four million bits per second on

each of two channels. A projected use of this memory will allow the

user to "dump" large quantities of programs and data into this storage

medium for leisurely review at a later time; hard copy output can

optionally be made from files within the laser memory.

--Nigwil (talk) 09:29, 20 October 2013 (UTC)

What was the contribution of the Illiac IV?

This video The Illiac-IV lecture - Bay Area Computer History Perspectives Series provides a long list of the problems that the Illiac IV solved and how the results were still used beyond the life of the machine. — Preceding unsigned comment added by Nigwil (talk • contribs) 09:25, 20 October 2013 (UTC)

Location, style, polish, WP:NOR

I changed "outside San Francisco" in the lead to "Moffett Airfield in Mountain View, California". Lots of things (luckily) are "outside San Francisco", including the Taj Mahal.

I added the detail of ILLIAC IV having been housed in B. N-233 at Ames and included a reference for this. I made a couple of stylistic changes.

I'm far from the most doctrinaire editor, but this page's references are shockingly few. It seems to have been written from editors' personal recollections. What a surprise! This is a problem with very many articles on the history of computing and computing machinery, where corroborating documents are often thin on the ground but where some grizzled principals intent on preserving history haven't yet snuffed it. I have no doubt that many of the article's assertions are correct but surely more of them can be connected with some reasonable reference than currently appear.

Many other puzzling and/or clumsy bits appear, e.g.,

“...the best they could muster was 250 MFLOPS, with peaks of 150.”

“The machine was never delivered to Illinois, arriving in 1972.”

“Rumor has it that...”

“However keeping with the ploughing analogy consider what you would want behind your tractor would you want...”

and so on. Run-on sentences are one thing but "[highest] was 250 with peaks of 150"?

Maybe I'll rewrite for grammar and style, and just add a lot of "cites needed"? Rt3368 (talk) 14:51, 25 September 2014 (UTC)

“The machine was never delivered to Illinois, arriving in 1972.” I've adjusted the prose around that item, but kept the 1972 delivery date for now; note that our article intro gives a 1971 delivery date.

Of course, as a one-of-a-kind research machine, the process of installing the many components, assembling the SIMD configuration, and performing incremental testing along the way undoubtedly took some time, so there is no single date (or even single year) we can quote; we may need to give a range, and/or list any milestones which we can find sources for.

 Unician ∇  20:13, 26 September 2014 (UTC)

Orthogonality

The wording at the beginning of the "Background" section:

... adding as many instructions as possible to the machine's CPU, a concept known as "orthogonality" ...

is incorrect, cf. its link target, Orthogonality#Computer_science. Quite to the contrary, an "orthogonal" instruction set lacks redundancy, i.e. is as small as possible for the given purpose. Would some computer scientist (I am not) kindly improve the wording? Thanks, --HReuter (talk) 01:37, 13 December 2014 (UTC)

External links modified

Hello fellow Wikipedians,

I have just added archive links to one external link on ILLIAC IV. Please take a moment to review my edit. If necessary, add {{cbignore}} after the link to keep me from modifying it. Alternatively, you can add {{nobots|deny=InternetArchiveBot}} to keep me off the page altogether. I made the following changes:

• Added archive https://web.archive.org/20070610002849/http://images.library.uiuc.edu/projects/gcm/ccm/Detail.asp?Image__imageID=133&decade=All to http://images.library.uiuc.edu/projects/gcm/ccm/Detail.asp?Image__imageID=133&decade=All

When you have finished reviewing my changes, please set the checked parameter below to true to let others know.

This message was posted before February 2018. After February 2018, "External links modified" talk page sections are no longer generated or monitored by InternetArchiveBot. No special action is required regarding these talk page notices, other than regular verification using the archive tool instructions below. Editors have permission to delete these "External links modified" talk page sections if they want to de-clutter talk pages, but see the RfC before doing mass systematic removals. This message is updated dynamically through the template {{source check}} (last update: 18 January 2022).

• If you have discovered URLs which were erroneously considered dead by the bot, you can report them with this tool.
• If you found an error with any archives or the URLs themselves, you can fix them with this tool.

