Supercomputer - SpeedyLook encyclopedia

History

The first supercomputers appeared in the Années 1960, conceived by Seymour CRAY on behalf of the company Control Dated Corporation (CDC), first world manufacturer of supercomputer until in the Années 1970. CRAY Research, founded by Seymour CRAY after its departure of CDC, then took the advantage on CDC and its other competitors until 1990. In the Années 1980, to the image of what had occurred on the market of the minicomputers one decade earlier, from many small firms launched out on this market, but the majority disappeared in the “crash landing” from the market from the supercomputers in the middle of the Années 1990. Nowadays the supercomputers are generally designed like single models by “traditional” computer makers like IBM and HP which they on the matter have behind them a long tradition (IBM) or that they repurchased in the Années 1990 specialized companies then in difficulty to acquire experiment in this field.

The term supercomputer itself remains rather vague and evolutionary, because the most powerful computers of the world at a given time tend to be equalized then exceeded by machines of current use. First supercomputers CDC were simple computers mono Processeur S (but having sometimes to ten peripheral processors for the input-outputs) approximately ten times faster than competition. In the Années 1970 the majority of the supercomputers adopted a array Processor, which carries out the decoding of an instruction only once to apply it to a whole series of operands. It is only towards the end of the Années 1980 that the technique of the parallel systems massively was adopted, with the use in the same supercomputer of thousands of processors. Nowadays some of these parallel supercomputers use Microprocesseur S RISC conceived for computers of series, like the PowerPC or PA-RISC. Others use processors of lower costs of external appearance CISC, but microprogramed in RISC in the chip (AMD, Intel): the output is raised a little less by it, but the channel of access to the memory - often bottleneck - is requested much less.

Use

The supercomputers are used for all the tasks which require an enormous computing power like the weather forecasting, the study of the Climat, molecular modeling (calculation of the structures and chemical properties of compounds…), physical simulations (aerodynamic simulations, calculations of resistance of materials, simulation of explosion of nuclear weapon, study of the nuclear Fusion…), the Cryptanalyse, etc

The civil and military institutions of research count among the largest users of supercomputers. In France, one finds these machines in the national centres of calculations academic such as the IDRIS, the MOVIES, but also with ECA. Currently (January 2006), the most powerful French supercomputer is TERA-10 developed by Bulls and generating 60 teraflops.

Previously, the most powerful French supercomputer was the AlphaServer SC45 1 GHz pertaining to ECA, is classified 41e in November 2004 (it was 4th semi-2002)

Design

The supercomputers draw their superiority on the conventional computers at the same time from a innovating architecture enabling them to carry out several tasks simultaneously, and from a study pushed to improve all the components of the computer. They are generally conceived specifically for certain types of task (generally scientific numerical calculations) and offer performances very limited for other tasks.

Their mémorielle architecture is the special attention object to provide uninterrupted sufficient data to be treated with each processor in order to exploit to the maximum its computing power. The higher performances memory (better components and better architecture) explain to a large extent the advantage of the supercomputers on the traditional computers.

Their system of input/output is generally conceived to provide broad a Band-width, the Latence being less important since this type of computer is not conceived to treat transactions.

As for any parallel system, the Loi of Amdahl applies, and the originators of supercomputers devote many efforts to eliminate the parts not parallélisables of the software and to develop material improvements to remove the remaining bottlenecks.

Technical main obstacles

the supercomputers produce a great quantity of heat and must be cooled to function normally. The cooling of these computers often poses a large problem of Climatisation.
information can circulate more quickly only the Speed of light between two parts of the computer. When the size of a supercomputer exceeds several meters, the latency time between certain components amounts of tens of nanoseconds. The elements are thus laid out to limit the length of the cables which connect the components. On CRAY-1, for example, they were laid out in Cercle.
These computers are able to treat and communicate very important volumes of data in very little time. An important work of design is necessary to make sure that these data can be read, transferred and stored quickly. In the contrary case the computing power of the processors under-would be exploited.

History of the records

Operating systems for supercomputers

Linux becomes massively multiprocessors today, and is the system equipping the very great majority (85,20 %) of the 500 most powerful supercomputers of planet;
Unix gradually loses ground vis-a-vis Linux, but still occupies a choice place on the market of the supercomputers (6 %);
Windows can manage only to the maximum 2 processors, except in certain specific versions “waiter”, enabling him to manage according to versions 4,8 or 32 processors. Only 6 of the 500 most powerful supercomputers of planet make turn Windows, that is to say 1,20 %;
BSD and Mac OS is very little represented (0.40 % each one);
others (" Mixed") represent 6.80 %;

Notes and references of the article

Random links:

Allelic frequency | Coat of mail | KwaNdebele | Uharte-Arakil | Sverre Kolterud