Control device

External devices

RAM

Drawing - Connections between devices

Almost all of von Neumann’s recommendations were subsequently used in machines of the first three generations; their totality was called “von Neumann architecture.” The first computer to embody von Neumann's principles was built in 1949 by the English researcher Maurice Wilkes. Since then, computers have become much more powerful, but the vast majority of them are made in accordance with the principles that John von Neumann outlined in his 1945 report.

New cars of the first generation replaced each other quite quickly. In 1951, the first Soviet electronic computer MESM, with an area of ​​about 50 square meters, began operation. MESM had 2 types of memory: random access memory, in the form of 4 panels 3 meters high and 1 meter wide; and long-term memory in the form of a magnetic drum with a capacity of 5000 numbers. In total, the MESM had 6,000 vacuum tubes, and it was possible to work with them only after 1.5-2 hours after turning on the machine. Data input was carried out using magnetic tape, and output was carried out using a digital printing device coupled to memory. MESM could perform 50 mathematical operations per second, store 31 numbers and 63 commands in RAM (there were 12 different commands in total), and consumed power equal to 25 kilowatts.

In 1952, the American EDWAC machine was born. It is also worth noting the English computer EDSAC (Electronic Delay Storage Automatic Calculator), built earlier, in 1949, the first machine with a stored program. In 1952, Soviet designers commissioned the BESM, the fastest machine in Europe, and the following year, Strela, the first high-class production machine in Europe, began operating in the USSR. Among the creators of domestic cars, the names of S.A. should be mentioned first. Lebedeva, B.Ya. Bazilevsky, I.S. Bruka, B.I. Rameeva, V.A. Melnikova, M.A. Kartseva, A.N. Myamlina. In the 50s, other computers appeared: “Ural”, M-2, M-3, BESM-2, “Minsk-1” - which embodied more and more progressive engineering solutions.

The projects and implementation of the Mark-1, EDSAC and EDVAC machines in England and the USA, MESM in the USSR laid the foundation for the development of work on the creation of computers of vacuum tube technology - serial computers of the first generation. The development of the first electronic production machine, UNIVAC (Universal Automatic Computer), began around 1947 by Eckert and Mauchli. The first model of the machine (UNIVAC-1) was built for the US Census Bureau and put into operation in the spring of 1951. The synchronous, sequential computer UNIVAC-1 was created on the basis of the MENIAC and EDVAC computer. It operated with a clock frequency of 2.25 MHz and contained about 5,000 vacuum tubes.

Compared to the USA, USSR and England, the development of electronic computer technology in Japan, Germany and Italy was delayed. The first Japanese Fujik machine was put into operation in 1956; mass production of computers in Germany began only in 1958.

The capabilities of the first generation machines were quite modest. Thus, their performance according to modern standards was low: from 100 (Ural-1) to 20,000 operations per second (M-20 in 1959). These figures were determined primarily by the inertia of vacuum tubes and the imperfection of storage devices. The amount of RAM was extremely small - on average 2,048 numbers (words), this was not enough even to accommodate complex programs, not to mention data. Intermediate memory was organized on bulky and low-speed magnetic drums of relatively small capacity (5,120 words for BESM-1). Printing devices and data input units also worked slowly. If we dwell in more detail on input-output devices, we can say that from the beginning of the appearance of the first computers, a contradiction emerged between the high speed of central devices and the low speed of external devices. In addition, the imperfections and inconvenience of these devices were revealed. The first data carrier in computers, as is known, was a punched card. Then perforated paper tapes or simply punched paper tapes appeared. They came from telegraph technology after the beginning of the 19th century. Chicago father and son Charles and Howard Crums invented the teletype.

The first generation of computers, these tough and slow-moving computers, were the pioneers of computer technology. They quickly disappeared from the scene, as they did not find wide commercial use due to unreliability, high cost, and difficulty in programming.

To simplify the process of setting programs, Mauchly and Eckert began to design a new machine that could store a program in its memory. In 1945, the famous mathematician John von Neumann was involved in the work, who prepared a report on this machine. In this report, von Neumann clearly and simply formulated the general principles of the functioning of universal computing devices, i.e. computers. This was the first operational machine built on vacuum tubes and was officially put into operation on February 15, 1946. They tried to use this machine to solve some problems prepared by von Neumann and related to the atomic bomb project. She was then transported to Aberdeen Proving Ground, where she operated until 1955.

