JPH11195007A - Distributed processing system/method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To standardize task quantity allocated to respective slave processors and to reduce the number of communication times in a system where a master processor and plural slave processors are connected through a network. SOLUTION: The master processor 10, the slave processor A12, the slave processor B14 and the slave processor C16 are connected through the network 100. The master processor 10 allocates tasks which are to be processed in the slave processors. A task allocation management part 10b in the master processor 10 allocates the tasks of same quantity to the respective slave processors and changes task quantity allocated to the task requests of one time in accordance with the number of task request times from the slave processors within prescribed time. To put it concretely, task quantity allocated in accordance with the task requests of one time is increased and the number of communication times is reduced as the number of task request times within prescribed time is larger.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a distributed processing system and a distributed processing method, in particular, a plurality of processors are connected via a network, a specific processor (master processor) manages all corresponding tasks, and other processors are used. The present invention relates to a system and a method for performing parallel processing of tasks by assigning tasks to (slave processors).

[0002]

2. Description of the Related Art Conventionally, in order to process a given program, a plurality of processors are connected via a network, and a master processor among the plurality of processors is configured to perform tasks (programs) with other plurality of slave processors. A system for performing parallel processing by allocating a group of independently executable processes has been proposed.

In such a distributed processing system,
In order to process tasks efficiently, it is important to distribute the load on a plurality of slave processors. When the performance of a plurality of slave processors is uniform, the load can be distributed by allocating the same amount of tasks to all the slave processors.

On the other hand, when the performances of a plurality of slave processors are not uniform with each other, or when the performances of the slave processors are uniform but each slave processor is simultaneously processing a task constituting another program, However, if tasks are equally assigned to each slave processor, there is a problem that the overall performance is suppressed by slave processors having low processing performance.

[0005] Therefore, tasks to be assigned are generated in a fine granularity (unit), a given number of tasks are first given to all slave processors, and the remaining processors are processed again for slave processors that have finished processing the given tasks. It is conceivable to assign tasks. As a result, more tasks are allocated to the slave processors with high processing performance, and the load is distributed.

Of course, if the performance of each slave processor is predetermined and the master processor can grasp the performance of each slave processor in advance,
It is possible to distribute the load by determining (statically determining) the assignment of tasks to each slave processor from the beginning, but in a multitask environment in which each slave processor processes a task constituting another different program. Then, it is difficult for the master processor to completely predict the load state of the slave processor at the time of execution.

Therefore, even if the performance of the slave processors is uniform or not uniform, but the performance of each slave processor is known in advance, the technique for dynamically allocating tasks as described above is not available. It will be effective.

[0008] Such a load distribution processing system is disclosed in, for example, Japanese Patent Laid-Open No. 7-152702.

[0009]

However, a fixed number of tasks are first given to all the slave processors, and the remaining tasks are assigned again to the slave processors which have completed the processing of the given tasks. In task assignment, communication from the master processor to the slave processor is required at the time of task assignment, and communication (task request communication) from the slave processor to the master processor is also required at the end of task processing. That is, although the amount of processing executed by the slave processors can be equalized and leveled by reducing the granularity (unit) of the task, there is a problem that the communication between the master processor and the slave processors increases with the task allocation. . In particular, since a task request is frequently transmitted from a slave processor having high processing performance to the master processor, there is a problem that communication between the slave processor having high processing performance and the master processor is significantly increased.

The present invention has been made in view of the above-mentioned problems of the related art, and has as its object to efficiently assign a task to each slave processor without unnecessarily increasing communication between a master processor and a slave processor. It is an object of the present invention to apply a distributed processing system and a distributed processing method for realizing efficient processing of a task (that is, task processing in a short time) by allocating.

[0011]

According to a first aspect of the present invention, a plurality of processors are connected via a network, and a master processor among the plurality of processors is connected to another plurality of slave processors. Wherein the master processor has an assignment management unit that changes a task amount assigned to the slave processor according to a status of the slave processor. Here, changing according to the situation means, for example, dynamically changing the task amount itself to be assigned at one time in response to a task request from a slave processor.

In a second aspect based on the first aspect, the assignment managing means changes a task amount to be assigned according to a task processing capability of the slave processor. Efficient processing of tasks while reducing the number of communication of task requests by assigning a larger amount of tasks at a time to slave processors with large processing capacity by changing the task amount according to the processing capacity (More tasks are processed by a slave processor having a large processing capability).

In a third aspect based on the second aspect, the assignment managing means evaluates the task processing capability based on a time interval of a task request from the slave processor. Evaluate the processing capacity based on the task request time interval.Specifically, the shorter the time interval is, the higher the processing capacity is, and by assigning a larger amount of tasks at a time, a separate This communication is unnecessary, and the number of times of communication can be reduced.

