The birth of the Soviet missile defense system. The end of modular machines
As we remember, the development and practical adoption of 5E53 was accompanied by a sincere spiritual and scientific upsurge of the entire staff of the SEC.
The basic problems of modular arithmetic were successfully solved, the machine was accepted, the prototype worked perfectly, the algorithms were written. So far, no one could have imagined how everything would turn out, and in the wake of euphoria, the employees decided not to stop at the project that had already been successfully completed (as it seemed to them) and to build something else.
In this article, there are many illustrations of similar Western projects, unfortunately, this is a forced measure, since it was not possible to find photographs related to the late works of Yuditsky and the SVC team in any open source (and it is not known whether photographs of these secret prototypes existed at all).
When creating 5E53, a monumental groundwork was erected for further research - the mathematical foundations of working with SOCs, methods of increasing fault tolerance were finally polished, periphery was created (and forming it in the USSR was a separate feat, comparable to the development of a supercomputer, they complained about the monstrous quality of Soviet drives, printers and other things even fanatical communists). For the firmware, a convenient 256-bit diode matrix on a dielectric substrate, DMR-256, was manufactured and its production was organized at Mikron, methods for assembling hybrid ICs of increased density were tested.
Let us not forget that Yuditsky's employees did not differ in senile ossification characteristic of many official Soviet scientific schools, there was no smell of gerontocracy there, everyone was (like their boss) young and daring and wanted to create further. They planned to improve the next version 5E53, build it on real microcircuits and include a number of even more progressive technical solutions there (fortunately, in the preliminary design of the new supercomputer, there were no longer any restrictions on working specifically with missile defense algorithms). A real revolution in the field of supercomputers was conceived, fueled by a friendly competition with Kartsev's group - Yuditsky understood that his machine was objectively slower, he was burned by the desire to squeeze out the same number of operations on a modular computer.
The intended monster was radical and progressive even by the standards of the astounding Western architectures of the 1980s (when nothing else was tried to improve performance). In addition to modular arithmetic, it had to be modular, reconfigurable (!) And with a hardware-microprogram implementation of Java (!) USSR) or an even weirder language - IPL (Information Processing Language, developed by Allen Newell, Cliff Shaw and Herbert A. Simon) of the RAND Corporation and the Carnegie Institution around 1: possibly the first functional programming language in the world, assembly type, oriented to work with lists).
It is worth talking about these very unorthodox innovations separately, especially since the idea of hardware support for nuclear devices was successfully borrowed by Burtsev for his "Elbrus". It all started in the United States with the famous Burroughs company, which in the early 1960s decided to enter the mainframe market.
Thus was born the B5000 - a great machine that outstripped the development of computer architecture by a decade.
Developed by a team led by Robert Barton in 1961, the mainframe was the first in a series of Burroughs large systems, produced until the late 1980s, with a unique architecture never seen before by any other manufacturer. Renowned computer scientist John R. Mashey lists her as one of his most admired inventions:
Burroughs machine expert, University of Virginia professor Alan Bateson, in an interview for "Whatever Happened to the Seven Dwarfs?", Put it this way:
In the 1960s-1980s, many companies produced various lapel badges in honor of themselves, their products and any technologies; these badges were worn by company employees, they were distributed at exhibitions and presentations, and used in advertising. Nowadays, many of them have considerable collection value. The first large computer Burroughs so impressed the imagination of even its creators that in honor of him was released a badge with perhaps the most original advertising inscription: "I touched the B5000" (I touched the B5000). Also called is the book of memoirs published in 1985 by the famous computer scientist and developer, specialist in the theory of languages and compilation, who worked at Burroughs on its version of Algol Richard Waychoff (Richard Edwin Waychoff).
The same badge (from the personal collection of Ole Hagen Jensen)
What is so unique that Bob Barton and his team built?
They knew they were entering the large systems market much later than their competitors at IBM, RCA, and others, so they needed to offer something truly exceptional.
In addition, they had the opportunity to study the architecture of the systems already presented and realized that they all have certain common disadvantages, concentrated in the area of what the most advanced computer is useless without - software. Machines were supplied separately from software, and all programs were written, as a rule, by a company that bought a computer from scratch and for themselves, no one even thought about the convenience of development, architecture was separated from software. Computers were developed by special groups of engineers who did not even think about how easy or difficult it would be to develop programs for their machines. The B5000 was supposed to provide answers to all of these questions.
It was the first computer in the world that was developed as a single hardware and software complex and an integral system, in contrast to S / 360, for which IBM did not even manage to bring to mind the originally planned OS / 360 (moreover, this practice of separate design was preserved in IBM and onward and eventually became dominant, including the x86, PPC, and ARM instruction systems). Designing a machine without regard to executable code led to numerous crutches in the implementation of software and operating systems in the 1980s, which indirectly affects the quality of software today.
For the B5000, things were different. From the very beginning, it was designed in conjunction with the language, OS and system software. Its main architectural feature is full hardware support for high-level languages, based on two innovations - the stack processor and tag-descriptor memory.
The B5000 did not have an assembler, its processor was able to directly execute the JAVU instructions. During the development, the question arose, which of the languages to choose as the main one? There were only a few of those in those years, but the choice fell on the most powerful academic language, for which a new standard has just appeared - Algol-60. It became the main system language, and there was good support for Cobol (which mainly involved working with powerful string operators) and Fortran. Thus, the entire architecture of the B5000 was built around a very advanced language, for example, long before the #define directive in C appeared, a similar mechanism was used in the B5000, moreover, it was built into the language itself, and was not a preprocessor command.
Most other computer manufacturers could only dream of implementing the Algol compiler due to its complexity and extremely slow software implementation. It was believed that it was unrealistic to achieve an acceptable speed when using it, and if you do not use hardware support, this was the case (in particular, this is one of the reasons why Algol, as a language, did not gain wide popularity at that time). The then young student, the legendary Donald Knuth, who had previously developed Algol-58 programs for their early machines, worked on the implementation of the language at Burroughs for several months of summer vacation.
The Burroughs Algol compiler was very fast - it made a tremendous impression on the famous Dutch scientist Edsger Dijkstra. During the B5000 tests in Pasadena, his program was compiled at the speed of reading from punch cards, which was an amazing result for the time, and he immediately ordered several machines for the Eindhoven University of Technology in the Netherlands, where he worked. Hardware support and sufficient RAM allowed the compiler to operate in single-pass mode (even though early machine assemblers almost always used multi-pass compilation at the time).
B5500 and its components.
1 - printer (B320, B321 or B325), 2 - punched card reader (B129), 3 - control console with 4 - teletype for entering commands (for example, Teletype Corp Model 33 KSR), 5 - puncher (B303 or B304), 6 - magnetic tape drives (B422 / B423 –15 Mbit), 7 - processor (one of two possible), 8 - I / O coprocessor, 9 - terminal controller, 10 - display and debug console. Also, the photo did not fit 2 cabinets with RAM and a cabinet of the second processor (photo of the University of Tasmania http://www.retrocomputingtasmania.com)
We dwell on all these advantages in such detail precisely because similar ideas occurred to Yuditsky's group, and later to Burtsev's group (except that Yuditsky, unlike Burtsev, did not have a living Burroughs at hand to study). As a result, many things described as unique and unparalleled in the world, implemented in Elbrus, actually appeared much earlier, including advanced protection mechanisms.
Even the improved Burroughs Algol did not include many of the insecure constructs required by the operating system and other system software. To support them, a special extension Espol (Executive Systems Problem Oriented Language) was developed. Espol was used to write the operating system kernel - Burroughs MCP (Master Control Program) and all system software. Espol's hardware support made it easy to implement virtual memory, multiprocessing, fast context switching and procedure calls, memory protection, and code sharing. The B5000 was the first commercial virtual memory machine. In addition, due to this, complete reentrancy of the code was realized in a natural way, without additional efforts on the part of programmers. The Espol language was replaced in the late 1970s by the more advanced Newp (New Executive Programming Language).
