The Director's Letter

The Museum is in a time of change: location, staff and exhibits. But our plan is to keep this Report in its familiar form enabling us to communicate our activities to you. ,

One of the greatest changes has been the departure of Jamie Parker, the Museum's first employee and developer of all the exhibits. She left in August to get married and join her husband in Geneva. In her four years with the Museum, she used her photographic memory to conceptualize exhibits. Jamie had an intuitive feeling for the artifacts and how they could be exhibited even though her education was in art history not computer science. While with the Museum, she cataloged and put three times as much in the warehouse as we had on the floor. One of Jamie's last chores was to organize our yard sale.

The yard sale allowed Jamie to weed our "warehouse." In her first years, she accepted everything because that was her job. The Museum ended up warehousing a number PDP-12s, 338 display systems and PDP-6s. Since Jamie knew what was what and what was best, she selected the items to sell, thus cutting down our storage costs and providing the members with a good day of poking through old junk and taking apart computers. The cover photo is a tribute to Jamie: one of the yard sale customers is carrying off his loot and inspecting the display of the ENIAC, an exhibit put together by her.

A new crew of exhibit and archives employees will help us plan the space for Museum Wharf. Meredith Stelling has taken over as the Coordinator. She has been with the Museum for a year handling publications and archives. Meredith, Greg Welch and Bill Wisheart are the main exhibit staff and will be joined in January by Oliver Strimpel, on leave from The Science Museum in London.

In September, the new space at the Wharf seemed vast and barren, except for chalk marks on the floor indicating where the new exhibits would be positioned. But the space is already beginning to fill out with two truck loads of the SAGE (30,000 pounds), an IBM 1401 card system and a collection from the University of Illinois.

Reviews of exhibit plans started in September. Sheila Grinnel, developer of ASTC's travelling "Chips and Changes" exhibit, Bruce McIntosh, a designer, and Paul Tractman, senior editor, The Smithsonian, spent a day consulting on the proposed organization. Then on October 13th, board members Brian Randell and consultant Dick Eckhouse reviewed the next iteration.

Successive refinements bring our plans in line with reality. The SAGE system will form the fulcrum of the exhibits leading into the computer generations on one floor, and backward in time to the revolutionary one-of-a-kind computers on the other. The process of moving has now started and the enormity of the task ahead is clear. But the team is together and progress can be seen.

Gwen Bell
Director

Harvard Mark III. Magnetic drum storage was pioneered on the Harvard Mark III. the drum rotated at about 3,600 rpm and its randon access time was 17 micro{milli}seconds. By modern standards that was quite slow, however, it was the only way to have moderately priced memory in any quantity in the early 1950s. With its tapes and plugboards the Harvard Mark III covered 40 square feet, and was one of the first hardwired assemblers that transformed mathematical symbols into machine code.

In 1955, Martin Weik compiled a "Survey of Domestic Electronic Digital Computing Systems," providing a remarkable snapshot of the computer population. The survey briefly describes and gives specifications for about 100 different machines in existence as of December 1955.

Weik's inventory supplied the base to compile a fundamental reference for collecting and research at The Computer Museum. Records for each machine were gathered from contemporary historical accounts in recent books and journals, operating manuals, and in some cases the machines themselves. Then the findings were checked against those appearing in Weik's original survey.

This research was done by Paul E. Ceruzzi, assistant Professor of History at Clemson University with the aid of Rod McDonald of Rider College, and Greg Welch of The Computer Museum. Different specifications and descriptions have been given to the same machines over time for various reasons. Rather than arbitrarily selecting one description, the data was collected and explained.

These differences occurred for a variety of reasons. Specifications liven in one account often do not agree with those given in another, because a computer's characteristics usually changed from the time of its early design to its final days of operation. The characteristics of some were entered after they had been redesigned and rebuilt, (e.g. SEAC) and others before such redesign (e.g. Johnniac). Nomenclature was also a problem-one manufacturer's "rapid access registers" might be another's "accumulators"these differences were reconciled through research.