Cheers. —^{cyberbot II}_{Talk to my owner:Online} 05:13, 17 October 2015 (UTC)

5 level code

As mentioned in the ORDVAC article and quoted here by Charlie Jones (discussion platform entry by Charles Richmond) on the web some (all?) of these computers used a 5 level code for encoding hex digits to 0-9 and "KSNJFL" (for todays A-F) to a punch tape. (The letter sequence was described there using memory hooks like "king size numbers just for laughs" or "kind souls never josh fat ladies".)

Here is the 5-level code used by Illiac, transcribed from THE ILLIAC
MINIATURE MANUAL, by John Halton, Digital Computer Laboratory File 260,
University of Illinois, Urbana, 1958, page 3.  I have preserved the
layout as much as is possible using ASCII:


                             THE TAPE CODE
                             -------------
           | Characters | n for 92             | Characters | n for 92
Tape Holes | F/S | L/S  |  Orders   Tape Holes | F/S |  L/S |  Orders
----------------------------------  ----------------------------------

   |  o   |    0     P         2F     |O o   |    Delay  Delay      3F
   |  o  O|    1     Q        66F     |O o  O|    $(Tab) D         67F
   |  o O |    2     W       130F     |O o O |    CR/LF  CR/LF    131F
   |  o OO|    3     E       194F     |O o OO|    (      B        195F
   |  oO  |    4     R       258F     |O oO  |L/S=Letter-Shift    259F
   |  oO O|    5     T       322F     |O oO O|    ,      V        323F
   |  oOO |    6     Y       386F     |O oOO |    )      A        387F
   |  oOOO|    7     U       450F     |O oOOO|    /      X        451F
   | Oo   |    8     I       514F     |OOo   |    Delay  Delay    515F
   | Oo  O|    9     O       578F     |OOo  O|    =      G        579F
   | Oo O |    +     K       642F     |OOo O |    .      M        643F
   | Oo OO|    -     S       706F     |OOo OO|F/S=Number-Shift    707F
   | OoO  |    N     N       770F     |OOoO  |    '      H        771F
   | OoO O|    J     J       834F     |OOoO O|    :      C        835F
   | OoOO |    F     F       898F     |OOoOO |    x      Z        899F
   | OoOOO|    L     L       962F     |OOoOOO|    Space  Space    963F

To me those code looks like being then defined by some comfortable wiring method for the keyboard - but this is just a personal guess. Sorry, but i have no good clue right now what the last column is meant to mean - not even after having a quick look into THE ILLIAC MINIATURE MANUAL by John Halton from 1958 (8 pages). (I am storing the wholte table and header words here for a better long term reference.) --Alexander.stohr (talk) 16:39, 11 October 2016 (UTC)

• This additional source just came to my attention: MODIFICATION OF TELETYPE EQUIPMENT FOR USE WITH THE ILLIAC by R. E. Miller, March 11, 1954. This makes it very clear that the Illiac character set was not the native character set of the Teletype they used, but rather, they had to significantly modify stock Teletypes to work with this character set. The rationale for this eludes me. Douglas W. Jones (talk) 15:35, 20 July 2017 (UTC)

These apply to the ILLIAC I (and II?), not the ILLIAC IV of this article. Maury Markowitz (talk) 20:59, 5 December 2017 (UTC)

GA Review

This review is transcluded from Talk:ILLIAC IV/GA1. The edit link for this section can be used to add comments to the review.

Reviewer: David Eppstein (talk · contribs) 23:18, 27 January 2018 (UTC)

On the whole the prose is of good quality (criterion 1a), but it relies heavily on acronyms some of which are defined well after they are used (IAS, ILLIAC, ARPA) or are never defined (PC board). I think it would be better to avoid this style and spell out technical terms like "control unit" wherever they appear.

The lead section should consist of a summary of the material from other sections, and when doing so it does not need citations (MOS:LEAD, criterion 1b). But instead the claims in the first paragraph (that it was one of the earliest massively parallel computers, that the original design had 256 FPUs and a single CPU, that it could process large array data, and that its instruction set made it SIMD) are unsourced, do not appear to be summaries of anything later in the article, and in some cases contradict the article (which says that there were 256 units that could be partitioned into multiple FPUs each, and that theere were four control units). Similarly, the claim in the second paragraph that Slotnick's original idea for this specific machine was in 1952 is not a summary and is unsourced. And the end of the second paragraph ("instead of 1024") contradicts the claim in the first paragraph that the original plan was for 256 processors. The claim in the third paragraph about a new facility is sourced, but not a summary. And again, many claims in the fourth paragraph are neither sourced nor summaries.