All unsafe constructs in a program are rejected by the Newp compiler, unless a block in RAM is specifically tagged with a special tag to enable these instructions. This block marking provides a multi-layered security mechanism. In addition, Newp programs that contain unsafe constructs are not initially executable. A special system security administrator can make them executable, but ordinary users cannot. Even privileged users, who usually have root access, cannot start them without explicit administrator permission. Only compilers designated by the operating system could create executable files with extended commands, and only MCP itself could designate a program as a compiler (via the Security Admin console command).
Newp was so advanced that it was supported by the original Unisys ClearPath mainframe architecture, the heirs of the B series, until 2014, when the x86 migration began. Also, long before bash in Linux, a separate command line language WFL (Work Flow Language) was developed to effectively manage MCP. On IBM mainframes, its counterpart was the famous Job Control Language (JCL).
Computers designed specifically for Java were very complex, but later developed along the path beaten by Burroughs until the mid-1980s (among them are the Lilith workstation of Nicklaus Wirth, the father of Pascal and the famous LISP machines), when they were superseded by the x86 architecture and General purpose RISC processors.
The tag-protected descriptors in the B5000 are checked in hardware at every memory access at every data change step. In addition, the system does not need manual management of memory allocation, and moreover, this is generally impossible. Every segment of protected data, such as code, cannot even be read, let alone modified, in an uncontrolled manner, which makes most attacks impossible and errors impossible.
Of course, a suitable privileged process can explicitly change the bits of the tag and thus change itself, but only the ESPOL compiler can generate such code, while the MCP will refuse to execute anything that it identifies as ESPOLCODE, no matter what privilege level the person who tries to start it has ... Such programs must be installed as part of the operating system at the very beginning, and it is fundamentally impossible to add or change them in the process.
As a result, Burroughs mainframes remained the most secure and safest machines on the planet for the next thirty years, which is why the US Federal Reserve System chose them for many years as its banking computer standard. As we already said, this architecture (naturally, continuously improved by new models) was supported in hardware until recently, and only since 2014 there has been a transition to standard x86 servers.
One of the few real problems with the B series was that its parent, the B5000, ended up with an extremely complex processor and memory subsystem. In the era of transistor machines, one could turn a blind eye to this, but this moment greatly complicated the integration of subsequent models. In the years when all manufacturers switched to single-chip models with a custom processor, the Burroughs series of large machines were still available in multi-chip designs.
The first version of the mainframe-on-a-chip, SCAMP, did not appear until the late 1980s, when it was too late, although this processor and its successors were used by Unisys until the mid-2000s.
Reliability was an additional problem with the enormous complexity of the processor and the sheer number of transistors and early ICs.
However, Burroughs computers could not break - the company had a reputation as one of the best suppliers of high availability systems, their machines traditionally worked for years without an emergency stop (also, by the way, by the way, by the way, by the way, their adding machines, rightly considered the most reliable in the industry). In order for the B5000 to meet stringent quality criteria, considerable redundancy and flexibility have been incorporated into the system.
Hardware modules could be turned off and installed on the fly without stopping work or losing data, which was something fantastic at the time. In order to monitor the state of all nodes of the machine and reconfigure the system in the course of calculations, bypassing bad sections, a special MDLP coprocessor (Maintenance Diagnostic Logic processor) was added. It was also used by engineers to diagnose all system components.
As a result, despite the fact that the B5000 was an order of magnitude more complex than traditional machines of those years, its reliability not only did not suffer, but also significantly surpassed most computers of this class.
The credibility of the company in the banking environment was so high that in 1973, when the Society for Worldwide Interbank Financial Telecommunications (SWIFT) was created, it was Burroughs who built its backbone switching systems in 4 years of work. And to this day, Burroughs' successor, Unisys Corporation, is the largest provider of the SWIFT network.
The "desktop mainframe" Unisys Micro A (1989) is actually the first server in the modern sense of the word. Its processor is SCAMP-A (top) and later is SCAMP-D (1997, bottom). By order of Unisys, the chips were first produced by IBM in their factories, then by LSI. The Unisys ClearPath Libra 6200 Server (2012-2015) is the latest to support the B5000 architecture (Photo by the University of Tasmania http://www.retrocomputingtasmania.com, from the private collection of John Culver https://www.cpushack.com/ and the official Unisys ClearPath prospectus )
The B5000 was used by NASA, US Air Force, Carrier Corporation, University of Washington, University of Denver, Caltech, Stanford University, Monash University in Australia (they were Burroughs loyal to the end and had all their machines in sequence, up to the B7800), Drexel Institute of Technology in Montreal , British Post and American Bureau of Mines.
Also in 1964, Burroughs built the B8300 for real-time applications such as airline reservations. A rather rare version of Algol 60, Jovial, was chosen as the system language. It was developed in 1959 as a new high-level programming language for real-time systems at SDC by a group led by Jules I. Schwartz and was originally a dialect of Algol-58, as indicated by its jocular name (Jules Own Version of the International Algebraic Language).
At first, it was intended to program the electronics of combat aircraft, but in the 1960s it became an important part of a series of US military projects, in particular SACCS (Strategic Automated Command and Control System - a system that controlled nuclear weapons USA) and of course SAGE. Approximately 95% of the SACCS software (co-developed by ITT and IBM) was written by SDC in Jovial. Development took two years (about 1400 man-hours), more than twice as fast as SAGE software.
In the late 1970s, when developing the standard architecture for the military processor MIL-STD-1750A, it was decided that Jovial would remain the primary language for this architecture. Many companies provided their compilers for it - Advanced Computer Techniques (ACT), TLD Systems, Proprietary Software Systems (PSS) and others. The last standard for this language, MIL-STD-1589C, was adopted in 1984, currently three dialects of this standard are still used: J3, J3B-2 and J73. Jovial was only discontinued in 2010, although compilers continue to be released.
As with Cobol, most of the software implemented in Jovial is mission-critical, and maintenance is becoming increasingly difficult, with partial replacements starting in 2016, although sometimes the choice is more than odd. For example, the software for the famous B-2 bomber was ported from Jovial to Pure C (!), Which can hardly be considered an effective solution in terms of safety and ease of support.
Approximately the same architecture was proposed by the SVC engineers, but their supercomputer had one more unique feature - it was, as we have already said, modular!
Yuditsky's new machine was supposed to include central processing subsystems (up to 16 central processors), input-output (up to 16 input-output processors), memory (up to 32 sections of 32K x 64-bit RAM) and a powerful modular system for dynamic switching of the listed modules in complex graph (any CPU could be connected to any PVV and any section of RAM). The overall performance of the computer was estimated at an absolutely monstrous 200 MIPS - the 1 Cray-1977 produced 160! In the processor, of course, a table implementation of arithmetic was planned.
As a result, Yuditsky noted with pleasure that, despite the extremely atypical system of residual classes for a supercomputer, his new project would be able to beat Kartsev's M-10! It really was an absolutely unique hybrid, which absorbed all the most advanced world developments in the field of computers of those years, the parallel matrix architecture from the M series, hardware support for the B5000 YED and, of course, the proprietary technology from Yuditsky himself - SOK.
The most striking thing is that the result did not at all look like a cross between a hedgehog and a snake - it was an absolutely working, extremely comfortable and most powerful machine in the world of those years, the closest American competitors lagged behind by a generation. Plus, she was incredibly reliable.
In general, we all already understand that it would not have been possible to implement it in a series in the USSR even at the cost of the chief designer's life.
To implement tabular arithmetic, the machine needed a new compact, large-capacity permanent memory. For several years now, a division of S. A. Garyainov has been developing it at the SVTs. The essence of the work was to create unpackaged diode arrays, as well as the design and manufacturing technology of devices based on them.
It was for this purpose that they wanted to adapt the already mentioned DMR-256. On the basis of the matrix, a corresponding original structural system was developed: the DMR crystals were mounted on a sital board, the boards were assembled into a seven-story MFB stack (multifunctional unit), the stacks were installed on a large printed cross-board. Several backplanes were mounted in a metal sealed block case filled with freon. To remove heat from the block, heat pipes were installed in it.