Different metrics were often used for speed: the time it took to fetch a number from memory in a drum machine may have been given as the fastest possible, the slowest possible, the average, or the fetch time using optimum coding techniques. A time frame for each machine was established to provide a subjective though reasonable assessment of its historical significance.

The first phase of the survey is complete: the data is stored on disks, and printouts are available for scholars. The next phase is to build the collection, define additional research topics and to develop a very accurate map of computing up to 1955.

Gwen Bell

What did computing look like during its "first generation"-the time from the dedication of ENIAC in 1946 to the mid-fifties?

The variety was astonishing. Experimental one-of-a-kind computers, each with its own unique character, ruled, even though most incorporated vacuum tubes and drum memories, stored programs and data internally, and communicated via Flexowriters.

While most were built with vacuum tubes, many also used relays and crystal diodes.

For memory, they relied on delay lines, cathode ray tubes, drums, magnetic tape loops, paper tape, punched cards, magnetic wire, and toward the end of the period, magnetic cores.

For input and ouput, they used teletypes, punched cards, other paper tape readers, and CRT displays as well as Flexowriters.

Their sizes ranged from that of a small desk to several large rooms full of equipment bays with consoles one could walk into. And their speed ranged from one to tens of thousands of operations per second.

Preliminary Findings: Technology

Most first-generation computers did use vacuum tubes, but not all in the same way. After ENIAC's dedication, designers saw the advantage of tubes for speed, but sought to minimize their number. Those computers used fewer tubes in their circuits, and thus were more reliable and compact. Solid state diodes, not tubes, performed logical operations. This was pioneered in SEAC in 1950, after which only a few computers, such as the Circle and Monrobot, continued to use tubes for logic as ENIAC did.

Between 1946 and 1955, at least a dozen relay computers were built, an indication that some designers did not agree with the prevailing view of the superiority of vacuum tubes. One such person was Howard Aiken, who on visits to Continental Europe in the 1950's influenced the choice of relays for several computers. Konrad Zuse's computer company also produced a line of successful relay computers installed mainly in Continental Europe. Some of the relay computers, like the Bell Labs 5 and 6, were based on se quence calculator designs of a decade earlier. Others, like ERAs "Abel" and the British "ARC," were designed along the lines of stored-program electronic computers, but used relays to save money or to get a prototype working quickly.

By late 1955, a few transistors already were finding their way into computer circuits: in Bell Labs' TRADIC, the IBM 608 Calculator, and perhaps one or two others.

Memory

A wide range of memory devices were used in first-generation computers. None of the mass storage techniques available in the early 1950's was clearly superior; the choice always involved a trade-off of access time versus reliability. This unsettled situation persisted until the end of this period, when the magnetic core memory was perfected.

Drum from the English Electric Deuce. Built in 1957, the Deuce drum stored 8K x 32-bit words on 256 track: of 32 words each. It measured four inches by six inches; most first generation drums were eight to 20 inches in diameter and two to four feet in length. The Deuce drum is on exhibit at The Computer Museum.

The drum was by far the most common memory device. A third of the stored- program computers used it for their primary memory, and most of the others used it for secondary storage. The most popular of the early computers, the IBM 650 with several thousand installations, was a drum machine. A drum is fundamentally an electromechanical device; its reliability, high capacity, and relatively low cost made it the most successful medium.

The designers of the first stored program computers had high hopes for purely electronic, parallel memories. Williams tubes were widely available, but their performance was erratic. Developed in Manchester, England in 1948, they were used on the IBM 701 and in a variant form on the Whirlwind.

John von Neumann, unsatisfied with their reliability, contracted with Jan Rajchman at RCA to produce a electronic, parallel memory, but von Neumann had to make due with Williams tubes on the IAS machine and its offspring in Los Alamos and elsewhere. Finally Jan Rajchman's Selection was completed and installed, but worked well on only one machine, the Johnniac at the Rand Corporation.

