Dragon in use

The British impression

A report about Dragon 1 was written after the war by Gil Hayward, who worked with Tommy Flowers at Dollis Hill and was the designer of the Mark 2 Tunny machine. He was less than enthusiastic about it!

"Sam's [Sgt. T. L. Collins - Sam Scram was the nickname given to him] `baby`, as he called it, was new to us ... Broadhurst [Sid Broadhurst, head of the Tunny design team] came along for the switch-on, interested to see how the American cross-point relays, which were unfamiliar to us, would perform. Sam threw the main switch, causing the most deafening clatter and grinding noise we had ever heard coming from an assembly of relays. It sounded like a monster munching its way through sheets of tin plate. The load on the power supply was so great that the room lights dimmed as several hundred relays pulled in at once, and then brightened again on their release. Sam seemed to think this was normal."

"Dragon, as we dubbed it, crashed and clanged on for about three weeks. Then I came in one morning to find a disconsolate Sam, who figured his baby was very sick. The thing spluttered on for a short time longer and then fell silent - Sam's Dragon was dead. Sid Broadhurst came in to examine the wreckage. The relays that the designers had used were intended for telephone exchanges, where they would operate once per telephone call. In Dragon they operated about once per second, continuously. Their contacts had cut into each other, to the point where they no longer met. Sid took me on one side and said quietly, in that dry way of his, 'I think we could make one of those, don't you?'. He went back to DH and in the remarkably short time of about three weeks had produced his own version of Dragon, which we installed in the same room as Sam's defunct baby."
Sam was late in arriving the morning when our Dragon was switched on. He came in and said in his usual breezy way, 'Hi, Cappy, when are you gonna run this thing?' I told him that it was already running. For once Sam was speechless. Except for a very faint pitter-patter from the relays, our Dragon was practically silent. Sam dismantled his machine, and it was taken away. 1

However, details from other reports, specifically the weekly Dragon and liaison reports, seem to show that the Dragon did not break after three weeks, but was still running up to ten weeks later so it's possible that Gil Hayward's account is somewhat embellished. It does appear likely that there were reliability issues with the Dragon 1 and that the Dragon 2, which actually took around eleven weeks to build rather than three, was definately the more dependable of the two machines.

The American view

An alternative version of the above report can be found in the NR 4628 Special Fish Report (Box 1417) by Albert W. Small, an American cryptanalyst in the US Signal Corps who was seconded to Bletchley Park to report back on the work breaking the Lorenz cipher.

[Dragon] is indeed proving itself to be quite worth while. The British are not using it to break psi's; and they are not using it to set psi's on days of high motor dottage, since this can be done quickly by hand.
But when they want to set psi's on days when motor dots are low, Dragon gets answers in a truly astounding fashion. For this reason it is kept busy on low-motor-dottage traffic. It solves an average of 4 or 5 transmissions a day which might not have been set otherwise.

It is reported that Dragon managed to break 31 de-chis on low motor dottage by the end of October, the month it arrived. Unfortunately for Dragon though, it appears that the increasing complexity of the Colossus machines along with the speedy work of the codebreakers in the Testery were taking much of Dragon's most profitable material.

Photographs courtesy of National Cryptologic Museum/National Security Agency.

Virtual Dragon FAQs

Q: Did Dragon work?

From working on Virtual Dragon, and based on the few actual Lorenz message that are available, it appears that the Dragon was able to do it's job quite well, but obviously, it depends on someone selecting the right crib to be found in the message and the assumption that all contractions are legitimate ones.

Q: Was it successful

It is reported that a number of the codebreakers working in the Testery, for example people like Captain Jerry Roberts and Peter Edgerly, were fantastically fast at spotting the tell-tale signs of a break into a de-chi and could easily outpace the Dragon on a number of messages, getting a break within a few minutes. Dragon did get to succesfully work on a number of messages where the codebreakers had failed to find a break but it's delivery probably came too late in the war to be of huge consequence overall.

Q: How fast did it run?

The speed of the machine was likely to be based on the standard speed of the TeleType input rather than the speed of the relays themselves, so unless they changed the TeleType to run faster than normal, we are looking at a speed of 50 baud which is around 6-7 characters per second.

Q: Why use Dragon, why not find the PSI wheel starts on Colossus?

Colossus is quite capable of finding the PSI start positions, but to do this takes quite a while as it is looking for a statistical count across the whole message so has to search the whole tape multiple times to find the result. Say for example we're searching through a message that is 5000 characters long, to find the best result for the first two Psi wheels means counting through the whole 5000 characters for each of the settings on both wheels. The first Psi wheel has 43 settings and the second Psi wheel has 47 settings meaning Colossus has to search the tape 2,021 times (43x47) and while Colossus can read the tape at an incredible 5000 characters per second, this still means it will take about 30 minutes to check them all! This is just for the first two Psi wheels, it requires at least a further 2 runs to find the other Psi wheel settings. (The Colossus Mark 2 could run, if configured, to run 5 characters simultaneously though, which drastically reduces this time to under 10 minutes.)

Now compare this to Dragon - it only runs at 6-7 characters per second, but as it's just doing a simple check on each row in parallel, it can check for one crib through the whole 5000 characters in about 12 minutes and didn't necessarily have to check the whole tape - it could be stopped sooner if it finds a matching possibility which could then be passed straight onto the Testery. This means that using Dragon for the Psi wheels could then release one of the Colossus machines to begin a break on the Chi wheels on another new message.

The end of the Dragon

At the end of the war, GCCS asked SSA if they would like them to dismantle the Dragon for scrap as it was no longer working. The reply was "DRAGON'S return undismantled desired." It was returned to the SSA at Arlington Hall by 15 September 1945.

What remains of Dragon is currently to be found in storage at the National Cryptologic Museum (NCM), Ft. Meade, Maryland. It is, as far as is known, probably the only surviving piece of World War II special purpose Rapid Analytic Machinery left in the world.

Photographs courtesy of National Cryptologic Museum/National Security Agency.


1 Colossus - The Secrets of Bletchley Park's Codebreaking Computers. Chapter 23 - The British Tunny Machine.