The preliminary project of a unique computer, simply indexed with Roman numerals "IV", was completed in early 1973. "IV" was conceived as a prototype for subsequent developments of the SVC. However, even before the project was completed, it seemed to be put to good use.
At the end of 1971, the Sukhoi Design Bureau Kulon applied to the SVC with an order for the development of CAD systems for aircraft. High and promising requirements were imposed on CAD, which outstripped any capabilities of Soviet computers of those years.
The system was supposed to support more than 700 automated workstations for the developers of the aircraft and its components. Each AWP was a terminal with a plotter, and the calculations had to be done on the main supercomputer (in those days, the annual output of even simpler AWPs in the USSR was no more than half a thousand). The draft design was completed and accepted with satisfaction by the customer, but the Ministry of Radio Industry (headed by whom it is known) refused to produce the car, citing a lack of funding (despite the fact that the project was intended for the military Sukhoi Design Bureau, and we did not spare money for the defense industry).
However, an even more interesting use for "IV" appeared almost immediately, in early 1972. Then the SVC received an order from the GRU itself for the development of a preliminary design of a supercomputer for processing data structured in a special way (translating from the GRU language into a human one - for breaking ciphers), which received the code name "Machine 41-50".
A 64-bit computer had to have a speed of at least 200 MIPS, 16 MB of RAM and advanced peripherals. The SVC decided to build a vector computer with a system of commands working on arrays and focused on the implementation of the customer's algorithms. In this case, the problem of dynamic parallelization was solved at the hardware-microprogram level. Draft project 41-50 SVC was carried out jointly with the Institute of Cybernetics of the Academy of Sciences of Ukraine, another underestimated Soviet genius, one of the world's best specialists in parallel computing and director of the IC, Academician V.M.Glushkov, was involved in the work.
It makes sense to somehow start a conversation about Glushkov separately - he was one of the world's largest scientists in the field of computer science (during the 15th edition of the British-American Encyclopaedia Britannica in 1973-1974, an article on cybernetics was commissioned to Glushkov!), But his projects (and there were absolutely amazing things, for example, the Soviet Internet) they drowned so mercilessly that (according to the tradition of outstanding Russian specialists in the field of computers) he did not live to be 60 years old, having died from a heart attack.
Glushkov was appointed scientific leader of the project, and two special subdivisions (a branch of the SIC) were created in the IC, headed by Z. L. Rabinovich and B. N. Malinovsky. The chief designer was Yuditsky.
The design of 41-50 began with studying algorithms for solving customer problems and trying to fit them into modular arithmetic (as we can see, in all projects of SOK machines, the work was based on algorithms - in fact, this was the drawback of this class of computers - a huge binding to specific tasks, making the car almost highly specialized). The work was headed by V. M. Amerbaev - as a mathematician and the main author of modular arithmetic, and L. G. Rykov - as a circuit engineer implementing these algorithms.
L.G. Rykov recalls:
The result of the research was collected in the work RTM U10.012.003 "Machine algorithms for two-stage non-positional arithmetic", and on the whole it was disappointing. The fact is that in the tasks of the GRU, the percentage of non-modular operations was colossal, it was impossible to reduce them to SOC, and it was stupid to constantly convert back and forth and drive them into a regular coprocessor.
As a result, the performance of an ultra-complex and powerful computer would not exceed an ordinary supercomputer of traditional architecture. In general, the RNS system gave bonuses due to reliability, ease of implementation of table arithmetic and reduction in the amount of equipment, but Yuditsky was not a fanatic and understood that modular arithmetic was not a silver bullet. There are cases in which it simply does not fall on the algorithms, despite all the tricks.
In the final, after discussions and discussions, the SIC decided to abandon the SOC while maintaining the general vector-modular scheme of the machine and revised the project. Such flexibility favorably distinguished them from many Soviet design bureaus, which, having once found a more or less successful technical solution, continued to fanatically stamp it (like transistor versions of BESM in all versions and its own command system, which is extremely successful on some tasks and to the same extreme curve - on others).
They decided to make the machine, of course, on IC and as a basis they took the most powerful then in the USSR emitter-connected logic of the 100 series. Before it was stolen, it was called Motorola MC10000 (aka MECL - Motorola emitter coupled logic) - a series of quite powerful and fast ECL microcircuits, developed in 1962 (MECL I). The series had several generations - I, II, III and 10000, released in 1971. However, it differed from the 1968 version only in the resistor ratings. After 7 years it was mastered to be copied in the USSR as the IS100, it was intended for the most powerful computers such as Elbrus.
Unfortunately, the microcircuits of this series turned out to be extremely difficult for the Union and had huge problems with quality and stability, for which they are notorious (we will talk about the IS100 in the part about the A-135 and Elbrus, the devil will break his leg in copying powerful ESLs in the USSR, and this topic needs to be dealt with separately, it is closely linked with the commercial relations of the two giants - Motorola and Fairchild).
In the West, Motorola 10k was not the most popular choice of supercomputer builders, for these purposes they most often used an ESL from a competitor - Fairchild, the Fairchild F100K series (later they tried to copy it with a 10-year delay for Electronics SS BIS - the K1500 series, the result was, well, let's just say - not very successful, this is also the subject of a separate conversation). It was on F100K (3 microcircuits out of 4 types used - 11C01, F10145, F10415 and only one MC10009 for the address sampling circuit, Cray used a cheaper one in the place where it was not critical) Cray-1 was assembled.
The great and terrible Cray-1, one of its processor boards based on Fairchild F100K chips and the Soviet clone Motorola 10k - 100 series of the development period (photo https://cdn.britannica.com/, https://en.wikipedia.org/, https://ru-radio-electr.livejournal.com/)
IS100 production was mastered at Mikron, Vent in Vilnius, Svetlana in Leningrad and Integral in Minsk. Then problems began, the composition of the series did not provide for vector chips, as a result, additional ICs were required, which were not in the release program.
It was decided to join the program by developing the missing microcircuits for it. And the topic "Yukola" was opened, within the framework of which the composition of the ICs requiring development was determined (there were quite a few of them - 14, note that the fully vector Cray was assembled, in general, on only 4 types of microcircuits, and only one type was used in the ALU ) and developed their functional and schematic diagrams. The design and technological development of these ICs was planned to be carried out jointly with NIIME as part of the preparation of working project 41-50.
The preliminary design of the computer was accepted by the state commission with a high assessment and with a recommendation to continue the work. One of the ideologists of 41-50 N.M. Vorobyov recalls the final of the events as follows:
Naturally, this was not without the Ministry of Radio Industry, since it was planned to produce 41-50 at their facilities.
This is how one more project of the SVTs on the creation of a supercomputer ended.
A. I. Abramov, a representative of the general customer in the SVC, recalls its finale:
Note that the development of BESM-10 ITMiVT actually failed, without doing anything workable on the topic, the Lebedev school did not know how to work with supercomputer technologies at all.
Two of their highest achievements are BESM-6 (which everyone could not get enough of, because they had nothing else), with a performance of only about 1-1,5 MIPS and with an extremely ugly and inconvenient command system, not to mention the absence of even integer arithmetic (Lebedev was never an outstanding system architect of computers), and the controversial "Elbrus" Burtsev, which was clearly better than the creations of his boss, but no less inconvenient and far from as productive as the work of the SVC. In addition, the quality of manufacturing of machines developed by ITMiVT was terrible, we will also talk about this further.
System 41-50 was the latest development of supercomputers at SVC.
Three projects in a row were failed, and by the same ministry - 5E53 due to the fact that a machine created specifically for missile defense algorithms (and adopted by the military acceptance and PROSNIK!) Is allegedly unable to implement these algorithms, "IV" - under the pretext of lack of money, and even the terrible GRU was forced to be content with the "Elbrus" thrust into its teeth, unable to push through the party bureaucrats 41-50, which, again, was enthusiastically accepted by them and fully corresponded to their terms of reference.
The last case was, in general, outrageous - the Ministry of Radio Industry, in fact, refused to release a computer for no reason at all, having fought off the scouts as from annoying schoolchildren. We will not buy you a beautiful car, play with a cast iron one.