SWAC Williams Tube. The Williams tube was invented by Sir Frederick Williams at the University of Manchester in 1948. It was the first purely electronic parallel memory, but it was unreliable. Although magnetic- core memories superseded the Williams tube by 1954, the Williams tube was still faster than drum memory and delay lines. Unlike the earlier version of the Williams tube, the Williams tube from the SWAG (Standards Western Automatic Computer) was more compact and featured higher reliability.

It enabled the calculator from the SWAC to fully utilize the speed of the Williams tube memory by completing arithmetic operations in a few microseconds. Instead of handling numbers as a train of pulses, there were parallel circuits in the SWAC that transfered numbers almost instantly. This transferring of numbers in parallel made it possible to do computations at many times the speed of serial computers. The SWAC was the first Williams tube computer to be completed in the United States. Its rate of success was also dramatic, producing useful results seventy percent of the time. The Williams tube from the SWAC is on exhibit at The Computer Museum.

IBM 650. The IBM 650 was the most widely used first-generation computer. Hundreds were delivered between 1955 and 1959. Although the 650 teas faster than other magnetic drum computers, its high success rate was a result of a well-integrated, punchedcard input and output and its adapt ability to existing punched-card systems.

Huskey Lecture. Harry Huskey giving a lecture next to his Bendix G15 at The Computer Museum in December 1982. He said: "In 1952 and 1953 while at Wayne University (Detroit), 1 dusted off the ideas and designed a computer which the Bendix Corporation elected to build, the Bendix G15. The memory was a magnetic drum with separate read and write heads. All information was read, erased and rewritten every drum rotation just like the mercury delay lines. This gave some technical advantages-the read heads and the write heads could each be optimized for their functions."

Some 15 first-generation computers used mercury delay lines for their main memory. The delay line was more reliable but slower than the Williams tube, while it was less reliable but faster than a drum. The UNIVAC's delay line memory, for example, could access a number in 400 microseconds, compared to 25 microseconds for IAS's Williams tube store, and 2,500 microseconds for the IBM 650 drum. Delay line computers included many historically significant machines: the Cambridge EDSAC, the EDVAC, the SEAC, the Pilot ACE, and the UNIVAC. A few other machines, such as the Pegasus, used magneto-strictive delay lines.

The development of magnetic core memory finally gave computer designers a memory that was reliable, fast and parallel, but expensive at the outset. In 1953, core memories were installed on the Whirlwind computer at MIT and the ENIAC at the Ballistic Research Lab. By 1955, only two commercial computers, the RCA BIZMAC and ERA 1103A, used core memory. Without the new manufacturing technology to build cores, manufacturers of machines based on drums, delay lines, and other devices continued to plan and build these architectures until the price of core fabrication fell.

Harry Huskey, who designed a superior version of the Bendix G15, says: "Bendix made more than four hundred of the G15's- in fact the fittings on number 400 were gold plated. Bendix did plan a transistor version of the G15 but the declining costs of magnetic cores and their improved reliability marked the end of the cyclic memory computers."

Input/Output

Nearly all first-generation computers used a Flexowriter or comparable electronic typewriter with a paper tape reader attached for both input and output. The Flexowriter was simple and rugged, but slow. Photoelectric readers, pioneered on EDSAC and quickly adopted in the United States, read paper tape 20 times faster. A photoelectric reader could input data at 120 characters per second (cps) instead of the six cps that a mechanical reader could handle.

Other computers used punched cards or teletype. The CRT display, so familiar to modern computer users, first appeared on one or two experimental computers like the Whirlwind, and finally on a commercial computer, the ERA 1103, in 1955.