ENIAC became the first representative of the 1st generation of computers. Any classification is conditional, but most experts agreed that generations should be distinguished based on the elemental base on which the machines are built. Thus, the first generation appears to be tube machines.

The structure and operation of a computer according to the “von Neumann principle”

It is necessary to note the enormous role of the American mathematician von Neumann in the development of first-generation technology. It was necessary to understand the strengths and weaknesses of ENIAC and make recommendations for subsequent developments. The report by von Neumann and his colleagues G. Goldstein and A. Burks (June 1946) clearly formulated the requirements for the structure of computers. Let us note the most important of them:

· machines using electronic elements should operate not in the decimal, but in the binary number system;

· the program, like the source data, must be located in the machine’s memory;

· the program, like numbers, must be written in binary code;

· the difficulties of the physical implementation of a storage device, the speed of which corresponds to the speed of operation of logical circuits, require a hierarchical organization of memory (that is, the allocation of RAM, intermediate and long-term memory);

· an arithmetic device (processor) is constructed on the basis of circuits that perform the addition operation; the creation of special devices for performing other arithmetic and other operations is impractical;

· the machine uses a parallel principle of organizing the computational process (operations on numbers are performed simultaneously in all digits).

The following figure shows what the connections between computer devices should be according to von Neumann's principles (single lines show control connections, dotted lines show information connections).

Almost all of von Neumann’s recommendations were subsequently used in machines of the first three generations; their totality was called “von Neumann architecture.” The first computer to embody von Neumann's principles was built in 1949 by the English researcher Maurice Wilkes. Since then, computers have become much more powerful, but the vast majority of them are made in accordance with the principles that John von Neumann outlined in his 1945 report.

New cars of the first generation replaced each other quite quickly. In 1951, the first Soviet electronic computer MESM, with an area of ​​about 50 square meters, began operation. MESM had 2 types of memory: random access memory, in the form of 4 panels 3 meters high and 1 meter wide; and long-term memory in the form of a magnetic drum with a capacity of 5000 numbers. In total, the MESM had 6,000 vacuum tubes, and it was possible to work with them only after 1.5-2 hours after turning on the machine. Data input was carried out using magnetic tape, and output was carried out using a digital printing device coupled to memory. MESM could perform 50 mathematical operations per second, store 31 numbers and 63 commands in RAM (there were 12 different commands in total), and consumed power equal to 25 kilowatts.

The capabilities of the first generation machines were quite modest. Thus, their performance according to modern standards was low: from 100 (Ural-1) to 20,000 operations per second (M-20 in 1959). These figures were determined primarily by the inertia of vacuum tubes and the imperfection of storage devices. The amount of RAM was extremely small - on average 2,048 numbers (words), this was not enough even to accommodate complex programs, not to mention data. Intermediate memory was organized on bulky and low-speed magnetic drums of relatively small capacity (5,120 words for BESM-1). Printing devices and data input units also worked slowly. If we dwell in more detail on input-output devices, we can say that from the beginning of the appearance of the first computers, a contradiction emerged between the high speed of central devices and the low speed of external devices. In addition, it was revealed

imperfection and inconvenience of these devices. The first data carrier in computers, as is known, was a punched card. Then perforated paper tapes or simply punched paper tapes appeared. They came from telegraph technology after the beginning of the 19th century. Chicago father and son Charles and Howard Crums invented the teletype.

The first generation of computers, these tough and slow-moving computers, were the pioneers of computer technology. They quickly disappeared from the scene, as they did not find wide commercial use due to unreliability, high cost, and difficulty in programming.

ENIAC consisted of 42 blocks measuring approximately 2.75-0.7-0.3 m, in which 30 separate devices (units) were located: power system; device for starting and stopping the machine; clock generator (cycling unit); the central programming device is a patch board (typesetting field), the individual sockets of which are connected by plugs; 20 accumulator registers, which played the role of RAM and adding (subtracting) device; multiplier; division/square root device; three replaceable function tables; a relay buffer device that communicated between the machine and the punched card reader; the so-called master programmer (“control programmer”) and some others.

