Talk:Colossus computer

Template:CryptographyReader

Turing's Colossus participation

According to History of computing, Alan Turing was a major participant in the development of Colossus

He was a major participant in the code breaking efforts at Bletchley Park and contributed several mathematical insights, both to breaking the Engima code and the Lorenz code. He also built the "bombe" there, an electromechanical code breaking device for the Enigma code. He didn't play a major role in design or construction of Colossus; that was Newman's deal.

See http://www.computer50.org/mark1/turing.html and http://www.turing.org.uk/turing/scrapbook/ww2.html --AxelBoldt

--- Okay, I'll update Alan Turing's article to include that paragraph...

not a general purpose computer

IIRC, describing the Colossus as "general-purpose" and "programmable" is stretching things considerably. --Robert Merkel 14:14, 1 Sep 2003 (UTC)

That's true. The main alterable feature was the set of Boolean functions used in the cross-correlation algorithm, which didn't really amount to a "program". The rest of the functions were fixed. It would be more accurate to say that it was a fixed-program, single-purpose computer with variable coefficients, and certainly not a stored-program computer, which is what most people mean today by "computer". -- Heron 14:57, 22 Jul 2004 (UTC)

Turing complete not important

This prargraph is not convincing:

Whilst Colossus featured (limited) programmability and was the first of the electronic digital machines to do so, it was not a true general purpose computer, not being Turing-complete. It was not then realized that Turing completness was significant, and most of the pioneering modern computing machines (eg, the ABC machine, the Harvard Mark I electro-mechanical relay machine, the Stiebitz Bell Labs relay machines, Konrad Zuse's first two designs, ...) weren't either.

Being Turing complete is not a big deal. Even Conways game of life has been shown to be Turing Complete. It has been shown that the Zuse Z3 was Turing complete. This machine did not differ that much from the Analytical engine or the ABC. It is quite possible that the Colossi were Turing complete. At least theorectically. In that it could have been possible to program the machine to perform an arbitary processing task. The issue of theoretical Turing Completion is over rated. The real issue is practical Turing Completion. In that respect all these machines failed to make the grade and so did the Zuse Z3. It was not until the EDSAC that a machine became practically Turing Complete

I would tend to doubt that the Colossi were Turing-complete, actually. For one, the only test/compare capability they had was whether or not to print out counters at the end of each loop if they were greater than a certain hand-set value. For another, there was no real program, and certainly no conditional branch. They really were incredibly specialized to the job of collecting statistical information used to break a particular coding machine. Very ground-breaking technology, but not ground-breaking architecture. Noel (talk) 02:12, 16 Jan 2005 (UTC)

However one may judge the importance of Turing completeness in this or that early machine, the Colossus were in fact the first programmable digital electronic computers. They were anticipated in each of these dimensions by other designs (programmability by the Analytical Engine; electronic by the ABC machine and some of the analog computers; digital by the Analytical Engine and the ABC machine), but in combination, Newman and crew were first. Their programability (even if limited) made them general purpose machines in a non-trivial sense.

The point about Turing Completeness is important only by way of getting across a distinction amongst various computing designs -- one that is important theoretically if not in practice (for all machines since the EDVAC). Readers may benefit, if marginally, from inclusion of this point. At least they will have links to follow if they're curious. Thus, I judge it to be 'mostly harmless'. I would keep the paragraph. ww 16:31, 5 Apr 2004 (UTC)