Almost from the beginning of this era, designers recognized the advantages of magnetic tape as a medium for bulk input/output, but tape was slow in being adopted. The use of metallic tape was pioneered on the UNIVAC while the SEAC used magnetic wire mounted in compact cassettes for off-line storage.

Size The smallest stored-program computer was probably one built by Hughes Aircraft for aircraft guidance and control. It measured about two feet by one foot, used a drum memory, and was installed aboard a C-47 airplane in 1953. The largest was perhaps the Whirlwind, which occupied 55,000 square feet. Other large-scale installations that could claim the honor of "biggest" include the IBM 701, the RCA BIZMAC, and the Harvard Mark II, which filled a large room at the Naval Proving Ground in Dahlgreen, Virginia.

Commercial drum computers were generally quite small, ranging in size from that of a small desk to several large cabinets. The cost of development and construction ranged from a few thousand dollars for a prototype Circle Computer (surely the cheapest) to several million for Whirlwind. However, the Whirlwind was more than a single computer, it was an ongoing project involving computers, memories and applications programming.

Architecture

Quite a few computers without a stored- program design were produced and sold into the 1950's. The advantages of the stored program design were slow in being accepted, and many companies built computers of both types. ERA, for example, built a "Logistics Computer" in 1952, which incorporated a fixed program for certain types of problems.

Computer Research Corporation built a general-purpose drum computer, the CRC 102, and also produced the popular CRC 101, a special-purpose machine called a Digital Differential Analyzer. The aircraft industry, a big customer for digital differential analyzers, kept the market alive and several companies were the suppliers. Several externally- programmed drum computers installed in Continental Europe reflected the design of Howard Aiken's Harvard Mark III and Mark IV

Of the stored program computers, about an equal mix handled numbers serially, digit by digit, and in parallel, a word at a time. Similarly, they were equally mixed between binary and decimal machines, with some commercial models like the CRC 102 available either as a binary or a decimal machine.

Core Memory Stack. This core memory stack from the Whirlwind, which is on exhibit at The Computer Museum, measures 17 x 10 x 9 inches. Each core memory plane is arranged in an array of 32 x 32 cores. The first core memories were designed by Jay Forrester for the Whirlwind in 1953 at MIT. Computer access time dropped from twenty-five microseconds for tube storage to nine microseconds for magnetic cores.

A wide range of instruction sets also existed, from CALDIC with only a dozen or so instructions, to the RAYDAC with a four-address code and built-in fixed and floating point instructions. When random access core memory replaced serially-accessed magnetic drums or delay lines, the "von Neumann" architecture of binary arithmetic, single- address instructions, and parallel memory prevailed.

Reports by Burks, Goldstine, and von Neumann on the IAS computer discussed the stored-program principle in detail, especially with regard to modifying the address field of an instruction during a program's execution. Several first- generation computers used special index registers to accomplish the same thing. These were called "B-lines" on the Ferranti Mark I, the first machine to use them, and the name stuck. In the United States, the Consolidated Engineering 30- 201 and its descendents had B-lines. Descriptions of computer architectures nearly always mentioned the stored program in connection with indexing. Some descriptions, including one by Alan Perlis, point out that computers with B- lines were superior in many ways to the simpler IAS design.

Programming

The first generation of computers were programed in machine language, typically by binary digits punched into a paper tape. Activity in higher-level programming was found on both the large-scale machine and on the smaller commercial drum computers.

High-level programming languages have their roots in the mundane. A pressing problem for users of drum computers was placing the program and data on the drum in a way that minimized the waiting time for the computer to fetch them.

It did not take long to realize that the computer could perform the necessary calculations to minimize the so called latency, and out of these routines grew the first rudimentary compilers and interpreters. Indeed, nearly every drum or delay line computer had at least one optimizing compiler. Some of the routines among the serial memory computers include SOAP for the IBM 650, IT for the Datatron, and Magic for the University of Michigan's MIDAC.

Parallel memory machines had less sophisticated and diverse compilers and interpreters. Among the exceptions were SPEEDCODE developed for the IBM 701, JOSS for the Johnniac, and a number of compilers and interpreters for the Whirlwind.

Use

The list of computing installations up to 1955 reveals dominance of the military, followed by laboratory and then business use. In 1954, a Magnefile was installed for inventory control at B. Altman & Co. in New York, and a MODAC 404 was used by Reader's Digest for keeping track of subscriptions, but these were exceptions to the rule.

Installations found at air force or army bases often had not just one, but several computers. Though not a "typical" installation, the Ballistic Research Lab at Aberdeen, Maryland illustrates how military agencies commanded the greater fraction of all computing power in the mid- 1950's. It included: ENIAC; a Bell Labs Model V Relay Computer; EDVAC (a stored program, serial computer); ORDVAC (a stored-program, parallel computer); several digital differential analyzers; punched card multipliers; analog computers; desk calculators, and other computing devices of various shapes and sizes.

Conclusion

The "milestones" of the first generation were brought about by many people who continue to be leaders in the field. Grace Hopper worked on the UNIVAC; Maurice Wilkes on the EDSAC; Joe Weizenbaum and Harry Huskey on the Bendix G-15; Gene Amdahl on his dissertation machine, the WISC; Max Palevsky on the Bendix D-12 Digital Differential Analyzer; An Wang on the Wedilog; Ken Olsen on MIT's memory test computer; and Seymour Cray on the ERA 1103.

Computing was about to change rapidly. In the next few years installations jumped to the thousands. Serially-produced, commerciallymanufactured, standardized machines became the rule. Over the years, experimentation has continued, but never with the diversity of ideas about the basic architecture of this inaugural era.

Paul Ceruzzi, with Rod McDonald and Gregory Welch.

A manufacturing process for core memories was developed by Lincoln Labs in 1952. Core memories were always strung by hand, and production of the first cores was complex and expensive. The following picture story is from the unclassified manual, Ferrite Cores For Computer Memories. These cores were used in the Whirlwind and the Memory Test Computer.

1. Core Pressing. After five days of getting the material ready for making cores, core pressing was done automatically by a Stokes press which was capable of 60 pressing operations per minute.

2. Dimensional Check. The machine die and the weight of the pressed cores had to be continually monitored to insure maximum uniformity of core size. Before each press run a dimensional check was made with a tool- maker's microscope in order to assure quality control.

3. Firing. Firing was the most critical operation of core production. The firing temperature was approximately 2400 F, and elaborate controls were necessary to maintain the correct temperature.

4. Cooling. After the cores left the tun- nel of the kiln, they were still at an elevated temperature of 500 F Cooling took place quickly in the open air, and then the cores were ready for counting and electrical testing.

5. Electrical Testing. Core drivers helped in electrical core testing. The cores, which were temperature sensitive, were tested at a uniform 25 C. The temperature was controlled by core handlers in temperatureregulated boxes or airconditioned rooms.

6. Pulse Testing. A sample of 50 cores from each lot was used for hysteresis-loop measurements. The test equipment for pulse testing and semiautomatic selection testing consisted of an electronic core counter, an evaluation pulse tester, fully automatic and semiautomatic core testers, and a plane tester.

7. Evaluation Test. Evaluation pulse testing was performed on a sample of 20 cores. The data obtained from the hysteresis-loop tests and the evaluation pulse tests yielded important information concerning the performance of core lots in a memory. It was at this step where lots could be rejected on the basis of the evaluation test.

8. Stringing. After core testing had been completed, the magnetic cores which had been accepted were hand strung into memory planes of 4096 cores each.

9. Final Test. The cores in the plane were then given a final pulseresponse test in order insure their acceptability. If damaged, removal of defective cores from a plane was easy at this stage.

10. Finished Product. The final operation in the construction of a plane was the insertion of the inhibit winding and sensing wire which linked all the cores in the plane.