Appendix B. Cryptanalytic Machines in NSA (May 1953)
The following transcription describes Bombes, Scritchers, and ERA's GOLDBERG. (Wheatley, LeRoy H. "Cryptanalytic Machines in NSA." May 30, 1953, PDF pages 3-8.)
Dr. Howard Campaigne
The author is Dr. Howard Campaigne, who began his 1983 oral history interview "Well, when Pearl Harbor came along, December 7th, 1941, I was teaching at the University of Minnesota; teaching mathematics." (Farley, Robert D. "CAMPAIGNE, Howard, Dr. Oral History Interview." NSA, June 29, 1983, pages 1-2.)
Dr. Campaigne, a mathematician and cryptographer, served during WWII at OP-20-G and for a short tour at Bletchley Park. Following World War II, he joined the Navy Security Group, AFSA and NSA, in the R&D element as a mathematician researcher.
Dr. Campaigne is a retired Navy Captain in the Navy Security Group Reserve. He served the bulk of his career in the Research and Development elements of NSG, AFSA, and NSA.
Dr. Campaigne concluded his interview with "a little comment." (Farley, "CAMPAIGNE Oral History Interview," page 130.)
Campaigne: When I went into the analytic equipment, which I did almost the first week I was there and stayed there, I had visions. These would be labor-saving devices, and we wouldn't need a lot of people around. And it's been a continual disappointment that we had so damn many people around. Of course, what we've done is to use these devices to do more rather than to do what we were doing before more economically. But I still feel we ought to be able to do with fewer people. More machines and fewer people.
Farley: It's coming to that, isn't it?
Campaigne: Well, I hope so. I think it should.
Farley: What happens to the people who are displaced?
Campaigne: Well, they join the "buggy whip" manufacturers. Retire.
Bombes, Scritchers, and GOLDBERG
Most cryptanalysis reduces to counting, comparing, rewriting, and referring. Each of these operations by itself is simple and easily done by the proper type of machinery. Where there is need to do them in combinations, or especially if a choice of methods must be made, mechanization is not so simple. For some operations the advantages of machines is evident. Some steps can be done faster, more accurately, and with better organization by machine.
Theoretically with enough men and enough time anything a machine can do could be done by hand. But to do certain very routine computations, such as that done by SUPERSCRITCHER, by means of a crew of thousands of people would raise tremendous personnel problems, so that it might be impossible to actually carry this out. The old joke about solving a simple substitution by a crew of 26 factorial Chinamen is no more than a joke.
Until 1935 practically all cryptanalysis, both by the Navy and the Army, was done by hand. About that time the possibilities of accounting equipment, such as IBM (International Business Machines) and Powers, was realized and some was procured for experimental use. It was a success, and over a period of years many special techniques, unorthodox for accounting, were developed, some requiring modifications of the machines. Many special devices or gates were developed to do specialized analytic devices, the first of which was called the GEEWHIZZER and applied to columnar transposition systems.
Vannevar Bush of Massachusetts Institute of Technology, undertook to develop a special machine for cryptanalysis. With the aid of graduate assistants this was done and the machine was shipped to the Navy in Washington in 1941. Later, two of the graduate assistants came too, John Howard and Lawrence Steinhardt. John Coombs went to the Naval Computing Machine Laboratory in Dayton to build BOMBES. The design of Bush's machine was ambitious, and provided for photoelectric comparisons of two texts for coincidence, monographic, digraphic, etc., up to nine letter repeats, for special patterns and for isomorphic repeats. In operation it proved to be slow (it printed each comparison, no matter how uninteresting) and full of "bugs". Some of the functions were abandoned after operational and maintenance experience, such as the isomorphic repeat search. At a later time improved models were built, and still exist as the 70mm (the tape width) Comparators. They have had long and useful service.
The advent of the war in late 1941 gave great impetus to procuring mechanical aids for the cryptanalyst. The I.C. MACHINE [see Editor note below] was made to compare two texts and measure coincidences. It could compare texts up to 600 letters long at all offsets in a few seconds, and the machine was small enough to sit on a desk. In practice it did not work this way, however; the device was simple enough, but the preparation of the text onto photographic plates was not, so the machines had to be operated as a battery near the camera and dark rooms.