The amount of programmability was really limited. About the only thing that was "programmable" (configurable, really) in any general way was which bit conditions (and combinations thereof) would cause counters to increment. The rest was all very specialized to the FISH cipher; e.g. the ability to configure what the bit patterns were on the various wheels which generated the keystream. Hardly "progamming" as it's thought of today. It is really incorrect, IMO, to compare the programmability of Babbage's Analytical engine (which was real programmability, in the modern sense) with the limited configurability (which I think is a much more appropriate word) of the Colossus. Noel (talk) 02:12, 16 Jan 2005 (UTC)

never turned off claim not plausible

Having been one whose experience with electronic hardware goes far enough back to have burned fingers on hot tubes, I find this claim rational, but not very believable. It is certainly true that thermal stresses at power on and power off time reduced tube life, and so avoiding those stresses by avoiding power cycles is a rational approach. However, real equipment must be powered down to make repairs, add new attachments, fiddle around and so on. Since nearly all tube equipment uses high voltages, working on powered up tube gear is most always dangerous. I cannot believe the Colossi were turned on and stayed on till the end of the War. Unless someone can provide some evidence for this, I'll come back and revise after a decent interval. ww 15:21, 7 Jul 2004 (UTC)

Several websites (including [1] (http://news.bbc.co.uk/1/hi/technology/3754887.stm)) repeat the claim. — Matt 15:49, 7 Jul 2004 (UTC)

That may be so, but the ones I can trace all seem to stem from a claim reported from an interview with Sale. I suspect an underlying exagerration for effect taken (wrongly) by a reporter. Tube equipment had regular failures -- it ran hot (induces failures, perhaps most inevitably in economically manufacturable capacitors), it used considerable current (induces failures, largely from heating effects), high voltages (induces failures), it was dangerous to work on if not switched off (heat, high current, and high voltages), and so on. My doubts are not allayed, alas (and alack too). ww 17:49, 9 Jul 2004 (UTC)

Alas for your alack, this claim is also repeated in Randall (1977):

"Flowers' belief, which turned out to be fully justified, was that given the appropriate design practises, switching circuit networks involving even a large number of valves could be made to work reliably, if the equipment could be left on permanently." (emphasis mine)

The same claim is repeated in Johnson, The Secret War:

"There was some opposition to the proposal. Valves had a reputation for unreliability, but Flowers pointed out that from Post Office and BBC experience valves that were never switched off were very reliable." (emphasis mine).

His source was an interview with Flowers himself! So I'd say that this one is pretty solid. Noel (talk) 04:31, 17 Dec 2004 (UTC)

Have just noticed Flowers comments. I still suspect some skew in the meanings here. Flowers' observation is quite correct. But note that he is not reported to have said 'and so we never turned the Colossii off'. I still think that this is engineering hyperbole with regard to the claim about perpetual power on for the machines. Great resistance to turning them off, fix and fiddle as much as possible next to really hot and high voltage stuff, and so on, but... Since Flowers has passed on (and I suppose so have all the others who would know) I think we'll have to remain in ingnorance of the exact fact. But I still doubt it. Those things get hot, and if you look at the photos (and Sales' site for more detail) this is not a machine that is laid out for absolute minimum of little tiny places that are hard to reach. It looks as though they quite sensibly used some sort of ready made rack system and took what they got for serviceablity access. Rarely turned off (and great relutance to turn off, and rules against routine turn off for any reason) seems to me to be about right. ww 19:29, 28 Feb 2005 (UTC)

Since I wrote the above, I have gotten a bunch more Colossus material, including the 3 articles (by Flowers, Coombs, and Chandler, respectively) in Annals of History of Computing, and I can now solidly back up the claim that they were in fact not turned off:

"Most of the machines [out of 10 total - JNC], having been installed and handed over, had their valve heaters on literally until the end of the war. There is little doubt that expansion and contraction of valve heaters being switched on and off caused many valve failures in computing equipment built soon after the war." (from Chandler, Installation and Maintainence of Colossus, pp. 262).

This continual uptime was no doubt in part due to the intense operational requirements, but I'm pretty sure that even without that, Flowers' views on attaining reliability would have kept them on. Station X by Michael Smith contains a direct quote from Flowers on this issue:

"I knew that if you never moved them and never switched them off they would go on forever." (pp. 193)

Given the very large number of valves in these things (2,400 in the Mark II's - by way of comparison, the Pilot ACE had only about 800) I'm sure they didn't want to have to look for dead tubes!

