Chapter 8 (U) Courage and Chaos: SIGINT and the Computer Revolution

\(U) It Wasn\'t Safe at the Cutting Edge

\(U) Well before the plans for Goldberg and Sled had matured, OP-20-G,

joined a bit later by SIS, started a great adventure. They became part

of what many see as one of the most important technological revolutions

in history. The SIGINT agencies became active players in the attempt to

make a technical fantasy come true - to build a universal machine. \"G\"

and the SIS, along with several other military and civilian agencies,

became prime movers in the early stages of the computer revolution.

Establishing that historic foothold was not easy for either SIGINT

group.

\(U) An idea had emerged and became somewhat formalized outside of the

intelligence community by 1945. It was going to be possible to have a

high-speed electronic computer that could mimic any mathematical or

logical process. With a rapidly changeable program, it had the potential

to be a machine for every purpose, from calculation to machine control.

The key to the machine\'s flexibility was its simplicity. It was to have

very, very few hardwired functions, perhaps just the four basic

arithmetic ones, and a few that allowed the movement of data between the

input-output components, memory, and the single central processor. That

and the organization of the machine around the binary system would, it

was hoped, make it relatively inexpensive and allow it to become a

massproduced product. With one piece of hardware that could be made to

imitate any machine through an inexpensive and easily changed set of

instructions, the new computer had a great future. It would replace all

other calculation and, perhaps, data processing devices.

\(U) The ideas for the universal computer that began to take definite

shape in England and the United States in 1946 were very appealing. As

soon

as they heard of them, mathematicians and engineers within \"G\" and the

SIS pleaded with their superiors to make programmed computers part of

the SIGINT arsenal. They were persuasive. By 1947 both agencies had

committed themselves to acquiring general-purpose \"computers.\"

\(U) Neither agency realized what traumas they would have to go through

to obtain them, however. Especially in the case of the SIS, the postwar

experience was as anxiety-filled as the trials that Hooper and Wenger

had gone through in the mid-1930s when they sought Vannevar Bush\'s

help.

\(U) Because Wenger had been able to set up a semicaptive engineering

corporation in 1946, OP20-G had an easier time than the army did. But

even \"G\" and its Engineering Research Associates had some very tough

moments trying to make the new computer come to life.

\(U) That had not been foreseen in 1946. After learning of the

possibilities of the new architecture, each agency had expected that

outsiders would provide all that was needed. That was naive. The SIGINT

agencies soon found it necessary to do much, much more than they

anticipated. Because of the chaos that marked the development of the

computer industry in the postwar era, both had to create their own

machines.

\(U) An Idea Differed

\(U) In 1945 while the ambitious Goldberg\'s technology, if not its

architecture, shifted with the appearance of technical innovations, and

while machines like O\'Malley were being constructed for immediate

problems, another and more adventuresome project began at \"G.\"1

I UP bblKKH/COMINTJmEL USA, AU&, LAN OBR AND NZU/XI

\(U) Duenna and the other \"electronic\" machines of the last two years

of the war, combined with the knowledge of what other computer projects

in the nation were attempting, gave the \"M\" group some ideas about a

general-purpose computer. It was to be one much more flexible than

Bush\'s older Rockefeller Analyzer or even his purely electronic Rapid

Arithmetic Machine.

\(U) When they had a few moments for reflection in 1944 and 1945,

Engstrom and others in \"M\" speculated about what they could accomplish

if they could find a large and fast memory, such as the vastly improved

versions of the delay lines they were already experimenting with, to add

to an electronic processor. While RCA\'s Jan Rachman\'s new idea for an

all-electronic computer was rejected as almost \"screwball,\" \"M\'s\"

men kept thinking about the future. If a large memory with a speed that

came close to that of the electronic processor could be found, then they

thought a general-purpose computer was a possibility.

\(U) But unless there was a high-speed memory, electronic processors

would have to remain as special-purpose devices. Until the software

could keep up with the speed of the processor, there was little need for

electronics. If an electronic computer depended upon tape readers or the

like for its directions, it could be no faster than the slow mechanical

components.

\(U) The input speeds of the best tape and card readers of the era were

orders less than electronic processors. That limitation was compounded

by the serial nature of both technologies. It was impractical to ask

tape and card systems to back up to previous positions and repeat the

reading of data or \"instructions.\" A universal computer needed a

memory that could support \"go to\" commands because tapes and cards

could not fulfill that need.

(TO// 3D The limitations imposed by the absence of high-speed memory

were one of the reasons why the Sled architecture seemed so appealing.

With special \"boxes\" hooked together through plugboard programs,

processing was not dependent upon the nonexistent memory. The absence of

memory was also one of the reasons why IBM and other business machines

manufacturers confined their postwar electronic offerings to limited and

special-purpose attachments, such as multipliers and dividers that

hooked onto tabulators. 2

\(U) In 1945, any engineer who thought about moving further than the

Aiken-IBM combination of motors, shafts, and tape readers, or the Moore

School\'s set ofENIAC special-purpose boxes \"programmed\" through

resetting huge electrical cables, had to have a great deal of faith. He

had to believe that some technological hints would soon become viable

and affordable hardware. There were some indications that such dreams

might come true. But in 1945-6 they were just indications.