As a result, Yuditsky realized that it makes sense to develop only what can be produced at the facilities of the SVC itself - 16-bit minicomputers. Naturally, the use of modular arithmetic did not promise any particular advantages for them, and the SOC project was completely abandoned forever.
There is a legend quoted by Academician V.M. Amerbaev and known only from his words:
V.S.Linsky recalls this case, and possibly another:
This is actually quite strange story, modular arithmetic could help with banking computers, but the main manufacturer of banking computers was Burroughs, who relied on completely different principles of system architecture. Perhaps it could be some kind of company that wanted to overthrow the monopolist, but there are not so many serious players in this market. A small company would not have had that kind of money; a large one, such as IBM, basically developed everything by itself, and, again, was as conservative as possible. In addition, all the information (well, except for the ABM algorithms) about the SOC was already in the open press, even without the chipboard stamp. Getting it right is not a lot of time for a few good mathematicians.
Well, in general, the United States knew very well that the USSR was very, very interested in extracting Western technology by all means (from disassembling a sample of radios donated privately to various diplomats, to buying licenses and outright theft), but in principle it would not sell any high-tech.
The release of the Setun computer, for example, was banned even for their native Czechoslovakia, although the Czechs begged almost on their knees, promised huge gesheft from the sale to Western Europe and were already ready to build a production line (although there is a strong suspicion that the reasons here were not connected with politics, but rather with the magic words "cut" and "rollback", quite relevant in Soviet times, as we remember, certain circles in the Czech Communist Party also pressed their own developments, throwing out millions of crowns to purchase from the French they themselves did not desired Bull mainframes). So the negotiations here were initially doomed, and it would be foolish not to understand this.
A. V. Pivovarov recalls another case:
This story is already much more realistic, the USSR collaborated surprisingly a lot and fruitfully with France both in the field of fundamental science, especially mathematics, and applied science, including pharmaceuticals, to France, as well as to the Federal Republic of Germany, where our scientists were released more often and more willingly, the exchange of technology though limited, was also present.
On the subject of SOK, Yuditsky published more than 60 monographs and articles, becoming its largest theorist, many patents were obtained for all nodes and algorithms, some even in Germany, France, Great Britain, Italy and the USA, so that the absolutely clear message of the Ministry of Radio Industry “sit and do not stick your head out , nothing that you do will never be released ”led to a serious psychological shock and huge disappointment for the entire staff of the SVC. Let's remember how much time and effort was spent on development, how many rework, night shifts, vigils until the morning with a soldering iron and an oscilloscope, how many hopes and expectations when the developments will be embodied in metal ...
Three major failures in a row, and through no fault of their own, is a lot for any research group.
As a result, the scientific activity of the SVC dropped to almost zero, while the team was recovering from the battle with the Ministry of Radio Industry. As a result, the topic of modular arithmetic in the USSR was completely curtailed, according to some sources, foreign scientists who observed this (and did not know, of course, the real reasons for the events), decided that this was from the complete futility of the entire direction and also sharply reduced the intensity of work on SOC machines ...
In the Union, modular computers were completely forgotten, in Russia - even more so until 2005, when the 50th anniversary of the first publication of Wallach and Svoboda on this topic was celebrated. Then the surviving employees of the SVC decided at the same time to remember their contribution to this direction, to honor the memory of everyone who took part in the design of modular computers, and to find out if any similar projects were realized somewhere else?
And they initiated a special conference "50 years of modular arithmetic" in Zelenograd. It was very successful, 49 delegates took part, representing 32 firms from Russia, Belarus, Kazakhstan, Ukraine and the United States, who made 44 reports, a collection of works almost a thousand pages thick was published.
Currently, variants of modular arithmetic are widely used in microcontrollers of access cards with a high level of protection for the implementation of cryptoalgorithms, according to the ISO / IEC 10118-4: 1998 standard (section Hash-functions using modular arithmetic). These keys are mainly produced by STMicroelectronics. In addition, cryptographic microcontrollers have been or are being produced by M-Systems (SuperMAP controller), Emosyn LLC (a division of ATMI, Theseus Platinum chip), Hifn, and others.
V. M. Amerbaev and A. L. Stempkovsky from IPPM RAS also worked in the early 2010s on alternative versions of non-positional systems, for example, the so-called logarithmetic, in which the representation of numbers is multiplicative - a pair of the sign bit and the binary logarithm of the modulus of the number is used. With such a representation of numbers, division and multiplication operations are greatly simplified, which is logical, but the digital implementation of additive operations - addition and subtraction - becomes more complicated. As a result, even more exotic hybrids arose, for example, the modular LG code. It uses prime numbers as bases and uses the logarithmic representation of the residues for each simple base. From a hardware point of view, such a scheme can be used to build extremely efficient DSPs, since LG code greatly accelerates one of the main operations of such a processor, the Fourier transform.
In addition, serially modular processors were used in the systems of special processors AFK "Vychut-1" and "Vychut-2" (information on them was practically not found and it is not known what they were and what they were used for) and means of cryptographic protection of communication lines - products CRYPTON-4M7 and SECMOD-K. Information on "CRYPTON" is modest, but available. This is a cipher extension to the telephone, its basis is a modular 32-bit DSP, which implements the functions of speech encryption and transmission at a speed of 2400-12000 Baud.
At present, articles on modular chips periodically appear in Russia (for example, Kalmykov I.A., Sarkisov A.B., Yakovleva E.M., Kalmykov M.I. Caucasian Federal University No. 2 (35) / 2013), but rather sluggishly, and the matter did not advance beyond theoretical developments.
History has shown that SOC is amazingly convenient for rather narrow applications - fault-tolerant systems, public key cryptography and digital signal processing, and not very convenient for everyone else. As such, it is now used abroad, nevertheless, it is annoying that the outstanding pioneers in this area - Soviet engineers, were forgotten for a long time, and their unique works did not bring either glory or benefit to their homeland.
The basic problems of modular arithmetic were successfully solved, the machine was accepted, the prototype worked perfectly, the algorithms were written. So far, no one could have imagined how everything would turn out, and in the wake of euphoria, the employees decided not to stop at the project that had already been successfully completed (as it seemed to them) and to build something else.
In this article, there are many illustrations of similar Western projects, unfortunately, this is a forced measure, since it was not possible to find photographs related to the late works of Yuditsky and the SVC team in any open source (and it is not known whether photographs of these secret prototypes existed at all).
When creating 5E53, a monumental groundwork was erected for further research - the mathematical foundations of working with SOCs, methods of increasing fault tolerance were finally polished, periphery was created (and forming it in the USSR was a separate feat, comparable to the development of a supercomputer, they complained about the monstrous quality of Soviet drives, printers and other things even fanatical communists). For the firmware, a convenient 256-bit diode matrix on a dielectric substrate, DMR-256, was manufactured and its production was organized at Mikron, methods for assembling hybrid ICs of increased density were tested.
Let us not forget that Yuditsky's employees did not differ in senile ossification characteristic of many official Soviet scientific schools, there was no smell of gerontocracy there, everyone was (like their boss) young and daring and wanted to create further. They planned to improve the next version 5E53, build it on real microcircuits and include a number of even more progressive technical solutions there (fortunately, in the preliminary design of the new supercomputer, there were no longer any restrictions on working specifically with missile defense algorithms). A real revolution in the field of supercomputers was conceived, fueled by a friendly competition with Kartsev's group - Yuditsky understood that his machine was objectively slower, he was burned by the desire to squeeze out the same number of operations on a modular computer.
The intended monster was radical and progressive even by the standards of the astounding Western architectures of the 1980s (when nothing else was tried to improve performance). In addition to modular arithmetic, it had to be modular, reconfigurable (!) And with a hardware-microprogram implementation of Java (!) USSR) or an even weirder language - IPL (Information Processing Language, developed by Allen Newell, Cliff Shaw and Herbert A. Simon) of the RAND Corporation and the Carnegie Institution around 1: possibly the first functional programming language in the world, assembly type, oriented to work with lists).
