PART FOUR: COMMUNICATION AND CALCULATION 702 1943; it used 1500 valves (called vacuum tubes in the US). While not a stored- program machine, Colossus may have been the world’s first true automatic digital computer. Credit for this invention has become a matter of dispute, much of the development work being hidden because of the Second World War. As early as 1931, Konrad Zuse in Germany had constructed a mechanical calculator using binary components. This ‘Z1’ was followed by a Z2 in 1939, but it was not until 1949 that he built the Z3 which used electromechanical relays. In the late 1930s, Dr John V.Atanasoff sought a device to perform mathematical calculations for twenty degree candidates at Iowa State College. Rejecting mechanical solutions, he outlined ideas for memory and logic elements and built the first electronic digital computer with the assistance of Clifford Berry. It was called ABC—Atanasoff-Berry-Computer. This very early vacuum tube system seems to have had no influence on subsequent developments, but in 1974 a Federal judge ruled Dr Atanasoff the true inventor of the concepts required for a working digital computer. In 1937, George R.Stibitz at the Bell Telephone Laboratories conceived a binary calculator using telephone relays; with S.B.Williams and other colleagues, their Complex Computer was completed in 1939, and performed useful work at Bell Labs until 1949. Between 1939 and 1944, Harold H.Aitken, of Harvard University, with engineers of the International Business Machines Corporation, built the electromechanical Automatic Sequence Controlled Calculator, or Mark I. Between 1942 and 1946, the valve-based Electronic Numerator, Integrator and Computer (ENIAC) was built at the Moore School of the University of Pennsylvania to compute ballistic tables for the US Army Ordnance Corps. This was followed by the Electronic Discrete Variable Automatic Computer (EDVAC), built between 1947 and 1950. Also starting about 1947, J.Presper Eckert and John Mauchly of the Moore School conceived the Universal Automatic Computer (UNIVAC), which was built for the US Census Bureau and put into operation in 1951; eventually 48 UNIVAC I computers were made. Their Eckert-Mauchly Corporation was absorbed into Remington Rand and Sperry Rand’s Univac became the main rival to IBM for several decades. John von Neumann, a Hungarian immigrant at the Institute for Advanced Study in Princeton, became a consultant to the atomic bomb project at Los Alamos. In 1944, Herman H.Goldstein, working for the Ordnance Corps, told von Neumann about the Moore School’s work on ENIAC, and he began an active collaboration with them. A report written by von Neumann with Goldstein and Arthur W.Burks of the University of Michigan, first described the design of an electronic stored-program digital computer. The Whirlwind computer was constructed between 1944 and 1945 at MIT’s Lincoln Laboratory, under the direction of Jay W.Forrester. It was built INFORMATION 703 for the US Army Air Force to simulate flight, and was the first system to use data communications and interactive graphics. In 1946, Alan Turing joined the UK’s National Physical Laboratory (NPL), and immediately set about designing a computer. He presented NPL’s executive committee with what some claim as the first detailed design for a storedprogram electronic digital computer (others claim von Neumann). Turing’s proposal gave a cost estimate for construction of £11,200, and envisaged a plan for national computer development. Approval was given and funding obtained for the Automatic Computing Engine (ACE), but by 1948 not a single component had been assembled, and Turing resigned. Only a Pilot ACE was built, but it had a complex, 32-bit instruction format; 800 vacuum tubes; and mercury-delay-line storage of 128 32-bit words. A few years earlier, M.V.Wilkes and his colleagues at Cambridge built the Electronic Delay Storage Computer (EDSAC), using ultrasonic storage; and F.C.Williams and his group at the University of Manchester built their Mark 1 based on Williams’s cathode-ray tube (CRT) storage. In 1948, John Bardeen, Walter Brattain and William Shockley of the Bell Telephone Laboratories published the results of their work on solid-state electronic devices, which began just after the war in 1945. This paper described the invention of the point-contact transistor, which has almost completely displaced the valve as the active component in electronic circuits. For this work, the three shared a Nobel Prize in 1956. Also in 1948, Claude Shannon of Bell Labs developed his mathematical theory of communication (also called information theory), the foundation of our understanding of digital transmission. In 1949, Forrester invented magnetic-core storage for the Whirlwind, which became the standard internal memory for large-scale digital computers until semiconductor memory was introduced in the mid-1960s. Computer operation and programing During the 1950s, medium to large-scale systems, known as mainframes, were