In a fourth aspect based on the second aspect, the assignment management means evaluates the task processing capability based on the number of task requests from the slave processor within a predetermined time. . The processing capacity is evaluated based on the number of task requests within a predetermined time. Specifically, the processing capacity is evaluated by assigning a larger amount of tasks at a time by evaluating that the processing capacity is larger as the number of task requests is larger. Therefore, it is possible to reduce the number of times of communication by eliminating the need for separate communication.

In a fifth aspect based on the first to fourth aspects, the master processor further has a task queue means for holding a task to be assigned to the slave processor, and the assignment management means comprises: The task amount to be assigned is changed according to the remaining task amount held in the task queue means.
The assigned task amount is changed in accordance with the remaining task amount. Specifically, the remaining task amount is large, and in the initial stage of time, the assigned task amount is increased to reduce the number of communication of task requests, and the remaining tasks are reduced. At a later stage in time when the amount is small, the amount of tasks to be allocated is reduced, the load of processing by each slave processor is leveled, and all tasks can be processed efficiently. Note that the present invention includes an embodiment in which the assigned task amount is temporally changed according to the processing capacity of each slave processor and the remaining task amount.

A sixth aspect of the present invention is a distributed processing system in which a plurality of processors are connected via a network, and a master processor among the plurality of processors assigns a task to be processed by another plurality of slave processors. The master processor includes a task queue unit that holds a task to be assigned to the slave processor, and an assignment management unit that changes a task amount assigned to the slave processor according to a remaining task amount held in the task queue unit. Means. In the early stage where the remaining task amount is large, the assigned task amount is increased to reduce the number of communication of task requests, and in the later stage where the remaining task amount is small, the assigned task amount is decreased and the The load processed by the slave processor is leveled, and all tasks can be processed efficiently.

A seventh invention is a distributed processing method in which a plurality of processors are connected via a network, and a master processor among the plurality of processors assigns a task to be processed by another plurality of slave processors. The task amount assigned to the slave processor is changed according to the status of the slave processor.

An eighth invention is characterized in that, in the seventh invention, the task amount to be allocated is determined according to a task processing capability of the slave processor. Efficient processing of tasks while reducing the number of communication of task requests by assigning a larger amount of tasks at a time to slave processors with large processing capacity by changing the task amount according to the processing capacity (More tasks are processed by a slave processor having a large processing capability).

In a ninth aspect based on the eighth aspect, the task processing capability is determined based on a time interval between task requests. Evaluate the processing capacity based on the task request time interval.Specifically, the shorter the time interval is, the higher the processing capacity is, and by assigning a larger amount of tasks at a time, a separate This communication is unnecessary, and the number of times of communication can be reduced.

In a tenth aspect based on the eighth aspect, the task processing capability is determined based on the number of task requests within a predetermined time. The processing capacity is evaluated based on the number of task requests within a predetermined time. Specifically, the processing capacity is evaluated by assigning a larger amount of tasks at a time by evaluating that the processing capacity is larger as the number of task requests is larger. Therefore, it is possible to reduce the number of times of communication by eliminating the need for separate communication.

According to an eleventh aspect, in the seventh to tenth aspects, the task amount to be allocated is determined according to a remaining task amount that has not been allocated yet. The assigned task amount is changed in accordance with the remaining task amount. Specifically, the remaining task amount is large, and in the initial stage of time, the assigned task amount is increased to reduce the number of communication of task requests, and the remaining tasks are reduced. At a later stage in time when the amount is small, the amount of tasks to be allocated is reduced and leveled, and all tasks can be processed efficiently. It should be noted that the present invention also includes one that changes the assigned task amount according to the processing capacity of each slave processor and the remaining task amount.

A twelfth invention is a distributed processing method in which a plurality of processors are connected via a network, and a master processor among the plurality of processors assigns a task to be processed by another plurality of slave processors. The task amount to be assigned to the slave processor
It is characterized in that it is changed according to the remaining task amount that has not been allocated yet.

[0023]

Embodiments of the present invention will be described below with reference to the drawings.

<First Embodiment> FIG. 1 is a block diagram showing the configuration of a distributed processing system according to this embodiment. The master processor 10 is connected to the slave processors A12, B14, and C16 via the network 100. The master processor 10 has a task queue 10a for holding tasks to be processed and a task assignment management unit 10b for managing task assignment to each slave processor. The task assignment management unit 10b includes the slave processors A, B, A task request from C is received, and the task amount assigned to each of the slave processors A, B, and C is dynamically changed. Note that "dynamically changing" in the present embodiment does not mean that a predetermined amount of tasks are sequentially assigned to each slave processor in response to a task request as in the related art, but that each slave processor A, B , C
Means changing the task amount to be assigned at one time in response to a single task request from. In other words, it means that the task amount assigned to a slave processor at one time at a certain time may be different from the task amount assigned to the same slave processor at another time at another time.

