R. Michael Hord

The most recent addition to the Museum's Hall of Super Computers is the Illiac IV, an advanced computer designed and developed at the University of Illinois in the mid-1960's by Professor Daniel Slotnick and sponsored by the Defense Advanced Research Projects Agency. On loan from NASA Ames where it was delivered in 1971 and used in computational fluid dynamics research, the Illiac IV exhibit at the Museum includes the central unit, the processing unit cabinet with eight processing units and two Burroughs disks. The following article is excerpted from R. Michael Hord's Illiac IV The First Supercomputer, published in 1982 by the Computer Science Press. The book is available at the Museum store. (Reprinted with permission from the author.)

Project History

It was during the spring of 1970 that the Illiac IV computer project reached its climax. Illiac IV was the culmination of a brilliant parallel computation idea, doggedly pursued by Daniel Slotnick for nearly two decades, from its conception when he was graduate student to its realization in the form of a massive supercomputer. Conceived as a machine to perform a billion operations per second, a speed it was never to achieve, Illiac IV ultimately included more than a million logic gates-by far the largest assemblage of hardware ever in a single machine.

Until 1970, Illiac IV had been a research and development project, whose controversy was limited to the precise debates of computer scientists, the agonizing of system and hardware designers, and the questioning of budget managers. Afterward, the giant machine was to become a more or less practical computational tool, whose disposition would be a matter of achieving the best return on a government investment of more than $31 million.

Illiac IV was funded by the U.S. Department of Defense's Advanced Research Project Agency (ARPA) through the U.S. Air Force Rome Air Defense Center. However, the entire project was not only conceived, but to a large extent managed, by academicians at the University of Illinois. Finally, the system hardware was actually designed and built by manufacturing firmsBurroughs acted as the overall system contractor; key subcontractors included Texas Instruments and Fairchild Semiconductor.

Perhaps the greatest strength of Illiac IV as an R&D project, was in the pressures it mounted to move the computer state of the art forward. There was a conscious decision on the part of all the technical people involved to press the then-existing limits of technology. Dr. Slotnick [. . .] made it clear to his coworkers that the glamour and publicity attendant to building the fastest and biggest machine in the world were necessary to successfully complete what they had started.

Design History

The story of Illiac IV begins in the mid-1960's. Then, as now, the computational community had requirements for machines much faster and with more capacity than were available. Large classes of important calculational problems were outside the realm of practicality because the most powerful machines of the day were 'too slow by orders of magnitude to execute the programs in plausible time. These applications included ballistic missile defense analyses, reactor design calculations, climate modelling, large linear programming, hydrodynamic simulations, seismic data processing and a host of others.

Designers realized that new kinds of logical organization were needed to break through the speed of light barrier [186,000 miles per second] to sequential computers. The response to this need was parallel architecture. It was not the only response. Another architectural approach that met with some success was overlapping or pipelining wherein an assembly line process is set up for performing sequential operations at different stations within the computer in the way an automobile is fabricated. The Illiac IV incorporates both of these architectural features.

The Illiac IV is the fourth in a series of advanced computers from the University of Illinois; its predecessors include a vacuum tube machine completed in 1952 (11,000 operations per second), a transistor machine completed in 1963 (500,000 operations per second) and a 1966 machine designed for automatic scanning of large quantities of visual data. The Illiac IV is a parallel processor in which 64 separate computers work in tandem on the same problem. This parallel approach to computation allows the Illiac IV to achieve up to 300 million operations per second.

The logical design of the Illiac IV is patterned after the Solomon computers. Prototypes of these were built in the early 1960's by the Westinghouse Electric Company. This type of computer architecture is referred to as SIMD, Single Instruction Multiple Datastream. In this design there is a single control processor which sends instructions broadcast style to a multitude of replicated processing units termed elements. Each of these processing elements has an individual memory unit; the control unit transmits addresses to these processing element memories. The processing elements execute the same instruction simultaneously on data that differs in each processing element memory.

In the particular case of the Illiac IV each of the processing element memories has a capacity of 2,048 words of 64 bit length. In aggregate, the processing element memories provide a megabyte of storage. The time required to fetch a number from this memory is 188 nanoseconds, but because additional logic circuitry is needed to resolve contention when two sections of the Illiac N access memory simultaneously, the minimum time between successive operations is somewhat longer.

In the execution of a program it is often necessary to move data or intermediate results from one processor to another. One way of regarding this interconnection pattern is to consider the processing elements as a linear string numbered from 0 to 63. Each processor is provided a direct data path to four other processors, its immediate right and left neighbors and the neighbors spaced eight elements away. So, for example, processor 10 is directly connected to processors 9, 11, 2, and 18. This interconnection structure is wrapped around, so processor 63 is directly connected to processor 0.

Illiac IV functional diagram.

This routing diagram shows schematically the neighbor-to-neighbor linkages which form the 64 processing elements (PE) into a ring, as well as the connections of the PE's eight apart such that data can bypass intermediate PE's when the distance to be covered is large.