\(U) Some thought that delay lines, tubes filled with chemicals, could be

reengineered to serve as memories. The young experts at the University

of Pennsylvania who were building the ENIAC felt they could convince

delay lines from radar sets to behave well enough to hold programs as

well as the data needed for immediate processing. That was a courageous

commitment because those \"acoustic\" delay lines were very

temperamental. It was very difficult to regulate the timing of the

pulses that flowed through them. Slight changes in ambient temperature

caused serious distortions. Also, it was difficult to monitor the

behavior of the crystals that sensed the data \"pulses\" at each end of

the tube. Even when all the technical difficulties were eliminated, a

fundamental problem remained. The tubes could hold only a few \"bits.\"

\(U) There were some other memory possibilities being discussed at the

end of the war. One

was to use a variation of the emerging television technology to store

and recover \"dots\" of information on an oscilloscope-like screen. If

it could be made to work, it would be an ultra-fast memory. A computer

would not have to \"wait\" until the information it needed cycled past a

sensing station. It would run at electronic speeds and would allow

parallel data transmission. 3

\(U) There were more esoteric ideas for powerful memories, such as RCA\'s

Selectron and the use of magnetics, but they were even less ready than

the other alternatives.

f6} Although the engineers at OP-20-G knew of the technological limits,

they could not pass up a chance to at least survey universal computer

options. John Howard formalized some of the ideas in a June 1945

memorandum; then, along with the \"G\" mathematician, C. B. Tompkins,

toured all the East Coast computer projects looking for more ideas.4 But

little came of their trips. \"G\" was too busy to explore other than

cryptanalytic machines. That remained true for several months after the

war ended. Its workload even prevented \*G\" from sending a

representative to one of the earliest postwar computer meetings.

\(U) When Howard Engstrom received an invitation to participate in a

major navy symposium on computers, he replied that \"G\" had done little

of the type of work that was to be discussed and that he and his crew

were too busy to attend.

\(U) The urge to explore the possibilities of a general-purpose computer

continued. But little could be achieved. \"G\" found it difficult to

acquire connections to the outsiders, especially the academics, who

seemed to be taking the first major steps towards creating the modern

computer. \"G\'s\" old scholarly friend and go-between, Vannevar Bush,

had stepped back from OP-20-G when the war broke out and did not try to

reestablish the 1930s relationship. That left \"G\" without a

prestigious outside scientist who could

provide the critical endorsements speculative projects needed.

\(U) Bush also decided not to return to MIT. He remained in Washington,

acting as something of an academic elder statesman and high-level

science policy maker until his retirement. Among his many contributions,

he gave advice on the future of science in the military. In addition,

Bush was frequently called upon to make recommendations concerning the

integration of the nation\'s intelligence services. His role as a

science advisor to President Eisenhower also played an important part in

SIGINT mechanization in the 1950s. 5

\(U) Bush stayed at quite a distance from the computer developments of

the postwar era. He also stayed away from OP-20-G, except for a few

courtesy visits that Joseph Wenger arranged. One reason for Bush\'s

arm\'s length relationship was a very heated argument with the Bureau of

Ships about the Comparator. Soon after the war the bureau decided that

it should be protected by patents. Bush was sent all the necessary

paperwork to sign. He did so, but only after the deepest protests to the

navy about revealing precious secrets and about imposing upon him. 6

\(U) OP-20-G had lost another friend. Stanford C. Hooper was in

semiretirement. He was now old and ill, and he had to spend much time in

Florida. He was acting as a consultant to several small electronics

firms, including ERA, however. He still had the ear of many Washington

influentials, but he could no longer aggressively fight to link OP-20-G,

the scientific establishment, and the large corporations. In fact, he

had become a bit soured on the corporations and acidemia. He had come to

favor small private companies as the only guarantor of innovation and

responsiveness.

\(U) Meanwhile, the other part of OP-20-G\'s old university-computer

connection, Bush\'s \"boys,\" had migrated to the \"captive\"

corporation, ERA. Howard, Coombs, and Steinhardt were

keeping up with computer developments, but ERA\'s first contracts and

the imperative to develop a \"cryptanalytic\" machine kept them too busy

to act as computer innovators. As a result, their 1945 general-purpose

computer aspirations languished until mid-1946.

\(U) Then, \"G\" developed a new and energetic computer champion. At the

same time, it found someone with great enough scientific status to

validate its request to acquire something which, in the mid-1940s,

seemed more fanciful than Bush\'s 1930s machine.

\(U) Goodbye Dr. Bush, Hello Professor von Neumann

\(U) Just as the Goldberg project was launched in St. Paul and as

Wenger\'s own research group was deciding whether or not to have someone

build an electronic Super Bombe, one oP\'G\'s\" mathematicians, James T.

Pendergrass, enrolled in a summer institute on the programmable, digital

electronic computer. 7

\(U) His inclusion in the Philadelphia meeting was almost an

afterthought. Apparently \"G\" had not been asked to send someone until

a few weeks before the Moore School Lectures began. Pendergrass had

intended to spend much of the summer on vacation, but when his boss,

Howard Campaigne, called him, he found it impossible to refuse the

assignment. He rushed to the University of Pennsylvania and immediately

began sending reports to Campaigne.

\(U) Howard Campaigne was one ofthose bright young men who had been

brought into \"G\" early in the war. Like his friend, Joe Eachus, he

spent much time in England.8 And like Eachus he became deeply involved

with the RAM program. Deciding not to go to ERA he became a civilian

scientist within \"G.\" He helped shape and direct \"G\'s\" postwar

research agenda. By 1946 he was one of Joseph Wenger\'s right-hand men

and was respected enough to be allowed to act as a repre

sentative of \"G\" to the outside world. That was what caused him to

attend an important Navy Department conference in spring 1946.

