The Remington Rand Univac LARC

Chuck Cole

by Charles Cole

The Remington Rand Univac LARC (Livermore Advanced Research Computer) computer system was delivered to the Lawrence Livermore National Laboratory (then called the University of California Radiation Laboratory) in March of 1960. There had never been anything built on the scale of this computer. It was massive both in physical size and computational capability. A new building (now numbered B-117) was constructed just to house it. Physically, it consisted of four major cabinets. Each cabinet was approximately twenty feet long, four feet wide, and seven feet tall. In addition there was an array of twelve floating-head drums, each approximately four feet wide, three feet deep and five feet tall. There were eight tape units that used metallic tape (each tape weighed about ten pounds), a punched-card reader, and a large printer. Operation of the system was controlled from a central console with lots of flashing lights, switches, digital readouts, and pushbuttons. Teletype units (with paper tape readers / punches) provided direct communications with the computer. The control console also provided an array of toggle switches to feed direct commands to the system.

The four hardware cabinets consisted of: the I/O processor unit (one cabinet) where all information destined to and from the drums, tapes, printers, console, etc. was routed and controlled; the computing unit (one cabinet, where computational activity occurred); and memory (two cabinets, each with 16K of ferrite core memory). Later, a third memory cabinet was added to expand local memory.

Storage devices were state-of-the-art for the day. The drum units were rotating cylinders coated with a magnetic material. Storage density was 450 binary digits per inch. Each cylinder was 27-1/2 inches long and 24 inches in diameter, providing for a storage capacity of 250,000 LARC words (12 decimal digits each) per drum. Twelve drums were included in the Livermore system. The drum surface was divided into 100 circumferential bands. Each band was divided into 25 sectors (a sector could hold 100 LARC words). A data transfer could begin at any sector boundary. Each band was made up of six parallel tracks; five for the digit of information, and the sixth for a permanently recorded sector address. The single floating head could be positioned over any band for reading or writing. Moving the head across the drum from one band to the next took 70 milliseconds. If it was to be repositioned more than one band, the speed increased to 50 milliseconds per band. Reversing the direction of travel took 10 milliseconds. There were two independent controllers (called synchronizers) for read operations, and one for write. With the independent controllers one drum could be repositioning its head, while a different unit was reading data, thereby significantly improving data recovery time. A complete band of information could be transferred between the drum and main memory in about 90 milliseconds after positioning was completed.

The magnetic tape of choice on the LARC computer was made of metal. There were eight magnetic tape units on the LARC, although the system was designed to support up to forty units. Standard tape reels were either 1500 or 2400 feet in length (8-inch diameter, or 10-inch diameter reels). A 2400-foot reel would theoretically hold approximately 600,000 LARC words. The units could record at densities as high as 250 characters per inch, although on the LARC delivered to Livermore this density was limited to 208 characters per inch. Data was recorded in blocks of ten words each. There was a space between each recorded block on the tape so, in actual practice, a single tape would hold approximately 450,000 LARC words. Data was recorded in 8-track parallel mode; 7 tracks representing the character, and the eighth a control pulse. The tape moved at 100 inches per second. Full acceleration or deceleration took about 5 milliseconds.

The printer provided a maximum of 130 characters per line (ten characters per inch). It supported a set of 51 alphanumeric characters. A single line could be printed in approximately 80 milliseconds with a speed of 600 single-spaced lines per minute. Quite a respectable print rate considering that the console printers (there were two on the LARC system, called Flexowriters) provided only 10 characters per second. The paper feed mechanism moved the paper a line at a time, and each advance took 20 milliseconds.

Computationally, the LARC was a marvel or, maybe, a monster. Arithmetic was performed in decimal mode, which was the custom at Univac at the time (the Univac I also performed decimal arithmetic). The LARC employed twelve decimal digits. A five-bit register represented the numerical value in each digit. Arithmetic was performed using these coded digits in a dizzying array of temporary storage registers that saved the initial integer values as well as partially computed results. There were storage registers, shift registers, and result registers that could store information for repetitive calculations that would ultimately yield an answer. Calculation speed was also dizzying: a 12 x 12 digit addition or subtraction could be accomplished in 4 microseconds, while a 12 x 12 multiplication would complete in twelve microseconds. Division took a bit longer.

The LARC offered arithmetic to 22 decimal digits, and performed floating-point arithmetic using the uppermost two decimal digits to represent the power of the number in the remainder of the field. Thus, floating point operations offered values of ten to the power of plus or minus ninety-nine.

The LARC was the first "supercomputer," and demonstrated the exciting promise of computing as a scientific tool. It also was instrumental in demonstrating why binary arithmetic (rather than decimal) was a far superior option for computers. But, in those days, such issues were still being sorted out.

The LARC arrived at the Lawrence Radiation Laboratory in 1960, after a three- week truck ride from Philadelphia (a caravan of five 18-wheelers crossing the country in the wintry conditions of February and early March). Accompanying the hardware was a crew of approximately forty installers from Remington Rand. These included electricians, plumbers, air conditioning experts, computer engineers, technicians and so on. It took more than two months to complete the physical installation and another two months to get the computer working again. Following installation, there was the nerve-wracking task of passing a rigorous "acceptance" test. This test required that the computer satisfy certain pre- specified performance criteria before the Laboratory would accept it. An engineering / technician work crew of four individuals remained at Livermore for the additional seven months it took to pass the acceptance test.