There is much more of great interest in these articles (e.g. Chandler's note that the tubes were operated at only 25% of their maximum power dissipation, which reduced heat output and also helped with tube life), but I haven't the energy to type it all in! BTW, I have this recollection that Flowers would not be the best source for operational details of the Colossi - he was the chief designer, but was not at BP. Noel (talk) 03:43, 1 Mar 2005 (UTC)

Colossus priority comments

A question about priority for Colossus in certain respects has been discussed at Talk:Bletchley Park. I include this pointer for its relevance to this article. ww 14:28, 22 Jul 2004 (UTC)

I placed the following comment there: I reproduce it here as it's much more germane here:

A later note. After reviewing Colossus material in great detail (to far more depth than we have in our article), I would not describe Colossus as "programmable", at least in anything like the sense we now think of the word.

For one, there was no program, in the sense of an algorithm, with transfers of control, especially conditional transfers. The commonly-reported line about Colossus having some conditional testing is true only in a very limited sense; at the end of each pass of the tape loop, it could compare the contents of counters with pre-set values (in switches), and would print or not print the totals in the counters if they exceeded the preset values.

All Colossus could do, really, was read paper tape, perform certain plug-selected binary operations on one or more of the bits of each frame read from the tape (usually only 2 channels out of 5 were looked at), and depending on the output of that binary function, increment counters. That's all. It was really incredibly specialized to the job of doing statistical analyses on the key/cipher stream of the Tunny cipher.

The novelty of Colossus lies in the technology, more than the architecture: it was certainly the first large-scale electronic digital device, and thereby had a great influence on post-WWII computer development in the UK (since many of the leading lights in that field, such as Newman and Turing, were intimately associated with Colossus). Noel (talk) 04:11, 1 Mar 2005 (UTC)

T-52

I've moved the following from the article for now:

This is true, but is it on-topic? That is, was Colossus ever used for breaking T-52 traffic? — Matt 20:18, 6 Dec 2004 (UTC)

I have found no reference anywhere that it was used to attack the T-52. However, I haven't found a definite statement of the form "the Colossi were never used to attack Sturgeon", so I suppose it's still theoretically possible. However.... my take on this question would be "very probably not". If you go through the detailed description of the Colossus here (http://www.codesandciphers.org.uk/lorenz/colwalk/colossus.htm) (make sure to click on the "More Text" links on each image. to see the informative detailed text about that part of Colossus), it's incredibly specialized to attacking the SZ-40/42. E.g. a very large part of the machine (and tube count) was dedicated to generating the key-stream from the Chi wheels of the SZ-40/42 - e.g. the sizes of the various wheels were wired into the machine, etc. So to attack the T-52 you'd have had to basically rewire the whole machine. I consider it highly unlikely that they did so, given the pressure on them to break into Tunny traffic. Had they wanted to attack the T-52, they probably would have used the Robinson (or some tweaked variant thereof), because on those machines the key-stream was on a second paper-tape, i.e. it would have been a lot easier to adapt it to attack a different machine. Still, the latest reference I've found in the original documents to breaking into Sturgeon was in the summer of '44, at which point it was all being done by hand. I'll keep an eye out to see if I see any references to attacking it later with any machine help. Noel (talk) 02:28, 16 Jan 2005 (UTC)

Colossal clanger

Sorry, this gloating is a little gratuitous and unseemly, but, quoting from an online Microsoft Encarta article ("Computer"):

Now (and correct me and the Wikipedia article if I'm wrong), but I recall that 1) Alan Turing was never the head of a team attacking Tunny (although he did invent "Turingismus"/"Turingery", one type of attack on Tunny); 2) Turing's involvement in Colossus was very limited, if any (although Turing did play a large role in generating some of the ideas behind early computers); and 3) Colossus was never used to attack Enigma (it was used to help break Tunny traffic).