\(U) The conference was on the nature of large-scale computers. The major

address was given by the man who would soon equal or exceed Vannevar

Bush\'s status in the scientific-political realm, John von Neumann. 9

\(U) John von Neumann was perhaps the most famous of the new applied

mathematicians. He had migrated from Europe in the 1930s to join the

likes of Albert Einstein at America\'s only true research institute, a

place where scholars set their own agendas, von Neumann became one of

the \"scientifically anointed\" at the Institute for Advanced Study at

Princeton.

\(U) The first rumblings of war led the Institute and von Neumann to move

far beyond their abstract academic origins. During World War II, von

Neumann made important contributions to the atomic bomb project. As a

result of that involvement, he became entangled in the ENIAC computer

effort at the University of Pennsylvania.

\(U) The University of Pennsylvania\'s World War II contract with Army

Ordnance for the ENIAC had come almost by chance, just as the NDRC ended

its computer program, and as firms such as RCA rejected pleas to turn

their hardpressed engineers to computer projects. Ordnance was in need

of a way to speed the calculation of firing tables. With no other

alternative, the army accepted the proposal of two young engineers at

the Moore School. They promised to build an electronic version of

Bush\'s great Differential Analyser. Fortunately for the history of

computers, John Mauchly and Presper Eckert were given a great deal of

freedom and time. Their much delayed postwar delivery of the relatively

special-purpose ENIAC was not treated as a sign of failure, and their

plan for a programmable

universal electronic computer was quickly funded.

\(U) With the help of John von Neumann, they started the project (EDVAC)

and began seminars that attracted the pre- and postwar generations of

computer builders.10 von Neumann\'s stature in the scientific and

military communities had grown so much that his presence gave the Moore

School\'s computer efforts the highest credibility. While working on the

design of what is regarded as the first true universal computer, the

EDVAC, the original leaders of the ENIAC project, Mauchly and Eckert,

had become estranged from the university\'s administration and, to some

degree, from John von Neumann.

\(U) Von Neumann, whose importance increased in the postwar years, also

became alienated from the University of Pennsylvania. He decided to

found his own computer initiative. He was soon able to convince his old

academic home, the Institute for Advanced Study (IAS) at Princeton, to

accept several military and civilian grants and to create a center to

house his attempt to design and build his own computer. His \"IAS\"

machine was intended to serve the needs of applied mathematicians and

physicists.

\(U) Von Neumann did not confine himself to computer building. He became

a major figure in Cold War science and policy. He advised all of the

American leaders of the era, and he served on the most important

science-related boards. He even became a good friend of OP-20-G and

later NSA, serving on their expert panels. He gave them much technical

and political advice throughout the 1940s and 1950s. His contributions

included more than hints about new computer technologies. He frequently

urged the SIGINT agencies to sponsor fundamental electronic research to

be conducted by leading academics.\"

\(U) While von Neumann was forging his Cold War reputation, the Moore

School had begun its own machine, the EDVAC. Sponsored by Army

Ordnance, EDVAC was to have the simplest of architecture. Although it

was intended to be an operational machine for the Aberdeen Proving

Grounds, it was also something of a testbed. A central goal of the

project was to prove that a universal machine could be made to work and

to do it quickly. Therefore, EDVAC was designed as simply as possible.

\(U) EDVAC was a binary machine that depended upon a serial acoustic

delay-line memory. That memory was to hold both programs and data. The

acoustic technology limited the machine to about 1,000 words of fast

memory. Technological limits also dictated much of the EDVAC\'s internal

organization. Trying to avoid the problems caused by the high failure

rate of vacuum tubes, EDVAC\'s internal structure was made as sparse as

possible. It had just one-third the number of tubes used in the ENIAC.

\(U) To keep the number of components at an absolute minimum, the machine

had only a few built-in instructions. That was a wise decision. Each

\"instruction\" demanded dozens of tubes and hundreds ofhandwired

connections. And each increased the computer\'s cost and multiplied the

probability that it would experience a failure well before any

significant computational task could be completed.

\(U) In addition to keeping the number of components to a minimum,

EDVAC\'s designers limited the machine to the serial transmission and

processing of data (one bit at a time). Serial processing also reduced

the amount of failureprone electronics. But it carried the price of

slower processing rates.

\(U) EDVAC\'s designers made another tradeoff that favored simplicity

over speed. The machine\'s operations were based on \"fixed clock\"

timing. That meant that no matter how little time one operation took,

succeeding work had to wait until the next clock pulse.

\(U) EDVAC\'s planners tried to keep their task manageable by

concentrating on building a machine for mathematicians. EDVAC was not

intended to be a data processor. The EDVAC engineers did not try to

solve the many problems involved in making input and output rates

approach electronic speeds. Slow tape and card readers gave the machine

its data, and its even more primal cardpunches and teletypewriters

displayed results. Although one of the first computer programs written

by the ENIAC-EDVAC group was for sorting, EDVAC\'s builders never

pretended that it could replace tabulator equipment.

\(U) While the EDVAC\'s designs were being set, the ENIAC\'s parents,

Eckert and Mauchly, left the University of Pennsylvania and attempted to

found and keep afloat their own for-profit computer company. After more

than six years of anxiety and tragedy, they completed the UNIVAC

computer.