It is worth talking about these very unorthodox innovations separately, especially since the idea of hardware support for nuclear devices was successfully borrowed by Burtsev for his "Elbrus". It all started in the United States with the famous Burroughs company, which in the early 1960s decided to enter the mainframe market.
B5000
Thus was born the B5000 - a great machine that outstripped the development of computer architecture by a decade.
Developed by a team led by Robert Barton in 1961, the mainframe was the first in a series of Burroughs large systems, produced until the late 1980s, with a unique architecture never seen before by any other manufacturer. Renowned computer scientist John R. Mashey lists her as one of his most admired inventions:
"I always thought this was one of the most innovative examples of combined hardware and software design I have seen, far ahead of its time."
Burroughs machine expert, University of Virginia professor Alan Bateson, in an interview for "Whatever Happened to the Seven Dwarfs?", Put it this way:
"If you could now read the specs of this machine through the eyes of 1960s people, you would be amazed!"
In the 1960s-1980s, many companies produced various lapel badges in honor of themselves, their products and any technologies; these badges were worn by company employees, they were distributed at exhibitions and presentations, and used in advertising. Nowadays, many of them have considerable collection value. The first large computer Burroughs so impressed the imagination of even its creators that in honor of him was released a badge with perhaps the most original advertising inscription: "I touched the B5000" (I touched the B5000). Also called is the book of memoirs published in 1985 by the famous computer scientist and developer, specialist in the theory of languages and compilation, who worked at Burroughs on its version of Algol Richard Waychoff (Richard Edwin Waychoff).
The same badge (from the personal collection of Ole Hagen Jensen)
What is so unique that Bob Barton and his team built?
They knew they were entering the large systems market much later than their competitors at IBM, RCA, and others, so they needed to offer something truly exceptional.
In addition, they had the opportunity to study the architecture of the systems already presented and realized that they all have certain common disadvantages, concentrated in the area of what the most advanced computer is useless without - software. Machines were supplied separately from software, and all programs were written, as a rule, by a company that bought a computer from scratch and for themselves, no one even thought about the convenience of development, architecture was separated from software. Computers were developed by special groups of engineers who did not even think about how easy or difficult it would be to develop programs for their machines. The B5000 was supposed to provide answers to all of these questions.
It was the first computer in the world that was developed as a single hardware and software complex and an integral system, in contrast to S / 360, for which IBM did not even manage to bring to mind the originally planned OS / 360 (moreover, this practice of separate design was preserved in IBM and onward and eventually became dominant, including the x86, PPC, and ARM instruction systems). Designing a machine without regard to executable code led to numerous crutches in the implementation of software and operating systems in the 1980s, which indirectly affects the quality of software today.
For the B5000, things were different. From the very beginning, it was designed in conjunction with the language, OS and system software. Its main architectural feature is full hardware support for high-level languages, based on two innovations - the stack processor and tag-descriptor memory.
The B5000 did not have an assembler, its processor was able to directly execute the JAVU instructions. During the development, the question arose, which of the languages to choose as the main one? There were only a few of those in those years, but the choice fell on the most powerful academic language, for which a new standard has just appeared - Algol-60. It became the main system language, and there was good support for Cobol (which mainly involved working with powerful string operators) and Fortran. Thus, the entire architecture of the B5000 was built around a very advanced language, for example, long before the #define directive in C appeared, a similar mechanism was used in the B5000, moreover, it was built into the language itself, and was not a preprocessor command.
Most other computer manufacturers could only dream of implementing the Algol compiler due to its complexity and extremely slow software implementation. It was believed that it was unrealistic to achieve an acceptable speed when using it, and if you do not use hardware support, this was the case (in particular, this is one of the reasons why Algol, as a language, did not gain wide popularity at that time). The then young student, the legendary Donald Knuth, who had previously developed Algol-58 programs for their early machines, worked on the implementation of the language at Burroughs for several months of summer vacation.
The Burroughs Algol compiler was very fast - it made a tremendous impression on the famous Dutch scientist Edsger Dijkstra. During the B5000 tests in Pasadena, his program was compiled at the speed of reading from punch cards, which was an amazing result for the time, and he immediately ordered several machines for the Eindhoven University of Technology in the Netherlands, where he worked. Hardware support and sufficient RAM allowed the compiler to operate in single-pass mode (even though early machine assemblers almost always used multi-pass compilation at the time).
B5500 and its components.
1 - printer (B320, B321 or B325), 2 - punched card reader (B129), 3 - control console with 4 - teletype for entering commands (for example, Teletype Corp Model 33 KSR), 5 - puncher (B303 or B304), 6 - magnetic tape drives (B422 / B423 –15 Mbit), 7 - processor (one of two possible), 8 - I / O coprocessor, 9 - terminal controller, 10 - display and debug console. Also, the photo did not fit 2 cabinets with RAM and a cabinet of the second processor (photo of the University of Tasmania http://www.retrocomputingtasmania.com)
We dwell on all these advantages in such detail precisely because similar ideas occurred to Yuditsky's group, and later to Burtsev's group (except that Yuditsky, unlike Burtsev, did not have a living Burroughs at hand to study). As a result, many things described as unique and unparalleled in the world, implemented in Elbrus, actually appeared much earlier, including advanced protection mechanisms.
Even the improved Burroughs Algol did not include many of the insecure constructs required by the operating system and other system software. To support them, a special extension Espol (Executive Systems Problem Oriented Language) was developed. Espol was used to write the operating system kernel - Burroughs MCP (Master Control Program) and all system software. Espol's hardware support made it easy to implement virtual memory, multiprocessing, fast context switching and procedure calls, memory protection, and code sharing. The B5000 was the first commercial virtual memory machine. In addition, due to this, complete reentrancy of the code was realized in a natural way, without additional efforts on the part of programmers. The Espol language was replaced in the late 1970s by the more advanced Newp (New Executive Programming Language).
All unsafe constructs in a program are rejected by the Newp compiler, unless a block in RAM is specifically tagged with a special tag to enable these instructions. This block marking provides a multi-layered security mechanism. In addition, Newp programs that contain unsafe constructs are not initially executable. A special system security administrator can make them executable, but ordinary users cannot. Even privileged users, who usually have root access, cannot start them without explicit administrator permission. Only compilers designated by the operating system could create executable files with extended commands, and only MCP itself could designate a program as a compiler (via the Security Admin console command).
Newp was so advanced that it was supported by the original Unisys ClearPath mainframe architecture, the heirs of the B series, until 2014, when the x86 migration began. Also, long before bash in Linux, a separate command line language WFL (Work Flow Language) was developed to effectively manage MCP. On IBM mainframes, its counterpart was the famous Job Control Language (JCL).
Computers designed specifically for Java were very complex, but later developed along the path beaten by Burroughs until the mid-1980s (among them are the Lilith workstation of Nicklaus Wirth, the father of Pascal and the famous LISP machines), when they were superseded by the x86 architecture and General purpose RISC processors.
The tag-protected descriptors in the B5000 are checked in hardware at every memory access at every data change step. In addition, the system does not need manual management of memory allocation, and moreover, this is generally impossible. Every segment of protected data, such as code, cannot even be read, let alone modified, in an uncontrolled manner, which makes most attacks impossible and errors impossible.
Of course, a suitable privileged process can explicitly change the bits of the tag and thus change itself, but only the ESPOL compiler can generate such code, while the MCP will refuse to execute anything that it identifies as ESPOLCODE, no matter what privilege level the person who tries to start it has ... Such programs must be installed as part of the operating system at the very beginning, and it is fundamentally impossible to add or change them in the process.
As a result, Burroughs mainframes remained the most secure and safest machines on the planet for the next thirty years, which is why the US Federal Reserve System chose them for many years as its banking computer standard. As we already said, this architecture (naturally, continuously improved by new models) was supported in hardware until recently, and only since 2014 there has been a transition to standard x86 servers.
