A Few Words About This Picture
THE PHOTOGRAPH ABOVE SHOWS AN INSTALLATION of two IBM 7090 Electronic Data Processing Systems, manufactured by International Business Machines and installed at North American Aviation in Southern California sometime in the early 1960s. In colloquial terms the photo shows a large “mainframe” computer system. The word probably comes from the large metal frames, housed in the cabinets, on which the computer’s electronic circuits were mounted. Entire chassis could swing out, like refrigerator doors, for maintenance. That distinguishes mainframes like the 7090 from today’s minicomputers, which typically have all their processing circuits mounted in a single cabinet, or microcomputers, whose processing circuits are contained on a single chip. Many regard the 7090, introduced by IBM in 1959, as the classic mainframe computer.
The installation shows a number of rectangular cabinets beneath overhead fluorescent fixtures that provide harsh, bright light. The entire room has its own climatecontrol system, probably separate from that of the rest of the building. Typically the computer room was much more comfortable than anywhere else in the building, but few people were authorized to enter it.
The row of cabinets in the left rear corner of the room houses the main processor circuits. These circuits used discrete transistors, mounted and soldered (along with resistors, diodes, jumper wires, inductors, and capacitors) onto printed circuit boards. A spaghetti-like tangle of wires connected the boards to one another. The density of these circuits was about ten components per cubic inch. The use of discrete transistors classifies the 7090 as a second-generation machine; the first generation used vacuum tubes, and the third used integrated circuits. (The integrated circuit, or chip, was invented a few years before this picture was taken. Integrated circuits first appeared in commercial computers in 1965.) Each cabinet in this group contains specialized circuits for different sections of the processor: the various registers, the control circuits, devices for communication with the rest of the computer, and so forth. Today each cabinet would correspond to a corner of a microprocessor chip whose total area might be less than a square inch.
The 7090 was intended mainly for scientific customers, although it could also handle other tasks, including routine business data processing. At this installation North American Aviation was using it for the development of the massive Saturn rocket engines that later in the decade would propel astronauts to the moon and back. The computer could carry out about 50,000 to 100,000 floating-point arithmetic operations per second, making it among the fastest of its day. Comparisons with modern computers are difficult, as the yardsticks have changed, but it was about as fast as the IBM AT personal computer, which used an Intel 80286 processor. Its thirty-six-bit word length made it well suited for scientific calculations requiring many digits of precision, and it had the further advantage of allowing the processor to address a lot of memory directly.
The thirty-six bits gave it an advantage analogous to the telephone company’s providing ten-digit numbers instead of seven-digit ones. With only seven-digit numbers, each person’s phone could not have a unique designation. Similarly, a thirty-six-bit word length allowed the 7090 to have direct access to the large quantities of data generated from instruments monitoring a test of a rocket engine. By comparison, early personal computers were hamstrung by an eight-bit word length. The IBM PC used sixteen-bit words, and the current generation of PCs uses thirty-two-bit words—still shorter than the 7090.
In the center of the photo, between the two men standing in the back of the room, is the control console for one of the 7090s. Another control panel is in the lower left corner of the picture. The console is festooned with an impressive array of blinking lights, buttons, and switches; to those who grew up on the science fiction movies of the 1950s, the 7090 looks like a “real” computer.
What do they all do? The rows of small lights indicate the status of each bit of each register making up the computer’s central processor. In the event of a hardware malfunction or programming error, the operator could stop the machine and read the contents of each register directly in binary. He or she could also execute a program one step at a time, noting the contents of the registers at each step. If desired, the operator could directly alter the bits of a register by flipping switches. Such “bit twiddling” was exceedingly tedious, but it gave the operator an intimate command over the machine that few people since that time have enjoyed. Notice that there is no video display terminal, so common in the personal-computer age. A few of these were in use in the mid-1960s, but most mainframe computers did without them.
Most of the time the operator had simpler tasks to perform. Programmers, none of whom are visible in this picture, were the ones who really “controlled” the computer. Programmers submitted their work in the form of decks of punched cards, passed into the room through a window to preserve the climate control. Programmers seldom, if ever, saw the machine itself. Unless there was a problem, the operator’s job consisted mainly of mounting and demounting tapes, pressing a few buttons to start a job every now and then, inserting decks of cards into a reader, and pulling off paper from the printer and delivering it to the world outside. It was not a very interesting or high-status job, though to the uninitiated it looked impressive. Two card readers are visible at the far right of the picture. These were the primary way one fed data and programs to the machine.
The room on the whole looks austere and serious—no whimsy permitted here. There aren’t any stray pieces of paper or coffee cups lying around, and no posters on the wall either, just a clock. No doubt the room was cleaned up for the photographer, but in practice most mainframe installations did look pretty much like this. Academic and corporate users often showcased their expensive computer installations, and it would not do if the room looked messy. (The photo unfortunately does not show one exception to this rule that was found taped to the wall of nearly every mainframe installation: a crude picture of Snoopy printed out with X ’s and O ’s.)
One of the cabinets to the left of the console housed the computer’s memory, or random-access store. This was made up of thousands of doughnut-shaped cores of magnetic material, hence the name core memory for a mainframe’s primary storage device. A set of wires threaded through each core magnetized it in one direction or another, to represent the binary number one or zero. Another wire read the information. A typical 7090 was delivered with 32,768 words of core, each word being 36 bits long. In modern terms that corresponds to about 150 kilobytes of memory, which is roughly what came with the IBM PC when it first appeared in 1981. Core memory was expensive, not only in the making and testing of each individual core but in the manual labor involved in threading the cores into an array. By the end of the 1960s memories made up of silicon chips began to appear, although one occasionally still hears the term core to refer to a computer’s memory.
A single 7090 installation could be rented for about $30,000 a month or purchased for about $1.6 million, and there are two in this room. At more than $80 an hour (based on a twelve-hour day), it was imperative that the machine never be left idle. Today one typically leaves one’s computer switched on while going to lunch; to waste that many computer cycles on the 7090 would have been scandalous. In practical terms this meant that after programmers had submitted their decks of cards, the operator gathered them in piles called batches and ran them through the machine all at once. A programmer had to wait until his batch was run to get the results. If he then found a mistake or needed to refine his problem further, he had to submit a new deck and wait once more. Programmers longed to speed things up by sitting at the console themselves during debugging, but that was out of the question; it might have required letting the processor sit idle for a few minutes while the programmer mulled over the latest run. Batch operation was another defining characteristic of the mainframe era.
Dominating this photograph are the rows of tape drives. They provide a more capacious, though slower, memory than the core unit. To get at data on a tape, you had to first find the tape, then mount it on the drive, and then spool to the place where the data you wanted was stored. Having about twenty drives connected, as in this picture, made it more likely that the tape you were looking for was already mounted, but the presence of numerous reels of tape piled on the desks in the foreground shows that quite often it was not. Not only was core memory permanently in place, but you could go to any piece of data in core as easily as any other—like using a CD or vinyl record instead of a cassette for music. That’s why such memory is now called “random access.” On the other hand, one reel of tape could hold a lot more than the 7090’s entire core memory.
Just to the right of the console is the printer, on which the man in the checked shirt is leaning. It was a large and noisy device with a moving chain that clattered away, uppercase only, on fan-fold paper seventeen inches wide. Before printing a job, the printer spewed out several mostly blank pages that contained a few lines of cryptic symbols indicating the time the job took, what resources it required, and similar information. Then it printed the results of the computation. These prolegomena were of little use to the person submitting the problem, but they were guaranteed to impress or intimidate all but the most computer-savvy users. These printouts, too, were defining characteristics of the mainframe era.
Perhaps most interesting about the installation shown here is what you cannot see: the software. For a mainframe installation like this one, there were two kinds of programs: those submitted by users and staff to solve problems, and the system software intrinsic to the machine, which controlled its operations. This latter software was exceedingly difficult to write and consumed precious memory, but without it the whole machine was useless. One of the great advantages of batch operation was that it required far less system software than modern time-shared or networked systems, which allow many users to work with the central computer at the same time. True, you did have to wait until the machine got to your deck of cards, but when it did, nearly all the computer’s forces were focused on your problem and yours alone.
Included in the 7090’s system software were an operating system called IBSYS and the programming language FORTRAN, which allowed users without much expertise in computer science to write useful programs on their own. There was little applications software of any kind commercially available. The free-market forces that drive today’s personal-computer software business, ensuring users a steady stream of new products, were absent. There were not enough computers around to make a lucrative market for independent programmers, and in any case IBM strongly encouraged customers to use its own proprietary software.
Most of today’s software products are far superior to anything available on a 1960s mainframe. Consider the electronic spreadsheet, for example. To do the required calculations on the 7090 would have required a customized program in FORTRAN and several expensive runs before the user finally got desired results. Instead users generally had the computer perform raw calculations and assembled the results into useful form by hand. Needless to say, I have found no record of anyone using a 7090, which had no video display and whose standard printer had only uppercase characters, for word processing as we now practice it.
With the advent of personal computers and workstations, one hears that mainframes are dinosaurs that will not survive the turn of the century. Certainly the 7090 has disappeared. There may possibly still be one running somewhere, but it is very doubtful. Unlike the old pumping stations, machine tools, and electric generators often showcased in this magazine, there is little reason to keep an antique computer running. A month’s electric bill alone could pay for a replacement that would have more power. IBM, Amdahl, Unisys, and other companies continue to make and sell mainframe installations, however. These, of course, have much more memory capacity and processing speed than the 7090 shown here, but otherwise they do not look all that different. These survivors toil away at routine but data-intensive tasks like generating a payroll, calculating taxes, and managing an insurance company’s records. Some of these jobs could be done on one’s desktop, but the customer may feel that the expense of rewriting the software is too great. Whether they will survive in an ecological niche of the computer environment is an open question, but the golden age of the mainframe, as illustrated by this picture, is over.