\(U) The UNIVAC was also a delay-line, fixedclock machine, but it went

far beyond the EDVAC in terms of power and sophistication. One reason

for that was the UNIVAC\'s attempt to become \"the\" new business

machine, one to replace hundreds of tabulators. That called fcr the

development of much-enhanced I/O technology. Anew data processing

capability was to some extent achieved through the creation of magnetic

tape systems, a development that helps explain why the first UNIVAC did

not appear until 1951.

\(U) The goal of building a computer to replace the tabulators led to a

very historic decision by Eckert and Mauchly. Because they wanted to

maximize the speed of data processing, which typically demanded little

calculation on a great deal of information, they deviated from a purely

binary representation of numbers within the UNIVAC. It had what was

called at the time a \"decimal\" representation. Although UNIVAC used

binary circuits, a decimal format was imposed to speed the input-output

functions.

\(U) Eckert and Mauchly\'s commercial computer aspirations, as well as

John von Neumann\'s academic ones, were just emerging when the

University of Pennsylvania decided to host its historic summer 1946

Moore School conference. All those who had made contributions to

computing during the war were invited to hear presentations by von

Neumann and others who were outlining the computers of the future.

\(U) A Summer in Philadelphia - an Exciting One

\(U) It was probably Howard Campaigned attendance at an earlier (May)

navy symposium that made him aware of the Moore School conference. The

Washington meeting was where he first made contact with von Neumann and

where he realized that the general-purpose computer was going to be

built, with or without OP-20-G. Campaigne decided that \"G\" should at

least have a chance to be one of its sponsors.

\(U) He hurriedly arranged for some funds and then called his assistant,

James Pendergrass, asking him to attend the coming Philadelphia

symposium. Campaigne was unable to tell him much about what was to take

place in Philadelphia. As a result, much of what Pendergrass encountered

surprised as well as thrilled him.

\(U) During the Moore School\'s summer program, Pendergrass studied the

designs ofthe ENIAC and those for the much more advanced EDVAC. He

listened to the presentations ofthe other men who had begun to develop

universal electronic machines.

\(U) Coming from a physical sciences background and being an advocate for

applied mathematics, Pendergrass was especially taken with John von

Neumann\'s ideas, including his version of a programming language. When

von Neumann outlined the concept for his new Institute for Advanced

Study machine, Pendergrass became convinced that \"G\" had to have a von

Neumann

lOMybCKblJJCUMINIJJkEL US/

\\MD NZU/X1

(IJ) UNIV& C

type of computer. He thought the von Neumann design was the best, the

one that would be the first to appear in hardware, and the one most

likely to be cloned by a manufacturer. Well before the Philadelphia

conference was concluded, Pendergrass convinced Howard Campaigne that

OP-20-G should have one of the \"new\" computers, specifically one with

a von Neumann architecture.

\(U) Pendergrass was not the only one who believed in the IAS design. The

mathematical and applied physics community took it as the model for

computers for the 1940s and early 1950s. The emerging von Neumann

architecture was especially attractive to mathematicians because it

promised to provide a much faster and more precise computer than other

designs of the era. It had the potential to be faster than the serial

type of machines by factors in the hundreds. At least five IAS computers

were copied in American universities and advanced research centers.

\(U) The IAS machine was not going to send or process data bit by bit,

nor was it going to make one operation wait for a \"clock.\" It was

going to send data simultaneously and would initiate an

operation as soon as the previous one finished. As or more important, it

was not going to be based on the limited delay-line memory, von Neumann

bet that RCA would keep to its pledge and develop the very advanced

Selectron tube within a few months. The Selectron promised to be a fast

memoiy that would maximize the potentials of electronic speed and

parallel data-transmission and processing.

\(U) To many in the computer field, however, von Neumann seemed too far

ahead of the technology. He also appeared to be naive about how much of

the complexity of his proposed machine could be mastered by his handful

of engineers. His critics thought the EDVAC group was taking a more

sensible course: creating a minimal and reliable computer that had a

possibility of being completed on time.

\(U) Given the ambitious nature of von Neumann\'s computer, Pendergrass

and Campaigne sensed that it was not going to be easy to persuade \"G\"

that scarce resources should be devoted to a machine that was not yet

fully designed, let alone built. No one, in fact, could

predict when any of the new computers would be completed.

\(U) Despite von Neumann\'s reputation and the accolades that the atomic

energy community was awarding to the IAS design, Pendergrass and

Campaigne knew they would have to prove that \"a\" machine could compete

with al! the specialpurpose devices that were in place at \"G\" as well

as those that were being planned. And they would have to show, without

insulting anyone, that the new computer would be as good as or better

than Goldberg, the perhaps-universal comparator. Pendergrass and

Campaigne were in a situation quite analogous to Hooper\'s in the early

1930s: How could they convince \"operational\" types that scientists had

a better and practical grasp of the future?