Editor note. "I.C." refers to a technique called the Index of Coincidence, which I explain in Nobody but Us: A History of Cray Research's Software and the Building of the World's Fastest Supercomputer chapter 20, "Engineering Research Associates".
Another photographic device, TESSIE, was started early in 1942. It compared texts for repeats. Subsequently the Army developed a super photoelectric device, the 5202, COMPARATOR, which had much greater capacity and flexibility. These along with others formed a distinct series of photoelectric comparators, the last of which was AMBER, completed in 1947. All of these machines, except the last, contributed definitely to the prosecution of the war. In recent years photoelectric techniques have been used less than digital electronic.
The only analytic machines ever built in large quantity were the BOMBES. These were designed in 1942 with advice from the British. The Navy designed a 16-unit model of which 125 copies were made, nearly identical. To operate and maintain these around the clock took a trained crew of 800 people.
The Army built a single machine called MADAME X, consisting of 144 units which could be run as several separate machines with smaller numbers of units. It also had the advantage of trying the wheel orders in automatic succession. These BOMBES were used against the ENIGMA, the cipher machine used by the Germans for 90 per cent of their enciphering, and consequently were of the greatest importance. One estimate by the Navy was that, costing less then a cruiser, the BOMBE installation had caused the sinking of 60 German submarines. The successes against the German Army and Air were even more important.
The introduction by the Germans of new reflectors with unknown wiring led to the invention and construction of several machines, called "SCRITCHERS", able to do the BOMBE problem without having all the enciphering elements. These machines came quite close to being digital computers and were probably the most ingenious machines built during the war. The fundamental idea of scritching is credited to the British.
After the war the comparator series continued to develop. The Navy product was GOLDBERG, which was the first machine designed to hold its data on a magnetic drum. The Army built CONNIE which used punched teletype tape. Both these comparators were influenced by the British war time machines called ROBINSON and COLOSSUS. In fact, the name GOLDBERG is an American version of ROBINSON, since the cartoonist Rube Goldberg drew weird gadgets just as Heath-Robinson did.
From Connie was developed the more special ROBIN for making round-robins, or all comparisons. The sonic delay line machine DELLA uses a new medium, sound waves in mercury, to continue the line.
Since 1946 there have been three main lines of new developments, the exhaustive trial devices, the dictionary machines, and the cryptanalytic computers.
The exhaustive trial devices include HECATE and WARLOCK. They distinguish themselves by having very high operation rates, and by being large and working only by exhaustive trials.
The dictionary machines look up weights, meanings, etc., in a large memory. They are physically large and limited in their abilities. Their rates of operation are not so fast as HECATE, but are high nevertheless. They started with Navy's MERCURY (a war time development, long defunct) and Army's SLIDE-RUN MACHINE, and now include DEMON I, II, and III, SKATE I and II, and SLED I and II. This last can also do such operations as dragging cribs and reading depths.
The computers were inspired by developments at Harvard and the University of Pennsylvania. They are extremely flexible, able to do almost any logical process by breaking each problem into minute steps. As a consequence of the minuteness of the steps some operations are not nearly so fast as are some more specialized devices. The machines developed at NSA are not copies of standard computers but specially designed machines with much more logical flexibility. They are ATLAS I and II, ABNER I and NOMAD.
In retrospect these machines have seemed to create more work than they accomplish. In the original planning it was expected that the burden of hand work would be lightened and the need for personnel decreased. Although many things formerly impossible are now done, there are in fact more hand jobs than before. These require more analytic ability, and bring more pressure on the people in order to make best use of the machines.
The reason for this can be seen by an example. A certain process called a "pass" is needed to solve a cryptanalytic problem, BOOTSTRAPS. To do this by hand takes several hundred man hours, valued at nearly $1,000. This is exorbitant, so BOOTSTRAPS was deferred in favor of more feasible jobs.
Then a method of doing a pass was devised on card equipment. By this means the cost was approximately $32.50, and several hundred passes were made. The result of a sequence of passes is the material from which cryptanalysts can proceed to a solution. The new method led therefore to more work for cryptanalysts. Each solution opened new jobs to do as well.
Then a program for making a pass on ATLAS became operational and the cost of a pass became $1.25. This is such a bargain that all available data was run through the process, making a tremendous job for cryptanalysts (and plain text). This is the way analytic machinery makes more work for the analyst rather than less.