One of the few real problems with the B series was that its parent, the B5000, ended up with an extremely complex processor and memory subsystem. In the era of transistor machines, one could turn a blind eye to this, but this moment greatly complicated the integration of subsequent models. In the years when all manufacturers switched to single-chip models with a custom processor, the Burroughs series of large machines were still available in multi-chip designs.
The first version of the mainframe-on-a-chip, SCAMP, did not appear until the late 1980s, when it was too late, although this processor and its successors were used by Unisys until the mid-2000s.
The pinnacle of large banking machines. Burroughs B7900 (1984) became the last classic mainframe of the B series, in 1986 they merged with Sperry: this is how Unisys appeared, which still exists today (photo of the University of Tasmania http://www.retrocomputingtasmania.com)
Reliability was an additional problem with the enormous complexity of the processor and the sheer number of transistors and early ICs.
However, Burroughs computers could not break - the company had a reputation as one of the best suppliers of high availability systems, their machines traditionally worked for years without an emergency stop (also, by the way, by the way, by the way, by the way, their adding machines, rightly considered the most reliable in the industry). In order for the B5000 to meet stringent quality criteria, considerable redundancy and flexibility have been incorporated into the system.
Hardware modules could be turned off and installed on the fly without stopping work or losing data, which was something fantastic at the time. In order to monitor the state of all nodes of the machine and reconfigure the system in the course of calculations, bypassing bad sections, a special MDLP coprocessor (Maintenance Diagnostic Logic processor) was added. It was also used by engineers to diagnose all system components.
As a result, despite the fact that the B5000 was an order of magnitude more complex than traditional machines of those years, its reliability not only did not suffer, but also significantly surpassed most computers of this class.
The credibility of the company in the banking environment was so high that in 1973, when the Society for Worldwide Interbank Financial Telecommunications (SWIFT) was created, it was Burroughs who built its backbone switching systems in 4 years of work. And to this day, Burroughs' successor, Unisys Corporation, is the largest provider of the SWIFT network.
The "desktop mainframe" Unisys Micro A (1989) is actually the first server in the modern sense of the word. Its processor is SCAMP-A (top) and later is SCAMP-D (1997, bottom). By order of Unisys, the chips were first produced by IBM in their factories, then by LSI. The Unisys ClearPath Libra 6200 Server (2012-2015) is the latest to support the B5000 architecture (Photo by the University of Tasmania http://www.retrocomputingtasmania.com, from the private collection of John Culver https://www.cpushack.com/ and the official Unisys ClearPath prospectus )
The B5000 was used by NASA, US Air Force, Carrier Corporation, University of Washington, University of Denver, Caltech, Stanford University, Monash University in Australia (they were Burroughs loyal to the end and had all their machines in sequence, up to the B7800), Drexel Institute of Technology in Montreal , British Post and American Bureau of Mines.
Also in 1964, Burroughs built the B8300 for real-time applications such as airline reservations. A rather rare version of Algol 60, Jovial, was chosen as the system language. It was developed in 1959 as a new high-level programming language for real-time systems at SDC by a group led by Jules I. Schwartz and was originally a dialect of Algol-58, as indicated by its jocular name (Jules Own Version of the International Algebraic Language).
At first, it was intended to program the electronics of combat aircraft, but in the 1960s it became an important part of a series of US military projects, in particular SACCS (Strategic Automated Command and Control System - a system that controlled nuclear weapons USA) and of course SAGE. Approximately 95% of the SACCS software (co-developed by ITT and IBM) was written by SDC in Jovial. Development took two years (about 1400 man-hours), more than twice as fast as SAGE software.
In the late 1970s, when developing the standard architecture for the military processor MIL-STD-1750A, it was decided that Jovial would remain the primary language for this architecture. Many companies provided their compilers for it - Advanced Computer Techniques (ACT), TLD Systems, Proprietary Software Systems (PSS) and others. The last standard for this language, MIL-STD-1589C, was adopted in 1984, currently three dialects of this standard are still used: J3, J3B-2 and J73. Jovial was only discontinued in 2010, although compilers continue to be released.
As with Cobol, most of the software implemented in Jovial is mission-critical, and maintenance is becoming increasingly difficult, with partial replacements starting in 2016, although sometimes the choice is more than odd. For example, the software for the famous B-2 bomber was ported from Jovial to Pure C (!), Which can hardly be considered an effective solution in terms of safety and ease of support.
Approximately the same architecture was proposed by the SVC engineers, but their supercomputer had one more unique feature - it was, as we have already said, modular!
Yuditsky's new computer
Yuditsky's new machine was supposed to include central processing subsystems (up to 16 central processors), input-output (up to 16 input-output processors), memory (up to 32 sections of 32K x 64-bit RAM) and a powerful modular system for dynamic switching of the listed modules in complex graph (any CPU could be connected to any PVV and any section of RAM). The overall performance of the computer was estimated at an absolutely monstrous 200 MIPS - the 1 Cray-1977 produced 160! In the processor, of course, a table implementation of arithmetic was planned.
As a result, Yuditsky noted with pleasure that, despite the extremely atypical system of residual classes for a supercomputer, his new project would be able to beat Kartsev's M-10! It really was an absolutely unique hybrid, which absorbed all the most advanced world developments in the field of computers of those years, the parallel matrix architecture from the M series, hardware support for the B5000 YED and, of course, the proprietary technology from Yuditsky himself - SOK.
The most striking thing is that the result did not at all look like a cross between a hedgehog and a snake - it was an absolutely working, extremely comfortable and most powerful machine in the world of those years, the closest American competitors lagged behind by a generation. Plus, she was incredibly reliable.
In general, we all already understand that it would not have been possible to implement it in a series in the USSR even at the cost of the chief designer's life.
To implement tabular arithmetic, the machine needed a new compact, large-capacity permanent memory. For several years now, a division of S. A. Garyainov has been developing it at the SVTs. The essence of the work was to create unpackaged diode arrays, as well as the design and manufacturing technology of devices based on them.
It was for this purpose that they wanted to adapt the already mentioned DMR-256. On the basis of the matrix, a corresponding original structural system was developed: the DMR crystals were mounted on a sital board, the boards were assembled into a seven-story MFB stack (multifunctional unit), the stacks were installed on a large printed cross-board. Several backplanes were mounted in a metal sealed block case filled with freon. To remove heat from the block, heat pipes were installed in it.
The preliminary project of a unique computer, simply indexed with Roman numerals "IV", was completed in early 1973. "IV" was conceived as a prototype for subsequent developments of the SVC. However, even before the project was completed, it seemed to be put to good use.
At the end of 1971, the Sukhoi Design Bureau Kulon applied to the SVC with an order for the development of CAD systems for aircraft. High and promising requirements were imposed on CAD, which outstripped any capabilities of Soviet computers of those years.
The system was supposed to support more than 700 automated workstations for the developers of the aircraft and its components. Each AWP was a terminal with a plotter, and the calculations had to be done on the main supercomputer (in those days, the annual output of even simpler AWPs in the USSR was no more than half a thousand). The draft design was completed and accepted with satisfaction by the customer, but the Ministry of Radio Industry (headed by whom it is known) refused to produce the car, citing a lack of funding (despite the fact that the project was intended for the military Sukhoi Design Bureau, and we did not spare money for the defense industry).
However, an even more interesting use for "IV" appeared almost immediately, in early 1972. Then the SVC received an order from the GRU itself for the development of a preliminary design of a supercomputer for processing data structured in a special way (translating from the GRU language into a human one - for breaking ciphers), which received the code name "Machine 41-50".
A 64-bit computer had to have a speed of at least 200 MIPS, 16 MB of RAM and advanced peripherals. The SVC decided to build a vector computer with a system of commands working on arrays and focused on the implementation of the customer's algorithms. In this case, the problem of dynamic parallelization was solved at the hardware-microprogram level. Draft project 41-50 SVC was carried out jointly with the Institute of Cybernetics of the Academy of Sciences of Ukraine, another underestimated Soviet genius, one of the world's best specialists in parallel computing and director of the IC, Academician V.M.Glushkov, was involved in the work.