PES} Pendergrass got to work during the summer and continued on, with

Howard Campaigne\'s enthusiastic help, through the remainder of the

year. They composed two persuasive technical reports. 12

(TS} The first was sent to \"G\'s\" higher-ups in October, the second in

December. A great deal of effort had gone into both reports to ensure

they would convince the cryptanalysts that, like the proposed Sled, the

von Neumann machine would end the horror of having to wait two years

while a requested special machine was constructed.13 The reports did not

refer to any particular experience, but Pendergrass knew that many in

his audience had gone through the frustrations of World War II when

almost all the RAMs (and the Bombes) had arrived too late.

f¥S} In one of the classic statements in the history of computers,

Pendergrass wrote: \"It is not meant that a computer would replace all

the machines in Building \#4, nor is it meant that it could perform all

the problems as fast as the existing special purpose machines. It is

however, the author\'s contention that a computer could do everything

that any analytic machine in Building

\#4 can, and do a good percentage of these problems more rapidly.\"14

(TS) The text of the report reflected both Pendergrass\' orientation and

the nature of mid19408 computers. After explaining the logic of the von

Neumann machine and admitting that it might be some time before any such

computer would be available, he outlined what he thought was to be a

standard programming language, one based on von Neumann\'s \"one

address\" concept. 13

\(U) Von Neumann envisioned a machine that would be used for very precise

calculation and little data processing. To speed calculating, it was to

be a pure binary machine. To further improve performance, he had turned

away from the original EDVAC idea of a four-address instruction. He had

come to believe that the most efficient instruction format should

include only one place to get or put data. That would allow, given the

word size of the computer, more precise calculations without additional

hardware.

\(U) He argued that because his machine would be busy with much internal

work, such as multiplication and division, it would be more efficient to

have \"get\" and \"place\" addresses in separate statements. Only if a

computer was to be used for much I/O and little calculation would a

multi-address instruction be reasonable.

\(U) Von Neumann also believed that his machine should have very few

commands. The smaller the number of commands, the less internal

circuitry that would be needed. Following his mandates, his engineers

were able to reduce the number of components in the IAS machine. It had

only two-thirds the number of tubes of the EDVAC. Von Neumann\'s

mathematical focus also meant that he gave little thought to the I/O

problem. What happened within the computer was more important to him

than handling masses of data.

(T9//gf) Pendergrass agreed with all of von Neumann\'s ideas, but in his

reports he suggested that a few additions be made to von Neumann\'s set

of minimal instructions. They were to be ones which, like multiplication

and mod 2 commands, would be needed to meet special cryptanalytic needs.

Especially important to him were those which would speed the analysis of

baudot traffic.

(=fi\$) The politically important parts of the Pendergrass reports were

the sections in which he and Campaigne presented computer programs for

major cryptanalytic attacks. Demonstrating how the machine would perform

the attacks was critical if\'G\" was to be persuaded to invest in a

computer. Of course, the reports reflected an implicit faith that

programming would be much, much less of a problem than building a

specialpurpose machine. No mention was made of how long it took to write

the programs.

(TS//SI) The first report included software programs for a Generalized

Copperhead problem, a Four-Wheel Enigma Grenade Problem, and a Hagelin

attack. The December report was intended to impress any holdouts. It

contained programs to imitate two of the grand achievements of World War

I, the Duenna and the Mercury. Asa concluding argument, the report

showed how to end the great crypto-disappointments of the war years.

There was a program that could imitate a cipher wheel. It showed that a

general-purpose computer might act as an electronic Super Bombe. 16

\(U) Buy a Computer, Now

(£} The reports made their point, at least with farsighted men like

Joseph Wenger. He took action even before the Pendergrass-Campaigne

report of December was completed. To reinforce Pendergrass\' arguments,

he immediately arranged for ERA\'s John Howard to conduct a computer

feasibility study and assigned Pendergrass to continue to survey the

computer field.

\(U) Pendergrass did not waste time. In November he informed \"G\" that

many military and some civilian agencies were very interested in digital

computers and that projects at the National Bureau of Standards and RCA

(with the IAS) stood a good chance of producing machines by 1948. The

IAS-RCA project, he thought, had the best design and best chance of

success. 17

\(U) After attending another major computer conclave at Harvard in

January, Pendergrass forwarded a new survey of America\'s and Britain\'s

computer ambitions. He cited the emergence of more computer projects,

most of which were supported by government agencies. The navy\'s Bureau

of Ordnance and the ONR, he showed, had already established quite a

foothold, as had Army Ordnance. Even the Census Bureau had become

involved.18 Among the six active projects19 (the one under John V.

Atanasoff had just been cancelled by the Naval Ordnance Laboratoiy), the

IAS computer, Pendergrass reported, continued to be the best option. It

remained much closer to completion than the proposed Whirlwind at MIT,

and it was more suited to cryptanalytic work than the upcoming

UNIVACortheEDVAC.

\(U) The only nonpositive things that Pendergrass had to say about the

Princeton efforts were that he had discovered that RCA and Princeton did

not have a formal agreement binding the corporation to build a computer

and that its valuable Selectron was still in the \"uncertain\" category.

Neither seemed critical to Pendergrass, however. He expected a working

IAS machine by mid-1948. He assured \"G\" that if the Selectron were not

perfected, an electrostatic memory, such as the one proposed byMFPs Jay

Forrester, would serve as a fully acceptable substitute.

fS-) Pendergrass\' surveys were read by Howard Campaigne, then sent to

Joseph Wenger. Wenger trusted Pendergrass, and he believed that OP-20-G

should gain a foothold in computers before one of the other branches of

the navy

established a monopoly. Without waiting for John Howard and C. B,

Tompkins to submit their ERA report (it arrived in February 1947),

Wenger made a commitment to acquire a von Neumann type of computer. \*°

(Tg//8f) Even before a contract was let, plans were laid to use the new

machine on major operational problems.21 \"G\" even put aside the idea

of building an electronic Super Bombe, at least until the potentials of

the new universal computer were explored.

\(U) In January Wenger was so enthusiastic that he ordered his men to

establish project \"Atlas,\" although he did not yet have the funds to

design and acquire a machine. az The name \"Atlas\" was picked because a

comic strip used it as a name for a \"mental giant,\" but a reference to

raw courage would have been as appropriate. Wenger still had to gain

formal approval for the \"G\" computer.

(•S\^As Wenger struggled to find the money he needed, additional

crypto-studies reinforced the initial enthusiasm, and went beyond it,

perhaps raising expectation a bit too high: This opens tremendous

possibilities in the field of clinical attack by speeding this attack up

to the point where large volumes of traffic may be so processed. With

sufficient skill in preparing the logical control, it seems possible

that the machine may be made to perform any cryptanalytic operation now

done by hand, which does not require intuition.\"23

(■S} Wenger did everything he could to make sure the \"G\" computer

proposal would be funded. He had Pendergrass assigned as a !iais3n to

the Office of Naval Research. It was exploring computers and was

intensely committed to furthering applied mathematics. With Pendergrass

in touch with ONR\'s experts, they would be unlikely to block the \"G\"

request on technical grounds. Other mathematicians in the agency were

sent to important computer seminars: Eachus,

Campaigne, Blois, Tordella, and others met with the \"greats\" of

computer history, such as Alan Turing and M. V. Wilkes.

dSfr The contacts and investigations soon started to pay off. \"G\" was

gaining a reputation as one of \"the\" centers of computer expertise.

Other development projects, such as Whirlwind at MIT, gladly shared

design information.

\(U) At least in terms of computer architecture, \"G\" was well

integrated with high science. \"G\" became committed to the atomic

scientists\' favored way of sending data within the machine: all the

bits at one time in parallel, rather than one bit at a time (serial

mode) as in the EDVAC. As important, \"G\" wanted Atlas to have a single

memory, one to hold both data and instructions. That was in contrast to

some architecture, such as those of Howard Aiken at Harvard, who thought

separate memories, concurrent processing, and dozens of registers made

for a more powerful computer.24 Without any hesitation, \"G\" favored a

pure binary system for its computer. The idea that became embodied in

the UNIVAC, that some decimal representation was more efficient was

rejected.

\(U) While his research crew defined Atlas, Wenger worked on the politics

of acquisition. He convinced the CNO of the need for Atlas, gained an

extremely high priority rating for it,25 and then sidestepped some

serious objections from the Bureau of Ships.

\(U) In response to hints there were already enough navy computer efforts

and that long term research should be left toothers, Wenger informed the

bureau that \"G\" needed to acquire a \"special analytical machine.\"

The word \"special\" gave OP-20-G the opening it needed to avoid a

worst-case situation in which it would be forced to wait for and accept

a machine it might not want. It also gave \"G\" the chance to play a

positive role in the emergence of the computer industry. 26

EO 3.3(h)(2) P.L. 86-36

\(U) Well before authorization had been granted, \"G\" began a more

detailed design and made evaluations of possible computer manufacturers.

With all the other government agencies sponsoring research in the field

and with the interest shown by several private companies, \"G\'s\"

experts did not anticipate that a large investment would be required for

the design or for the hardware. \"G\" still thought RCA would enter the

market. The National Bureau of Standards also seemed ready to build a

computer. Wenger expected to have Atlas at the Nebraska Avenue complex

in approximately two years. 27

\(U) Whatever the options, Wenger wanted quick action. Even though a

\"special\" machine had been approved and although Monogram funds were

available, there was always the chance that the White House might decide

that computers were a luxury. Even the \$100,000 to \$300,000 for the

machine might be seen as too much for a peacetime intelligence agency. 28

\(U) Laying out the general specifications for Atlas was relatively

painless. Pendergrass had done his technical homework, and his

recommendations were only refined, not changed. Beginning in March 1947,

when \"G\" decided to take more responsibility for designing its Atlas,

Campaigne, Eachus, Pendergrass and many others at \"G\" began to meet to

detail the functional characteristics of their newest \"analytical

machine.\" They even began to write programs. The enthusiasm was so

great that many worked nights and weekends on their problems. 29

■{S} \"G\" was to have a von Neumann computer, not asouped-up version of

its older devices. Suggestions that Pendergrass\' original sketches be

altered by adding special-purpose attachments were adamantly rejected,

as were those recommendations that the machine have control switches and

plugboards. Software, driving elemental circuits, was to be the only

control mech

{S}- But the number of commands built into the machine was to be

expanded beyond von Neumann\'s original list, and Pendergrass\' early

recommendations. By 1947 close to forty commands were in the design. The

expansion was aimed at easing cryptanalytic processing, as had been the

alteration in the fundamental word size in the machine to six digits.

That would allow letters as well as numbers to be analyzed.

(9} The additional commands were at the fundamental level of the

machine. There were \\no suggestions that complex sequences to imitate

entire processes be wired into Atlas. A series of multiplication

commands and a divide instruction were included, however, as were shift

commands and noncarry arithmetic capabilities\* Shifts were especially

useful when rotor or wheel stepping was required.

anism

30

Also, there were hopes that a random number generator could be devised.

\(U) Campaigne, Eachus, and the others on the design team had bright

hopes for Atlas. But there were limits to the aspirations. They accepted

the fate that plagued the first computer generation: They did not

attempt to write a compiler or a high-level language for the machine.

The only treat the \"G\" group gave programmers was the luxury of

writing in octal rather than binary notation. That provided some relief,

but it did not allow a programmer to avoid specifying the location of

memory addresses in \"absolute\" terms. There was no software to

automatically keep track of where instructions or variables were

located.