The references are consistently formatted (criterion 2a) and appear reliable (criterion 2b).

It would have been helpful to have a link for the Falk reference (the actual title appears to be "What went wrong V: Reaching for a gigaflop: The fate of the famed Illiac IV was shaped by both research brilliance and real-world disasters" and it can be found at https://doi.org/10.1109/MSPEC.1976.6367550). The Slotnick's link is ok, but https://doi.org/10.1109/MAHC.1982.10003 is I think more permanent. And similarly the Barnes link is ok (and has the advantage of not needing subscription access) but the official link would be https://doi.org/10.1109/TC.1968.229158

Detailed checking found the following additional issues with sourcing (criteria 2c and 2d) beyond the unsourced claims in the intro:

• "It is also credited with being the first supercomputer to use solid-state memory", "Generally considered a failure": not found in Falk p.65 (footnote 1). The closest I can find to the "failure" claim is on p.68, and credited to "one of the men" on the project rather than as a general opinion.
• "The IAS machine was...": the MacKenzie source is good for the claims that it read words one at a time and operated on them in bit-parallel fashion, but not for "fairly conventional" (MacKenzie suggests instead that it was a trend-setter rather than a follower), nor the 40-bit word size, nor the comparison with present-day architectures, nor the 80-track drum size, nor the 1024-word main memory, nor the comparison of scaling behavior between word-serial bit-parallel and word-parallel bit-serial. In conclusion, much of the "Origins" section is falsely sourced, with footnotes that have nothing to do with the claims they appear to source.
• "left the IAS in February 1954 to return to school for his PhD" is I think too closely paraphrased from the source "left Princeton in February 1954 to return to school for a Ph.D."
• "a college, John Cocke": he's an institution of higher learning? (The correct word would have been "colleague".)
• "ended up at IBM in 1958": the source says that this is when Slotnick started thinking about parallel computers again, not when he joined IBM.
• "After a short time at IBM and then another at Aeronca Aircraft, Slotnick ended up at Westinghouse's Air Arm division, which worked on radar and similar systems. Under a contract from the US Air Force's RADC, Slotnik was able to build a team": none of these things are in the only source for the paragraph, MacKenzie.
• "SOLOMON's CU would read instructions from memory, decode them, and then hand them off to the PE's for processing.": not in the source, and potentially misleading. I am not convinced from what I see at the source that the program store for SOLOMON was the same thing as the memory it used for its data, as it would be in a modern computer.
• "the PE Memory module, or PEM": these names are not in the source.
• "The CU could access the entire memory via a dedicated memory bus": not in the source, and very confusing. What is the CU supposed to do with data values stored in the memory? Its function is to control the other units, not to make calculations.
• "Although there are problems, known as embarrassingly parallel, that can be handled by entirely independent units, these problems are generally rare.": off-topic for a discussion of the detailed hardware design of an ILIAC predecessor, and entirely unsourced.
• "A single PE using this design was built in 1963.": the source says "We built experimental PEs" (plural) and gives the broader date range 1962–1963.
• "This evolution toward a smaller number of more complex PEs would continue under ILLIAC IV.": not in the source.
• "no further funding was forthcoming": the source talks about "the shutoff of most of our DOD funds", not quite the same thing.
• "who at that time had been at the forefront of supercomputer purchases": contradicted by the source, which instead says "only slowly became important customers".
• "Westinghouse management considered it too risky...take on the development costs.": the footnote says p.118 of MacKenzie but this material is on p.119.

At this point I got tired of finding problem after problem and stopped looking for more. Almost every single footnote and almost every single footnoted claim up to this point has something wrong with it. The article needs a thorough sentence-by-sentence check of whether what it says can be justified by what's in the sources. Only then will it be ready for GAN.

The article is on a specific topic so criterion 3a isn't really an issue, but there is a lot of component-by-component detail of machines that are not ILIAC IV; is this really necessary (criterion 3b)?