The other major control feature that characterizes the Illiac N is the enable/ disable function. While it's true that the 64 processing elements are under centralized control, each of the processing elements has some degree of individual control [provided] by a mode value. For a given processor [it] is either 1 or 0, corresponding to the processor being enabled "on" or disabled "off". The 64 mode values can be set independently under program control, depending on the different data values unique to each processing element. Enabled processors respond to commands from the control unit; disabled elements respond only to a command to change mode. Mode values can be set on specific conditions encountered during program execution. For example, the contents of two registers can be compared and the mode value can be set on the outcome of the comparison. Hence iterative calculations can be terminated in some processors while the iteration continues in others, when, say, a quantity exceeded a specific numerical limit.

In addition to the megabyte of processor element memory, the Illiac IV has a main memory with a sixteen million word capacity. This main memory is implemented in magnetic rotating disks. Thirteen fixed head disks in synchronized rotation are organized into 52 bands of 300 pages each (an Illiac page is 1,024 words). This billion- bit storage subsystem is termed the Illiac IV Disk Memory or 14DM. The access time is determined by the rotation rate of the disks. Each disk rotates once in 40 milliseconds so the average access time is 20 milliseconds. This latency makes the access time about 100,000 times longer than the access time for processor element memory. The transfer rate, however, is 500 million bits per second.

This memory subsystem, the input/ output peripherals and the management of the other parts of the system [were] under the direction of a Digital Equipment Corporation PDP-10 conventional computer. A Burroughs B-6700 computer compiles the programs submitted to the Illiac into machine language.

This Burroughs Disk exhibited at The Computer Museum is only one of the thirteen synchronously rotating fixed head disks that comprised the 16M word main memory of Illiac IV.

Circuitry

Initial plans for Illiac IV circuitry envisioned bipolar emitter-coupled logic (ECL) gates capable of speeds of the order of 2-3 ns. The ECL circuits were to be packaged with 20 gates per chipa level of complexity that later would be called medium scale integration. [Texas Instruments was chosen as the subcontractor for these circuits.] Illiac IV initial specifications called for a 2,048-word, 64- bits-per-word, 240-ns cycle time memory for each of its processing elements. In 1966, the only technology that seemed to meet the requirements was the thinfilm memory. At that time, a few developmental semiconductor memory chips were being studied, but no computer manufacturer would yet consider them seriously for main memory use.

[However, a change] to smaller ECL circuit chips proved a death blow to thin-film memory. When the smaller chips' requirements for added space on circuit boards and interconnections were taken into account, it turned out that there was not enough room for the smallest feasible thin- film memory configuration. Strangely, the failures of the ECL circuits and thin-film memories also set the stage for a brilliant hardware success: Illiac IV was to be one of the first computers to use all semiconductor main memories. Slotnick chose Fairchild as the semiconductor memory subcontractor.

Called for were 2,048 words (64 bits/ word) of memory for each of the 64 Illiac processing elements, a total of 131,072 bits per processing element. The memory was to operate with a cycle time of 240 ns and access time of 120 ns. Slotnick recalls the development proudly: "I was the first user of semiconductor memories, [and] Illiac IV was the first machine to have all-semiconductor memories. Fairchild did a magnificent job of pulling our chestnuts out of the fire [. . .] the memories were superb and their reliability to this day is just incredibly good."

Results

The end results this pioneering [project] had on computer hardware were impressive: Illiac IV was one of the first computers to use all semiconductor main memories; the project also helped to make faster and more highly integrated bipolar logic circuits available; in a negative but decisive sense, Illiac IV gave a death blow to thin- film memories; the physical design, using large, 15-layer printed circuit boards, challenged the capabilities of automated design techniques.

Jay Patton

Jay Patton, Manager of Installation Planning at Burroughs Corporation, coordinated the initial set up of the Illiac IV at NASA Ames in 1970 and came to the Computer Museum in December to reinstall it. Comments made during his gallery talk follow, conveying an idea of the massive size of the computer and its capabilities.

"In 1970, ARPA (Advanced Research Project Agency) determined that the Illiac IV parallel architecture could best be tested in an environment that had research programs requiring the potential power of the machine. A new wing was built to house Illiac IV It took one month to disassemble the unit from our testbed in Paoli, which had 100 tons of air conditioning built into it. The computer totalled 53' in length, and took 11 40' vans to house it, weighing 99 tons. One truck alone had only power supplies in it.

Illiac IV had a total of 11,739 pc boards. You can imagine what the spares problem was, and projecting what the failure rate would be. There was a group of people who did nothing but work on equations such as the mean time between failure rate. Inside each pc board were 12 layers of pc material. Each of the boards is coded with a letter code at the top, and a number code at the bottom. You cannot physically put a wrong board in the wrong spot.

From the control unit to each one of the processing extenders (which is a separate computer all in itself) there were belted cables in the back running the length-in one unit alone, there's over 85 miles of cable. The cooling air was 45,000 cubic feet of air per minute. It used over a half a megawatt of power. When we turned it on, we had to do it by sections, not all at once.

The disk system had a transfer rate of 500 x 106 bits per second, when you had two disks running in parallel. The parallel concept for Illiac was used to bypass the speed of light limitation, because you could do 64 additions, subtractions, or multiplications simultaneously. The maximum speed intended by the design was 200 x 106 operations per second; it actually achieved an effective speed of over 60 million instructions per second on some applications.

You can imagine the traumatic experience I had when I compared the 1970 National Geographic photograph of the Illiac IV and the recent National Geographic (October 1982) photograph of Illiac being torn apart and having an autopsy done on it. Then you can imagine how I felt when a call came from Marcie Smith [NASA Ames] to tell me that the Computer Museum was going to ask me to help put Illiac back together - she asked me to control my laughter. The computer really was the dinosaur of the sixties. What you see in the museum are the skeletal remains of a once-proud unit."

As the Museum has evolved, it has established a close relationship with its members and friends-engineers, computer scientists and history buffs -who are responsible for many donations. Often they refer the department to an available artifact, or make a donation from their own collections. When an object is offered to the collection, they act as curators, illuminating the importance of the acquisition, and sometimes preparing text for an exhibit. While not actually employed by the Museum, they act in its behalf as the experts in computing technology.

The collections policy outlines the process of acquiring artifacts. A deaccessioning clause clarifies to donors at the piece they donate today may not always be part of the permanent collection for reasons of space, a lessening of historical value, or duplication. The deaccessioning policy contributes to our habitual "squirrelling" of artifacts; the donor has agreed that the piece may be taken off the catalog listing and traded with another Museum for another piece, or its parts, if it is a duplicate, could be sold to other collectors through the Museum store. Very little is ever scrapped.

After determining the significance of an acquisition, the artifact is pursued. Most acquisitions require a little detective work and some phone calls to ensure shipment, while a few others are more elusive. In June of 1981, Greg Mellen from Univac in St. Paul called to say he had located a part of the 1956 NTDS (Naval Tactical Data System) in an office in St. Paul. Seymour Cray was the director of development for the NTDS project, the first automated command and control system within the Navy. Initial letters were mailed and calls made to guarantee the CP-642's release to the Museum. It was not until June of 1982 that the paperwork arrived in a large package from the Navy. In order to clear the CP-642, the Navy needed several letters of intent and background from the Museum, all of which had to be notarized, establishing ourselves as a reputable agency for the preservation of computing history. Another six months later, after several follow-up calls, the Navy wrote that they needed a statement from the state of Massachusetts that the Museum was, indeed, tax exempt. In January, 1983, the Navy informed us that the CP-642 was in an office in St. Paul, presumably not due to be shipped until April, 1983, almost two full years after the process started.

When an acquisition arrives at the Museum, it is checked for damage and suitability for immediate display (this usually involves climbing through 40 foot trucks, removing quilted covers and making some on-the-spot decisions). When the nine tons of Illiac IV arrived completely disassembled on the shipping dock-with no Illiac IV experts available in Marlboro-most of the machine, with the exception of the skeleton and several processing units, was sent to storage. Through a contact at NASA Ames, we located Jay Patton at Burroughs, who had originally installed the computer at NASA. Jay spent two days at the Museum, retrieving what had been mistakenly shipped away, and piecing Illiac back together.

A sequential identification number is assigned, with the last two digits representing the year of the donation. Each artifact is catalogued by manufacturer, serial number, physical description, date, and place in computing history, donor name and address, special characteristics, and a brief explanation of the artifact. It is cross referenced to its archival documentation if any exists. An acknowledgement letter, collections policy and receipt for tax purposes are sent to the donor for his records.

The Museum's archives and library began with active solicitation of documentation of collected machines. The understanding was that original manuals would be worthwhile research materials in years to come. This has evolved to the point where relevant photographs, theses, books, films and videotapes are also collected. In collecting archival material, the leads of the Museum's friends and donors are investigated. Contacts for archival material include libraries who wish to donate surplus material from their shelves, and individuals going through personal document collections. On the night of Maurice Wilkes' "Pray, Mr. Babbage" premiere, Mary Hardell donated volume one, number one of the ACM Journal and Bill Luebbert donated a full set of the videotapes from the Los Alamos computer conference. A new acquisition, such as Illiac IV, precipitates outside interest and donations. People who worked on the machine or at the University of Illinois are going through file drawers and attics to collect supplementary materials for us.

This summer's Report lists the whole collection by appropriate categories. Only one-third of the permanent collection is exhibited, with all material that is in storage documented and available for research purposes. As the collection and exhibitions grow, the ratio will probably remain the same. Some parts of the collection are better developed than others, but by looking at what has been collected, it is easier to determine what should be pursued. The collection's growth reflects a new understanding of the importance of preserving computer history, and the many milestones within the computer industry. Active involvement from members, friends and experts in certain areas of computing technology is an invaluable resource in this development.