It makes sense to somehow start a conversation about Glushkov separately - he was one of the world's largest scientists in the field of computer science (during the 15th edition of the British-American Encyclopaedia Britannica in 1973-1974, an article on cybernetics was commissioned to Glushkov!), But his projects (and there were absolutely amazing things, for example, the Soviet Internet) they drowned so mercilessly that (according to the tradition of outstanding Russian specialists in the field of computers) he did not live to be 60 years old, having died from a heart attack.
Glushkov was appointed scientific leader of the project, and two special subdivisions (a branch of the SIC) were created in the IC, headed by Z. L. Rabinovich and B. N. Malinovsky. The chief designer was Yuditsky.
The design of 41-50 began with studying algorithms for solving customer problems and trying to fit them into modular arithmetic (as we can see, in all projects of SOK machines, the work was based on algorithms - in fact, this was the drawback of this class of computers - a huge binding to specific tasks, making the car almost highly specialized). The work was headed by V. M. Amerbaev - as a mathematician and the main author of modular arithmetic, and L. G. Rykov - as a circuit engineer implementing these algorithms.
L.G. Rykov recalls:
Akushsky was more a theoretician and did not descend to such concepts as delay time, pulse race and other circuit technical troubles. Vilzhan Mavlyutinovich is a completely different person. He did not disdain our problems and always tried to find such a variant of the mathematical solution, which is most successfully implemented in hardware.
The result of the research was collected in the work RTM U10.012.003 "Machine algorithms for two-stage non-positional arithmetic", and on the whole it was disappointing. The fact is that in the tasks of the GRU, the percentage of non-modular operations was colossal, it was impossible to reduce them to SOC, and it was stupid to constantly convert back and forth and drive them into a regular coprocessor.
As a result, the performance of an ultra-complex and powerful computer would not exceed an ordinary supercomputer of traditional architecture. In general, the RNS system gave bonuses due to reliability, ease of implementation of table arithmetic and reduction in the amount of equipment, but Yuditsky was not a fanatic and understood that modular arithmetic was not a silver bullet. There are cases in which it simply does not fall on the algorithms, despite all the tricks.
In the final, after discussions and discussions, the SIC decided to abandon the SOC while maintaining the general vector-modular scheme of the machine and revised the project. Such flexibility favorably distinguished them from many Soviet design bureaus, which, having once found a more or less successful technical solution, continued to fanatically stamp it (like transistor versions of BESM in all versions and its own command system, which is extremely successful on some tasks and to the same extreme curve - on others).
They decided to make the machine, of course, on IC and as a basis they took the most powerful then in the USSR emitter-connected logic of the 100 series. Before it was stolen, it was called Motorola MC10000 (aka MECL - Motorola emitter coupled logic) - a series of quite powerful and fast ECL microcircuits, developed in 1962 (MECL I). The series had several generations - I, II, III and 10000, released in 1971. However, it differed from the 1968 version only in the resistor ratings. After 7 years it was mastered to be copied in the USSR as the IS100, it was intended for the most powerful computers such as Elbrus.
Unfortunately, the microcircuits of this series turned out to be extremely difficult for the Union and had huge problems with quality and stability, for which they are notorious (we will talk about the IS100 in the part about the A-135 and Elbrus, the devil will break his leg in copying powerful ESLs in the USSR, and this topic needs to be dealt with separately, it is closely linked with the commercial relations of the two giants - Motorola and Fairchild).
In the West, Motorola 10k was not the most popular choice of supercomputer builders, for these purposes they most often used an ESL from a competitor - Fairchild, the Fairchild F100K series (later they tried to copy it with a 10-year delay for Electronics SS BIS - the K1500 series, the result was, well, let's just say - not very successful, this is also the subject of a separate conversation). It was on F100K (3 microcircuits out of 4 types used - 11C01, F10145, F10415 and only one MC10009 for the address sampling circuit, Cray used a cheaper one in the place where it was not critical) Cray-1 was assembled.
The great and terrible Cray-1, one of its processor boards based on Fairchild F100K chips and the Soviet clone Motorola 10k - 100 series of the development period (photo https://cdn.britannica.com/, https://en.wikipedia.org/, https://ru-radio-electr.livejournal.com/)
IS100 production was mastered at Mikron, Vent in Vilnius, Svetlana in Leningrad and Integral in Minsk. Then problems began, the composition of the series did not provide for vector chips, as a result, additional ICs were required, which were not in the release program.
It was decided to join the program by developing the missing microcircuits for it. And the topic "Yukola" was opened, within the framework of which the composition of the ICs requiring development was determined (there were quite a few of them - 14, note that the fully vector Cray was assembled, in general, on only 4 types of microcircuits, and only one type was used in the ALU ) and developed their functional and schematic diagrams. The design and technological development of these ICs was planned to be carried out jointly with NIIME as part of the preparation of working project 41-50.
The preliminary design of the computer was accepted by the state commission with a high assessment and with a recommendation to continue the work. One of the ideologists of 41-50 N.M. Vorobyov recalls the final of the events as follows:
In the process of developing the draft design, we worked closely with the colonels from the customer on the algorithms for processing their specific information: in essence, it was a joint work. Therefore, both we and they were keenly interested in the fate of the project. They were clear supporters of 41-50, since, in fact, participating in the development of the project, they knew it perfectly, made all the decisions they needed and were confident in the result.
However, unexpectedly for us, there was a long, several months, unexplainable pause. The conclusion of a contract for the development of a technical design was postponed. What was happening there, we did not know - the GRU is a serious organization. But when the decision was finally made, the colonels specially came to explain the situation to us. The men with sadness and apologies (“we stood for you as a mountain as we could”) told us the sad news: there will be no continuation of the work.
It was decided to use the Elbrus adapted for their tasks, but they do not believe in the possibility of adaptation, since there are no levers to ensure its implementation.
However, unexpectedly for us, there was a long, several months, unexplainable pause. The conclusion of a contract for the development of a technical design was postponed. What was happening there, we did not know - the GRU is a serious organization. But when the decision was finally made, the colonels specially came to explain the situation to us. The men with sadness and apologies (“we stood for you as a mountain as we could”) told us the sad news: there will be no continuation of the work.
It was decided to use the Elbrus adapted for their tasks, but they do not believe in the possibility of adaptation, since there are no levers to ensure its implementation.
Naturally, this was not without the Ministry of Radio Industry, since it was planned to produce 41-50 at their facilities.
Final of the SVC project
This is how one more project of the SVTs on the creation of a supercomputer ended.
A. I. Abramov, a representative of the general customer in the SVC, recalls its finale:
Once I was summoned by the chief engineer of the ordering department of the Ministry of Defense Colonel S. F. Sereda and instructed me to prepare proposals on the use of the results of the Leader theme. By that time, the decision to terminate work on the creation of the 41-50 supercomputer had already been made. After consulting with our curator at the GU MO, V. M. Kapunov, we proposed a draft design (a stack of books over a meter thick) to transfer to ITMiVT for use in the BESM-10 project. And that was done.
Note that the development of BESM-10 ITMiVT actually failed, without doing anything workable on the topic, the Lebedev school did not know how to work with supercomputer technologies at all.
Two of their highest achievements are BESM-6 (which everyone could not get enough of, because they had nothing else), with a performance of only about 1-1,5 MIPS and with an extremely ugly and inconvenient command system, not to mention the absence of even integer arithmetic (Lebedev was never an outstanding system architect of computers), and the controversial "Elbrus" Burtsev, which was clearly better than the creations of his boss, but no less inconvenient and far from as productive as the work of the SVC. In addition, the quality of manufacturing of machines developed by ITMiVT was terrible, we will also talk about this further.
System 41-50 was the latest development of supercomputers at SVC.
Three projects in a row were failed, and by the same ministry - 5E53 due to the fact that a machine created specifically for missile defense algorithms (and adopted by the military acceptance and PROSNIK!) Is allegedly unable to implement these algorithms, "IV" - under the pretext of lack of money, and even the terrible GRU was forced to be content with the "Elbrus" thrust into its teeth, unable to push through the party bureaucrats 41-50, which, again, was enthusiastically accepted by them and fully corresponded to their terms of reference.
