JP2002123497A - Parallel calculating device, analyzing method and recording medium with analysis program recorded thereon - Google Patents

Abstract

PROBLEM TO BE SOLVED: To obtain a parallel calculating device capable of realizing the acceleration of analysis processing without increasing the communication load of a master unit even if realizing high paralleling by increasing the number of slave units. SOLUTION: The process to be carried out by each slave unit includes a reception waiting shifting step in which shift is made to a reception waiting state, an analysis operation step for performing an analysis operation for a self area, a boundary data transmitting step in which boundary data obtained from operation results are transmitted to all slave units for analyzing adjacent areas, a repeating operation step in which the analysis operation of the self area is performed again by using all boundary data after receiving the boundary data form all of the slave units and boundary data transfer and the analysis operation are hereafter repeatedly performed until the boundary value of each area is converged, and an analysis end notification transmitting step in which notification for notifying the end of the analysis processing is transmitted to the master unit.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a parallel computer device comprising a master device and a plurality of slave devices, and more particularly to a parallel computer device for analyzing heat flow of a boiler or the like and navigation performance analysis of a ship or the like. The present invention relates to a computer and an analysis method thereof.

[0002]

2. Description of the Related Art A conventional parallel computer will be described below. For example, in order to promote the differentiation of combustion equipment such as a boiler, it is necessary to further enhance the heat flow analysis, but as the advancement proceeds, the calculation load increases accordingly. More specifically, the current combustion calculation requires, for example, two to four weeks using a workstation (DEC Alpha) per case. In order to shorten the time required for the calculation, it is necessary to upgrade the computer. However, since a high-performance computer is expensive, it is urgently necessary to prepare a computer environment for performing a low-cost and high-speed operation.

Therefore, it is effective to configure an inexpensive and high-speed computer with a PC cluster having high cost performance. However, in order to realize high speed, high parallelization is required. In order to realize this, it is necessary to develop a communication method that supports high parallelism.

[0004] The operation of a highly parallel computer using a conventional PC cluster will be described below. Basically, conventionally, each slave device started from a computer operating as a master device performs a heat-fluid analysis operation on a predetermined region in a boiler assigned to the own device (region division method). In this method, the communication partner of the slave device is fixed to the master device, and the order of transmission and reception is fixed, so that deadlock does not occur. FIG. 13 is a flowchart showing a conventional analysis method.

[0005] First, the master device transmits an initial value necessary for performing a heat and fluid analysis to all slave devices (step S101), and a response from the slave device, that is, a value assigned to the slave device. It waits for reception of boundary data of the region (which will be input information of a slave device that performs a heat flow analysis of a region adjacent to the region) (step S102).

Thereafter, when the master device receives the boundary data from the first slave device (step S10).
(3, No), and further waits for reception of boundary data from the remaining slave devices (step S102). When it is confirmed that boundary data has been received from all slave devices (step S103, Yes), the received boundary data is received. The data is rearranged (step S104), and the boundary data is transmitted to all slave devices (step S10).
5).

[0007] Thereafter, until the heat and fluid analysis using the boundary data is completed (step S106, No), the processing of the above steps S102 to S105 is repeatedly executed.
When the boundary value of each area has converged and the analysis result has been obtained (Yes in step S106), the processing in the master device ends.

On the other hand, after receiving the initial value from the master device (step S111), each slave device performs a thermal-hydraulic analysis operation on the area assigned to the slave device (step S112). Then, the boundary data obtained from the calculation result is transmitted to the master device (step S1).
13), and receives the latest boundary data from the master device as a response (step S114).

At this stage, each slave device checks whether or not the boundary value of each area has converged. If the boundary value has not converged (step S115, No), the above-described slave apparatus continues until the boundary value of each area converges. Step S112 to step S11
The processing of Step 4 is repeatedly executed, and when the convergence is achieved (Step S115, Yes), the processing in the slave device ends.

[0010]

SUMMARY OF THE INVENTION However,
In the conventional parallel computing device, as the number of slave devices increases, the number of transmission destinations of the master device increases, and as a result, the communication load on the master device increases. There was a problem that speed could not be obtained.

More specifically, in the conventional parallel computing device, since the transfer of boundary data is all controlled by the master device, the waiting time of the slave device becomes longer,
As a result, a desired communication speed cannot be obtained.

The present invention has been made in view of the above, and even when the number of slave devices is increased to achieve high parallelism, the speed of analysis processing can be increased without increasing the communication load on the master device. It is an object of the present invention to provide a parallel computing device capable of realizing the above.

[0013]

Means for Solving the Problems The above-mentioned problems are solved,
In order to achieve the object, in the parallel computing device according to claim 1, a predetermined analysis is performed by adopting a region division method and transferring boundary data between slave devices to which adjacent regions are assigned. Processing, and
The master device transmits an initial value necessary for performing the analysis processing to all slave devices, while the slave device that has received the initial value transmits the initial value from another slave device that analyzes an adjacent area. The state shifts to a reception waiting state for waiting for boundary data, and thereafter, an analysis operation of the own area is performed, and boundary data obtained from the operation result is transmitted to all slave devices analyzing the adjacent area, and the adjacent area is analyzed. After receiving the boundary data from all the slave devices, the analysis operation of the own area is performed again using all the boundary data, and thereafter, the boundary data is transferred until the boundary value of each area converges. And repeatedly execute the analysis operation, and at the stage when the boundary value of each area has converged, transmit a notification for notifying the end of the analysis processing to the master device. , The master device, at all stages that received the notification from the slave device, characterized in that the whole process is terminated.

According to the present invention, boundary data is transmitted and received between slave devices without passing through the master device. Specifically, each slave device performs an analysis operation on an area assigned to itself, and transmits boundary data obtained from the operation result directly to all slave devices that analyze an adjacent region. I do.

In the parallel computing device according to the second aspect, the master device creates a transmission / reception partner table for transferring boundary data by directly designating a communication partner between the slave devices at the time of initialization. After that, the initial value including the table is transmitted to all slave devices,
The slave device transmits boundary data obtained from the calculation result to all slave devices that analyze an adjacent area based on the table.

According to the present invention, the boundary data obtained from the analysis operation result is transmitted and received between the slave devices based on the transmission / reception partner table.

In the parallel computing device according to the third aspect, the master device measures the load of the CPU in each slave device before transmitting the initial value, and the measurement result is defined in advance. If it is smaller than the predetermined value, it is determined that analysis processing is possible. Conversely, if it is larger, it is determined that the CPU is being used by another user, and a necessary number of analysis processing CPUs are secured. , Wherein the load measurement is repeatedly executed.

According to the present invention, at the time of starting the analytic operation processing, a search for available CPUs is performed, and a necessary number of CPUs are secured in advance.

According to a fourth aspect of the present invention, in the parallel computing device, a predetermined analysis process is performed by exchanging boundary data between slave devices to which adjacent regions are allocated by adopting a region division method. And the master device transmits an initial value necessary for performing the analysis processing to all slave devices, while the slave device that has received the initial value transmits another initial value to another slave device that analyzes an adjacent area. Transitions to a reception waiting state waiting for boundary data from the plurality of CPUs, and a plurality of CPUs in the same slave device divide the allocated area and perform an analysis operation, and transfer boundary data obtained from the operation result to the adjacent area. After transmitting to all the slave devices analyzing the region to be analyzed and receiving boundary data from all the slave devices analyzing the adjacent region, Using the boundary data, the analysis operation of the region is performed again, and thereafter, the transfer and analysis operation of the boundary data are repeatedly executed until the boundary value of each region converges, and when the boundary value of each region converges, Transmitting a notification for notifying the end of the analysis process to the master device, and finally, when the master device receives the notification from all the slave devices, ends all the processes. And

According to the present invention, a computer having a plurality of CPUs and a shared memory performs arithmetic processing while sharing one area, and reduces overhead without performing communication between CPUs in the same computer.

In the analysis method according to the fifth aspect, an initial value transmitting step of transmitting an initial value necessary for performing the analysis process to all slave devices as a process executed by the master device. And an analysis end notification receiving step of receiving a notification for notifying the end of the analysis process from all the slave devices, and further, as a process executed by each slave device, another slave device that analyzes an adjacent area. A reception waiting transition step for transitioning to a reception waiting state for waiting for boundary data from the server; an analysis operation step for performing an analysis operation of the own area; and boundary data obtained from the operation result, all slave devices for analyzing an adjacent area. The boundary data transmission step for transmitting data to the After that, using all the boundary data, perform the analysis operation of the own region again, and thereafter, repeatedly execute the transfer and analysis operation of the boundary data until the boundary value of each region converges, and And transmitting a notification for notifying the end of the analysis processing to the master device when the boundary value of each area has converged.

According to the present invention, boundary data is transmitted and received between slave devices without going through the master device. Specifically, each slave device performs an analysis operation on an area assigned to itself, and transmits boundary data obtained from the operation result directly to all slave devices that analyze an adjacent region. I do.

In the analysis method according to the sixth aspect, in the initial value transmitting step, a transmission / reception partner table for transferring boundary data by directly designating a communication partner between the slave devices at the time of initialization is set. Create, after that, transmit the initial value including the table to all slave devices, further, in the boundary data transmission step, based on the boundary data obtained from the calculation result, based on the table, The transmission is performed to all slave devices that analyze adjacent areas.

According to the present invention, the boundary data obtained from the analysis operation result is transmitted and received between the slave devices based on the transmission / reception partner table.

In the analysis method according to the seventh aspect, before transmitting the initial value, a load of a CPU in each slave device is measured, and the measurement result is set to a predetermined value. If it is smaller, it is determined that analysis processing is possible. Conversely, if it is larger, it is determined that the CPU is being used by another user, and the required number of CPUs that can perform analysis processing are determined.
A load measurement step of repeatedly executing the load measurement until the time is secured.

According to the present invention, a search for available CPUs is performed at the time of starting the analysis operation processing, and a necessary number of CPUs are reserved in advance.

[0027] In the analysis method according to the eighth aspect, the initial value transmitting step of transmitting an initial value necessary for performing the analysis processing to all the slave apparatuses as the processing executed by the master apparatus. And an analysis end notification receiving step of receiving a notification for notifying the end of the analysis process from all the slave devices, and further, as a process executed by each of the slave devices, another slave that analyzes an adjacent area. A reception waiting transition step for transitioning to a reception waiting state waiting for boundary data from the apparatus, an analysis operation step in which a plurality of CPUs in the same slave apparatus divides an allocated area and performs an analysis operation, A boundary data transmission system for transmitting the obtained boundary data to all slave devices analyzing an area adjacent to the area. After receiving the boundary data from all the slave devices that analyze the adjacent region, the analysis operation of the region is performed again using all the boundary data, and thereafter, the boundary value of each region converges. Until the threshold value of each area converges, and at the stage when the boundary value of each area has converged, an analysis end to transmit a notification for notifying the end of the analysis processing to the master device. A notification sending step;
It is characterized by including.

According to the present invention, a computer having a plurality of CPUs and a shared memory divides one area to perform arithmetic processing, thereby reducing overhead without performing communication between CPUs in the same computer.

[0029] In the program according to the ninth aspect,
A reception waiting transition step of transitioning to a reception waiting state of waiting for boundary data from another slave device analyzing an adjacent area; an analysis operation step of performing an analysis operation of the own area; and boundary data obtained from the operation result. A boundary data transmitting step for transmitting to all the slave devices analyzing the adjacent region, and after receiving the boundary data from all the slave devices analyzing the adjacent region,
Using all the boundary data, the analysis calculation of the own region is performed again, and thereafter, until the boundary value of each region converges,
An iterative operation step of repeatedly executing the transfer and analysis operation of the boundary data, and an analysis end notification transmitting step of transmitting a notification for notifying the end of the analysis processing to the master device when the boundary value of each area has converged. And characterized in that:

According to the present invention, each slave device performs, for example, a thermal-hydraulic analysis operation on an area allocated to the slave device, and outputs boundary data obtained from the operation result.
Transmit directly to all slave devices that analyze adjacent areas.

[0031]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of a parallel computing device according to the present invention will be described below in detail with reference to the drawings. The present invention is not limited by the embodiment.

Embodiment 1 FIG. 1 is a diagram showing a configuration of a parallel computer device (PC cluster) according to the present invention.
In FIG. 1, 1 is a master device, and 2, 3, and 4 are slave devices. The parallel computer device according to the present invention comprises:
Unlike the conventional operation, the transfer of boundary data is performed by control between slave devices without having the master device control the transfer. Although the number of slave devices is three for convenience of description, this number is variable depending on the number of regions of the boiler to be subjected to the heat flow analysis.

FIG. 2 is a diagram showing a configuration of a refuse incineration boiler to be subjected to a heat flow analysis performed using the parallel computer device according to the present invention. Here, the boiler is divided as shown by a chain line in the figure, slave devices are assigned to individual regions, and each slave device performs a heat flow analysis of the region assigned to the own device.

Hereinafter, the operation of the master device and the slave device in the parallel computer device according to the present invention will be described with reference to the drawings. FIG. 3 is a flowchart showing the operation of the master device and the slave device, that is, a heat flow analysis method.

First, the master device transmits an initial value necessary for performing the heat and flow analysis to all the slave devices (step S1), and responds from the slave device, that is, finishes the heat and flow analysis. Wait for a notification for notification (step S2). Then, when the notification for notifying the end of the heat and fluid analysis is received, the processing as the master device ends.

On the other hand, in the slave device, after receiving the initial value from the master device (step S11), first, a transmission source (slave device for analyzing an adjacent area) is specified.
A reception wait for the number of communication partners is issued (step S1).
2). Thereafter, the slave device performs a heat-hydraulic analysis operation on the area assigned to the slave device (step S1).
3). Then, the boundary data obtained from the operation result is transmitted to all the slave devices analyzing the adjacent region (step S14), and further, the boundary data of the adjacent region is transmitted from all the slave devices analyzing the adjacent region. It is confirmed whether or not the data has been received, and if not, it waits until the data is received (step S15). In the heat flow analysis calculation processing, the boundary data transmitted and received between the slave devices are parameters such as temperature, pressure, and flow velocity.

Thereafter, the slave device that has received the boundary data from all the slave devices analyzing the adjacent area (step S16, No) uses all the received boundary data (parameters) again to execute step S1.
2 to execute the processing of step S15.

At this stage, each slave device checks whether the boundary value of each area has converged (step S16).
If not converged (step S16, No),
Until the boundary value of each area converges, the above steps S12 to S12 are performed.
The process of step S15 is repeatedly executed, and when the convergence is reached (step S16, Yes), a notification for notifying the end of the heat and fluid analysis is transmitted to the master device (step S17), and the process in the slave device ends. .

FIG. 4 is a diagram expressing the operation of the above flowchart in a time-series manner. Here, for convenience of explanation, a case where there are two slave devices will be described. First, the master device transmits to the slave devices (1) and (2) initial values necessary for performing the heat flow analysis.

After receiving the initial value from the master device, the slave device (1) first designates the slave device (2) for analyzing an adjacent area, and issues a reception wait. Thereafter, a heat flow analysis operation is performed on the area assigned to the own device, and boundary data obtained from the operation result is transmitted to the slave device (2) that analyzes the adjacent region. Furthermore, as an operation of waiting for reception completion, it is continuously checked whether or not boundary data of an adjacent area has been received from the slave device (2), and the boundary data is received at a position A shown in the figure.

After that, the slave device (1) that has received the boundary data from the slave device (2) issues a reception wait again as described above, and uses all of the received boundary data to perform a heat-fluid analysis calculation. Execute the process. Or later,
The slave device (1) repeatedly executes the processing shown in FIG. 3, and transmits a notification for notifying the end of the heat and fluid analysis to the master device when the boundary value of each region has converged.

On the other hand, also in the slave device (2), after receiving the initial value from the master device, first, the slave device (1) for analyzing the adjacent area is designated, and a reception wait is issued. After that, a heat flow analysis operation is performed on the region assigned to the own device, and boundary data obtained from the operation result is transmitted to the slave device (1) that analyzes the adjacent region. Furthermore, as an operation of waiting for reception completion,
It is checked whether or not the boundary data of the adjacent area has been received from the slave device (1). Here, since the boundary data has already been received at the position B shown in FIG.

Thereafter, in the slave device (2),
In the same manner as described above, a wait for reception is issued, and a heat flow analysis operation is performed using all the received boundary data. Thereafter, the slave device (2) repeatedly executes the processing shown in FIG. 3 and, at the stage when the boundary value of each region has converged, transmits a notification for notifying the end of the heat flow analysis to the master device.

FIG. 5 is a diagram showing a comparison result between the conventional parallel computing device and the parallel computing device of the present invention. As can be seen from FIG. 5, when the heat-fluid analysis calculation is performed by the above operation, the communication speed is improved as compared with the calculation result by the conventional parallel computing device. In the parallel computing device of the present invention, the boundary data is transferred between the slave devices, unlike the related art, so that the waiting time of the slave device is reduced and a desired communication speed is obtained.

FIG. 6 is a diagram showing a configuration of a general computer which operates as the slave device shown in the above embodiment and can realize the above-mentioned heat flow analysis.

This computer comprises a control unit 101 including a CPU, a memory unit 102, a display unit 10
3, an input unit 104, a CD-ROM drive unit 105, and a disk unit 106,
These units are connected via a system bus A. In FIG. 6, the control unit 101 executes the above-described heat flow analysis calculation processing of the boiler or the like. The memory unit 102 includes memories such as a RAM and a ROM, and stores programs to be executed by the control unit 101, necessary data obtained in the course of processing, and the like. The display unit 103 includes a CRT, an LCD (liquid crystal display panel), and the like, and displays various screens for a user of the computer system. The input unit 104 includes a keyboard, a mouse, and the like.
Used to input various information. Also, the illustrated C
The D-ROM 200 stores a program describing the above processing shown in the right side of FIG. 3 (slave device).

In the computer configured as described above, first, the program is installed in the disk unit 106 from the CD-ROM 200 set in the CD-ROM drive unit 105. Then, the program read from the disk unit 106 when the computer is started is stored in the memory unit 102. In this state, the control unit 101 (CPU) executes the processing shown in FIG. 3 according to the program stored in the memory unit 102.

In the present invention, the CD-ROM 2
00, a program describing each process is provided. However, the recording medium of the program is not limited to this. For example, a magnetic disk such as a floppy disk, an optical disk, etc. Magnetic disk, magnetic tape, ROM and RA in computer
It is also possible to use another recording medium such as M. Also,
The program recorded on the recording medium is applicable to all slave devices.

As described above, in the present embodiment, the boundary data is transmitted and received between the slave devices without the intervention of the master device.
The configuration is such that the heat flow analysis operation is performed on the area assigned to the own device, and the boundary data obtained from the operation result is directly transmitted to all slave devices analyzing the adjacent region. In this case, even when high parallelism is realized, the analysis processing can be speeded up without increasing the communication load of the master device.

Although the present embodiment has been described with reference to the parallel computing device for executing, for example, a heat-fluid analysis calculation process of a refuse incineration boiler, the parallel computing device of the present embodiment is not limited to the above-described process. Instead, it is effective for all analysis processes using the area division method. For example, it is also possible to perform a navigation performance analysis of a ship. For example, FIG.
FIG. 8 is a diagram showing a case in which a flow around a straight-ahead ship is calculated, and FIG. 8 is a diagram showing a case in which a chord calculation of a skewed and turning ship is performed.

Embodiment 2 FIG. 9 is a flowchart illustrating a heat flow analysis method according to the second embodiment. The description of the same configuration as in the first embodiment is omitted. Here, only portions different from the first embodiment will be described.

Hereinafter, the operation of the master device and the slave device in the parallel computer device of this embodiment will be described with reference to FIG.

First, at the time of initialization, the master device creates a transmission / reception partner table required for direct communication between the slave devices by designating the communication partner (step S2).
1) Then, an initial value including the transmission / reception partner table is transmitted to all slave devices (step S).
1) Wait for a response from the slave device, that is, a notification for notifying the end of the heat flow analysis (step S2).
Then, when the notification for notifying the end of the heat and fluid analysis is received, the processing as the master device ends.

On the other hand, in the slave device, after receiving the initial value from the master device (step S11), first, a slave device for analyzing an adjacent area is designated, and a reception wait for the number of communication partners is issued (step S11). S12). afterwards,
The slave device performs a thermal-hydraulic analysis operation on the area assigned to the slave device (step S13). And
The boundary data obtained from the calculation result is transmitted to all the slave devices analyzing the adjacent area based on the transmission / reception partner table (step S14). It is confirmed whether or not the boundary data of the adjacent area has been received (step S15). The subsequent operation is the same as in the first embodiment.

As described above, in the present embodiment, the same effects as those of the first embodiment can be obtained, and further, the boundary data obtained from the heat-fluid analysis calculation result is
Since transmission / reception is performed between the slave devices based on the transmission / reception partner table, arithmetic processing in the slave devices can be reduced. In the present embodiment, the parallel computing device that executes the heat-fluid analysis calculation processing of a boiler or the like has been particularly described. However, the parallel computing device according to the present embodiment is not limited to the above-described processing, and may be a region division method. This is effective for all the analysis processes using. For example, it is also possible to perform a navigation performance analysis of a ship (similar to the first embodiment).

Embodiment 3 FIG. 10 is a flowchart illustrating a heat flow analysis method according to the third embodiment. The description of the same configuration as in the first embodiment is omitted. Further, the processing of this embodiment is the same as that of the first embodiment.
3 and FIG. 9 are applicable. Here, Embodiment 1 or 2 described above is used.
Only the parts different from the above will be described.

The operation of the master device and the slave device in the parallel computer according to the present invention will be described below with reference to FIG. FIG. 10 (a) is a diagram showing a flow of CPU allocation at the time of executing the heat and flow analysis.
(B) is a diagram showing an example of a computer list and an execution list.

In the parallel computing device according to the present embodiment, before executing the program of FIG. 3 or FIG.
The processing shown in FIG. First, in the master device,
A computer list listing the host names of the computers as shown in FIG. 10B is opened (step S31), and a host is appropriately searched therefrom (step S32).
If no host remains in the computer list (step S32, no host), the execution list creation process ends. On the other hand, when the host remains (step S32, read), the master device reads the host and measures the load on the CPU (step S33).

As a result, when the calculated load is smaller than a predetermined value (step S34,
Yes), the master device determines that the host is vacant, and adds the host name to the execution list. Step S3 is repeated until there are no more hosts in the list.
The processes from 2 to S35 are repeatedly executed.

On the other hand, when the obtained load is larger than a predetermined value (step S34, N
o), the master device determines that the host is being used by another user, and thereafter repeatedly executes the processing of steps S32 to S35 until a required number of CPUs is secured.

Thereafter, in the present embodiment, the above-described program shown in FIG. 3 or FIG. 9 is executed by using the execution list created as described above and describing the hosts usable for the analysis processing.

As described above, in the present embodiment, the same effects as those of the first or second embodiment can be obtained, and
Further, when the heat flow analysis calculation process is started, the available C
Since the required number of CPUs are reserved in advance by searching for PUs, a plurality of users can use the same parallel computing device at the same time, and the operating rate of the CPU can be improved. Further, for the same reason, a faulty CPU can be excluded in advance, so that fault tolerance can be greatly improved. In the present embodiment, the parallel computing device that executes the heat-fluid analysis calculation processing of a boiler or the like has been particularly described. Is effective for all analysis processes using
It is also possible to perform a navigation performance analysis of a ship (similar to the first embodiment).

Embodiment 4 FIG. 11 is a diagram showing a configuration of the present embodiment. In FIG. 11, reference numerals 11 and 12 denote computers (slave devices).
Is a CPU, and 17 and 18 are shared memories. The overall configuration of the present embodiment is described in the first embodiment.
This is the same as FIG.

In the embodiments described above, parallel execution by multi-process, that is, one CPU in a computer
Has performed the heat-fluid calculation processing of one area (see the lower right of FIG. 11). Such a method is excellent in versatility because each processing is performed in an independent address space. P
In the distributed memory method, which is the basic form of the C cluster,
Since each CPU is executed in a separate OS, that is, in an individual address space, parallel execution by multi-process is inevitable.

Therefore, in the present embodiment, FIG.
As shown in the lower left, one process is assumed to be executed in the shared memory, and the processes are executed in parallel by multiple threads. That is,
A computer with two CPUs and a shared memory
Heat flow calculation processing of two areas is performed. In this case, since the threads in the same process are executed in the same address space, they must be shared memories.

FIG. 12 is a flowchart showing a heat flow analysis method according to the fourth embodiment. Hereinafter, the operation of the slave device in the parallel computer device according to the present invention will be described with reference to FIG. The operation of the master device is the same as that of the first, second, or third embodiment.

First, each slave device issues a reception wait for the number of communication partners (step S41), and then creates a thread for an area allocated to the own device.
That is, a heat flow analysis operation is performed on the area assigned to the own device (step S42). Specifically,
The CPU 13 starts the thread (1) (step S5).
1) A heat flow analysis calculation is performed for an even-numbered row in a predetermined area (step S52). Further, the CPU 14 starts a thread (2) (step S53), and performs a heat-fluid analysis operation on an odd-numbered row in a predetermined area (step S54).

After each thread is terminated (step S
43) In the slave device, the boundary data obtained from the operation result is transmitted to all the slave devices analyzing the adjacent region (step S44), and further, from all the slave devices analyzing the adjacent region, It is checked whether or not the boundary data of the adjacent area has been received, and if not received, it waits until the data is received.

After that, in the slave device that has received the boundary data from all the slave devices that analyze the adjacent area (step S45, No), using the received all the boundary data, steps S41 to S41 are performed again.
The processing of 44 is executed.

At this stage, each slave device checks whether the boundary value of each area has converged (step S45).
If it does not converge (step S45, No),
Until the boundary value of each area converges, the above steps S41 to S41 are performed.
The process of step S44 is repeatedly executed, and when the convergence is reached (step S45, Yes), a notification for notifying the end of the heat flow analysis is transmitted to the master device, and the process in the slave device ends.

As described above, in the present embodiment, 2
Since a computer having two CPUs and a shared memory is configured to perform heat-fluid operation processing by sharing the area, communication between CPUs in the same computer is not required, and overhead can be reduced as compared with the above-described embodiment. And, as a result, performance can be improved. In the present embodiment, the parallel computing device that executes the heat-fluid analysis calculation processing of a boiler or the like has been particularly described. This is effective for all the analysis processes using. For example, it is also possible to perform a navigation performance analysis of a ship (similar to the first embodiment).

[0072]

As described above, according to the parallel computing device of the present invention (claim 1), boundary data is transmitted and received between slave devices without passing through the master device. Specifically, each slave device performs an analysis operation on an area assigned to itself, and transmits boundary data obtained from the operation result directly to all slave devices that analyze an adjacent region. Configuration. As a result, even when the number of slave devices is increased to achieve high parallelism, a parallel computing device capable of realizing high-speed analysis processing without increasing the communication load of the master device can be obtained. The effect is as follows.

Further, according to the parallel computing device of the present invention, the boundary data obtained from the result of the analysis operation is transmitted and received between the slave devices based on the transmission / reception partner table. As a result, it is possible to obtain a parallel computing device capable of reducing the arithmetic processing in the slave device.

Further, according to the parallel computing device of the present invention, a search for available CPUs is performed at the time of starting the analytic operation processing, and the necessary number of CPs is previously determined.
U was secured. As a result, there is an effect that a parallel computing device capable of improving the operation rate of the CPU can be obtained. Further, since the failed CPU can be excluded in advance, there is an effect that a parallel computing device capable of greatly improving fault tolerance can be obtained.

Further, according to the parallel computing device of the present invention (claim 4), a computer having a plurality of CPUs and a shared memory is configured to perform arithmetic processing by sharing an area. This eliminates the need for communication between CPUs in the same computer and reduces overhead, so that a parallel computing device capable of improving performance as a result can be obtained.

According to the analysis method of the present invention (claim 5), boundary data is transmitted and received between slave devices without passing through the master device. Specifically, each slave device performs an analysis operation on an area assigned to itself, and transmits boundary data obtained from the operation result directly to all slave devices that analyze an adjacent region. I do. As a result, even when the number of slave devices is increased to achieve high parallelism, an effect is achieved in that the analysis processing can be speeded up without increasing the communication load of the master device.

Further, according to the analysis method of the present invention (claim 6), the boundary data obtained from the analysis calculation result is transmitted and received between the slave devices based on the transmission / reception partner table. Thereby, there is an effect that the arithmetic processing in the slave device can be reduced.

Further, according to the analysis method of the present invention (claim 7), at the start of the analysis operation processing, a search for available CPUs is performed, and the necessary number of CPUs are previously determined.
To secure. Thereby, there is an effect that the operation rate of the CPU can be improved. Also, the failed CPU
Can be excluded in advance, so that there is an effect that fault tolerance can be greatly improved.

Further, according to the analysis method of the present invention (claim 8), a computer (slave device) having a plurality of CPUs and a shared memory performs arithmetic processing while sharing the area.
This eliminates the need for communication between CPUs in the same computer and can reduce the overhead, resulting in an effect that performance can be improved as a result.

Further, according to the recording medium of the present invention (claim 9), the boundary data is transmitted and received between the slave devices without passing through the master device. Specifically,
Each slave device performs an analysis operation on an area assigned to itself, and transmits boundary data obtained from the operation result directly to all slave devices analyzing an adjacent region. As a result, it is possible to obtain a program capable of realizing high-speed analysis processing without increasing the communication load on the master device even when the number of slave devices is increased and high parallelism is realized. It works.

[Brief description of the drawings]

FIG. 1 is a diagram showing a configuration of a parallel computer device according to the present invention.

FIG. 2 is a diagram showing a configuration of a boiler to be subjected to a heat flow analysis performed using the parallel computer device according to the present invention.

FIG. 3 is a flowchart illustrating a heat flow analysis method according to the first embodiment.

FIG. 4 is a diagram expressing the operation of the heat flow analysis process in a time-series manner.

FIG. 5 is a diagram illustrating a comparison result between the conventional parallel computing device and the parallel computing device according to the first embodiment;

FIG. 6 is a diagram showing a configuration of a general computer capable of realizing a heat flow analysis.

FIG. 7 is a diagram showing a case where a surrounding flow calculation of a straight-ahead ship is performed.

FIG. 8 is a diagram showing a case where both strings of a skewed and turning ship are calculated.

FIG. 9 is a flowchart showing a heat flow analysis method according to the second embodiment.

FIG. 10 is a flowchart showing a heat flow analysis method according to the third embodiment.

FIG. 11 is a diagram showing a configuration of a fourth embodiment.

FIG. 12 is a flowchart illustrating a heat flow analysis method according to the fourth embodiment.

FIG. 13 is a flowchart showing a conventional analysis method.

[Explanation of symbols]

1 Master device, 2, 3, 4 Slave device, 11, 1
2 Computer (slave device), 13, 14, 15, 16
CPU, 17, 18 Shared memory.

Claims (9)

[Claims] 1. A parallel computer device that adopts a region division method and performs predetermined analysis processing by transferring boundary data between slave devices to which adjacent regions are assigned. An initial value necessary for performing the analysis process is transmitted to the slave device that has received the initial value, and waits for boundary data from another slave device that analyzes an adjacent area. After that, an analysis operation of the own area is performed, and boundary data obtained from the operation result is transmitted to all slave apparatuses that analyze the adjacent area, and from all the slave apparatuses that analyze the adjacent area. After receiving the boundary data, the analysis operation of the own region is performed again using all the boundary data, and thereafter, the boundary value of each region converges. Then, the boundary data transfer and the analysis operation are repeatedly executed, and when the boundary value of each area has converged, a notification for notifying the end of the analysis processing is transmitted to the master device. , When the notification is received from all the slave devices, all the processes are terminated. 2. The master device creates a transmission / reception partner table for transferring boundary data by directly specifying a communication partner between slave devices at the time of initial setting, and thereafter, the master device creates a transmission / reception partner table for all slave devices. An initial value including a table is transmitted, and the slave device transmits boundary data obtained from the calculation result to all slave devices that analyze an adjacent area based on the table. The parallel computing device according to claim 1. 3. The master device measures a load on a CPU in each slave device before transmitting the initial value, and when the measurement result is smaller than a predetermined value, analysis processing can be performed. On the contrary, when the CPU is large, it is determined that the CPU is being used by another user, and the load measurement is repeatedly executed until a required number of CPUs capable of performing analysis processing are secured. The parallel computing device according to claim 1. 4. A parallel computer device which adopts a region division method and performs predetermined analysis processing by transferring boundary data between slave devices to which adjacent regions are assigned, wherein a master device is connected to all slave devices. An initial value necessary for performing the analysis process is transmitted to the slave device that has received the initial value, and waits for boundary data from another slave device that analyzes an adjacent area. Then, a plurality of CPUs in the same slave device divide the assigned region and perform an analysis operation, and transfer the boundary data obtained from the operation result to all slave devices analyzing the region adjacent to the region. After receiving boundary data from all slave devices that transmit and analyze adjacent areas, The analysis operation of the region is performed, and thereafter, the transfer of the boundary data and the analysis operation are repeatedly executed until the boundary value of each region converges, and when the boundary value of each region converges, the end of the analysis processing is notified. And transmitting the notification to the master device, and finally, when the master device receives the notification from all the slave devices, terminates all the processes. 5. An analysis method in which a predetermined analysis process is performed by exchanging boundary data between slave devices to which an adjacent region is assigned by adopting an area division method, which is executed by a master device. As the processing, an initial value transmitting step of transmitting an initial value necessary for performing the analysis processing to all slave devices, and an analysis end notification reception receiving a notification for notifying the end of the analysis processing from all slave devices And a reception waiting transition step of transitioning to a reception waiting state of waiting for boundary data from another slave device that analyzes an adjacent area, as a process executed by each of the slave apparatuses. An analysis operation step of performing an analysis operation; and boundary data obtained from the operation result, Boundary data transmission step to be transmitted to the slave device, and after receiving the boundary data from all the slave devices analyzing the adjacent region, using all the boundary data, perform the analysis calculation of the own region again, Thereafter, until the boundary value of each region converges, a repetitive operation step of repeatedly performing the transfer and analysis operation of the boundary data, and at the stage when the boundary value of each region converges, a notification for notifying the end of the analysis process is provided. An analysis end notification transmitting step of transmitting to the master device. 6. In the initial value transmitting step, at the time of initialization, a transmission / reception partner table for transferring boundary data by directly designating a communication partner between the slave devices is created. Transmitting an initial value including the table to the apparatus; and furthermore, in the boundary data transmitting step, the boundary data obtained from the operation result is analyzed based on the table to analyze all adjacent regions. The analysis method according to claim 5, wherein the transmission is performed to a slave device. 7. The method further comprises: before transmitting the initial value, measuring a load on a CPU in each slave device, and if the measurement result is smaller than a predetermined value, it is determined that analysis processing is possible. Conversely, when the CPU is large, it is determined that the CPU is being used by another user, and the load measurement is repeatedly executed until a required number of CPUs capable of performing analysis processing are secured. The analysis method according to claim 5 or 6, wherein: 8. An analysis method in which a predetermined analysis process is performed by exchanging boundary data between slave devices to which an adjacent region is assigned by adopting an area division method, which is executed by a master device. As the processing, an initial value transmitting step of transmitting an initial value necessary for performing the analysis processing to all slave devices, and an analysis end notification reception receiving a notification for notifying the end of the analysis processing from all slave devices And a reception waiting transition step of transitioning to a reception waiting state of waiting for boundary data from another slave device for analyzing an adjacent area as processing executed by each of the slave devices. An arithmetic operation step in which a plurality of CPUs in the computer divides an allocated area and performs an analysis operation; Transmitting the obtained boundary data to all slave devices analyzing the region adjacent to the region, and receiving boundary data from all slave devices analyzing the adjacent region. An analysis operation of the area is performed again using all the boundary data, and thereafter, a boundary operation step of repeatedly performing the transfer and analysis operation of the boundary data until the boundary value of each area converges; An analysis end notification transmitting step of transmitting a notification for notifying the end of the analysis processing to the master device when the values have converged, the analysis method comprising: 9. A reception waiting transition step for transitioning to a reception waiting state for waiting for boundary data from another slave device for analyzing an adjacent area; an analysis operation step for performing an analysis operation of the own area; A boundary data transmitting step of transmitting the obtained boundary data to all slave devices analyzing the adjacent region; and after receiving the boundary data from all the slave devices analyzing the adjacent region, all the boundary data are received. Again, the analysis operation of the own region is performed again, and thereafter, the boundary data of each region is repeatedly executed until the boundary value of each region converges. At this stage, an analysis end notification transmission step for transmitting a notification for notifying the end of the analysis process to the master device is performed. When, characterized in that it comprises a computer-readable recording medium analysis program executed by the slave device.