JP2013196174A - Processing sharing control device, calculation device, calculation system, processing sharing control method, and processing sharing control program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a processing sharing control device capable of predicting a communication time with high accuracy so as to execute optimum processing sharing even in a state in which communication is unstable.SOLUTION: A processing sharing control device includes: a processing resource estimation part for holding first and second individual processing resource estimation values, each of which is an estimation value of a processing resource necessary for completing individual processing, which is included in entire processing and respectively executed by first and second arithmetical units; a communication resource estimation part for holding a first communication resource estimation value estimated on the basis of a communication source used for past communication with the first arithmetical unit, and a second communication resource estimation value estimated on the basis of a communication resource used in past communication with the second arithmetical unit; and a processing sharing pattern generating part for determining sharing of individual processing to the first arithmetical unit and the second arithmetical unit on the basis of the first individual processing resource estimation value, the second individual processing resource estimation value, the first communication resource estimation value and the second communication resource estimation value.

Description

  The present invention relates to a processing sharing control device, a computing device, a computing system, a processing sharing control method, and a processing sharing control program for controlling the sharing of processing to a plurality of computing devices. The present invention relates to a processing sharing control device, a processing device, a processing system, a processing sharing control method, and a processing sharing control program that determine processing sharing based on the processing.

  2. Description of the Related Art A technique for sharing and executing a process using an arithmetic system including a plurality of arithmetic devices is known. Depending on how the processes are shared, the processing performance of the entire computing system is greatly affected. Therefore, in order to maximize the processing performance of the entire computing system, optimization of the sharing of processing is required.

  However, in general, the processing performance of each of the plurality of arithmetic devices and the amount of usable resources are not necessarily equal. Moreover, in order to share processing, communication of information between arithmetic devices is required.

  Therefore, when determining the sharing of processing among a plurality of arithmetic devices, the performance and resources of each arithmetic device must be taken into consideration. Patent Document 1, Patent Document 2, and Patent Document 3 disclose methods for sharing processing among a plurality of calculation means in an arithmetic device in which a plurality of calculation means are connected by communication means.

  In the technique of Patent Literature 1, when the types of terminals and servers are limited, processing sharing among a plurality of processing units is performed by setting processing sharing patterns in advance. At that time, the processing time of the processing unit is taken into consideration.

  In the technique of Patent Document 2, when processing is performed in each of a plurality of CPU (Central Processing Unit) cores, the time required for the processing is estimated. Then, based on the estimated processing time, the processing is shared among the plurality of CPU cores so that the processing time is minimized as a whole. At that time, communication time between CPU cores and processing time of each CPU are taken into consideration.

  In the technique described in Patent Document 3, the amount of communication between communication devices (nodes) and communication time, the time required for the operation device to execute operations, and the like are acquired, and the performance of each task allocation pattern is predicted using them. To do. Then, task allocation is performed based on the predicted performance.

Japanese Patent Laid-Open No. 63-181082 (page 3, FIG. 3) JP 2010-79622 A (pages 11-13, FIGS. 10, 11) JP 2010-257056 A (page 6-7, FIG. 2)

  The techniques of Patent Documents 1 to 3 have problems in handling “communication time” or in its estimation method.

  In the technique described in Patent Document 1, processing is shared using only estimation of processing speed in the processing unit, and communication time is not taken into consideration. As described above, in order to share processing, communication such as transmission / reception of input / output data with other processing units is necessary.

  Whether the process is shared or not can be evaluated using, for example, the process time as a whole, which is the sum of the process times for each process. For this reason, the processing time as a whole is not necessarily minimized unless processing sharing is performed in consideration of not only the processing speed of the processing unit but also the time required for communication. As described above, the technique of Patent Document 1 has a problem that it is impossible to perform optimal processing sharing because the “communication time” is not considered.

  In the technique described in Patent Document 2, it is described that the communication time between CPU cores is taken into account when determining a processing pattern. If the communication time to be considered is the communication time between CPU cores in the same chip, the same substrate, or the same device, it is considered possible to predict with high accuracy. However, it is not always easy to predict the communication time between CPU cores in different apparatuses. For example, when the terminal and the server are connected by wireless communication, communication may become unstable due to the characteristic of wireless communication. For this reason, the communication time between the CPU in the terminal and the CPU in the server may vary every time communication is performed. Therefore, although the technology of Patent Document 2 determines the processing sharing in consideration of the communication time, there is a problem that the optimal processing sharing may not be obtained when the communication is not stable.

The same applies to the technique described in Patent Document 3, and there is a problem in application when communication is not stable.
(Object of invention)
The present invention was made in view of the technical problems as described above, and even in a situation where communication is unstable, it is possible to predict the communication time with high accuracy and perform optimal processing sharing. A processing sharing control device, a computing device, a computing system, a processing sharing control method, and a processing sharing control program are provided.

  The processing sharing control apparatus of the present invention includes a first individual processing resource estimated value that is an estimated value of processing resources necessary for the first arithmetic device to complete the individual processing included in the overall processing, and a second arithmetic operation. Used for past communication between a processing resource estimation unit that holds a second individual processing resource estimation value, which is an estimation value of processing resources necessary for the device to complete individual processing, and the first arithmetic unit. The first communication resource estimated value, which is a necessary communication resource between the first arithmetic device and the past communication with the second arithmetic device, was estimated based on the communication resource. A communication resource estimation unit that holds a second communication resource estimation value that is a necessary communication resource between the second computing device and the first individual processing resource estimation value; 2 individual processing resource estimation values, first communication resource estimation values, and second communication resource estimation values Based on, characterized in that it comprises a processing sharing pattern generation unit for determining the sharing of individual processing to the first computing device and second computing device.

  The computing device of the present invention includes a computing unit, a first individual processing resource estimated value that is an estimated value of processing resources necessary for the computing unit to complete individual processing included in the overall processing, and an external computing device. Based on communication resources used for past communication between a processing resource estimation unit that holds a second individual processing resource estimation value that is an estimation value of processing resources necessary to complete individual processing, and an arithmetic device A communication resource estimation unit that holds a communication resource estimation value that is a necessary communication resource with the computing device, a first individual processing resource estimation value, a second individual processing resource estimation value, and communication A processing sharing pattern generation unit that determines sharing of individual processing to the calculation unit and the calculation device based on the resource estimation value, and a processing sharing control device including the processing sharing pattern generation unit.

  The computing system of the present invention is a first computing device, a second computing device, and a first computing device that is an estimated value of processing resources necessary for the first computing device to complete individual processing included in the overall processing. A processing resource estimation unit that holds the individual processing resource estimation value and a second individual processing resource estimation value that is an estimation value of the processing resource necessary for the second arithmetic unit to complete the individual processing; A first communication resource estimation value, which is a communication resource necessary for the first arithmetic device, estimated based on communication resources used for past communication with the device, and a second arithmetic device A communication resource estimation unit that holds a second communication resource estimation value that is a necessary communication resource with the second arithmetic device, estimated based on communication resources used for past communication with And a first individual processing resource estimation value, a second individual processing resource estimation value, and a first communication A processing sharing control device including a processing sharing pattern generation unit that determines sharing of individual processing to the first computing device and the second computing device based on the source estimated value and the second communication resource estimated value. It is characterized by that.

  The processing sharing control method of the present invention estimates a processing resource value necessary for the first arithmetic unit to complete the individual processing included in the overall processing to obtain a first individual processing resource estimated value, Communication resources used for past communication with the first computing device by estimating the estimated value of processing resources necessary for the computing device to complete the individual processing to obtain the second individual processing resource estimated value Based on the above, the necessary communication resources with the first arithmetic device are estimated, the first communication resource estimation value is obtained, and the communication resources used for the past communication with the second arithmetic device are obtained. Based on this, a necessary communication resource with the second arithmetic unit is estimated, a second communication resource estimated value is obtained, a first individual processing resource estimated value, a second individual processing resource estimated value, a first Based on the communication resource estimated value and the second communication resource estimated value, to the first arithmetic device and the second arithmetic device And determining the allocation of individual processing.

  The processing sharing control program of the present invention is a first individual processing that is an estimated value of processing resources required for the computer included in the processing sharing control device to complete the individual processing included in the overall processing. A processing resource estimation unit that holds a resource estimation value and a second individual processing resource estimation value that is an estimation value of a processing resource necessary for the second arithmetic device to complete individual processing; and a first arithmetic device; A first communication resource estimated value, which is a communication resource required between the first computing device and the second computing device, estimated based on communication resources used for past communication between Communication resource estimation means for holding a second communication resource estimation value, which is a communication resource necessary for communication with the second arithmetic unit, estimated based on communication resources used in the past communication; 1 individual processing resource estimate, 2nd individual processing resource estimate Functioning as a process sharing pattern generation means for determining the sharing of individual processes to the first arithmetic device and the second arithmetic device based on the value, the first communication resource estimated value, and the second communication resource estimated value It is characterized by.

  According to the present invention, even when communication is unstable, it is possible to predict communication time with high accuracy and perform optimal processing sharing.

It is a block diagram which shows the structure of the process allocation control apparatus of 1st Embodiment. It is a block diagram which shows the hardware constitutions of the process allocation control apparatus of 1st Embodiment. It is a specific example of the individual processing resource estimation value table of the first embodiment. It is a specific example of the individual communication resource estimated value table of 1st Embodiment. It is a specific example of the process share pattern figure of 1st Embodiment. It is an example of the flowchart which shows operation | movement of the process assignment control apparatus of 1st Embodiment. It is a specific example of the resource estimation request | requirement process of 1st Embodiment. It is an example of the flowchart which shows the production | generation operation | movement of the process allocation pattern by the dynamic programming in the process allocation control apparatus of 1st Embodiment. It is a block diagram which shows the structure of the process allocation control apparatus of 2nd Embodiment. It is a block diagram which shows the specific example of the internal structure of the process share pattern management part of 2nd Embodiment. It is a specific example of the processing allocation pattern diagram for the individual apparatus of the second embodiment. It is an example of the flowchart which shows operation | movement of the process assignment control apparatus of 2nd Embodiment. It is a block diagram which shows the structure of the software which the arithmetic unit of 3rd Embodiment performs. It is an example of the flowchart which shows operation | movement of the arithmetic unit of 3rd Embodiment. It is a block diagram which shows the structure of the arithmetic system of 4th Embodiment.

(First embodiment)
Next, embodiments of the present invention will be described in detail with reference to the drawings. In the present embodiment, a description will be given of a process sharing control apparatus that generates a process sharing pattern, which is data that specifies how each processing apparatus is to share processes.

  FIG. 1 is a block diagram illustrating a configuration of a processing sharing control apparatus according to the first embodiment. Hereinafter, the software configuration included in the processing sharing control apparatus 1 will be described.

  The processing sharing control apparatus 1 of this embodiment includes a processing resource estimation unit 11, a communication resource estimation unit 12, and a processing sharing pattern generation unit 13.

  FIG. 2 shows an example of a hardware configuration of the processing sharing control apparatus of FIG. The processing sharing control device 1 includes a CPU 5, a memory 6, and a communication unit 7. The processing sharing control device 1 reads a program and data stored in the memory 6 at any time using calculation means such as a CPU 5 and performs communication processing with an external arithmetic device using the communication means 7 as necessary. .

  The processing resource estimation unit 11 holds an estimated value of a computing resource that is a resource necessary for performing the process to be assigned. The estimated value of the calculation resource is specifically an “individual processing resource” that is a processing resource required to individually process each partial process that is a partial process divided in advance in data processing. . In the description of this embodiment, it is assumed that the data processing is divided into P (individual processing) in advance (P is an integer of 1 or more). Hereinafter, when it is necessary to express in contrast with individual processing, data processing before division is referred to as “overall processing”.

  Note that the individual processes to be shared may be a part of the entire process, and it is not always necessary that all of the entire processes be shared objects. That is, the individual process to be shared may be included in the overall process.

  Further, when there are a plurality of individual processes, each individual process may be processed in parallel or sequentially. As will be described later, a criterion for determining whether or not each process is shared is arbitrarily set according to requirements required for the system. Accordingly, there is a possibility that the sharing of the individual processes may change between the case where the individual processes are processed in parallel and the case where the individual processes are performed sequentially.

  The individual processing resource obtained by the estimation is a processing resource that is estimated to be necessary when the processing device actually shares the processing and actually performs the processing. This individual processing resource can be estimated as follows. That is, the individual processing is experimentally executed for each of the arithmetic devices to be shared, and the resources required for the processing are actually measured. For example, when the resource is processing time, it can be estimated that an equivalent processing time is required even during actual operation by actually measuring the time or number of clocks required for processing by each arithmetic device.

  A specific example of the individual processing resource estimation value table indicating the estimation result of the individual processing resources is shown in FIG. The individual processing resource estimated value table shows a result of estimating how much computing resources are required for each partial process constituting the entire process. A specific example of the estimated value will be described later.

  The communication resource estimation unit 12 estimates an individual communication resource, which is a resource necessary for transmitting and receiving data necessary for switching, when switching processing between arithmetic devices according to the processing sharing pattern generated by the processing sharing pattern generation unit 13. To do.

  When other processing devices are assigned to individual processing, input information used for processing or input information such as control information for instructing execution of processing may be transmitted. When the individual process is completed, output information such as output data as a result of the process or control information for notifying the end of the process may be received. Therefore, the individual communication resource may be estimated for both the communication before the start of the individual process and the communication after the completion of the individual process.

  The individual communication resource obtained by estimation is a communication resource estimated to be necessary when actually performing communication. This individual communication resource can be estimated as follows. That is, communication is experimentally performed between the processing sharing control device 1 and the arithmetic device, and the resources required for the processing at that time are actually measured. For example, when the resource is communication time, it can be estimated that the same communication time is required even during actual operation by actually measuring the communication time between the arithmetic devices.

  By the way, the communication time varies depending on the amount of data. The amount of input information for individual processing that is transmitted before sharing with other arithmetic devices, and the amount of output information that is received as a result of individual processing after the completion of individual processing is determined from the contents of the individual processing. Can be estimated. Therefore, the communication time when transmitting / receiving data of the estimated data amount may be measured. Alternatively, the communication time may be estimated by measuring the communication time per unit data amount between arithmetic devices and converting it with the estimated data amount.

  A specific example of the individual communication resource estimation value table showing the estimation results of the individual communication resources is shown in FIG. The individual communication resource estimated value table is a communication resource necessary for transmitting input information before the start of each partial process, and output information after the end of each partial process, that is, another arithmetic device performs a subsequent process. The result of estimating communication resources necessary for receiving information used at the time is shown. A specific example of the estimated value will be described later.

  The process sharing pattern generation unit 13 inputs estimated values from the processing resource estimation unit 11 and the communication resource estimation unit 12. Furthermore, the estimated values of the individual processing resources and the individual communication resources may be input from the processing resource estimation unit 11 and the communication resource estimation unit 12 included in the external arithmetic device.

  The process sharing pattern generation unit 13 determines a process sharing for optimizing a predetermined evaluation criterion and generates it as a “process sharing pattern”.

  In addition, the process sharing pattern generation unit 13 sets a device ID (Identification Data) necessary for generating the process sharing pattern. The device ID uniquely associates the processing sharing control device 1 with the processing devices connected to the processing sharing control device 1 (the processing devices 2,..., N in FIG. 1 (N is an integer of 1 or more)). Is set. The association between the arithmetic device and the device ID may be given in advance, or may be set each time the operation of the processing sharing control device 1 is started or during operation.

  Note that dynamic programming can be used as an example of specific means for determining processing sharing, but is not limited thereto.

An example of the processing sharing pattern is shown in FIG. The process sharing pattern is a numerical sequence in which P device IDs are arranged. The process sharing pattern associates an individual process with an arithmetic device that executes the process. That is, the i-th element (i is an integer satisfying 1 ≦ i ≦ P) from the left in the process sharing pattern indicates that the individual process i (1 ≦ i ≦ P) is shared. For example, in the example of FIG. 5, it is designated that the device with the device ID “1” is assigned to the first individual process, and the device with the device ID “2” is assigned to the P-th individual process. .
(Description of operation)
Next, the operation of this embodiment will be described.

  FIG. 6 shows an example of a flowchart showing the operation of the present embodiment. The process sharing control device 1 of the present embodiment includes a process sharing generation start request input process (step S11), a resource estimation request process (step S12), a resource estimation result acquisition process (step S13), and a process sharing pattern generation process S (step 14) It operates in order.

  First, the processing pattern generation unit 13 inputs a processing sharing generation request from the outside (processing sharing generation start request input, step S11).

  Next, the process pattern generation unit 13 performs a resource estimation request process (step S12). Specifically, step S12 includes an individual processing resource estimation request and an individual communication resource estimation request.

  An example of the operation of the resource estimation request process (step S12) is shown in FIG. Specifically, first, the processing pattern generation unit 13 requests the processing resource estimation unit 11 to estimate individual processing resources (step S121). Next, the processing pattern generation unit 13 requests the communication resource estimation unit 12 to estimate individual communication resources (step S122). Each of the processing resource estimation unit 11 and the communication resource estimation unit 12 requested to estimate outputs an estimated value held therein, or performs estimation by performing a predetermined measurement or the like at that time, and outputs the estimated value To do.

  The individual processing resource estimation (step S121) is performed for individual processing 1,. The estimation of the individual communication resource is performed for each of the external arithmetic devices 2,..., N connected to the processing sharing control device 1 via the communication means. There may be a plurality of external arithmetic units or a single unit.

  Reference is again made to the flowchart of FIG. Next, the process generation pattern generation unit 13 performs a resource estimation result acquisition process (step S13). That is, the process generation pattern generation unit 13 waits until reception of estimation results from the process resource estimation unit 11 and the communication resource estimation unit 12. Here, if a result is not returned after a predetermined time has elapsed since the resource estimation request was made, the standby may be aborted.

  Next, the process sharing pattern generation unit 13 generates a process sharing pattern (step S14). In order to obtain a processing sharing pattern, an optimization method such as a greedy method or a dynamic programming method is used by using the estimated calculation resources for each individual processing. Of course, the optimum processing sharing pattern may be obtained by examining the possibility of all processing sharing in a brute force manner.

Examples of a method for obtaining an optimal sharing process pattern include “greedy method” and “dynamic programming method”. Hereinafter, a specific method for determining a shared processing pattern by each method will be described.
(1) Greedy Method The “greedy method” is a method of obtaining a solution by dividing a problem into a plurality of subproblems, evaluating each of them independently, and selecting them in descending order of evaluation values. Therefore, the greedy method is a suitable method when individual processing 1 to individual processing P can be processed independently.

  In the following example, consider an example in which data on the processing sharing control device 1 is shared and processed by a plurality of arithmetic devices.

  In this case, the individual communication resource includes a communication time estimation value (individual communication resource estimation value) and an individual processing result necessary for transmitting data necessary for individual processing from the processing sharing control device 1 to each arithmetic device. It is the sum with the estimated value of the individual result communication resource, which is the communication time required for transmission from the arithmetic device to the processing sharing control device 1. The estimation of the individual result communication resource can be realized with the same configuration and operation as the estimation of the individual communication resource.

In order to determine an appropriate sharing of processing, for example, in the initial state, all the individual processes are all processed by the arithmetic unit 2. That is, the processing sharing pattern is (2, 2, 2,..., 2). Then, the process sharing pattern is sequentially changed, and the evaluation function value max (t2, t3, t4,..., Tn)
The processing sharing pattern is obtained when is as small as possible.

  Here, t2, t3,..., TN are totals of individual communication resources, individual result communication resources, and individual processing resources related to individual processing performed by the arithmetic device n (2 ≦ n ≦ N).

Since the processing sharing pattern is (2, 2, 2, ..., 2) in the initial state,
min (t2, t3, t4,..., tn) = t2 ≧ 0
And
t3 = ... = tN = 0
It is.

  In addition, when the communication resource is time and the processing resource is power consumption, when the unit expressing the amount of the resource is different, conversion may be performed using an appropriate coefficient to obtain the total.

Specifically, the following operation is performed.
1) Evaluation function value max (t2, t3, t4,..., TN) as processing sharing pattern (2, 2, 2,..., 2) In order to greatly reduce the value of, the sharing of one individual process is changed from the computing device 2 to the sharing of other computing devices.
2) From the partial processes 1 to P, the procedure (1) described above is sequentially repeated to change the sharing.
3) If there is no sharing change that reduces the evaluation function value, (2, 2, 2,..., 2) is set as the processing sharing result.

By repeatedly changing the above procedure 1) -3) with respect to the process sharing pattern, an arithmetic unit suitable for sharing is determined for each partial process, and a process sharing pattern can be finally generated. .
(2) Dynamic programming "Dynamic programming" breaks down a problem into multiple stages, breaks it down into multiple sub-problems that are sequentially processed at each stage, and discards all but the optimal solution up to a certain stage. However, this is a method for obtaining an optimal solution. Therefore, the dynamic programming method is suitable when the entire process is divided from the individual process 1 to the individual process P and each individual process needs to be performed in order.

  The following example considers an example in which the data on the processing sharing control device 1 is used as input data, the individual processing 1 to the individual processing P are sequentially processed, and the final result is returned to the processing sharing control device 1.

  Here, for a device whose device ID is 1 ≦ j ≦ P (j is an integer), the individual communication resource C (j, a (j), b (j)) and the individual processing resource E (j, b) (J)) is defined.

  C (j, a (j), b (j)) performs the individual processing j, and the result of processing by the arithmetic device a (j) in the individual processing (j−1) ) Indicates the value of the individual communication resource to be transmitted. E (j, b (j)) indicates the value of the individual processing resource for processing the individual processing j by the arithmetic device b.

  For convenience, before the execution of the individual process 1, that is, before the execution of the first individual process, the communication to the arithmetic device b (j) (b (j) is 2, 3,..., N) is performed. The time required (individual communication resource) is C (0, a (0), b (0)). After execution of the individual process P, that is, after the end of the final individual process, a time (individual communication resource) for sending the result to the process sharing control device 1 is C (P + 1, a (P + 1), b (P + 1)).

For a (0) ≠ 1, C (0, a (0), b (0)) is set to a sufficiently large number as compared to a value that minimizes the value of the evaluation function described later. This is expressed as C (0, a (0), b (0)) = ∞
Is written.

For b (P + 1) ≠ 1,
C (P + 1, a (P + 1), b (P + 1)) = ∞
It is.

  Note that E (0, b (0)) and E (j + 1, b (P + 1)) are set to 0, but are not limited thereto.

Here, as an evaluation function value, the sum of C (j, a (j), b (j)) + E (j, b (j)) from J = 0 to P + 1 Σ (C (j, a (j ), B (j)) + E (j, b (j)))
Ask for. However, regarding the combination of a (j) and b (j)
b (j-1) = a (j)
That is, the sum is obtained only for the combination in which the transmission source a (j) of communication performed prior to the individual processing j and the arithmetic device b (j−1) that performed the individual processing j−1 match.

  Then, b (j) (j = 1,..., P) that makes the value of the evaluation function as small as possible is obtained.

  Next, a processing resource V (j, a (j), b (j)) is defined. V (j, a (j), b (j)) is “communication from the arithmetic device a (j) to the arithmetic device b (j) and the individual processing j is processed by the arithmetic device b (j). The value of the minimum processing resource under the condition “Yes”.

V (j, a (j), b (j)) is V (j-1, a (j-1), b (j-1)) + C under the condition b (j-1) = aj. A (j) and b (j) when the value of (j, a (j), b (j)) + E (j, b (j)) is minimized. That is,
V (j, a (j), b (j))
= Min (a (j), b (j); V (j-1, a (j-1), b (j-1)) + C (j, a (j), b (j)) + E (j , B (j)))
It is. However, the function min (x; y) here means x when y is minimized.

  In order to obtain the sharing process pattern, first, V (j, a (j), b (j)) is sequentially calculated in order from j = 0 to P + 1 to obtain V (P + 1, a (P + 1). ), B (P + 1)).

Next, min (b (P + 1); V (P + 1, a (P + 1), b (P + 1)))
Find the value of. This b (P + 1) is a value that minimizes the sum of the individual communication resources and the individual processing resources.

  Next, a (j) and b (j) realizing minimization are calculated. This can be obtained by tracing the process of calculating V (j, a (j), b (j)) from j = P + 1 to j = 0. In this way, the sharing process pattern can be obtained as (b (1), b (2),..., B (P)).

  FIG. 8 is an example of a flowchart showing an operation when a process sharing pattern is generated using dynamic programming in the process sharing control apparatus of the present embodiment. FIG. 8 shows an example of individual communication resources and individual processing resources when there is one processing sharing control device and P = 2 computing devices. The device IDj is 1, 2 for the processing sharing control device and 2, 3 for the two arithmetic devices, respectively. Hereinafter, the respective devices are represented as a processing sharing control device 1 and arithmetic devices 2 and 3.

In the preprocessing, since the temporary processing result before the processing is transmitted from the processing sharing control device 1 to the arithmetic device 2 or the arithmetic device 3, the individual communication resources are:
C (0, 1, 2), C (0, 1, 3)
It becomes. Individual processing resources corresponding to each
E (0,2), E (0,3)
It is.

Before the execution of the processing 1, the processing device 1 that has performed the preprocessing executes the processing 1 again, or the other processing device transmits the result of the preprocessing.
C (1, 2, 2), C (1, 2, 3), C (1, 3, 2), C (1, 3, 3)
It becomes. Then, the arithmetic devices 2 and 3 execute the process 1.

The individual processing resource corresponding to each of the individual communication resources described above is based on the individual processing resource for processing process 1 and the device ID of the arithmetic device that performed the processing.
E (1, 2), E (1, 3) E (1, 2), E (1, 3)
It becomes.

  In the same manner, C (j, a (j), b (j)) and E (j, b (j)) are sequentially obtained until post-processing. Then, V (P + 1, a (P + 1), b (P + 1)) is obtained by sequentially calculating V (j, a (j), b (j)) in order from j = 0 to P + 1. , Min (b (P + 1); V (P + 1, a (P + 1), b (P + 1))).

  The arrows in FIG. 8 indicate the result of tracking the value that minimizes the sum of the individual communication resources and the individual processing resources. In this case, the processing pattern is (2, 3,..., 2).

  The degree of optimization can be evaluated by using the estimated calculation resources for each individual process and estimating the calculation resources necessary for the entire process. In this case, as a suitable example of the evaluation criterion for evaluating the degree of optimization, there is a method that employs minimization or maximization of the estimated total of calculation resources. In other words, there is a method of determining the sharing of individual processing so that the total amount of computing resources is minimized or maximized.

  A specific example of a resource estimation method performed by the processing resource estimation unit 11 and the communication resource estimation unit 12 will be described. The processing resource estimation method performed by the processing resource estimation unit 11 includes, for example, a method of acquiring processing resources from a table of estimated values prepared in advance. Furthermore, the acquired estimated value may be corrected and used in consideration of the current usage status of processing resources.

  As a communication resource estimation method performed by the communication resource estimation unit 12, for example, there is a method of acquiring communication resources from a table of estimated values prepared in advance. Furthermore, the acquired estimated value may be corrected and used in consideration of the current usage status of communication resources.

  The estimated value related to the individual processing resource and the individual communication time held in the estimated value table may be a value estimated in advance with reference to a past value.

  For example, when the processing time required for individual processing is used as the individual processing resource, the history of the individual processing time is stored. Then, the individual processing time is estimated from the past individual processing time at the time of actual processing.

Specifically, the number indicating the number of executions of individual processing is i (i is an integer of 1 or more), and the number of processing times is N (N is an integer of i or more). Σ (Individual processing time of i-th processing)) / N
Is obtained as an estimated value of the individual processing resource.

  In addition to the processing time required for the individual processing, a history regarding the CPU usage rate may be stored. In this case, the processing time expected when the CPU usage rate is 100% can be calculated and used as the estimated value of the individual processing resources.

Specifically, the number indicating the number of processes is i, the number of processes is N,
(Σ (individual processing time of i-th processing) × (CPU usage rate [%] during i-th processing) / 100)) / N
Is obtained as an estimated value of the individual processing resource.

  In addition, it is possible to obtain an estimated value of the individual processing resource with higher accuracy by using the value immediately before determining the sharing in the history of the processing time and CPU usage rate required for the individual processing.

  The same applies to the communication time. That is, an average value may be obtained from the past communication time history, and may be used as an estimated value of individual communication resources. At this time, if the communication time immediately before determining the sharing is used in the communication time history, an estimated value of the individual communication resource with higher accuracy can be obtained.

Next, the effect of this embodiment will be described. As described above, the processing sharing control apparatus 1 estimates processing resources and communication resources for the arithmetic device using the history of processing resources and communication resources. Then, the processing sharing is determined using the acquired estimated value of the resource. Therefore, there is an effect that it is possible to perform optimum processing sharing with high accuracy.
(Second Embodiment)
A process sharing control apparatus according to the second embodiment of the present invention will be described. In addition to the function of the processing sharing control device 1 of the first embodiment, the processing sharing control device B1 of the second embodiment outputs the generated processing sharing pattern to another arithmetic device connected by communication means. It has a function.

  The configuration and operation of the main part of the processing sharing control apparatus B1 of this embodiment are the same as the configuration and operation of the processing sharing control apparatus 1 of the first embodiment. Therefore, only differences between the configuration and operation of the processing sharing control apparatus B1 of the present embodiment and the processing sharing control apparatus 1 of the first embodiment will be described.

  FIG. 9 is a block diagram illustrating an example of the configuration of the processing sharing control apparatus B1 according to the second embodiment. The difference in configuration between the processing sharing control device B1 and the processing sharing control device 1 of the first embodiment is that the processing sharing pattern generation unit 13 of the processing sharing control device 1 is replaced with a processing sharing pattern management unit B13. It is only a point.

  FIG. 10 is a block diagram illustrating an example of the configuration of the processing sharing pattern management unit B13 according to the second embodiment. The processing sharing pattern management unit B13 includes a processing sharing pattern generation unit B131 and a processing sharing pattern distribution unit B132. Since the processing sharing pattern generation unit B131 is the same as the processing sharing pattern generation unit 13 in the configuration of the first embodiment, the description thereof is omitted.

  The process sharing pattern distribution unit B132 distributes the converted pattern after converting the generated process sharing pattern for the other arithmetic devices 2,..., N.

  This will be described using an example of processing sharing in FIG. For the arithmetic device 2 designated by the device ID = 2, only the elements in which the element is “2” in the processing sharing pattern, that is, the elements for which the arithmetic device 2 is designated as the device IS = 2 are picked up. To do. In the example of FIG. 5, since the device ID for performing the individual processing 1 and the individual processing 3 is “2”, the value of these two elements remains “2” and the values of the other elements are “0”. A simple pattern. Then, the generated pattern is distributed to the arithmetic device 2 specified by the device ID = 2. This pattern is referred to as “individual device processing sharing pattern”. FIG. 11 shows an example of the individual device processing assignment pattern distributed to the arithmetic device 2 specified by the device ID = 2. The processing for generating the individual device processing sharing pattern is performed for all the arithmetic devices 2,...

  Note that the process sharing control device 1 may include a calculation unit (not shown) having a calculation function, and a part of processing may be shared as a calculation device with a device ID = 1. In this case, the processing assignment pattern for the individual device may be transmitted to the calculation unit inside the processing assignment control device 1.

  Next, the operation of the second embodiment will be described. FIG. 12 is a flowchart showing the operation of the second embodiment. Here, the process sharing generation start request input process (step SB11), the resource estimation request process (step SB12), the resource estimation result acquisition process (step SB13), and the process sharing pattern generation process (step SB14) are shown in FIG. , Processing sharing generation start request input processing (step S11), resource estimation request processing (step S12), resource estimation result acquisition processing (step S13), and processing sharing pattern generation processing (step S14).

  The difference between the operation of the second embodiment and the operation of the first embodiment is that the process sharing pattern distribution process (step SB15) is performed after the process sharing pattern generation process (step S14) ends. In step (step SB15), the processing sharing pattern distribution unit B132 distributes the individual device processing sharing pattern to the other arithmetic devices.

The effect of this embodiment will be described. The processing sharing control apparatus according to the first embodiment includes only a function for generating a processing sharing pattern. On the other hand, the process assignment control device of the second embodiment can instruct each processing device to which the individual process is assigned which process should be assigned. Thereby, an individual process can be shared and performed between arithmetic units.
(Third embodiment)
Next, as a third embodiment of the present invention, the configuration and operation of the arithmetic device A1 that performs processing sharing by inputting the processing sharing pattern generated as described above will be described. The arithmetic device A1 shares the individual processing according to an instruction by the processing sharing pattern from the processing sharing control device B1.

  First, the configuration of the third embodiment will be described with reference to an example in which processing is realized mainly in the form of software. Since the configuration of the processing sharing control apparatus B1 in the present embodiment is the same as the configuration of the processing sharing control apparatus 1 in the first embodiment, the description thereof is omitted.

  Next, a software configuration of processing executed by the arithmetic device A1 will be described. FIG. 13 is a block diagram illustrating a configuration of software executed by the arithmetic device A1 according to the third embodiment. The processing of the arithmetic device A1 includes an individual device processing assignment pattern input unit A11 and an individual processing execution unit A12.

  The individual device processing assignment pattern input unit A11 inputs the individual device processing assignment pattern and requests the individual processing execution unit A12 to execute the processing allocated to the arithmetic device A1 in the individual processing assignment pattern. .

  The individual process execution unit A12 executes the process requested from the individual device process assignment pattern input unit A11.

  Next, the operation of the third embodiment will be described. FIG. 14 is a flowchart showing the operation of the third embodiment. First, the individual apparatus process assignment pattern input unit A11 performs the individual process assignment pattern input process (step SA11). Next, the individual device processing assignment pattern input unit A11 performs processing execution request processing (step SA12). Next, the individual process execution unit A12 performs the individual process execution process (step SA13).

  The arithmetic device A1 is preferably connected to the processing sharing control device B1 of the second embodiment and used. At this time, the same overall processing program may be installed in all the arithmetic devices connected to the processing sharing control device B1. In this case, in each arithmetic device, it is possible to omit a part of the processing in units of individual processing, or to replace with another processing having the same function but different performance. Further, if the same overall processing program is installed in all the arithmetic devices, it is not necessary to transfer the individual processing program assigned to the assigned arithmetic device each time the individual processing is assigned.

Next, the effect of this embodiment will be described. As described above, each arithmetic device shares individual processing according to the process sharing pattern generated by the process sharing pattern generation unit 13. Therefore, each arithmetic device can perform only necessary processing.
(Fourth embodiment)
Next, specific examples of the above-described embodiments will be shown.

  In the fourth embodiment, an example of an arithmetic system including the processing sharing control devices B1-1 and B1-2 of the second embodiment and two arithmetic devices A1-1 and A1-2 of the third embodiment. Indicates. FIG. 15 is a block diagram illustrating a configuration of a calculation system according to the fourth embodiment. The arithmetic device A1-1 and the processing sharing control device B1-1 are configured as an integrated arithmetic device C1 which is one unit. Similarly, the arithmetic device A1-2 and the processing sharing control device B1-2 are configured as an integrated arithmetic device C2 that is one unit. The processing sharing control device B1-1 and the arithmetic device A1-1, the processing sharing control device B1-1, and the arithmetic device A1-2 are connected by communication means.

  In the present embodiment, an operation using the integrated arithmetic device C1 as a host is performed. That is, it is a main body that calculates the integrated arithmetic device C1 processing assignment pattern, and also estimates the individual computing resources and individual communication resources of the arithmetic device A1-1.

  On the other hand, the integrated arithmetic device C2 estimates the individual computing resources and the individual communication resources of the arithmetic device A1-2. Then, the integrated arithmetic device C2 transmits the estimated values of the individual calculation resources and the individual communication resources of the arithmetic device A1-2 to the integrated arithmetic device C1.

  The integrated computing device C1 uses the estimated values of the individual computing resources and the individual communication resources of the computing device A1-1 and the estimated values of the individual computing resources and the individual communication resources of the computing device A1-2, A processing sharing pattern that specifies the sharing of processing to A1-2 is calculated.

  In this embodiment, processing time is considered as a resource. The individual calculation resource is a processing time in the arithmetic devices A1-1, A1-2 and the processing sharing control B1-2. The individual communication resource is a communication time between the processing sharing control B1-1, the arithmetic devices A1-1, A1-2, and the processing sharing control B1-2.

  The “individual processing time” is estimated as follows for each integrated arithmetic device by the processing resource estimation unit 11 included in the integrated arithmetic device C1 and the integrated arithmetic device C2. In this embodiment, as a table of estimated values related to processing resource estimation, estimated values when it is assumed that the CPU processing performance of each of the arithmetic devices A1-1 and A1-2 can be used 100% are stored in advance in a table format. I will keep it.

  For example, when the processing resource is processing time, the processing resource is estimated by causing the CPU included in each of the arithmetic devices A1-1 and A1-22 to execute the individual processing in advance and measuring the time required for the processing. It can be carried out. At this time, the CPU usage rate at the time of executing the individual processing is also measured. Then, the CPU usage rate in each arithmetic device is acquired when a processing resource estimation request is made. Then, a processing resource when the CPU usage rate is 100% is obtained and set as a processing resource estimated value. This is because the processing time measured in a state where the CPU usage rate is low does not accurately represent the actual load of individual processing. Therefore, when the CPU usage rate is 100%, it is converted into a processing resource estimated value.

  The above processing is added, and the estimated processing time is corrected based on the CPU usage rate.

Specifically, when the CPU usage rate during execution of individual processing in a certain arithmetic device is α [%], the processing time estimated to be necessary in this arithmetic device is:
(Individual processing time estimate to be held) × α / 100
Can be estimated. What is estimated is the individual processing time estimated value for the arithmetic device A1-1 and the individual processing time estimated value for the arithmetic device A1-2.

The “individual communication time” is estimated by the communication resource estimation unit 11 included in the integrated arithmetic device C1 and the integrated arithmetic device C2 as follows. In this embodiment, as a table of estimated values related to processing resource estimation, for example, processing time when the communication speed is 1 Mbps is held. Moreover, the communication speed in each arithmetic device is acquired at the time of a processing resource estimation request. That is, before determining the sharing of processing, dummy data having a size determined for each individual process is transmitted in advance, and the time required for transmission can be obtained, whereby the communication speed at that time can be measured. If the communication speed obtained here is β Mbps, the processing time required for individual communication is
(Individual communication time estimate to be held) / β
Can be estimated. The estimated values are the individual communication time estimation values for the arithmetic device A1-1 (individual communication time for communicating data from the arithmetic device A1-1 to the arithmetic device A1-2) and the individual communication time for the arithmetic device A1-2. It is an estimated value (individual communication time for communicating data from the arithmetic unit A1-2 to the arithmetic unit A1-1).

  Further, as described above, the amount of data for individual processing may vary depending on the content of the individual processing. Further, the amount of input information for individual processing or the amount of output information as a result of individual processing is not necessarily the same. Therefore, the above individual communication time estimated value may be handled as an estimated value when communication of unit data amount is performed, and the communication time may be estimated by multiplying the estimated value by the amount of data actually communicated.

  In order to determine the processing sharing pattern, the sum of the individual processing time and the individual communication time is used as an evaluation criterion. That is, an optimization method may be used so as to minimize the sum of the individual processing time and the individual communication time. As an optimization method at this time, for example, dynamic programming can be used. At that time, the CPU usage rate of each arithmetic device immediately before determining the sharing may be acquired from each arithmetic device and taken into consideration. That is, control such as not sharing the processing to the CPU having a high CPU usage rate or sharing the individual processing with a light load may be performed.

  Note that the individual communication time is taken into account, for example, when a certain individual process is executed in A1-1 and the subsequent individual process is executed in A1-2, the data necessary for the process is changed from A1-1 to A1-2. It is because it is necessary to transfer to. When the subsequent two individual processes are executed by the same arithmetic unit of the third embodiment, the individual communication time may be 0 or a sufficiently small value given in advance.

As described above, in the present embodiment, the extent to which processing resources and communication resources can be used is estimated before determining processing sharing. Therefore, even when the arithmetic device is in a high load state or communication is unstable, there is an effect that it is possible to determine the optimum processing sharing according to the situation at that time.
(Fifth embodiment)
When the entire process is image processing, the estimation processing time can be corrected using the properties of the image. In the fifth embodiment, a specific example of processing resource estimation when the entire process is image processing and an example of sharing of processing using the processing resource are shown. For example, when the processing time of the individual processing is proportional to the area of the image, the estimated value of the individual processing resources per unit area, that is, the estimated processing time may be held in a table format in advance. By multiplying the estimated processing time per unit area by the area of the image that can be acquired at the time of processing resource estimation request, the estimated processing time when the individual processing is performed on the image can be obtained.

  In addition, when calculating the number of feature points from an image by individual processing, processing similar to the above is possible even when processing time proportional to the number of feature points is required. In other words, the estimation processing time of individual processing for obtaining one feature point is prepared in advance, and the estimation when the individual processing is performed on the feature point of the image by multiplying the estimation processing time by the number of feature points Processing time can be determined.

  As described above, the case where the processing time of the individual processing is proportional to the image area and the number of feature points has been described as an example. However, the correspondence relationship is not limited to a proportional relationship as long as the correspondence relationship between the image feature amount such as the image area and the number of feature points and the increase / decrease in the processing time can be specified. Applicable.

  Further, the optimization problem can be applied as it is described in the first embodiment.

  Other than the above, it can be easily obtained by considering the memory capacity of each arithmetic device and solving a limited optimization problem related to the amount of memory. For example, when it is assumed that the memory capacity that needs to be held in advance for a certain individual process exceeds the memory capacity of the arithmetic device, the determination that the arithmetic device is not to perform the individual processing is also possible. Is possible.

  Further, instead of the individual processing time and the individual communication time, the power consumption required for the individual processing and the power consumption required for the individual communication may be adopted as the determination criteria (parameter, generation information). In this case, it is possible to generate a processing sharing pattern that achieves the lowest power consumption as a whole.

  As mentioned above, although the processing sharing control apparatus mentioned above has demonstrated as an example mainly implement | achieving with the form of software, you may implement | achieve part or all of a component in hardware.

  Each of the above programs may be stored in a non-transitory medium such as a ROM (Read Only Memory), a RAM (Random Access Memory), a semiconductor memory device such as a flash memory, an optical disk, a magnetic disk, or a magneto-optical disk. Good.

A part or all of the above-described embodiment can be described as in the following supplementary notes, but is not limited thereto.
(Appendix 1)
A first individual processing resource estimated value that is an estimated value of processing resources necessary for the first arithmetic unit to complete individual processing included in the overall processing, and a second arithmetic unit to complete the individual processing; A processing resource estimation unit that holds a second individual processing resource estimation value that is an estimation value of the processing resource required for
A first communication resource estimation value, which is a communication resource required between the first arithmetic device, estimated based on communication resources used for past communication with the first arithmetic device; And a second communication resource estimation value, which is a communication resource necessary for the second arithmetic device, estimated based on communication resources used for past communication with the second arithmetic device. A communication resource estimation unit holding
Based on the first individual processing resource estimation value, the second individual processing resource estimation value, the first communication resource estimation value, and the second communication resource estimation value, the first arithmetic device and the second A processing sharing pattern generation unit that determines the sharing of the individual processing to the two arithmetic devices;
A process sharing control apparatus comprising:
(Appendix 2)
The process sharing control apparatus according to claim 1, wherein the first communication resource estimated value and the second communication resource estimated value are obtained for each of the individual process before the start of execution and after completion of the individual process. .
(Appendix 3)
The first individual processing resource estimation value and the second individual processing resource estimation value correspond to the total processing capacity when each of the first arithmetic device and the second arithmetic device executes the individual processing. The processing sharing control apparatus according to appendix 1 or 2, characterized in that it is obtained based on a processing capacity usage rate that is a ratio of used processing capacities.
(Appendix 4)
The individual processing is processing of a predetermined share of input information of the overall processing,
4. The process sharing control apparatus according to any one of appendices 1 to 3, wherein the first individual process resource estimated value and the second individual process resource estimated value are obtained based on the shared amount.
(Appendix 5)
The supplementary note 4 is characterized in that the first communication resource estimated value and the second communication resource estimated value are obtained based on an output information amount output as a result of the individual processing with respect to the shared amount. Processing sharing control device.
(Appendix 6)
The processing sharing control device according to any one of appendices 1 to 5, further comprising a processing sharing pattern distribution unit that outputs the sharing to the first computing device and the second computing device.
(Appendix 7)
The first individual processing resource estimation value and the second individual processing resource estimation value are respectively the memory capacity that can be allocated for the first arithmetic device and the second arithmetic device to complete the individual processing. An estimate,
The first communication resource estimated value and the second communication resource estimated value are estimated values of memory capacity that can be allocated for communication with the first arithmetic device and the second arithmetic device, respectively. And
The processing sharing pattern generation unit is configured to store a memory based on the first individual processing resource estimation value, the second individual processing resource estimation value, the first communication resource estimation value, and the second communication resource estimation value. The process sharing control device according to any one of appendices 1 to 6, wherein the process sharing pattern is generated so that allocation of the process is possible.
(Appendix 8)
The first individual processing resource estimation value and the second individual processing resource estimation value are estimated values of power consumed by the first arithmetic device and the second arithmetic device, respectively, for completing the individual processing. And
The first communication resource estimated value and the second communication resource estimated value are estimated values of power consumed to perform communication with the first arithmetic device and the second arithmetic device, respectively. ,
The process sharing pattern generation unit is configured to determine the first individual processing resource estimated value, the second individual processing resource estimated value, the first communication resource estimated value, and the second communication resource estimated value, 7. The process sharing control apparatus according to any one of appendices 1 to 6, wherein the process sharing pattern is generated so that power consumed for completing the individual process is minimized.
(Appendix 9)
An arithmetic unit;
A first individual processing resource estimated value that is an estimated value of processing resources necessary for the computing unit to complete individual processing included in the overall processing, and an external computing device necessary for completing the individual processing. A processing resource estimation unit that holds a second individual processing resource estimation value that is an estimation value of the processing resource;
A communication resource estimation unit that holds a communication resource estimation value that is a necessary communication resource with the arithmetic device, estimated based on communication resources used in past communication with the arithmetic device;
A processing sharing pattern for determining the sharing of the individual processing to the computing unit and the computing device based on the first individual processing resource estimated value, the second individual processing resource estimated value, and the communication resource estimated value A generator,
Including a processing sharing control device,
An arithmetic device comprising:
(Appendix 10)
A first arithmetic unit;
A second arithmetic unit;
A first individual processing resource estimated value that is an estimated value of processing resources required for the first arithmetic device to complete individual processing included in the overall processing, and the second arithmetic device completes the individual processing. A processing resource estimation unit that holds a second individual processing resource estimation value that is an estimation value of a processing resource necessary to perform the processing, based on communication resources used for past communication with the first arithmetic device Based on the estimated first communication resource estimation value, which is a necessary communication resource with the first arithmetic device, and the communication resource used for past communication with the second arithmetic device. A communication resource estimation unit that holds a second communication resource estimation value that is a necessary communication resource with the second arithmetic device, the first individual processing resource estimation value, and the second Individual processing resource estimate, the first communication resource estimate and the first Based of the communication resource estimates, the a processing sharing control unit including a processing sharing pattern generation unit for determining the sharing of individual processing to the first computing device and the second computing device,
An arithmetic system comprising:
(Appendix 11)
The first arithmetic unit estimates a processing resource value necessary for completing the individual processing included in the overall processing to obtain a first individual processing resource estimation value,
A second arithmetic unit estimates an estimated value of processing resources necessary to complete the individual processing to obtain a second individual processing resource estimated value;
Estimating necessary communication resources with the first arithmetic device based on communication resources used for past communication with the first arithmetic device to obtain a first communication resource estimate ,
Based on communication resources used for past communication with the second arithmetic device, necessary communication resources with the second arithmetic device are estimated to obtain a second communication resource estimate. ,
Based on the first individual processing resource estimation value, the second individual processing resource estimation value, the first communication resource estimation value, and the second communication resource estimation value, the first arithmetic device and the second 2. A processing sharing control method, wherein the sharing of the individual processing to the two arithmetic devices is determined.
(Appendix 12)
A computer provided in the processing sharing control device,
A first individual processing resource estimated value that is an estimated value of processing resources necessary for the first arithmetic unit to complete individual processing included in the overall processing, and a second arithmetic unit to complete the individual processing; Processing resource estimation means for holding a second individual processing resource estimation value that is an estimation value of processing resources required for
A first communication resource estimation value, which is a communication resource required between the first arithmetic device, estimated based on communication resources used for past communication with the first arithmetic device; And a second communication resource estimation value, which is a communication resource necessary for the second arithmetic device, estimated based on communication resources used for past communication with the second arithmetic device. Communication resource estimation means for holding
Based on the first individual processing resource estimation value, the second individual processing resource estimation value, the first communication resource estimation value, and the second communication resource estimation value, the first arithmetic device and the second A processing sharing control program that functions as processing sharing pattern generation means for determining sharing of the individual processing to two arithmetic devices.

  As an application example of the present invention, there is optimization of cooperation processing between a client and a server. That is, the entire client and server are regarded as one arithmetic device, and the arithmetic device is caused to execute predetermined processing. At this time, as long as the processing (overall processing) performed by the arithmetic device as a whole can be divided into partial processing (individual processing), the content of the processing is arbitrary. For example, the entire process may be a process of collating an input image, and the individual process may be a process of collating a part of the image.

DESCRIPTION OF SYMBOLS 1, B1, B1-1, B1-2 Processing share control apparatus 2, ..., N, A1, A1-1, A1-2 Arithmetic apparatus 11 Processing resource estimation part 12 Communication resource estimation part 13, B131 Process sharing pattern Generation unit A11 Individual device processing sharing pattern input unit A12 Individual processing execution unit B13 Processing sharing pattern management unit B132 Processing sharing pattern distribution unit C1, C2 Integrated arithmetic device 5 CPU
6 Memory 7 Communication means

Claims (10)

A first individual processing resource estimated value that is an estimated value of processing resources necessary for the first arithmetic unit to complete individual processing included in the overall processing, and a second arithmetic unit to complete the individual processing; A processing resource estimation unit that holds a second individual processing resource estimation value that is an estimation value of the processing resource required for
A first communication resource estimation value, which is a communication resource required between the first arithmetic device, estimated based on communication resources used for past communication with the first arithmetic device; And a second communication resource estimation value, which is a communication resource necessary for the second arithmetic device, estimated based on communication resources used for past communication with the second arithmetic device. A communication resource estimation unit holding
Based on the first individual processing resource estimation value, the second individual processing resource estimation value, the first communication resource estimation value, and the second communication resource estimation value, the first arithmetic device and the second A processing sharing pattern generation unit that determines the sharing of the individual processing to the two arithmetic devices;
A process sharing control apparatus comprising: 2. The process sharing control according to claim 1, wherein the first communication resource estimation value and the second communication resource estimation value are obtained for each of before and after execution of the individual process is started. apparatus. The first individual processing resource estimation value and the second individual processing resource estimation value correspond to the total processing capacity when each of the first arithmetic device and the second arithmetic device executes the individual processing. The processing sharing control apparatus according to claim 1, wherein the processing sharing control apparatus is obtained based on a processing capacity usage rate that is a ratio of used processing capacities. The individual processing is processing of a predetermined share of input information of the overall processing,
4. The process sharing control apparatus according to claim 1, wherein the first individual process resource estimated value and the second individual process resource estimated value are obtained based on the shared amount. 5. The first communication resource estimated value and the second communication resource estimated value are obtained based on an output information amount output as a result of the individual processing with respect to the shared amount. Processing sharing control device. The processing sharing control device according to claim 1, further comprising a processing sharing pattern distribution unit that outputs the sharing to the first computing device and the second computing device. An arithmetic unit;
A first individual processing resource estimated value that is an estimated value of processing resources necessary for the computing unit to complete individual processing included in the overall processing, and an external computing device necessary for completing the individual processing. A processing resource estimation unit that holds a second individual processing resource estimation value that is an estimation value of the processing resource;
A communication resource estimation unit that holds a communication resource estimation value that is a necessary communication resource with the arithmetic device, estimated based on communication resources used in past communication with the arithmetic device;
A processing sharing pattern for determining the sharing of the individual processing to the computing unit and the computing device based on the first individual processing resource estimated value, the second individual processing resource estimated value, and the communication resource estimated value A generator,
Including a processing sharing control device,
An arithmetic device comprising: A first arithmetic unit;
A second arithmetic unit;
A first individual processing resource estimated value that is an estimated value of processing resources required for the first arithmetic device to complete individual processing included in the overall processing, and the second arithmetic device completes the individual processing. A processing resource estimation unit that holds a second individual processing resource estimation value that is an estimation value of a processing resource necessary to perform the processing, based on communication resources used for past communication with the first arithmetic device Based on the estimated first communication resource estimation value, which is a necessary communication resource with the first arithmetic device, and the communication resource used for past communication with the second arithmetic device. A communication resource estimation unit that holds a second communication resource estimation value that is a necessary communication resource with the second arithmetic device, the first individual processing resource estimation value, and the second Individual processing resource estimate, the first communication resource estimate and the first Based of the communication resource estimates, the a processing sharing control unit including a processing sharing pattern generation unit for determining the sharing of individual processing to the first computing device and the second computing device,
An arithmetic system comprising: The first arithmetic unit estimates a processing resource value necessary for completing the individual processing included in the overall processing to obtain a first individual processing resource estimation value,
A second arithmetic unit estimates an estimated value of processing resources necessary to complete the individual processing to obtain a second individual processing resource estimated value;
Estimating necessary communication resources with the first arithmetic device based on communication resources used for past communication with the first arithmetic device to obtain a first communication resource estimate ,
Based on communication resources used for past communication with the second arithmetic device, necessary communication resources with the second arithmetic device are estimated to obtain a second communication resource estimate. ,
Based on the first individual processing resource estimation value, the second individual processing resource estimation value, the first communication resource estimation value, and the second communication resource estimation value, the first arithmetic device and the second 2. A processing sharing control method, wherein the sharing of the individual processing to the two arithmetic devices is determined. A computer provided in the processing sharing control device,
A first individual processing resource estimated value that is an estimated value of processing resources necessary for the first arithmetic unit to complete individual processing included in the overall processing, and a second arithmetic unit to complete the individual processing; Processing resource estimation means for holding a second individual processing resource estimation value that is an estimation value of processing resources required for
A first communication resource estimation value, which is a communication resource required between the first arithmetic device, estimated based on communication resources used for past communication with the first arithmetic device; And a second communication resource estimation value, which is a communication resource necessary for the second arithmetic device, estimated based on communication resources used for past communication with the second arithmetic device. Communication resource estimation means for holding
Based on the first individual processing resource estimation value, the second individual processing resource estimation value, the first communication resource estimation value, and the second communication resource estimation value, the first arithmetic device and the second A processing sharing control program that functions as processing sharing pattern generation means for determining sharing of the individual processing to two arithmetic devices.

Images