Of course, it's not as if Wikipedia articles are always error-free, but I couldn't resist the chance to note this clanger! — Matt Crypto 12:54, 21 Feb 2005 (UTC)

Mark I and Mark II

One of our "to-do" items is to describe the difference between Mk I and Mk II. Here are the results of some work I've done on this topic. The General Report on Tunny says (pp. 329) that:

"Colossus 2 possessed from the first, quintuple testing, a generous switch panel (including not-not), a versatile plug-panel, spanning, a double bedstead, and a greatly increased simplicity of operation."

and

"The chief modifications introduced later were the rectangling gadgets, devices to reduce the effect of doubtful cipher letters, and devices to make wheel-breaking easier."

A few words about these:

• "quintuple testing" - This refers to the 5-deep shift register through which past values of the (electronically simulated) rotors were passed, so that each character read from the tape could be tested against 5 characters of the key-stream at once, speeding up runs by a factor of 5. ("Colossus remembers characters of the wheel opposite places on the tape 1,2,3,4, back; in particular characters of Psi' {the Psi wheels after the variable stepping control - JNC} not of Psi . ... Thus when Colossus is examining a particular place on Z {the cipher-text - JNC}, it has available for comparison: (i) on the multiply tested wheel, the present character and the characters 1,2,3,4, back. ... (ii) on Z, and on all other wheels, only the present character." - General Report on Tunny, pp. 345) This feature had associated controls as well - "undesignated keys are used in multiple testing wherein five successive positions are tested at once and the wheel-patterns stepped five notches each time; in such cases these keys must be thrown down and other plugging done on the jack control board to provide for five positions stepping" (Special Fish Report, pp. 111).

• "not-not" - This refers, I believe, the the provision of two levels of negation in the boolean logical network which produced pulse outputs to the counters, to allow "OR" logical testing as well as "AND". The Special Fish Report says (pp. 111) that it was possible to count (i.e. test for) conditions such as "((character EQUALS value1) OR (character EQUALS value2)" by use of the logical expression "NOT ((character NOT EQUALS value1) AND (character NOT EQUALS value2)".

• "a versatile plug-panel"
  • I believe this refers, in part, to the ability to set the Chi wheel patterns at a panel on the front of Colossus. In the Mark I, these patterns had to be set using jumpers on a panel in the back of the machine, which was particularly inconvenient when the Colossus was used for "wheel-breaking"; the Mark II included a much more convenient panel on the front for setting wheel patterns. One of these patterns was hard wired to have a single 1 bit, and the rest all 0, for use in "rectangling" (a particular cryptanalytic technique used in "wheel-breaking"). ("On wheel-breaking Colossi there is the inestimable boon of a panel on the front of the machine, carrying an ordinary Chi trigger and a special pattern Chi trigger: in place of U-shaped pins, easily inserted plugs are used: they are so much easier that they are often used for setting. Each of the 5 Chi wheels has its ordinary and special patterns adjacent and each is controlled by a 3-way switch whose positions are: down - ordinary and special patterns in; normal - all out; up - single cross in the last position of the ordinary pattern." - General Report on Tunny, pp. 335)
  • There were other pluggable operations available as well, some of which were duplicated by switches (e.g. the logical operations, and selection of counters, in the point above) - "Colossus has a jack board for plugs, to be used in controlling it; and also a key board. The key board may be used instead of the plugs for most runs and is more convenient, or in conjunction with the plugs." (Special Fish Report Page, pp. 108). However, without detailed operation manuals for the Mark I and Mark II Colossi (which probably never existed), I doubt it will ever be possible to say for sure.

• "spanning" - "permit the selection of a stretch or span of tape for study, eliminating all before and after the counting" (Special Fish Report Page, pp. 108); i.e. for a given tape, the counters would only be enable for specified stretches of the tape. This was used when parts were known to be garbled, or (even worse) had characters missing (which would have thrown the key-stream and encrypted data out of synch). It was apparently also used for cases in which the stepping of the Psi wheels depended on the data ("introduced originally for P5 limitation" - General Report on Tunny, pp. 329).