introduced for business use by IBM, Univac and others; they cost from hundreds of thousands to millions of dollars; required special housing and air- conditioning; demanded a sizeable staff; and called for a high degree of security. From the beginning they were operated as a closed shop; that is, users never got near the actual hardware. Punched cards or paper tape were the primary input media; users entered their programs and data by keypunching. A deck of cards or roll of paper tape had to be physically delivered to the mainframe location. Hours, sometimes days, later—if there were many big jobs in the queue—users would get their output, usually in the form of fan-folded paper printouts. Frequently, finding some slight mistake, they would have to resubmit the whole job again. PART FOUR: COMMUNICATION AND CALCULATION 704 The earliest computers had to be programmed in machine language, which meant that both instructions and data had to be input in binary form, and the location of each address in memory had to be specified (absolute addressing). For humans this procedure is extremely tedious and prone to error. A big improvement was assembly language, which assigned alphabetical codes to each instruction, allowed the use of decimal numbers, and permitted relative addressing. However, the relationship between the lines of code and the machine’s memory was still one-to-one. Many users wanted to be able to specify their instructions to the computer in a language closer to algebra, and in 1954, John Backus of IBM published the first version of FORTRAN (Formula Translator), the first ‘higher-level’ computer language. In 1959, under the urging of the US Department of Defense, a standardized higher-level language for business use was developed, known as COBOL (Common Business-Oriented Language). By the mid-1960s, 90 per cent of all scientific and engineering programs were being written in FORTRAN, and most business data-processing in COBOL. Even though personnel responsible for running mainframes believed that they must maintain closed-shop service for security reasons, an important advance in user convenience was made in the 1960s—remote job entry (RJE). RJE was still batch-processing and users had to wait several hours for their results. However, instead of having to carry decks of cards and pick up printouts at the mainframe, they could go to a nearby RJE station where a card reader and high-speed printer were connected by dedicated line to the central computer. Most mainframe computers sold to businesses were not intended to be used interactively by anyone except their operators, who sat at a large console composed of a video display unit (VDU) and keyboard. Operators monitored the job-stream, and mounted magnetic tapes on which user programs and data were stored when the VDU alerted them. Computing was still a scarce resource, and users were charged for each task required to fulfil their job requests. The introduction of interactive access was largely user-driven; users put pressure on data processing departments to provide faster and easier access than was possible under batch processing. Also, relatively low-cost terminals (a few thousand dollars per station) had to become available to make this economically feasible. Special applications requiring real-time processing pioneered this development in the late 1950s and early 1960s—particularly air defence (the US Air Force’s SAGE system) and airline reservations systems. In 1959, Christopher Strachey in the UK proposed the idea of time-sharing mainframe computers to make better use of their capabilities for small jobs; each such job would be processed for only a second or two, but users, working interactively, would not experience noticeable delays because their interaction time is at least 1000 times slower than the computer’s. In 1961, MIT demonstrated a model time-sharing system, and by the mid-1960s access to INFORMATION 705 large, central computers by means of low-cost terminals connected to the public switched telephone service had become a commercial service. Terminals were usually Teletypes which, although designed for message communications, were mass-produced and inexpensive; so, rather than seeing results on a CRT screen, users would get their output printed on a roll of paper. Within large organizations, interactive VDU terminals became available to regular users. They resembled stripped-down operators’ consoles, and became known as dumb terminals because they were completely dependent on the mainframe for their operation, having no memory of their own. Communication with the mainframe was on a polled basis via high-speed, dedicated links, and transfer of information was in screen-sized blocks. Paralleling the evolution of mainframes towards open access, rapid technological developments were taking place to provide more computing power in smaller packages for less money. In 1963, the Digital Equipment Corporation (DEC) marketed the first minicomputer, the PDP-5, which sold