\(U) Like the TAS computer, Atlas was to be centered about what the

Princeton group considered \"the\" solution to the memory problem, the

RCA Selectron tube. It would allow an electronicspeed mass memory,

something needed to meet the potential of parallel data transmission and

processing. Hopefully, the Selectrons would support a large memory. In

1990s terms, Atlas was to have 64K. In terms of the longer word size of

the

Atlas, that was equal to 16,384 \"cells.\" That was orders greater than

what was planned for EDVAC.3\'

\(U) The Selectron was under development at RCA\'s research laboratory.

Rachman\'s tube promised to be much more powerful than the other types

of binary electrostatic storage devices that were under development. And

it was expected momentarily. Although some at OP-20-G had treated many

of Rachman\'s ideas as more than fanciful, because of his advanced work

during the war, he had become an ally of von Neumann, and his work

demanded respect. 32

\(U) The Selectron was a complex device, but it had a great advantage; it

was small and fast. Its size was one of its great attractions because

other high-speed memories of the period, such as delay lines or the

Wilkes electrostatic tube,33 took a great amount of space.

Unfortunately, the Selectron proved to be too complex.

\(U) It was based upon the principle that \"an insulated

secondary-electron emitter can be made to \'float\' at either of two

stable positions\...\" Deceptively simple, the principle demanded much

delicate hardware. Inside the three-byseven-inch tubes was a dielectric

target that was divided up by sixty-four metal bars and sixty-five

circular metal rings. They created 4,096 \"cells\" that were the storage

areas. When the four walls of a cell were all more positive than some

particular voltage, a \"bit\" was registered.

\(U) To von Neumann\'s and \"G\'s\" great disappointment, all that was

too much, even for the great Jan Rachman.34 By spring 1947 RCA had to

admit that it might be some time before the Selectron was ready. That

led to some technological soul searching in Princeton and Washington.

The IAS put more effort into a television-like electrostatic memory and

even explored the possibility of ultra-high-speed secondary memory based

on magnetic wire wound on bicycle wheel drives.

\(U) The news about the Selectron was only one indication that the

computer revolution was going to take much longer than had been thought.

RCA began to make it clear that it was pulling back from its hints of

becoming a manufacturer, the National Bureau of Standards program had

slowed to a crawl, and the probability that the exENIAC team, Eckert and

Mauchly, could deliver their promised computer to the Census Bureau in

time for the 1950 census sank to near zero.

\(U) By spring 1947 Atlas was on its own. If \"G\" were to have its

computer, it would have to take even more responsibility, perhaps even

for a very expensive failure. And it would have to make a critical

technological choice.

\(U) Little Thanks for Tliat Memory

\(U) In April 1947, after learning about the faltering industrial

commitments and the Selectron\'s possible stillbirth, \"G\" made two

very significant decisions. The first was to continue with the project

and the acquisition of a computer despite the absence of an \"industry\"

or even a university that seemed willing to build computers. The second

decision was perhaps more dramatic. 35

(T0//3I) When it was learned that the Selectron would not be available,

there was a critical meeting at \"G\'s\" Nebraska Avenue headquarters.

Some of those in attendance thought that without the high-speed memory

it would be senseless to continue more than very general design work.

What use would Atlas\' electronic circuits be if the memory was a slow

tape or similar device? Even looking for a manufacturer for Atlas did

not make sense to them. There were a few suggestions that the entire

project be put on hold.

(TC//9F) Howard Campaigne, perhaps worried that such a decision would

end chances of funding, put up a stubborn fight. He won half his battle:

The work was not canceled. But his victory seemed to open the door to

some dangerous

possibilities. His recommendation to go with what had always been the

\"fall back\" memory for Atlas and Goldberg,36 a magnetic drum, stood

the chance of making Atlas and \"G\" look rather foolish. It could make

Atlas very slow and perhaps very dumb.

£Rj//flI) Drums were much faster than tapes or cards, but they delivered

information at a rate of i/400th or less of delay lines. Some estimates

of the period gave the Selectron and electrostatic memories a 1,000-fold

advantage.37 If microfilm could have been made to be \"rewritable,\" it

could also have made a drum look antiquated. Seventymillimeter microfilm

held 12,000 bits per inch; drums had a density of from 100 to 200

bits. 38

GBS-) Although \"G\'s\" RAM group realized that such a memory would slow

the proposed machine manyfold, by a close vote its members decided that

a drum would be acceptable. It seemed a much better choice than

postponing the project and being left dependent on the whims of an

almost nonexistent computer industry.

-£BS} Campaigne and his associates realized they were taking a chance.

There were hosts of mechanical as well as magnetic-electronic challenges

to overcome. Whether the \"drums\" were long bars or three-foot

\"wheels\" covered with magnetic tape or sprayed with a magnetic

coating, the problems of milling, sensing heads, and drive motors

remained unsolved. Even ERA with a head start on drum construction

because of its connection to the earlier RAM projects, did not have a

finished and sure technology in hand. \^

(=F8\> \"G\" decided to take the risk. While the IAS group waited for

the Selectron\'s development or the appearance of another electrostatic

memory, \"G\" started to work on revised designs for a drum machine. It

also began a search for someone to build the newly defined Atlas. 40

CfS) No serious consideration seems to have been given to having, as

would many atomic ener

gy research groups, a university take charge of final design and

manufacture. And \"G\" did not spend much time investigating the few

companies that seemed willing to build computers. Thus, soon after the

critical April 1947 meeting, ERA was chosen even though \"G\" knew how

busy the young firm was with its first contracts.

\(8) There was some worry that Atlas might be a bit too much for the new

company and that some emerging problems with magnetic drums might not be

conquered. 41 But in August 1947 ERA was given a design contract. And it

was informed that \"G\" wanted a machine soon. ERA was not to wait for

the results of the several research projects OP-20-G and the SIS were

sponsoring to develop multifunction and ultrahigh-speed tubes and new

circuits. And there was no thought of delaying Atlas just because there

were not yet any high-speed printers suitable for an electronic

computer. 42

\(325) There were a growing number of reasons why \"G\" wanted ERA to

quickly prove the worth of a universal machine for cryptanalysis. Just

as ERA was put to work on the final designs, the Sled project with its

special architecture was being launched with much support from the

Bureau of Ships. In addition to having some competition, Atlas had to

face another possible trauma; there were well-grounded rumors that the

Monogram budget was to be cut severely so. 41

fFS} With a great deal of help from \"G\'s\" research group in

Washington, ERA was able to develop an acceptable design within a few

months. As requested, it matched the von Neumann concepts and was aimed

at avoiding manufacturing problems. Some rather useful ideas were

sacrificed to the needs of the production schedule. A second processor,

which would check results, was not included, and the suggestion to

develop a partitioned memoiy was dropped. Having as many as eight active

\"accumulators\" was also regarded as too much of a luxury.

fES} In early spring 1948, in return for promises to use as much

standard equipment as possible, ERA was awarded a construction contract.

There was a caveat, however. ERA was more than encouraged to build Adas

in a way that would allow the substitution of electrostatic (Selectron)

storage if and when it became available.

(T0//0I) ERA and \"G° were in a hurry. Adas was given an AA priority,

ERA borrowed much from MITs Whirlwind project, and ERA gave Adas as much

attention as possible even when it had to rush to complete some

special-purpose machines to attack Russian targets. 44

(S\> While the engineers in St. Paul were working on Atlas, the

mathematicians-turned-programmers in Washington built their own computer

to prepare for Adas\' arrival. They wanted programs ready to help prove

their electronic computer\'s operational value as soon as it was

delivered. Constructed within four months out of relays and a small

magnetic drum developed by ERA their Abel computer was a logical clone

of

Adas. It gave \"G\" more than a year\'s head start in training

programmers and in writing some operational programs. Its drum was not

large enough to perform all of Adas\' chores, and its relays were

hundreds of times slower than ERA\'s circuits, but it came to be almost

a \"pet\" of the research group \"

45

P Atlas 1

(•G-) Meanwhile, despite the growing pressures on ERA, it was able to

work something of a computer miracle: Adas was delivered to the navy in

early December 1950, fairly close to the anticipated delivery date. It

had taken ERA less than two years to construct the machine, perhaps

because so much time had been spent preparing for its production stage

and because of ERA\'s experience building special-purpose machines, such

as Goldberg. In fact, Adas was the thirteenth project for \"G.\"46

(€T Most of the design goals were accomplished. That made Atlas one of

the very first operational computers in the world. ERA also achieved

another sort of computer first: Adas worked and worked well for ; a

decade after it was sent to Washington. A very efficient testing and

maintenance schedule allowed replacement of tubes before they caused an

unexpected failure. That contributed to an almost unheard of 90 percent

\"up-time\" (availability), which made ERA very proud and very anxious

to transfer its new computer skills to the commercial marketplace.47 It

was also proud that it could have built one of the most powerful of all

the early computers using only 2,700 tubes and that its drum performed

reliably-48 Asa result, ERA and its follow

10\^ afcUKbl/ZlUMINlimfcL U&A, AU3, CAM ODR AMD HZU/X1

lEO 3.3(h)(2) P.L. 86-36

on companies became leaders in magnetic drum technology and gained a

reputation as supercomputer builders.

\(U) Saving a Reputation through Logic

PR8} But all the original Atlas goals were not attained. The machine had

cost perhaps three times the early postwar estimate; its delivery price

was just short of \$1,000,000. 49 More importantly, the drum held less

than one-third the amount of information that had been hoped for in

1947. But ERA was able to rotate it at an extremely high speed. Partly

by reducing its size from the dimensions of earlier drums (three feet in

diameter) to twenty-five inches long and eight inches in diameter,

Atlas\' drum was ten times as fast as the one installed on Goldberg.

\"\" The increased speed helped, but it did not solve the memory access

problem. The 1950 Adas began its life as a very slow machine because the

program, as well as data, had to be read from the drum. There did not

seem to be a viable technical save. Replacing the drum with the still

expensive and irritable electrostatic or delay-line memories seemed

impractical.

-ffS) The programmers atOP-20-G were charged with finding the best

solution they could. Perhaps to everyone\'s surprise, they came up with

an answer that made Adas competitive with other computers of the time.

i\^T The solution they devised was called \"interlacing.\" Combined with

very careful programming, it increased Adas\' speed by a factor of more

than 300. That meant that the drum-based Adas became approximately

two-thirds as fast as a similar machine using the new magnetic core

memory of the mid-1950s. In fact, Adas came close to being a match for

the IAS machine.5\*The increase in Atlas\' speed came at a high cost to

the early programmers, however.

(S\^The trick they had to pull off was to place instructions around the