The devices were connected to each other by two groups of 11-wire coaxial cables. One set of cables formed a digital backbone that carried trains of pulses representing numerical data. A separate conductor (core) in the cable corresponded to one decimal place (plus the core of the number sign), and the value of the transmitted digit was equal to the number of pulses passing through this conductor. The second group of cables was the program line and transmitted pulses that controlled the sequence of operations in various devices, depending on the settings of the plugs on the patch board. Each conductor in the cable was an independent program line (program channel) and carried a specific control signal from the clock generator (TG).

ENIAC-2 computer

ENIAC was a synchronous machine: the TG, whose pulses were continuously and simultaneously transmitted to all devices of the machine, coordinated its actions. The generator operated at a frequency of 100 kHz and every 200 μs produced a set of pulses, the duration of which was approximately 2 μs, and the time interval between them was 10 μs. The first of these pulses was called the central programming pulse (CPP) and specified the beginning and end of machine operations (a separate device, having completed its inherent operation, transmitted CPP as its output program pulse to another device, initiating its operation). The main machine cycle was equal to the time of one addition, which took 200 μs (i.e., 5000 additions were performed per second). The execution time for the remaining arithmetic operations was an integer addition cycle.

Program pulses sent simultaneously, each of which had its own purpose, made it possible to parallelize the execution of operations to some extent: for example, one accumulator performed addition, another received data from the function table, the third transmitted data to perforation, etc. (of course, with provided that the result of the calculation stored in the accumulator was not required for the next arithmetic operation).

The main electronic circuits of the machine were flip-flops, “and” cells, which acted as switches, and “or” cells, designed to combine pulses coming from different sources into one output. Ten flip-flops were connected in a ring to form a decimal (decadal) counter, which served the same role as the counting wheel in mechanical adding machines (thus requiring 20 triodes to represent one decimal digit). Ten such rings plus a trigger to represent the sign of a number constituted a storage register (there were 20 of them in ENIAC). Each of the registers was equipped with a circuit for transferring tens and was an accumulator, that is, it was used not only as memory, but also as an adder-subtractor. These operations were performed by counting pulses arriving at the input of the counters.

The multiplication operation was performed in a high-speed multiplying device (multiplier). It used four batteries and a built-in 9-9 multiplication table made on a resistive matrix (in cases where it was necessary to obtain a 20-bit product, six batteries were used). Two accumulators were used to store operands, two were used to store private products. When pulses corresponding to one bit of the multiplicand and one bit of the multiplier appeared at the inputs of the matrix, it generated pulses representing their partial product. The units digits of this product were sent to one accumulator, the tens digits to another. After completing the multiplication by the next digit of the multiplier, it was shifted one digit to the left and the multiplication was performed again. When all the digits of two 10-bit numbers were multiplied, one of the accumulators added the accumulated quotient products (the method of dividing these products into “unit” and “decimal” parts was therefore the same as in the Harvard Mark I - see Fig. "Electromechanical colossus" The entire process of multiplying two 10-bit numbers took 2.8 ms in ENIAC (or 357 multiplications per second).

The device for dividing and extracting square roots also consisted of four accumulators: the first contained the dividend (or radical expression), the second contained the divisor (or doubled square root), the third contained the quotient, and the fourth accumulator was used to perform the shift operation. During division, the divisor was subtracted from the dividend until the difference became negative. After this, the process was interrupted, the remainder of the dividend was sent to the fourth accumulator to be shifted one position to the left and then returned to the first accumulator and summed with the divisor until the sum became positive. In this case, either +1 or -1 was sent to the corresponding bit of the quotient accumulator (depending on whether the divisor was added or subtracted). The root extraction process was carried out in a similar way. Operating with 10-bit numbers, ENIAC performed 40 division operations and 3 root extraction operations per second.

The function tables also used resistive matrices and sets of switches that could be used to set 12 digits and 2 signs for each of the 104 independent arguments. Initially, function tables were conceived to store function values, but then they began to be used to store constants needed in calculations. When solving any problem, only one table was connected to ENIAC, while the other two were prepared by operators to solve the following problems (an idea borrowed from the design of IBM tabulators).

The initial data was entered into the machine from punched cards. A standard IBM reading device was used for this. Since the reading speed (about 2 numbers per second) was many times lower than the speed of arithmetic operations, in order to prevent batteries from being idle during data entry, the developers supplemented the machine with a buffer consisting of 1500 telephone relays ( it was developed by Samuel Williams, one of the designers of the Bell Labs machines.) The buffer, or, as it was called, the “constant transmitter,” converted the read number into a sequence of pulses equivalent to this number, and after receiving the CPP from ENIAC, sent data into batteries. The buffer also received the results of calculations from the batteries, freeing the latter to perform their inherent operations, and sent the received data to the final punch or (for printing) to the tabulator. In addition, the constant transmitter converted negative numbers represented by decimal's complement into ordinary form and had a set of switches on the front panel by which a series of constants could be entered into the machine.

Of course, the creators of ENIAC provided a set of measures for diagnosing individual devices of the machine. One of them was to apply single pulses from a clock generator to the batteries, which made it possible to determine a failed trigger (by the characteristic glow of neon bulbs connected to the batteries). Another type diagnostics was a step-by-step run of the test program.

Programming the machine - the developers called this process “setting up” - was carried out by a group of seven young women mathematicians (among them were the wives of Mauchly, Burks and Goldstein). It was carried out as follows. Firstly, using a patch board and plugs, devices that were supposed to be involved in solving a specific problem were connected to each other. Secondly, the so-called transceiver switches, located on the front panel of each of these devices, were set to the “on” position and formed local program-control circuits. The on position of the switches allowed the device to perform its actions after the arrival of a program pulse from the TG. In addition, a multi-pole stepper switches was installed on the device panel, which made it possible to repeat the same operations multiple times (up to nine times in a row).

To organize a given number of iterative cycles, connect individual sequences of calculations into a single circuit, change the order of execution of these sequences using the conditional jump command, a device was introduced into the machine, called the master programmer by the authors and containing ten 6-bit step counters connected to several decade counters.

The action described above was preceded by lengthy paperwork. Using the setup table, the sequence of operations necessary to solve a specific problem was described in detail. The table had 27 columns (one for each accumulator and function tables, for the control programmer, constant transmitter, etc.) and contained a time sequence of program settings, input and output pulses for each operation. Programming the machine was thus a labor-intensive and time-consuming process (it sometimes took days or even weeks). Any “installation” of the machine changed its configuration and turned it into a specialized device for solving a specific task, and the “program” became an internal, integral part of the ENIAC. This was, of course, a disadvantage compared to electromechanical machines controlled using perforated tapes.

Did the military get what they wanted? I think so. On a desktop calculating machine, calculating a 60-second projectile flight trajectory took 20 hours, a differential analyzer made it possible to obtain the same result (approximate) in 15 minutes, while ENIAC required only 30 seconds - half the flight time.

Throughout 1946, the car remained at the Moore School. Although the war ended, ENIAC continued to be used for military purposes - in calculating firing tables and calculations that were supposed to confirm the possibility of creating a hydrogen bomb (the machine successfully completed this task, which required processing about a million punched cards). However, she did not neglect peaceful tasks. ENIAC was transferred to Aberdeen early in 1947 and was brought back into service in August. Her subsequent work included solving problems for meteorologists and physicists who studied cosmic rays and studied the propagation of shock waves, etc.

The main engineering problem faced by the creators and users of ENIAC was the problem of frequent failures of electronic tubes. Later historians calculated that with almost 17.5 thousand tubes, which at that time were not highly reliable and operated simultaneously at 100 kHz, every second there were 1.7 billion situations in which at least one of them “flyed", which led to a malfunction of the entire colossus. Let us recall that then neither in radio engineering, nor in radars and decoding devices6 of such a “tube “abundance” was not even close, and many Mauchly-Eckert opponents doubted that the ENIAC could last even a few hours without failure (despite the fact that the electronic components intended for the machine were carefully tested, and special attention was paid to the quality of soldering).

The first years of operation of the machine almost confirmed the doubts of the skeptics (in 1946, the average time between failures of ENIAC was 5.6 hours). Later, the situation improved somewhat, primarily due to the fact that, on the advice of engineers from the RCA corporation, the supply voltage for the lamps was made less than the standard recommended by reference books, and the lamps were “trained” for a long time before being installed in the car. In addition, operating engineers have established that the most widespread failures of lamps occur when ENIAC is turned on and off (which is quite explainable by the physical processes occurring in them in transient mode). Hence, a radical, albeit expensive way to increase the reliability of the machine: never Turn it off!Following him, it was possible to ensure that ENIAC worked without failure for several years (the machine worked for a record time - 116 hours in a row in 1954).

Another source of headaches for operators, which was the influence of instability of industrial mains voltage on the operation of electronic units, was eliminated in 1950, when the machine began to be powered from an autonomous motor-generator system (the Soviet M-20, created much later, was also powered where the author worked).

Engineers and programmers at Aberdeen Laboratory made other useful changes to ENIAC. For example, in 1951, an electronic device for retrieving information from function tables was developed; the following year, a high-speed shift device was introduced into the machine, which significantly reduced the time it took to perform division and square root operations; finally, in July 1953, a 100-word ferrite core memory developed by Burroughs Corp. was built into the machine.

But perhaps the most important innovation was a modification of the programming process proposed in 1947 by John von Neumann and implemented the following year by BRL employee Dr. Richard H. Clippinger (1914-2003). An electronic unit was manufactured that, during the addition cycle, converted one of six dozen two-digit numbers installed on the function table into the corresponding number of CPP pulses, which were sent along the software highway and initiated the execution of one of 60 commands. Thanks to this innovation, there was no need for plug sets on the patch board, which significantly (up to several hours) reduced programming time and, in addition, simplified testing of any machine device. Thus, each function table turned into a small-capacity read-only memory device, and ENIAC became a sequential machine with an internally stored program. At the same time, however, it was deprived of the ability to execute several program steps in parallel, and its performance decreased by about six times. Subsequently, Eckert wrote that he and Mauchly assumed the possibility of such a modification already at the initial stage of designing the machine (which is confirmed by the same number of cores of “digital” and “software” cables).

The first universal computer worked for a total of 80,223 hours and ended its life on October 2, 1955 at 23:15. The fate of its main “builders” was not easy.

Mauchly and Eckert believed that the copyright to the machine belonged to them, since none of the MSEE executives took part in the implementation of the “PX Project”. So they asked the university president for permission to apply for a patent in their own name. The President agreed, but on the condition that the text of the application would say: “The authors will grant to the US government and the university a royalty-free license to manufacture such machines for non-commercial purposes.” Mauchly and Eckert refused to make any changes to the already prepared text and left the university on March 31, 1946 to organize their own company (which will be discussed in a future article).

They received a patent for ENIAC (No. 3120606 dated June 26, 1947) together with several members of their team, but their misadventures did not end there.

Thirty years after work on the machine began, on October 19, 1973, Federal Judge Earl Richard Larsen ruled in Minneapolis District Court after 135 days of deliberations: “Eckert and Mauchly did not first invent an automatic electronic digital computer, but extracted the essence of the concept from the invention.” Dr. John Vincent Atanasoff."

Did Larsen have any reason to make such a surprising decision for the computer world? More on this in the next article.

Notes

Founded in 1923 and named after cable manufacturer Alfred Fitler Moore, who donated a separate building to the Faculty of Electrical Engineering.

In 2004, a chip measuring 0.5 square meters. mm provided the same performance as ENIAC.

The problem of parallelizing programs was fully solved many years later.

Subtraction was performed as addition with decimal addition.

At the end of 1943, the Colossus computing and logical machine was put into operation in Britain, intended to decipher radio interceptions of messages from the fascist armed forces. The machine contained 1,500 vacuum tubes, but information about it was declassified only in the 1970s.

It should be noted that, thanks to its speed, ENIAC could perform in one hour a volume of calculations that a relay machine such as Bell’s Model V could handle in 15 days of continuous operation.

From the Series of Articles by Yu. Polunov “Historical Machines”.
The article was published in PCWeek/RE No. 13 dated April 19, 2006.