The last case was, in general, outrageous - the Ministry of Radio Industry, in fact, refused to release a computer for no reason at all, having fought off the scouts as from annoying schoolchildren. We will not buy you a beautiful car, play with a cast iron one.
As a result, Yuditsky realized that it makes sense to develop only what can be produced at the facilities of the SVC itself - 16-bit minicomputers. Naturally, the use of modular arithmetic did not promise any particular advantages for them, and the SOC project was completely abandoned forever.
There is a legend quoted by Academician V.M. Amerbaev and known only from his words:
In 1970-1971, US banking structures showed great interest in modular arithmetic. They needed high-performance tools for highly reliable self-correcting computations, which is what modular arithmetic is all about. According to the open press (articles, books, patents), they assessed the results of the work of I. Ya. Akushsky and D.I. The negotiations that had begun were thwarted by the KGB.
V.S.Linsky recalls this case, and possibly another:
While working at NIIFP-SVTs in 1966-1970, I openly expressed my negative attitude towards the SOC, right up to the appeal to the Military-Industrial Commission under the Council of Ministers of the USSR (MIC). VS Burtsev was familiar with my opinion, who expressed himself in the sense that a definite answer about the SOC is premature. To the question of the military-industrial complex employees about why the Americans want to buy the results of I. Ya. Akushsky and D. I. Yuditsky, I replied that, apparently, it is more profitable for them than to conduct research in this area themselves.
This is actually quite strange story, modular arithmetic could help with banking computers, but the main manufacturer of banking computers was Burroughs, who relied on completely different principles of system architecture. Perhaps it could be some kind of company that wanted to overthrow the monopolist, but there are not so many serious players in this market. A small company would not have had that kind of money; a large one, such as IBM, basically developed everything by itself, and, again, was as conservative as possible. In addition, all the information (well, except for the ABM algorithms) about the SOC was already in the open press, even without the chipboard stamp. Getting it right is not a lot of time for a few good mathematicians.
Well, in general, the United States knew very well that the USSR was very, very interested in extracting Western technology by all means (from disassembling a sample of radios donated privately to various diplomats, to buying licenses and outright theft), but in principle it would not sell any high-tech.
The release of the Setun computer, for example, was banned even for their native Czechoslovakia, although the Czechs begged almost on their knees, promised huge gesheft from the sale to Western Europe and were already ready to build a production line (although there is a strong suspicion that the reasons here were not connected with politics, but rather with the magic words "cut" and "rollback", quite relevant in Soviet times, as we remember, certain circles in the Czech Communist Party also pressed their own developments, throwing out millions of crowns to purchase from the French they themselves did not desired Bull mainframes). So the negotiations here were initially doomed, and it would be foolish not to understand this.
A. V. Pivovarov recalls another case:
Yuditsky had a contact with a French company, I don't remember its name, which wanted to buy a computer project.
Yuditsky came to me for permission to make such a deal, but I refused him for two reasons.
First, to complete such a transaction, it is necessary to manufacture a computer sample for the complete development of the technology, but there was nowhere to do it.
Secondly, why do we need to arm the French, then our potential military adversaries? Yes, even if I agreed, we would still not be allowed to do this by the higher authorities.
Yuditsky came to me for permission to make such a deal, but I refused him for two reasons.
First, to complete such a transaction, it is necessary to manufacture a computer sample for the complete development of the technology, but there was nowhere to do it.
Secondly, why do we need to arm the French, then our potential military adversaries? Yes, even if I agreed, we would still not be allowed to do this by the higher authorities.
This story is already much more realistic, the USSR collaborated surprisingly a lot and fruitfully with France both in the field of fundamental science, especially mathematics, and applied science, including pharmaceuticals, to France, as well as to the Federal Republic of Germany, where our scientists were released more often and more willingly, the exchange of technology though limited, was also present.
On the subject of SOK, Yuditsky published more than 60 monographs and articles, becoming its largest theorist, many patents were obtained for all nodes and algorithms, some even in Germany, France, Great Britain, Italy and the USA, so that the absolutely clear message of the Ministry of Radio Industry “sit and do not stick your head out , nothing that you do will never be released ”led to a serious psychological shock and huge disappointment for the entire staff of the SVC. Let's remember how much time and effort was spent on development, how many rework, night shifts, vigils until the morning with a soldering iron and an oscilloscope, how many hopes and expectations when the developments will be embodied in metal ...
Three major failures in a row, and through no fault of their own, is a lot for any research group.
As a result, the scientific activity of the SVC dropped to almost zero, while the team was recovering from the battle with the Ministry of Radio Industry. As a result, the topic of modular arithmetic in the USSR was completely curtailed, according to some sources, foreign scientists who observed this (and did not know, of course, the real reasons for the events), decided that this was from the complete futility of the entire direction and also sharply reduced the intensity of work on SOC machines ...
50th anniversary of modular arithmetic
In the Union, modular computers were completely forgotten, in Russia - even more so until 2005, when the 50th anniversary of the first publication of Wallach and Svoboda on this topic was celebrated. Then the surviving employees of the SVC decided at the same time to remember their contribution to this direction, to honor the memory of everyone who took part in the design of modular computers, and to find out if any similar projects were realized somewhere else?
And they initiated a special conference "50 years of modular arithmetic" in Zelenograd. It was very successful, 49 delegates took part, representing 32 firms from Russia, Belarus, Kazakhstan, Ukraine and the United States, who made 44 reports, a collection of works almost a thousand pages thick was published.
Currently, variants of modular arithmetic are widely used in microcontrollers of access cards with a high level of protection for the implementation of cryptoalgorithms, according to the ISO / IEC 10118-4: 1998 standard (section Hash-functions using modular arithmetic). These keys are mainly produced by STMicroelectronics. In addition, cryptographic microcontrollers have been or are being produced by M-Systems (SuperMAP controller), Emosyn LLC (a division of ATMI, Theseus Platinum chip), Hifn, and others.
V. M. Amerbaev and A. L. Stempkovsky from IPPM RAS also worked in the early 2010s on alternative versions of non-positional systems, for example, the so-called logarithmetic, in which the representation of numbers is multiplicative - a pair of the sign bit and the binary logarithm of the modulus of the number is used. With such a representation of numbers, division and multiplication operations are greatly simplified, which is logical, but the digital implementation of additive operations - addition and subtraction - becomes more complicated. As a result, even more exotic hybrids arose, for example, the modular LG code. It uses prime numbers as bases and uses the logarithmic representation of the residues for each simple base. From a hardware point of view, such a scheme can be used to build extremely efficient DSPs, since LG code greatly accelerates one of the main operations of such a processor, the Fourier transform.
In addition, serially modular processors were used in the systems of special processors AFK "Vychut-1" and "Vychut-2" (information on them was practically not found and it is not known what they were and what they were used for) and means of cryptographic protection of communication lines - products CRYPTON-4M7 and SECMOD-K. Information on "CRYPTON" is modest, but available. This is a cipher extension to the telephone, its basis is a modular 32-bit DSP, which implements the functions of speech encryption and transmission at a speed of 2400-12000 Baud.
At present, articles on modular chips periodically appear in Russia (for example, Kalmykov I.A., Sarkisov A.B., Yakovleva E.M., Kalmykov M.I. Caucasian Federal University No. 2 (35) / 2013), but rather sluggishly, and the matter did not advance beyond theoretical developments.
History has shown that SOC is amazingly convenient for rather narrow applications - fault-tolerant systems, public key cryptography and digital signal processing, and not very convenient for everyone else. As such, it is now used abroad, nevertheless, it is annoying that the outstanding pioneers in this area - Soviet engineers, were forgotten for a long time, and their unique works did not bring either glory or benefit to their homeland.
- Alexey Eremenko
- http://www.retrocomputingtasmania.com, https://www.cpushack.com/, https://cdn.britannica.com/, https://en.wikipedia.org/, https://ru-radio-electr.livejournal.com/
Information