for $27,000; two years later the more powerful and smaller PDP-8 was made available for only $18,000. This was the beginning of a new industry oriented towards more sophisticated customers who did not need or want the hand-holding service which was a hallmark of IBM. Such users, particularly scientists and engineers, preferred an open system into which they could put their own experimental devices, and develop their own applications programs. The micro breakthrough This trend culminated in the development of the integrated circuit (IC), popularly known as the microchip, or just ‘chip’, which enables digital-circuit designers to put essential mainframe functions into one or more tiny chips of silicon. The first patent for a microelectronic (or integrated) circuit was granted to J.S.Kilby of the United States in 1959. By the early 1960s, monolithic operational amplifiers were introduced by Texas Instruments and Westinghouse. In 1962, Steven R.Hofstein and Frederick P.Heiman at the RCA Electronic Research Laboratory made the first MOS (metal-oxide semiconductor) chip. Then Fairchild Semiconductors came out with the 702 linear IC in 1964, the result of a collaboration between Robert Widlar and David Talbert; soon the 709 was put on the market, one of the early success stories in the greatest technological revolution of the twentieth century. Widlar stressed active components in preference to passive ones. In 1964, Bryant (Buck) Rogers at Fairchild invented the dual-in-line package (DIP), which has come to be the universal packaging form for ICs. In the software area, John G.Kemeny and Thomas E.Kurtz at Dartmouth College developed a higher-level computer language for student use called PART FOUR: COMMUNICATION AND CALCULATION 706 BASIC; from its beginning in 1964, BASIC was designed for interactive use via ‘time-sharing’ on minicomputers. BASIC has become the standard higherlevel language for microcomputers, often being embodied in a ROM (read-only memory) chip, which is also known as firmware. As with the early development of the large-scale digital computer, invention priorities are often disputed, but, according to one source, the first electronic calculator was introduced in 1963 by the Bell Punch Company of the UK. This contained discrete components rather than ICs, and was produced under licence in the US and Japan. The first American company to make electronic calculators was Universal Data Machines, which used MOS chips from Texas Instruments and assembled the devices with cheap labour in Vietnam and South America. In 1965, Texas Instruments began work on their own experimental four-function pocket calculator to be based on a single IC, and obtained a patent on it in 1967. Westinghouse, GT&E Laboratories, RCA and Sylvania all announced CMOS (complementary metal-oxide semiconductor chips) in 1968. They are more expensive, but faster and draw less current than MOS technology; CMOS chips have made portable, battery-operated microcomputers possible. In 1969, Niklaus Wirth of Switzerland developed a higher-level language he called Pascal, in honour of the inventor of the first mechanical calculator. Pascal was originally intended as a teaching language to instil good programming style, but has become increasingly popular for general programming use, particularly on microcomputers where it is now the main rival to BASIC. However, Pascal is usually compiled (the entire program is checked before it can be run), whereas BASIC is interpretive (each line is checked as it is input). The idea of creating a simple, low-cost computer from MOS chips—a micro- computer-on-a-chip—seems to have occurred first to enthusiastic amateurs. In 1973, Scelbi Computer Consulting offered the Scelbi-8H as a kit for $565. It used the Intel 8008, the first 8-bit microprocessor, and had 1 kilobyte of memory. In January 1975, Popular Electronics magazine featured on its cover, the Altair 8800, proclaimed as the ‘world’s first minicomputer [sic] kit to rival commercial models’. Altair was the product of MITS, a tiny Albuquerque, New Mexico, firm, which sold it in kit form with an Intel 8080 microprocessor and a tiny 256 bytes of memory for $395. Users had to enter data and instructions bit-by-bit via sixteen toggle switches, and read the results of their computations in binary form on a corresponding array of sixteen pairs of panel lamps, but these early machines were received enthusiastically by dedicated ‘hackers’. (To put this price in context, mainframes cost a million dollars or more, and minicomputers tens to hundreds of thousands of dollars.) The microcomputer market developed rapidly, even though no software was available. Users had to develop their own applications using machine language which was (and remains) binary. However, in June 1976, Southwest INFORMATION 707 Technical Products Company offered the SwTPC M6800 with an editor and assembler, and shortly afterwards with a BASIC interpreter. By the end of 1976, the floodgates had opened; it was perhaps a fitting testament to this seminal year that Keuffel and Esser stopped production of slide-rules—donating the last one to the Smithsonian Institution in Washington. Microcomputers remained largely an amateur market (‘hacker’ came to be the preferred term because users were constantly tinkering with their innards) until the Apple II and Commodore PET came on to the market in 1977. For a couple of thousand dollars, customers were offered an alphanumeric keyboard for data entry and control; a colour display (their television set); a beeper which could be programmed to play simple tunes; a pair of paddles to play video games; a high-level language (BASIC) in ROM, and an inexpensive means of storing and retrieving data and programs (their audiocassette recorder). However, Apple and its rivals did not have much effect outside home and school environments until the floppy disk was miniaturized and reduced in price to a few hundred dollars in the late 1970s. A floppy diskette could store up to 100 kbytes and speeded access to data and programs by a factor of twenty. Thus, the entire contents of RAM (random-access memory), which disappeared when the power was turned off, could now be saved in a few seconds and loaded back quickly when needed again. Even so, the microcomputer remained a home or consumer product, largely for games, until useful applications programs were developed. In 1979, the VisiCalc program was developed by Dan Bricklinn and Bob Frankston specifically for the Apple II micro. VisiCalc enabled data to be entered on large (virtual) speadsheets; in that form, the data could be manipulated arithmetically en masse, with its results printed out in tabular form automatically. The business community now found that the technology of microcomputers had something unique to offer, and bought Apples in increasing numbers simply to use VisiCalc. In 1981, Adam Osborne, an Englishman who had emigrated to the US, introduced the Osborne 1, the first complete, portable microcomputer. Selling for less than $2000, it included a full-sized keyboard, CRT display, dual floppy drives, and was ‘bundled’ with some of the best software available at the time: the CP/M operating system; WordStar for word processing; SuperCalc for financial analysis; and interpretive and compiled BASIC systems. During all these developments, owners and operators of mainframes had taken little notice. Microcomputers were regarded as amazing but rather expensive toys. Mainframe services had been organized for computing specialists, and users had to employ these specialists—systems analysts and programmers—as intermediaries. In contrast, with only a few hours’ instruction, a non-specialist could learn a user-friendly microcomputer software package directly relevant to his work. It soon became apparent that, because microcomputers could be purchased without being scrutinized by a committee PART FOUR: COMMUNICATION AND CALCULATION 708 and sent out for competitive bids—the standard procedure for items costing hundreds of thousands of dollars or more—an employee could, and would, get one to put on his desk. At last the mainframe masters had to take notice. Almost from the beginning of the stored-programme digital computer, IBM has dominated the market to a remarkable extent. However, in no sense did they lead the introduction of micros, but entered the market only when it showed signs of maturity. Nevertheless, the IBM PC quickly took a dominant position in the market from its introduction in 1981. Since this ‘legitimatization’ of micros, software of great versatility and power has been introduced for the IBM PC and ‘clones’, as its imitators are known. Notable among these are integrated packages, such as Lotus 1–2–3, which was marketed in 1982. Lotus combined database, spreadsheet and graphics in a single program. As a result of their convenience, versatility and low cost, micros are already far more numerous than mainframes and minis, and will account for more than half the revenue of the entire computer market in the last half of the 1980s. Communications between mainframe and micro Because IBM dominates the mainframe market, many of its standards have become ad hoc world standards. However, most of these are inappropriate or meaningless in the microcomputer world. Mainframes use large ‘word’ sizes; that is, each such entity they handle is usually 32 bits (even 64 bits when high- precision computation is required); in contrast, early micros used 8-bit bytes, and more recent ones—led by IBM’s own PC—16-bit bytes. A few micros, particularly those emphasizing graphics, such as Apple’s Macintosh, now use 32-bit microprocessors. Also, IBM developed its own codes to interpret the bit patterns as numbers or letters. Their Binary-Coded Decimal (BCD) system—developed in response to the needs of business to encode each decimal digit separately—was extended to handle larger character sets by EBCDIC (Extended Binary Coded Decimal Interchange Code). With BCD, only four bits were used, providing a maximum of sixteen codes; the four codes not required for the digits 0–9 were used for the decimal point and plus and minus symbols. However, no letters could be represented directly. A 6-bit code was also developed which provided 64 codes, enough to handle upper-case letters, decimal digits and some special characters. However, as text processing developed, it became apparent that more than 64 codes would be needed. EBCDIC, IBM’s 8-bit system, provides 128 or 256 codes, allowing not just upper- and lower-case letters, but many special symbols and codes for control (nonprinting) purposes. At about the same time, other manufacturers got together and under the aegis of the American Standards Association developed ASCII (American INFORMATION 709 Standard Code for Information Interchange) which has since become a world standard under the auspices of the International Standards Organization (ISO). ASCII is also an 8-bit system with 256 possible codes, but assignments are entirely different from EBCDIC. However, even IBM, when it entered the microcomputer market, chose ASCII. Therefore, when any microcomputer-— IBM or another make—communicates with an IBM mainframe, code conversion is required at one end of the link or the other. Another source of incompatibility is the type of communications used between mainframes and their terminals. In the mid-1970s IBM developed a system to handle remote dumb terminals based on using an ‘intelligent’ device between communications lines and terminals: the cluster controller. A mainframe could poll remote terminal clusters to see if they had a request, but remote terminals could never initiate a dialogue. Also, communication links were established to run at fairly high speeds—9600 bits per second is common— because they had to poll every terminal on the system in turn (even if an operator was absent). The system IBM developed for this is known as Binary Synchronous Communications (BSC), and requires a synchronous modern (modulator-demodulator) at each remote cluster, as well as at the mainframe. However, when micros began to offer communications capabilities, manufacturers chose much less expensive asynchronous modems and protocols and, taking advantage of the built-in ‘intelligence’ of personal computers (PCs) as terminals, let the micro initiate the conversation. Still another source of incompatibility is that micros were first developed for home and hobbyist markets. They provided graphics (often colour) and audio capabilities at the time of their greatest market growth, which were essential for games, the most popular application at that time. Most mainframes, on the other hand, did not offer graphics and sound capability at remote terminals, only text. As more and more businesses encouraged the spread of micros into their day-to-day operations, they found that their most valuable data were often locked within mainframes. Therefore, unless someone laboriously re-entered these data from existing printouts, another way to transfer (download) from mainframe to micros had to be found. Also, the rapid development of micro technology led to smaller and smaller systems, and eventually battery-operated laptop portables. The pioneering systems were the Epson HX-20 and the Radio Shack TRS-80 Model 100, which were introduced in 1982 and 1983. The laptops provided essential microcomputer applications, including word processing and telecommunications, in a briefcase-size package for less than $1000. Users could now upload data gathered in the field to mainframes for processing. Laptop micros can communicate with remote systems wherever there is a telephone by means of a type of modem which does not have to be physically connected to the public switched telephone network (PSTN)—the acoustic PART FOUR: COMMUNICATION AND CALCULATION 710 coupler. Rubber cups which fit over the telephone handset provide a temporary connection. The purpose of any modem is to convert pulses, which cannot be passed through the PSTN, to audible tones of standard modem frequencies, which can. However, this type of communication is slow—usually 300 bits per second (bps) for acoustic couplers, and 1200 or in some cases 2400 bps for directly connected modems—and sensitive to noise and others types of interference which increase error rates. At 300 bps, it would take about twenty minutes to transfer all the data on a standard double-sided IBM diskette (360 kilobytes—equivalent to about 100 typewritten pages). The next step is to employ synchronous modems, which are used when terminals are attached to mainframes. If an IBM PC or compatible microcomputer is to be used, it must have a synchronous adapter card and 3270 emulation software. This hardware and software adds about $1000 to the cost of each PC, and provides terminal emulation, but not compatible file transfer between mainframe and micros. File transfer capability requires a coaxial communications card in the PC, which is connected to a controller. No modem is required at the PC, because the controller allows many devices to share a high-speed modem. Usually there are from 8 to 32 ports on a controller, which can support interactive terminals, remote-job-entry terminals, remote printers—and now PCs. However, special programs are required in the PCs, and often software must be added at the mainframe to support file transfer. Another problem in attaining true emulation is that the keyboard of an IBM PC is quite different from IBM 3278 terminals, which have a large number of special function keys. Operators who are familiar with terminal keyboard layouts find it difficult to learn PC equivalents, and particularly frustrating if they have to switch back and forth. With a PC emulating a 3270- type terminal, for instance, two keys may have to be pressed simultaneously, or even a sequence of keys struck, where one keystroke would do on the terminal keyboard. Still another problem is that, as has been mentioned, IBM mainframes and terminals communicate with a character code known as EBCDIC, whereas PCs use ASCII. Therefore, EBCDIC-ASCII conversion is required at one end or the other (it is usually done at the mainframe). The best emulation systems permit not just whole files, but selected records or even fields to be downloaded for processing by PC software. THE TELEGRAPH In the early 1600s, the Jesuit Flamianus Strada, impressed with William Gilbert’s research on magnetism, suggested that two men at a distance might communicate by the use of magnetic needles pointing towards letters on a dial; George Louis Lesage, a physicist of Geneva, did some experiments along these INFORMATION 711 lines in 1774, but electricity was still too poorly understood to provide a practical means of communication. Visual telegraphy A French abbé, Claude Chappe, with his three brothers invented and put into use the first practical system of visual (or aerial) telegraphy, the semaphore. They had first tried, and failed, with an electrical solution. In 1790, they succeeded in sending messages over half a kilometre, using pendulums suspended from two posts. However, public demonstrations were greeted with hostility by a revolutionary populace who thought the system would enable royalists to communicate with their king. Chappe, who had spent 40,000 livres of his own on the scheme, sought police protection for his towers. Chappe presented his invention to the French Assembly on 22 May 1792, and on 1 April 1793 the invention was favourably reported to the National Convention as a war aid. However, some vital questions were raised: would it work in fog? could secret codes be used? The Convention ordered an inquiry, for which it appropriated 6000 livres, and after a favourable committee report, Citizen Chappe was given the title of Telegraphy Engineer with the pay of a lieutenant, and ordered to construct towers from Paris to Lille. The 14km (8.7 miles) distance between towers was too far for best results, and Chappe’s business management was poor; however, the line was completed in August 1794 (Figure 15.2 (a)). Many messages were sent during construction, but the first to be recorded after completion was on 15 August, relaying the capture of Quesnoy from the Austrians; this was made known in Paris only one hour after troops entered the town. To operate a semaphore, the station agent manipulated levers from ground level, whose movements were faithfully followed by wooden arms placed three metres (10ft) above, on top of a stone tower. The central member, the regulator, was 4m (13ft) long, and arms of 1.8m (6ft) were pivoted from both ends. Regulator and arms could be placed in 196 recognizable positions in addition to those where the regulator was vertical or horizontal, because Chappe had decided that the only meaningful signals were when the regulator was in an inclined position. For communication at night, a lantern was placed at each end of the arms and at each of the pivots. There were 98 positions with the regulator inclined to the left for inter- operator communications, and another 98 with the regulator inclined to the right for dispatches. Chappe formulated a code of 9999 words with the assistance of L.Delauney, and the perfected list was published with each word represented by a number. These pairs were printed in a 92-page code book each page of which carried 92 entries, giving the somewhat reduced total of 8464 word-number equivalents. . on ENIAC, and he began an active collaboration with them. A report written by von Neumann with Goldstein and Arthur W.Burks of the University of Michigan, first described the design of an electronic. algebra, and in 1954, John Backus of IBM published the first version of FORTRAN (Formula Translator), the first ‘higher-level’ computer language. In 1959, under the urging of the US Department of Defense,. and lower-case letters, but many special symbols and codes for control (nonprinting) purposes. At about the same time, other manufacturers got together and under the aegis of the American Standards