Following acceptance, the computer was placed into production service, which meant that programmers and scientists could schedule "run time" on the system. This generally took the form of a programmer loading his problem on the system and then sitting at the console to run and monitor its execution. This was often a most frustrating activity as three separate significant factors conspired to weaken the likelihood of success: 1) The programmers were only beginning to learn how to use a computer for their very complex problems; 2) The operating system was newly developed by Laboratory Computation staff and still suffered occasional "glitches," and, 3) The hardware was temperamental. While most operating efforts were successful (generally yielding an acceptable result or some good clues as to what was going wrong), not all were clear. Fortunately, the system designers had constructed the operating console to allow for extensive bug shooting.

On infrequent occasions, major hardware modifications were required. The computer would be taken out of service and engineers and technicians would install the necessary upgrades. This often took the form of rewiring the backplane of the computing unit. Although the LARC was the first computer to utilize printed circuit technology (and, for that matter, transistor technology on a large scale), such processes were at a very rudimentary stage. Printed circuit interconnections existed only within a given circuit board. All connections between printed circuit boards were hardwired. This meant that in the backplane there was a wire for every point-to-point connection in the computer. Literally thousands and thousands of wires, all placed in as direct a line as possible from source point to destination point. The end result was a wiring mesh of many thousands of wires layered on top of each other, all on top of the backplane connection points. In places the stack of wires reached a depth of six to eight inches. A modification would require that some of those wires be removed or rerouted, and new wires installed. There was a unique set of tools for this purpose. A long hollow tube, with a light at the end and a magnifying glass (called a floroscope) was used by the technicians to burrow through the wiring mesh to the backplane to identify appropriate wires to remove, and to locate holes in the backplane where new wires were to be installed. Having found the appropriate wire or hole, the technician would then use long-neck pliers to remove a wire connection, or a long screwdriver-type tool modified for the purpose, to forcefully insert a new connection. This process required that the technician peer through the floroscope and guide the other tools to their proper position and execute the required action. This was a very tedious operation and could become quite time consuming. It was also fraught with obvious peril. An incorrectly removed or installed wire would not only prevent the new function from performing as intended, but would very likely create a problem with an operation that was functioning perfectly before the modification. Merely burrowing through the deep stack of wires would sometimes unintentionally dislodge other wires leading to further problems. The technicians most skilled in these modifications were very popular at the time.

Still and all, the LARC worked amazingly well, considering the nature of its radical new high-end computer design in the very early days of computer design and development. As mentioned above, the hardware was constructed of all discrete components: transitors, resistors, capacitors, and such. Maintenance of the individual circuit boards was performed at the component level. Individual transistors were replaced when faulty, and other electronic circuitry was repaired at that same micro level. Finding a failed part was quite a challenge as the fault had to be traced to the component level for repair. To aid in troubleshooting failures the printed circuit boards could be mounted on "extenders" and reinserted in the computer. This allowed observation of the inner working within the board circuitry. Unfortunately, extending the circuit board added several inches of length to the signal paths and, on rare occasions, would introduce a problem quite different from the one being analyzed.

But, despite all its flaws and shortcomings, the LARC was a useful new tool employed in the quest for advances in the scientific challenges of the Lawrence Radiation Laboratory. Livermore computer programmers developed the software to control data storage on the massive drum units. This experience proved most valuable in later efforts at developing software to control larger, more modern large-scale storage systems. Likewise, developing the software to control system functions was important training for the computer systems programmers that would later develop the operating systems for the Control Data computer line, and later the Cray computers. Perhaps most important was the experience garnered by the scientists who learned how to use the power of computing to simplify and visualize the problems they are still addressing today.

There are many milestones of note in the computing history of the Lawrence Livermore National Laboratory. The LARC computer system surely ranks alongside all the other major innovations in the "art" of supercomputing.

Chuck Cole joined Remington Rand Univac in 1959 just as the computer "revolution" was getting started. The Univac I, the first commercially available general-purpose computer, was in full production at Remington Rand. An early serial number had been delivered to Livermore. At the same time the Livermore Advanced Research Computer (LARC) was being built and debugged at eh Univac plant in Philadelphia, with scheduled delivery already several months late. Chuck was assigned to the LARC project, which eventually led to his assignment to Livermore Lab, and subsequently to his employment there (1963). Chuck left the Laboratory in 1972 to work on a project to install, checkout, and make operational the Illiac IV computer system, at Ames Reasearch Center in Mountain View. He worked on that project until 1975, when he returned to Lawrence Livermore.

In his 30+ years at Lawrence Livermore Lab, he enjoyed a widely diverse set of assignments, ranging from maintenance duties on the LARC, to an eventual appointment as a Deputy Associate Director in Computation Directorate. In between, the assignments included appointment as one of the initial members of a team that developed and deployed the "Octopus Network", a forerunner of the distributed computing environment of today. He was Supervisor of a computer operator group that, at one time, numbered more than a hundred. He formed the Laboratory’s first computer security organization and managed that group, and the growing demand for ever more rigorous computer security, for more than 10 years. In this assignment, he formed an important bridge between the DOE Contractor community and the various federal computer security management and oversight organizations, particularly within the Department of Energy. In 1989, he formed the Computer Incident Advisory Capability group to respond to the ever growing threats fostered by the Internet and global connectivity. In 1992, he served a brief appointment to DOE Headquarters supporting their computer security organization. He returned to the Laboratory to accept an appointment as Deputy Associate Director for Operations and Assurance under then Computation Associate Director Bob Borchers. He served in this assignment under a succession of Associate Directors until his retirement in December, 1997.

For more information: