The birth of the Soviet missile defense system. The greatest Soviet computer
Lebedev's monopoly
Let's make a reservation right away that the BESM-6 is a big machine, in every sense of the word. Therefore, to write about it briefly will not work. Due to the abundance of material, I had to cut a huge article into three parts, and each continues the previous one, so for a complete immersion it is recommended to study everything sequentially.
In 1958, Lebedev launched into series simultaneously two versions of his BESM - the military, faster, M-20, and the civilian, slower, BESM-2. At this moment, he actually took a legendary place, similar to Kalashnikov, among computer manufacturers.
The Soviet Union loved monopolies no worse than decaying capitalism, but in our country the schools of various general designers belonging to various ministries were a kind of analogue of corporations.
ITMiVT machines were built in a huge circulation by the standards of the USSR - a total of several thousand copies, including several hundred supercomputers. They stood everywhere - in Dubna, in Arzamas, in Moscow State University, in the MCC, in the A-135 missile defense system, in the S-300 - but it’s easier to list where they were not.
The Lebedev school has achieved such a monopoly that we still live with their legacy, as with the eternal and permanent AK.
How did ITMiVT come to this position?
By 1958, Lebedev had two favorite students and heirs - Vladimir Andreevich Melnikov and Vsevolod Sergeevich Burtsev. The second deserves a separate discussion in connection with "Elbrus", for now let's consider the fate and work of Melnikov.
In 1951, the first graduation of students in the specialty "Mathematical and calculating devices and devices" took place at the MPEI, and the department was created in the same year. The graduation took place because the students were already trained in the design of analog computers within the framework of the specialty "Automation and Telemechanics". At that time there was no talk of digital machines, the head of chief of staff professor G.M. Zhdanov, like almost all scientists of that generation, specialized only in analog machines, in 1956 his textbook "Mathematical machines and devices of continuous operation" was published.
Despite such conservatism, Zhdanov also attracted experts in computer architecture to teaching, in particular, two graduates of the radio engineering faculty of the Moscow Power Engineering Institute - our old friends Kartsev and Matyukhin. Hearing about such a case, many others did not stand aside - also known to us Kitov, Rameev and, naturally, Lebedev, who was just beginning his path to becoming a guru of domestic computer engineering.
As we have already mentioned, Lebedev was an excellent electrical engineer, initially he specialized in energy problems and even built an analog machine in 1945 to simplify network calculations, while working at the Department of Relay Protection and Automation of Power Systems at MPEI.
Back in 1939, in opposition to the dissertation of prof. A.V. Netushila (future dean of the Faculty of Automation and Computer Engineering) Lebedev studied her topic "Analysis of trigger elements of high-speed pulse counters" and thought about a rather unusual architecture of a digital computing cell.
The defense took place only after the war, and after 3 years Lebedev drove off to Kiev to build the MESM (by the way, Khrushchev was then the First Secretary of the Central Committee of the Communist Party (Bolsheviks) of the Ukrainian SSR, and this was how their friendship began). When Lavrentyev pulled him back to Moscow from there (in fact, before that he had invited him there, to the post of director of an entire institute, which was difficult to achieve in Moscow), Lebedev, as we remember, had sketches of the architecture of a large machine.
In parallel with the creation of BESM, he continued to teach at the MPEI and there he recruited his initial team among the first graduates - graduate students, to whom he gave assignments to design the nodes of his new computer (earlier, the diplomas of students were much more interesting: as a graduation work, to develop a node of one of the first computers in Europe is not a transformer box for a dacha village to calculate).
Among the most distinguished in 1951 were V.A. Melnikov and V.S. Burtsev, the latter's life was generally difficult, he lost his parents during the war and treated Lebedev like a father. In addition to this pair, A.G. Louth, I. D. Vizun, A.S. Fedorov and L.A. Orlov.
In general, Lebedev had two very strong sides.
Firstly, he was a first-class organizer who managed to acquire a human resource in the shortest possible time and skillfully adapt it to work, in this only Brooke could compare with him.
Secondly, even here no one could compare with him, despite the outward severity, Lebedev was an extremely charming, calm, flexible and tactful person - qualities that were lacking in transverse Kitov, Brook, Rameev, Kartsev, Yuditsky - and indeed to almost everyone domestic designers who lost to Lebedev's school.
Lebedev is the only one who managed to turn the development of computers into a kind of conveyor belt under his strict guidance. He himself designed (as best he could, in the spirit of the old school) basic logical cells and invented a system of commands (strictly classical in the spirit of the 1940s, and, in fact, he invented only two of them - one used in MESM / BESM / M- 20, and the second for BESM-6), and everything else was done by his students, actually building the machine.
He was also responsible for an extremely important part of any Soviet development - interaction with partocrats.
Lebedev was a kind of Jobs of the USSR - he, like Korolev, knew how to push virtually anything, anywhere, at all levels of the Soviet system, but, unlike Korolev, he was so beautiful and subtle that he did not make money until his death (unlike all other designers) not a single bloodline, and he was loved and respected even by those who lost him to school.
As a result, by the end of 1951, the basis of laboratory No. 1 was made up of graduates of the Moscow Power Engineering Institute, who by 1953 had been completing Lebedev's BESM. Melnikov was engaged in the development of a control device and was so successful that when it came to assembling BESM-2, he became its de facto chief designer, as well as the designer of BESM-6.
Almost nothing is known about the initial stages of the development of BESM-6 (it is not even clear why 6 and where 5 were), sensible references are found only since 1964.
Exploring comparable Western architecture can help shed a drop of light.
IBM 7030 Stretch
In 1959, an extremely powerful commercial computer for scientific computing, the IBM 7090, was released and the first two true supercomputers appeared - the IBM 7030 Stretch and the Remington Rand UNIVAC LARC. LARC special footprint in stories did not leave, but the IBM 7030 has become a real icon.
First of all, it was the first to achieve performance in excess of a million operations per second - 1,2 MIPS in 1959, a crazy figure; the first to use the SMS system - standard modular cards; the first, during the development of which the basic principles of hierarchical hardware design were discovered and applied, starting from the development of an optimal command system and ending with the implementation of individual modules on standard cells.
The Stretch was the first production computer to use standard ferrite memory modules, with oil immersion cooling, another innovation. In addition, Stretch could read and write into memory in six parallel threads, which made it possible to achieve a speed of working with RAM at the level of 2 MIPS (yes, in those days, a situation was possible when the memory was faster than the processor).
An eight-bit byte and 8/32/64 bit machine words, which is the most convenient and has become a standard until now (unlike the fanatical domestic machine words with a width of 22 to 50 bytes, often not even multiples of the size of the addressable memory cells), also appeared exactly in it, and the Stretch processor was completely 64-bit long before Intel Core. In addition, the computer had hardware support for working with bit fields, which made it possible to work with data types of variable length.
Stretch supported real arithmetic, integer arithmetic, and alphanumeric characters. Also, it was the first to implement multiprogramming and, within this framework, memory protection and standard interrupts.
Stretch was the first in the world to be built on emitter-connected logic circuits, super-powerful and requiring good cooling, but as fast as possible, ECL logic in various guises became the standard for supercomputer hardware until the 1980s.
First use of a XNUMX-stage standard pipeline in a processor (classic Fetch, Decode, Execute steps). We will talk about the conveyor separately below. For the first time in the world, the most flexible and effective (albeit not cheap) universal implementation of work with peripheral devices was applied, which has become the main feature of mainframes to this day.
A specialized data exchange coprocessor (ESC - exchange specialized computer, yes, then there was still no clear terminological distinction, in LARC, for example, each of the two processors was called a "computer", and the I / O subsystem was called a "processor") played the role of the main switch, providing communication between 32 I / O channels and RAM, thereby offloading the central processor. This scheme was so effective that it migrated to the IBM S / 360 and is still used today in large computers.
In addition, Stretch's value was in the fact that it was fully compatible with a colossal list of equipment from IBM - from magnetic drives of all types to typewriters and punchers, and thanks to the channel coprocessor, the whole economy always worked in parallel at full speed, and was connected by simple plugging desired cable.
Soviet engineers, who suffered when trying to cram the unsupported (since almost every non-trivial connection of a large computer with peripherals required monstrous crutches), would cry with relief if they happened to set up Stretch.
The Stretch command system was incredibly progressive, many of the ideas discovered during development by Gene Amdahl, Stephen Dunwell, Fred Brooks and John Cocke are literally used in modern microprocessors and have become the standard. since the 1960s: first for powerful systems, then everywhere. Among them are pre-decoding of instructions and pre-fetching of operands (the first-ever form of speculative execution based on branch prediction), reloading the pipeline after erroneous transitions, advanced index registers, and much more.
The decoding and prediction unit in the Stretch processor was actually a separate computer with its own pipeline. Due to this, while formally remaining uniprocessor, Stretch required only 4 instructions for matrix multiplication. Among other things - the machine turned out to be compact, being 35 times more powerful than the IBM 704, Stretch required the same machine. hall of 185 sq. m. and consumed about the same amount of energy.
In general, it takes a long time to describe the advantages of this machine, the most important thing is that all this is 1959, and all this is described in open sources of that time, as well as the design of UNIVAC LARC.
The most interesting part of the story
And now the most interesting part of the story - Lebedev, as part of a group of Soviet specialists, arrives in the United States in August 1959 at IBM to study the advanced experience of computer engineering. In two weeks, the Soviet delegation visited MIT and IBM factories, got acquainted with the scientific computer 7090 and, in general, with the organization of production and interaction "computer company - customers from universities." Most likely, it was at this moment that a super-idea took shape - to repeat the same in the USSR. To create a powerful computer architecture unified for scientific centers.
The role of IBM was to be played by ITMiVT, and the role of Amdall and Brooks was to be played by Lebedev and Melnikov. It remains to select a prototype car, the role of which was to be played by the future BESM-6.
By the way, Lebedev visited many places - both in Germany, and even in 1965 in Japan, and his protégé Melnikov - in India and China, and nothing has grown together with India, but in China he helped to master the Chinese clone BESM-2 in production ...
There was no talk of cloning Stretch - the machine was monstrously complex, the Soviet industry would not have pulled it. In addition, many details of the implementation were hidden, and in itself it occupied a slightly different niche that Lebedev aimed at and which the Soviets really needed.
The USSR desperately needed a powerful (much more powerful than anything that was in the scientific research institute at that time), universal, inexpensive (relatively) scientific computer that could be massively reproduced. There were many points of its application - the Kurchatov Institute, Moscow State University, Baumanka, Dubna, Novosibirsk, Arzamas, and so on, all related to nuclear research. BESM-2 and M-20 did not pull - all research centers demanded an analogue 7090, a car with a speed of at least 200 KIPS.
Under this case, Lebedev struck a business trip to the United States, everyone understood that it was necessary to study advanced experience.
To this end, he, the favorite of the Academy of Sciences, ministers and Khrushchev, knocked out funds and assembled a team. The mood was bright, everyone was on the rise - just about the USSR will learn the secrets of effective development of scientific computers and will finally create its own one, instead of the zoo "Minskov", "Nairi", "Uralov" and others, incompatible with each other, released small circulation and insufficiently powerful.
In a sense, Lebedev really wanted to become a Kalashnikov from a computer - to create a reference implementation, a symbol of Soviet power, which could be stamped in hundreds.
Has he coped with his task?
It will not be a spoiler to say that it is not quite.
So why and where did Lebedev fail, and where did the myth about BESM-6 come from?
To answer this question, you first need to go even further to the West.
So, in 1959, Lebedev was inspired to start developing his own architecture - not just a large machine, but a scientific supermachine, built on completely different principles than MESM and BESM, which looked like calculators from the general store against the background of the monstrous 7030.
Who would lead the development - there was no doubt, of course, Melnikov, who had proven himself in BESM and BESM-2.
It was also obvious who would develop the units, the Melnikovskaya group.
Lebedev, as always, took on organizational issues, knocking out funds and general support along the party line, as well as studying world models of computer engineering and developing a system of commands for a new machine. And there was something to study - it was necessary to decide on a prototype for copying.
In 1946, the US Navy created a small company ERA (Engineering Research Associates) from engineers who worked on decoders of Japanese naval codes during the war years. The company developed several civilian vehicles, including the ERA 1103, but Congress condemned the Navy as a commercial vehicle, and in 1952 the company was sold to Remington Rand. Then Sperry (which already owned UNIVAC by that time) in turn bought Remington and spun off the computer division of Sperry UNIVAC, releasing a redesigned ERA as a competitor to the IBM 704 - UNIVAC 1103.
CDC 1604
Sperry was apparently aiming for mass commercial mainframes, determined to become the second largest player in the market after IBM, and in 1957 a group of ex-military engineers got fed up with it. William Norris, Robert Perkins, William R. Keye, Howard Shekels, Robert Kisch and Seymour Roger Cray left Sperry, chipped in 5 dollars and registered their company that has become a legend - Control Data Corporation.
The chief engineer was Cray, one of the greatest computer architects in history, a man without irony called the Father of all Supercomputers.
However, his first work was a much simpler machine - the CDC 1604. Created as a scientific and commercial computer, in a lack of money (the start-up company did not even have the funds for normal transistors, Cray assembled a prototype from defective ones that he could get at a bargain price in local radio parts stores), he immediately uncovered all the incredible power of Cray's genius.
The CDC 1604, being several times cheaper than the monstrous scientific IBM 7090, turned out to be faster than it, for some time becoming the most powerful general-purpose computer on the planet with a performance of about 200 KIPS.
The CDC 1604 had the architecture typical of powerful American machines of the 1950s with an adder, and not the more progressive general-purpose registers (there were only the adder register itself, or the accumulator, as it is sometimes called, 6 index registers, a program counter and an arithmetic auxiliary register) and , respectively, was a unicast, 48-bit machine word contained two 24-bit instructions. The machine had integer and real arithmetic.
An interesting feature was the work indication. The top three bits of the battery could be read by the DAC and played back through the speaker using a tube amplifier built into the console. This scheme could be programmed for a variety of sound effects and warnings to the machine operator. Anyone who knew well the architecture of the CDC 1604 and the executable program could immediately understand by sound where the error occurred.
So, the prototype was found.
The CDC 1604 was cheaper and faster than the IBM 7090, its architecture was simpler, and in a year it sold a colossal circulation to American laboratories. This meant that by 1961 a huge array of Fortran programs had been written for it, among which the coveted software for nuclear scientists was recognized as the most valuable.
It is much easier to screw programs than hardware, so the goal of achieving binary compatibility with the CDC 1604 was from the very beginning.
Even in the USSR they understood that software is more important than hardware, computers can be developed, but where can one get millions of man-hours to write software for them?
In the 1960s, a revolution was supposed to happen - a Soviet computer was supposed to appear with a number of applications that were not inferior to the American ones (and so what if they were supposed to be coiled).
In principle, such a plan would have had a chance of success, but Cray failed.
In the midst of preparations for cloning the CDC 1604, on August 22, 1963, Control Data announced the CDC 6600, one of the greatest machines in history.
IBM was put to shame, they had not yet finished shipping their Stretch to all customers, and Cray's computer had already sent him into the stone age of computer science. By tradition, it was much more compact and cheaper than the IBM monster and 3,5 times faster than it - more than 3 megaFLOPS.
The first ever superscalar processor, 10 peripheral coprocessors, freon cooling (also the first in the world) of the most compact boards assembled using cordwood proprietary technology on the latest planar silicon transistors (400 pieces!), An advanced multitasking operating system SIPROS (Simultaneous Processing Operating System) - here are just a few of the innovations of this machine. The first computer buyers were the Atomic Energy Commission and the Weather Bureau, and by 000 1967 CDC 63s were in the hands of elite customers and became the backbone of scientific research at the time.
Atlant
At the same time, the third iconic supercomputer of that time appears in Great Britain - the famous Atlas (Atlas), developed and produced jointly by the University of Manchester, the cradle of all British computer science, and by Ferranti and Plessey companies commissioned by the government for use in the same difficult business as and CDS and BESM - the creation of nuclear weapons.
Atlas was built on fairly ancient germanium bipolar transistors, but had an incredibly progressive architecture, becoming the third pillar of modern machines, along with the IBM 7030 Stretch and CDC 6600. In total, 3 original Atlas were manufactured and two more upgraded Atlas 2 Titans.
Atlas used the second popular computer word scheme, which was also used in the CDC - the 2/12/24 bit format, instead of the IBM 48/8/16 bit standard (as we know, the more convenient IBM won). A 32-bit machine word could contain one floating point number, one instruction, two 48-bit addresses or signed integers, or eight 24-bit characters.
Among Atlas innovations there was a supervisor (three own registers for the program counter) and virtual memory (full-fledged), work with external devices was organized in a very original way, through separate registers for communication with I / O, the machine had an incredible number of index registers at that time - already 128. In addition, the Atlas processor had a unique asynchronous pipeline that worked on readiness, rather than being clocked as usual.
Because of this, it was difficult to evaluate its performance, but according to tests it roughly corresponded to Stretch (Atlas added two floating point numbers in about 1,59 μs, and Stretch in 1,38-1,5 μs). It was not until 1964, when the CDC 6600 appeared, that the Atlas was significantly surpassed, with Cray later admitting that it was the description of the prototype of this machine that prompted him to ideas that allowed the 6600 to be completed much earlier than originally intended.
Also a unique part of the architecture became the so-called. extracodes - what would now be called firmware, it was they who made it possible to build an operating system that surpasses the speed and functionality of an IBM machine. Extracodes were used to call mathematical procedures that would be too inefficient to implement in hardware, for example, sine, logarithm and square root, about 150 extracodes were responsible for supervisor functions, which significantly increased the productivity and security of working with the OS.
British scientists turned out to be so friendly that they themselves visited the USSR in 1963 and even gave several private lectures at ITMiVT on the Atlas machine, a small brochure was published on their results in the same year. As a result, work on BESM-6 went wild like a swan, cancer and pike. It began to be rewritten from CDC 1604 in order to take advantage of a huge software library, as evidenced by many of the architectural features of the original project that were included in the final version.
In both cases, the element base was transistors (which is normal for 1960, but as strange as possible for 1968), addressing was unicast, word width is 48 bits, command length is 24 bits, 2 commands are packed in a word, adder width is also 48 bits, address width 15 bits, general purpose registers 1 + 1 register-accumulator, accumulator circuit of addition, even the amount of RAM matched up to a bit - 32.768 words.
Naturally, all this was not an accidental coincidence - they began to design the architecture, focusing on CDC 1604. What is amazing - the size of the BESM-6 circuit boards was inch (specifically 6x8 inches), and in general, only hardware was metric. The architecture of the TEZ itself (a typical replacement element, as we called the elementary modules from which the machine was assembled) was also clearly developed with an eye on cordwood boards, albeit with about 6 times less installation density. In the USSR, it was difficult with transistors, so the logic itself was diode, and transistors were used only for amplifying and inverting the signal. As a result, only 60 of them were required, but 000 diodes (the CDC was assembled for 180 transistors).
And then Lebedev suffered, the fault of which was both his imagination and Atlas and CDC 6600, which appeared in the process of creating BESM-6.
From the first he desired extracodes and virtual memory, from the second - a superscalar pipelined processor with an orthogonal instruction system. Many technical solutions were also rejected - for example, the idea of channel work with peripherals, really successful in IBM machines. From 1960 to 1963, there were throwing - let's start one car, continue with the second, add the chips to the third.
As a result, horns and legs remain from the original CDC 1604 project.
Lebedev threw out integer arithmetic from the machine, since he could not achieve stable operation of the integer-real ALU, changed the format of real numbers (1 bit sign, 11 bit order, 36 bit mantissa for CDC 1604, 7 bit order, 1 bit sign, 40 bit mantissa for BESM-6) and command structure (6-bit opcode, 3-bit index or transition condition, 15-bit address or operand for CDC 1604 and two possible options: 4-bit index register, 6-bit opcode, 12-bit address / operand, or 4-bit index register, 4-bit opcode, 15-bit address / operand for BESM-6). Due to an attempt to organize virtual memory, the number of index registers was expanded from 6 to 15.
The command system has been completely redesigned.
The CDC 1604 had 11 integer instructions and 4 real, 9 shift instructions, 8 logic instructions, 15 memory instructions, 6 index arithmetic instructions, 4 jump instructions and 3 I / O instructions, 57 in total. BESM-6 had 12 real commands, 2 shift commands, 7 logical commands, 5 commands for working with memory, 8 index arithmetic commands, 7 jump commands and one (!) I / O command, 41 in total.
An interesting feature of BESM-6 were special bit commands, including “CLEARING THE NUMBER OF UNITS” and “CLEARING THE HOMEPA OF UNITS”. These teams were directly ripped from CDC 6600 and represent the so-called. "NSA Instruction" - instructions added at the request of the NSA to supercomputer processors for the convenience of cryptographers.
For example, calculating the number of ones is the command popcount, say popcount (10100110) = 4. It first appeared in the IBM Stretch processor and was then built into all older CDC and Cray machines until the 1980s and the end of the Cold War era and classic supercomputers ...
Why is it necessary?
Calculate the Hamming distance from the zero string in binary encoding. The NSA was cryptanalyzing the intercepted messages, and since the CDC 6600 had 60-bit words, one word was enough to store most of the alphabets of interest.
Cryptanalysts divided the message into lines, marked each unique character in the line with a single bit, calculated the Hamming distance using popcount and used it as a hash for further cryptanalysis. Unfortunately, it is not known whether at least one BESM-6 of the GRU or the KGB was used, the author strongly doubts this and thinks that these instructions were altered, rather, for presentation purposes - like look, our car can do that!
The instructions "ASSEMBLY" and "DISASSEMBLY" are a selection of certain bits by a mask, screwed specifically to organize at least some sane input-output of printed characters, as the old-timers mention, for example, these commands were used to transpose 80x12 matrices to work with punched cards. The transformation of a word into a textual octal representation was done by disassembling into groups of three bits per byte, in the GOST encoding, the codes of the numbers were immediately obtained. By assembly, the textual representation of octal numbers was converted into the actual number.
Naturally, it was impossible to assemble either Atlas or CDC on the Soviet element base, it was necessary to make a bunch of patches and crutches. One of the designers, V.N. Louth:
[Here, unfortunately, Louth is disingenuous, since they were no longer just there, but they were used to assemble serial computers, including in the USSR, just Lebedev did not consider it necessary for some of his philosophical reasons to use them ].
[Again, there were some pretty decent silicon ones, but ...]
These diodes had short switching times, many times better than those of transistors. However, the elements based on tunnel diodes turned out to have a poor load capacity, which led to the complication of the machine circuits, and we quickly abandoned them. The difficulty with using transistors was that they worked very slowly in saturation mode, and logic gates with unsaturated triodes were complicated due to the need to match the levels of input and output signals. And not only difficult, but also unreliable. For some time we have not seen a way out of the impasse. But then a completely new idea arose, never and nowhere previously described, at least for elements of computing technology.
In my opinion, A.A. Sokolov. The essence of the idea was to introduce an autonomous power source, galvanically not connected to other power circuits, into the well-known "current switch" element. For example, a miniature electronic clock battery could be used for this purpose.
The inclusion of a battery between the collector of the transistor and the collector load (resistor) made the switch an element with matched levels of input and output signals, and no particularly complex requirements were imposed on an autonomous power source. Of course, the battery could not be installed, since it will discharge over time, so in the real circuit it was replaced by a tiny rectifier consisting of a miniature transformer on a ferrite ring, two semiconductor diodes and a capacitor.
These rectifiers were called "suspended power supplies" (PPS). The paraphase outputs of the current switches, equipped with emitter followers, could operate on the inputs of the AND, OR logic circuits. The following circuitry emerged: passive combinatorial logic based on diode-resistor components
[The whole world had switched to TTL and ECL by then.]
In general, the element base of the first version of BESM-6, as we can see, was monstrous even at that time (and even by the standards of the USSR, which is even more striking!), But nevertheless, as usual, we have a good reason to be proud of how masterly we overcome the difficulties we ourselves have created.
Experience the Power of Effective Results
The result of all this was the emergence of a real mutant, externally (that is, in terms of bus width, machine word length, etc.) similar to CDC 1604, but assembled with elements of Atlas and CDC 6600, seasoned with a pinch of Lebedev's own unique vision and put on hardware implementation , perverted even by the standards of the USSR.
In 1963, students began to model the nodes of the future BESM-6, for a start, practicing the technology of working with transistors, which at the ITMiVT had no particular idea about at that time. It all ends with the fact that, according to the recollections of the younger student of Lebedev (with a smaller caliber than Melnikov) A.A. Gryzlov, they simply implemented the nodes from the M-20 on transistors, calling the resulting creation BESM-3.
Surprisingly, as we have already said, his initiative was supported by the head of the group and achieved launch into a mini-series, this is how a side sprout appears on the ITMiVT tree - BESM-4, to which Lebedev had nothing to do. According to the memoirs, Lebedev himself was not very pleased with such self-will, he did not care about BESM-4, all his attention and forces were absorbed by the BESM-6 project, but he did not interfere with the youth and even pressed a couple of levers in the party, which allowed him to launch the four in small batch with record speed.
In 1964, the institute had already assembled a prototype BESM-6, which had only one memory cube; slower transistors were used in the logic elements. In 1966, a prototype that had half-capacity random access memory (4 cubes instead of 8), but on modern transistors and diodes, operating at the design synchronization frequency, was already largely debugged, and factory tests were carried out on it in the fall, and in May 1967 year ended state tests.
BESM-6 was mass-produced from 1968 to 1987, a total of 355 cars were produced, the machine in Dubna was turned off in 1992, the penultimate (in the Mikoyan Design Bureau) was turned off and dismantled in 1995, but the very last ...
BESM-6 No. 345
BESM-6 No. 345 was manufactured in 1980, in 1981 it was registered in military unit 87286 (Sosnovy Bor, Leningrad Region), in 1982 it was set up by a team of the Moscow specialized chief assembly department, in 1983 it was introduced as a central computer of a complex full-scale simulator "Diana-Bars", developed by the specialists of the NITI them. A.P. Alexandrova. Its debugging continued until 1986.
Pay attention, by the way, to the typically Soviet imposing approach to installations, which says a lot about the quality of architecture and equipment. Installation of the machine took a year, another year was taken by unhurried debugging of the software (despite the fact that by the 1980s the experience and software had already been accumulated for 10 years!), Three more years were spent with the simulator, as a result, the machine was able to calculate something useful only after six (!) years after its release.
And this was considered a regular installation mode! At the same time, damned capitalism rotted to such a stage that the announcement of a delay in the installation of one or another machine for at least a couple of months became a reason to break the contract with heavy fines.
The fate of Cray-3 is indicative in this regard, it was supposed to be bought by Livermore in 1991, but the laboratory immediately revoked the contract as soon as it learned of the delay in delivery, and the failure to fulfill the contract hit Cray's reputation so much that he could only sell his supercomputer The US National Center for Atmospheric Research (NCAR), the military and laboratories refused to work with it, despite past merits.
In 1993, Cray-3 was installed in NCAR, but for a year it was not able to achieve stable operation, after which it was dismantled, and Cray Research went bankrupt.
In the USSR, the installation and debugging of even a serial computer that was already being produced for 13 years by that time could easily take five years of unhurried finishing work in place, and this was perceived as an absolute norm - the Union, it seems, was in no hurry.
Returning to BESM-6 No. 345, it was launched in 1986 and worked ... until 2008!
This is something beyond good, evil and common sense. It was used as a computer for a simulator of submarine crews, and for the last 10 years it has been working literally on an honest unprintable word; the personnel have been doing their best to keep the museum antiquity in working order far beyond its official service life.
The further fate of the rarity is unknown - it is possible that the BESM-6 of the Polytechnic Museum is she, perhaps the old woman ended her life in a furnace for scrap metal.
At the same time, pay attention to the magic of Lebedev - only 6 years passed from the development of the BESM-3 itself to the series, and for purely objective reasons, fiddling with the documentation for the car, debugging, etc., while Kartsev was driven from office to office by party officials six years and brought him to a heart attack, and after his death it took the staff another 1,5 years to finish the release of M-13!
Сonclusion
And finally, the burning question, did it happen?
Did you manage to create a niche computer covering the needs of Soviet research institutes? Equivalent to CDC 1604 that will run all the programs you need?
Alas, no, here Lebedev screwed everything up.
An attempt to focus on three sources and three component parts of the architecture at once led to failure - BESM-6 lost compatibility with CDC 1604 enough to proudly be considered a separate architecture and enough to stop running the coveted millions of lines of American code, for which everyone and started.
Lebedev was too clever, and as a result, it was not possible to achieve binary compatibility - Fortran programs, which compiled perfectly and worked on CDC, dropped out on BESM-6 in the most unexpected places. To correct them, even whole textbooks and monographs began to be written (for example, Borovin G.K., Komarov M.M., Yaroshevsky V.S. "Errors-traps when programming in FORTRAN"), but valuable time wasted away, the tasks stood.
As a result, the colossal project ended with mixed feelings, even during the testing phase in 1966.
The question remains - what to do now?
The consequences were as follows.
First, it was firmly decided not to be perverted anymore, but to simply and accurately copy the entire Western architecture in order to achieve binary compatibility. Lebedev himself, to his credit, realized the mistake and was no longer eager to design anything and, moreover, at the meeting of the Academy of Sciences supported the idea of borrowing S / 360 (this should be discussed separately).
Secondly, Melnikov was no longer allowed to develop independently. The BESM-10 project did not even begin, only descriptions and drafts were preserved, and "Electronics SS BIS", for which he was responsible until his death, was to become a clone of Cray-1.
Thirdly, in the country's most important nuclear center, Dubna, it was necessary to urgently deliver something that works with American software, as a result diplomacy got involved and attempts were made to buy or steal, exporting through neutral countries such as Switzerland, the real CDC 1604, and also better - CDC 6600. Attempts were only partially successful.
CDC 1604 was used by the US Navy and also to control the launch of the Minuteman I, therefore it was a military technology, but by 1968 it was discontinued and obsolete, so the Control Commission did not object to its supply. In 1968 (simultaneously with BESM-6) CDC also settled in JINR.
Of particular interest is that the CDC company itself was not opposed to selling anything anywhere, even to North Korea, and William Norris, its director, wrote a special letter to his Congressman Richard T. Hanna, asking him to exclude the corporation from accusations of cooperation with the communists:
All countries including the Socialists have a substantial base of computer hardware technology on which to build further advances in the state of the art. The major strength of the US in computer technology is its ability to market superior cost / performance computer systems for a wide range of applications. This does not mean that for any given application or group of applications, another country cannot build the equivalent as far as performance is concerned or even exceed what the United States has available. Also, there is no evidence to my knowledge that the USSR has ever been prevented from carrying out a military project because of the lack of adequate computer technology ...
We respectfully request that your Committee review the above points and consider incorporating them into the record. We would be pleased to have the privilege of appearing before your Committee to give you our more detailed views on these potential relationships with the Socialist countries and in stating our reasons in support of the Administration and Congressional trade initiatives and objectives.
Letter from William C. Norris, Chairman of
Control Data Corporation to Congressman
Richard T. Hanna, 1973.
The delivery with the machine included a compiler with Fortran, and with source codes, and the JINR programmer team under the leadership of Nikolai Nikolaevich Govorun was inspired by them and tried to write an analogue for BESM-6, since it refused to work directly.
As a result, I first had to write an assembler (the autocode with Lebedev's mnemonics was so inconvenient that in practice it was not used), then the loader, library support, and the rest of the operating system, which was logically named Dubna.
Naturally, this was a well-deserved reason for pride - the hard work of the professional programmers department was carried out by physicists and amateur engineers, the result was generally satisfactory.
A direct continuation of this story awaits us in the next article.
- Alexey Eremenko
- https://ru.wikipedia.org, https://www.ibm.com, https://www.computerhistory.org, http://www.vintchip.com/, http://www.histoire.info.online.fr, http://www.besm-6.su, http://www.chilton-computing.org.uk, https://www.1500py470.livejournal.com, https://ramlamyammambam.livejournal.com
Information