JPH06175987A - Network system for parallel computer - Google Patents

Abstract

PURPOSE:To surely perform data communication among the computers of the adjacent columns of a parallel calculation system, to assemble a structure at low cost while upscaling the parallel calculation system and to demonstrate calculation performance. CONSTITUTION:A general purpose compact computer is defined as a unit computer for calculation areas vertically and horizontally divided and discretized as prescribed, networks 11, 11' and 11'' are provided between the respective columns and the data communication among the unit computers 7, 7... arranged in the adjacent columns can be more surely performed. By obtaining the backup of softwares in a simple system in which the general purpose compact computers 7 such as simple and inexpensive personal computers or the like are arithmetically increased in design and increasing the number of the networks simply arithmetically, the calculation ability of the parallel calculation system can be unlimitedly improved and further, the function of simultaneous communication can be realized by a mechanism more inexpensive and more simple compared to a conventional form.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION The disclosed technology belongs to the technical field of a network system for mutual data communication among parallel computers as next-generation computers.

[0002]

2. Description of the Related Art As is well known, the prosperity of industrial society is largely owed to the extremely advanced science and technology. Especially, in the advanced industries such as the defense industry, so-called high-tech technology is used to promote the advanced development. ing.

Of these, computer-based advanced calculations require not only theoretical basic analysis but also complicated and high-speed aerodynamic calculations for aircraft, missiles, etc., but such complicated calculations were possible without a computer. Therefore, the importance of high-speed and complicated calculations by computers cannot be ignored in the future, but as is well known, the development of more complex and high-speed computers, including supercomputers, is actually extremely intense. Research and development is being conducted.

Moreover, as is well known to those skilled in the art as a powerful method for speeding up calculation processing, a so-called parallel computing system is extremely effective as disclosed in, for example, the invention disclosed in Japanese Patent Laid-Open No. 2-170260. It is targeted for research and development as a new technology.

The parallel computing system has been attracting attention as a computing system that enables parallel processing of 1000 units of microprocessors to perform information processing similar to human brain function. It is possible to realize ultra-high-speed data processing on the scale of 100 billion times, and by designing to construct thousands of parallel computers, parallel connection is possible by connecting 10,000 node processors to each other. Designs have been proposed, and they will occupy most of the theoretical calculation field of basic science in the future, and it is predicted that they will occupy a considerable weight in the business industry field as a whole. In the field of science and industry, it has been greatly highlighted.

[0006] The parallel computing system has attracted attention not only in the aerodynamic calculation of flying objects such as rockets but also in the nuclear industry, medical industry, electronic industry, etc., which require complicated and high-speed calculations, and intense R & D competition. Has been done.

As is well known, such a parallel computing system basically has a single computing area 1 as shown in FIG. Divide into 3,4,5 as shown in (b),
As shown in (c), in comparison with the calculation by the conventional computer 6 for the single calculation region 1, general-purpose small-scale computers 7, 7 ... By reducing the amount of calculation per computer by using, there is a merit that the calculation speed can be basically improved arithmetically by the excess number of general-purpose small-sized computers 7, and it is expensive. It has the advantage that it can be replaced with a super computer.

By covering the calculation area with a general-purpose small computer 7 such as a personal computer for each single calculation area, there is a potential point that the calculation amount per single unit computer can be reduced. Because they are doing it.

[0009]

However, compared with a super computer or the like, which has an excellent computing ability, but has a complicated structure and complicated software construction, and has a demerit that is costly. Despite having several advantages such as a general-purpose compact computer such as a personal computer, which is a commercially available product, can be expanded arithmetically, the amount of calculation is small, and the whole calculation can be discretized. Data communication between the unit general-purpose small-sized computers 7, 7 ... Is indispensable for comprehensively performing complex calculations, but as a basic common technology of parallel computers, the unit general-purpose unit corresponding to multiple divided calculation areas 2-8 is used. Since it is impossible to perform data communication only with the small-sized computers 7, 7, ... Of course, a network system for each general-purpose small-sized computer 7, 7 ... Is required. To, for example, there is a technique such as arranging special computer system is extremely complex,
There is a demerit that software development is complicated and it is not worth the cost.

In particular, since there is no effect on the use of commercially available general-purpose small computers and software, there is the inconvenience of not being on the profit base.

[0011]

OBJECT OF THE INVENTION The object of the invention of this application is to have the ability to calculate complex and sophisticated calculations at high speed using a next-generation parallel computer replacing the already developed supercomputer based on the above-mentioned prior art. However, the technical issue that should be solved is the problem that software backup and cost reduction cannot be made, and the inherent merit of parallel computers is to install multiple general-purpose small computers and to increase the number of arithmetic machines by adding them. It can be fully utilized, it is cheap in cost, conventional software can be backed up to the minimum necessary hardware, the calculation capacity can be improved, and the ability of simultaneous communication can be added to various industries. It is intended to provide an excellent parallel computer network system which is beneficial to the field of application of computing technology in.

[0012]

SUMMARY OF THE INVENTION In order to solve the above-mentioned problems, the structure of the invention of the present application, which is based on the above-mentioned objects and has the above-mentioned object, solves the above-mentioned problems at high speed and accurately. Set the calculation area vertically to
A large number of discretizations are made horizontally or vertically in a predetermined order and divided into a predetermined number of calculation areas, and a general-purpose small computer such as a personal computer is allocated to each unit calculation area, and a data communication network is used for each computer. By connecting them in pairs, the ring arrangement, the bus arrangement, the star arrangement, etc. are selectively adopted for the arrangement of the network so that the calculation data in the adjacent calculation areas can be communicated between the unit general-purpose small-sized calculations, For each network, a data communication function with a computer that controls relative calculation by installing another network between a parallel computer and a broadcast node for the network is provided, enabling simultaneous communication,
Furthermore, by increasing the number of general-purpose small computers, the total number of networks will be increased, and the arithmetic capacity and speed will increase.
The technical measures have been taken to improve the accuracy, to easily back up the software, to reduce the cost, and to have the function of simultaneous data communication.

[0013]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, an embodiment of the invention of this application will be described with reference to FIGS.
The explanation is based on the following.

The same parts as in FIG. 8 will be described using the same reference numerals.

FIG. 7 shows a mode of discretization by dividing the calculation area according to the invention of this application. As shown by a, the whole calculation area 1 is divided in a predetermined number of plural X directions, and each divided calculation area 8 is divided. , 8 ′, 8 ″, 8 ′ ″ are discretized, but are divided into the same number according to a predetermined order so as to have the same calculation weight, and further, in the Y direction as shown in e. , 91, 91 '..., 92, 92' ..., 93,
93 '..., 94, 94' ...

In FIG. 7, the calculation areas 91 and 9 are
1 ', 92, 92' ... Are different from each other in terms of area in the drawing, but are divided so that the calculation weights are the same.

Therefore, the division of FIG. 7 is only shown as an image.

When three-dimensional calculation is performed in the aerodynamic calculation in the aircraft industry or the calculation in the nuclear industry, the Z-direction division is performed in addition to the X-direction and Y-direction division in FIG. 7 to divide the optimum calculation area. In order to prevent the load of carrying calculation of each general-purpose small-sized computer 7 from being overloaded, the general-purpose small-sized computers 7, 7 ... By doing so, it will be possible to increase the total arithmetic calculation capacity by adding more.

Data communication in the mutual calculation processes of the general-purpose small-sized computers 7, 7 ... Is indispensable as described above. However, in the above-described mode, for example, the calculation area of the final divided array in FIG. Columns 91 ..., 92 ..., 93 ..., 9
4 ... Adjacent columns can perform data communication because of the arrangement, but columns that are not adjacent to each other, such as 91 columns ... And 93 columns ... 92 columns ... 94 columns ... There is no mutual information transfer, that is, no data communication occurs during execution of.

Therefore, the networks between the columns can be separated, and as an embodiment, the ring type networks shown in FIGS. 1 and 2 can be used. In the embodiment shown in FIG. 1, the unit computers 7 are arranged in the vertical direction. , 7 ... Between the networks 11, 11 ', 11''are provided to perform data communication between the unit computers 7, 7 ...

In addition, in the embodiment shown in FIG. 2, not only data between unit computers 7, 7 in a vertical array of one column, but between each vertical array, data between the unit computers 7, 7 in the first and last columns of the array. By providing a network 11 "" for communication, it is possible to perform data communication between all adjacent unit computers 7, 7, ...

If the respective embodiments shown in FIGS. 1 and 2 are imaged, the left end of the drawing, that is, the first row and the right end terminal row are connected in a cylindrical shape. It is possible to enable data communication between the unit computers 7, 7 ... In adjacent columns, and as another embodiment of the invention of this application, the unit computers 7, 7 ... Shown in FIG. Of the so-called bus type network system in which the unit computers 7, 7 ... Of each column arrangement are connected to the networks 11, 11 ', 11''of the column arrangement between the column arrangements of FIG. And the joints 12, 12 between each column
And a unit computer 7 in each vertical arrangement is connected to each central joint 12 to form a so-called star type network system.

Of course, the embodiment of the invention of this application is not limited to the above-described embodiments. For example, in order to realize a parallel computing system by adding a predetermined number of unit computers 7, a predetermined number of the plurality of units is realized. Unit calculator 7, 7 ...
It is possible that the network has such a configuration that the above-mentioned connections are possible.

As a common embodiment, the network system of each of the above-described embodiments can perform the calculation by nature, but the communication of data and programs common to each computer 7, 7 ... If you do this, you can use any computer 7,7 ...
Since the data communication is performed by the relay method in FIG. 5, as a mode for coping with the possibility that the overall operation efficiency may deteriorate due to the conditions such as the transmission efficiency of the relay type data communication, FIG. 6 can be adopted. In the embodiment of FIG. 5, a unit computer 7, 7, ... Is provided with a third network 14 to connect the host computer 13 as a control computer, It is possible to enable data communication between arbitrary computers 7, 7 ...
In the embodiment shown in FIG. 4, in order to prevent the load on each parallel computing computer 7 from increasing, the broadcast computing computer 15 or the host computing 13 is controlled by the communication dedicated computer 15 and the third set network 14 via the host computer 13. , And the simultaneous communication can be performed between the computers 7, 7 ....

[0025]

As described above, according to the invention of this application, basically, in a parallel computing system, it is possible to systematically add a commercially available small-sized general-purpose computer, so that the minimum necessary hardware can be used. It has an excellent effect that it can obtain a high performance by obtaining software backup and can be an optimal network system for dividing the calculation area.

Further, there is an effect that the total calculation capacity can be practically improved by increasing the number of computers and the number of networks.

Thus, even if the scale is increased,
Basically you can keep the whole structure simple by just increasing the hardware of the same structure, therefore
There is a merit that initial installation and maintenance are easy.

Further, as compared with the network system such as the conventional super computer, the excellent effect that the network system can be constructed at low cost is exhibited.

Further, as compared with the conventional mode, the excellent effect that the simultaneous communication function can be added with a simple mechanism or system is exhibited.

[Brief description of drawings]

FIG. 1 is a schematic view of an embodiment of the invention of this application.

FIG. 2 is a schematic view of another example of the same.

FIG. 3 is a schematic diagram of the embodiment.

FIG. 4 is a schematic view of yet another embodiment.

FIG. 5 is still another embodiment.

FIG. 6 is a schematic view of yet another embodiment.

FIG. 7 is a schematic diagram of a basic embodiment of the invention of this application.

FIG. 8 is a schematic diagram of a parallel computing system.

[Explanation of symbols]

 7 computer 11, 11 ', 11' ', 14 network 15 broadcast node 13 host computer

Claims (3)

[Claims] 1. A parallel computer network system in which a plurality of networks for mutual data communication of computers corresponding to a large number of discrete lattice points according to a predetermined order are connected to each of the computers independently of each other, and two sets of each computer are provided. A network system for parallel computers, characterized in that a communication network is connected. 2. The network system for a parallel computer according to claim 1, wherein the arrangement of the network is any of a ring type, a bus type, and a star type. 3. A network system for a parallel computer according to claim 1, wherein another network is provided between the broadcast node for the network and the host computer.