• "a double bedstead" - "positions for two tapes, but only one runs at once" (Special Fish Report Page, pp. 108); i.e. two paper tape transports, so that the operator could be mounting a tape on one, while the other was doing a run.

• "a greatly increased simplicity of operation" - I don't know whether there were additional features that were being thought of here, or just the combined effects of all of the above.

• "rectangling gadgets" - These are explained in the General Report on Tunny, on pp. 349-351, but they are so arcane (and deeply linked to the fine details of the cryptanalytic techniques used to break wheels) that I don't really fully understand them. Suffice it to say that they make minor changes in how counts are computed and printed.

• "devices to reduce the effect of doubtful cipher letters" - This refers to the "not-99" feature, described in the General Report on Tunny on pp. 343-344: "used to inhibit counting at doubtful letters of cipher replaced by Z {ciphertext - JNC} = 9. Such 9's rarely occur singly. Genuine 9's usually do occur singly. It is therefore only imposed at a 9 adjacent to another 9".

• "devices to make wheel-breaking easier." Not sure what this is (other than the rectangling stuff); it might refer to the ability to set the Chi wheel patterns from the front of the machine (referred to above). I'll look into this some more in a bit.

Anyway, this is probably way too much detail to put into the article; in particular, some of it is really specific to the arcane details of breaking Fish, and is simply not suitable for the article. Hopefully we can draw on this to see what we do want to put in, though.

Do we just want to describe it as "improvements to increase the speed and ease of use"? Noel (talk) 02:04, 11 Mar 2005 (UTC)

Nice research! I'll try and add some comments when I'm less tired, but I think this is once instance of a general question: we're going to have to work out how much "Colossus arcana" is appropriate for this article, what belongs in spin-off articles, and what can be left in the literature. Hmm...sleep needed...;-) — Matt Crypto 18:17, 11 Mar 2005 (UTC)

1990s reconstruction

"Even in 2004, Tony Sale notes that "Colossus is so fast and parallel that a modern PC programmed to do the same code-breaking task takes as long as Colossus to achieve a result!". " That's hard to believe. Is it actually true (the fact, not the fact that he said it)?

--Bubba73 22:28, 26 May 2005 (UTC)

• in fact, if someone could provide me with test data and a good description of what Colossus would do and how long it would take, I might try to verify that. --Bubba73 02:16, 29 May 2005 (UTC)

• • I have some doubts about that statement about the speed of the Colossus compared to a modern PC. In "The First Computers: History and Architectures", by Rojas and Hashagen, page 363 has a section "the performance of the Colossus". It was processing characters from a paper tape at 5,000 characters per second. Each character was encoded as 5 bits across the tape. For each character it could do up to 100 boolean calculations on each of the 5 channels. This is 2.5 million boolean operations per second.

• • He made the statement in 2004, when I got a 2.4 GHz P4, which was two steps down from the fastest one. It can perform a bitwise operation in 2 clock cycles, or 1.2 billion per second. Futhermore, it can do the same operation on 32 bits at once in the same amount of time. It seems likely that the Colossus was doing he same operation on 5 bits in parallel, since it was doing XOR on 5-bit characters. So even if the Pentium was used to work on 5 bits of its 43 bits at the time, that amounts to 6 billion boolean operations per second, over 1000 times as fast as the Coolossus.

• • The book quoted was from 2000 and mentions the same thing about a Pentium. In 2000 I had a 300 MHz Pentium II, and it wasn't the fastest one at the time. The original Pentium came out in late 1992 or early 1993 and ran at 60 MHz. I believe it could do a boolean operation on 32 bits in 2 clock cycles, or 30 million per second. Even if only 1 bit was worked on at the time (instead of 5 or more in parallel), that is still more than 10 times faster than the Colossus, at the very minimum.

• • Also, the book says that the Colossus had AND and OR gates. But the operation it was primarily doing was XOR. The Pentium has XOR built in, I believe, so it may have taken more than one operation for the Colossus to do a XOR. Bubba73 17:44, 14 Jun 2005 (UTC)