There are no issues with neutrality (criterion 4) or stability (criterion 5).

However, I think File:SISD, MIMD and SIMD computer processor designs.svg may be somewhat problematic (criterion 6). Essentially, it looks like a stealth way of introducing editorializations about the relative merits of different processing architectures into the article, without properly sourcing them. And at the size used for the article, it is completely illegible.

Conclusion: This is a long way from meeting criterion 2 (proper sourcing), and as such does not pass GA. Additionally I have significant concerns about criteria 1b (lead does not summarize article) and 3b (overly detailed about tangential topics). But in other respects the article looks pretty good, so once these issues are handled the article may be ready for another attempt at GA. —David Eppstein (talk) 01:13, 28 January 2018 (UTC)

Hi David, thanks for the notes.

• "The lead section should" - many edits here to clarify. Last para will have to wait until tomorrow.
• "The IAS machine was..." - wrong page number, was really 295. Changed to use Slotnick.
• "It is also credited with being the first supercomputer to use solid-state memory" - this is stated pretty directly in Falk, who states that such systems had been used only experimentally up to that point and no one had used it for the basis of a machine.
• "Generally considered a failure" - I believe this is accurate. Googling "illiac iv failure" turns up any number of useful direct hits in Books. Select any one you like and I'll use it instead.
• "I think too closely paraphrased" - I'm going to leave that.
• "a college, John Cocke" - fixed.
• "ended up at IBM in 1958" - fixed.
• "After a short time at IBM" - fixed.
• "I am not convinced ... program store for SOLOMON was the same thing as the memory it used for its data." - it wasn't and it doesn't say that. It very clearly states that program instructions were in one store and that "Each PE had its own memory for holding operands and results". This is mentioned in pretty much every ref.
• "the PE Memory module" - fixed
• "What is the CU supposed to do with data values" - ...reading and writing them, among other things; the CU was scheduling I/O, running some processing, and many other tasks. This is definitely mentioned in the source, and the new one I added. The new one has a diagram on 376 that shows how all o this worked.
• "Although there are problems" - removed.
• "This evolution toward a smaller" - removed
• "who at that time had" - MacKenzie describes the lab's involvement in computing as "epoch making" and has published a number of other papers widely making this statement.
• "the shutoff of most of our DOD funds" - I'm happy with the wording as it is, Westinghouse ended the project at that point and I don't think the wording is misleading. If the reader wants to know the complete details, they can read the (easily readable) reference.
• "only slowly became important customers" - "for massively parallel architectures". The statement in the article is talking about general computer purchases, not parallel machines specifically.
• "Westinghouse management considered" - fixed.

More, more! Maury Markowitz (talk) 14:47, 28 January 2018 (UTC)

• Let me clarify, since this doesn't seem to be reflected in your response. I don't just want you to fix the specific problems I identified. What I want is for you (or anyone else tempted to re-nominate this for GA) to do the same line-by-line checking of claims in the article against claims in the source that would be expected of a GA. You should find most of these problems yourself, rather than hoping that a reviewer would find them for you. The GA review process is supposed to be about recognizing articles that are already high-quality, not about how to revise your article so that it becomes high-quality. —David Eppstein (talk) 00:14, 31 January 2018 (UTC)

I understand, but as that task will take a while, I'm interested in more comments on the prose and organization. The talk page of the article says that the GA is closed, but looking on this page I do not seem to see that indication. In any event, I would like to leave this open to gather additional comments from other reviewers. Maury Markowitz (talk) 22:04, 3 February 2018 (UTC)

Assembly plant etc.

Everybody knows that the Illiac IV was implemented by Burroughs, and even in the late 1970s Burroughs used that fact in their public relations. I've just added a bit to the talk page of the ARPANET article quoting somebody's recollection of the plant etc. where this was done, since it looks as though it had one of the original ARPANET IMPS, back when there were only about a dozen nodes on the network. MarkMLl (talk) 07:29, 18 October 2019 (UTC)

Successor project?

Just come across http://www.silogic.com/PEPE/PEPE.html which was another Burroughs super, and- curiously- one that I never heard discussed when I worked for them. MarkMLl (talk) 17:33, 10 November 2022 (UTC)