On the other hand, each slave processor A, B, C
Is the task A assigned by the master processor 10.
1 to A3 are processed, but not only this assigned task but also other tasks can be processed simultaneously. For example, as shown in FIG.
12 executes only the task A1 assigned by the master processor 10, while the slave processor B14 simultaneously processes the task A2 assigned by the master processor 10 and the task C constituting another program.
The slave processor C16 can simultaneously process a task B constituting another program in addition to the task A3 assigned by the master processor 10.

In such a multitask environment, as described above, even if the processing capacities of the slave processors A, B, and C are the same, the slave processors A, B, and C are allocated from the master processor 10 according to the load of the task to be processed separately. There will be differences in task processing capabilities. In the present embodiment, even when the processing capacities of the plurality of slave processors are different from each other, the master processor 10 efficiently allocates the tasks to the respective slave processors.

In the present embodiment, such a multitask environment is not always essential, and the processing capabilities of the slave processors A, B, and C are not uniform. 10
The same can be applied to the case where only the tasks from are processed.

Hereinafter, a method of dynamically determining a task allocation amount for each of the slave processors A, B, and C, which is an operation of the task allocation management section 10b of the master processor 10, will be specifically described. As a task, for example, assuming image processing, the master processor 10 divides an image to be processed into a plurality of areas, allocates image data in each area to each of the slave processors A, B, and C, and executes image processing. It shall be. The processing of the image data in each assigned area can be independently executed by each slave processor, and the entire image processing ends when all the image processing of each divided area ends. The image data amount of each area is almost the same. That is, the tasks of the present embodiment are a group of processes that can be independently executed in a program, and one task is constituted by all tasks, and the computational performance required to execute each task is substantially equal.

FIG. 2 shows that the master processor 10 includes slave processors A, B, and C (hereinafter, slave A,
B and C) and task requests transmitted from each slave processor to the master processor are shown in chronological order. In the figure, (A) shows the processing from time 0 to time 100, and (B) shows the processing at time 1
The processing is from 00 to time 200.

For convenience of explanation, the performance ratio of each slave processor is slave A: B: C = 15: 5: 3 (when there is a difference in performance itself, and when the performance itself is uniform, other tasks are performed. And the case where there is a difference in the processing capability of the assigned tasks), and the total number of tasks to be assigned by the master processor 10 is 10
0, initial value of assigned task amount is 4 (4 tasks)
And Then, the master processor 10 counts the number of task requests from each slave processor, and when the number of task requests of all slave processors becomes 1 or more, that is, the first task request from the slowest slave processor. When there is, a task amount to be assigned is reviewed according to the number of task requests transmitted from each slave processor up to that point, and a new task amount is determined. In deciding the task amount to be assigned, the master processor performs assignment based on the following criteria.

That is, when the task request counts (communication counts) of the slaves A, B, and C are respectively a, b, and c. (1) When a> b = c, the allocation to the slave A is increased to a / b. , The assignments for slave B and slave C are maintained.

(2) When a>b> c The assignment to the slave A is increased to a / b, the assignment to the slave C is decreased to c / b, and the assignment to the slave B is maintained.

(3) When a = b> c The assignment to the slave C is reduced to c / b, and the assignment to the slaves A and B is maintained.

(4) When a = b = c The assignment to the slave A, slave B, and slave C is maintained.

The above criterion is applied even if A, B and C are exchanged. Hereinafter, the state of task assignment using these criteria will be described with reference to FIG.

At time t = 0, the master processor 1
From 0, four tasks are assigned to each slave processor as initial values. In the figure, the length of the assigned task differs depending on the slave, but this means that the processing capacity of each slave is different, so that it takes time to process the task by that length. I have. One task amount assigned to each slave is almost the same as described above. Slave A,
Since the processing capabilities of B and C are 15: 5: 3, at the time t1 (t1 = approximately 47) when the task request from the slave C, which has the slowest processing, is made to the master processor 10, the slave A and the slave B are processed. Has already made a task request to the master processor 10, and as shown in FIG.
, A total of three times, and the slave B makes a total of one task request. That is, a = 3, b = 1, and c = 1.

Therefore, the task assignment management unit 10b of the master processor 10 assigns a task to be assigned next according to the number of task requests from each slave processor in a predetermined time (time until there is a request from the slowest slave processor). Determine the amount. Specifically, since the criterion (1) among the above-mentioned criterion is satisfied, the allocation to the slave A is increased by 3/1 = 3 times, and the allocation to the slave B and the slave C is set to four initial values. Leave it alone. Therefore, after the time t1, when a task request from the slave A is transmitted, 4 × 3 = 12 tasks are assigned to the slave A, and the task requests from the slaves B and C are The current four tasks will be assigned. After determining the task amount, the task allocation management unit 10b resets the task request counter to 0 in preparation for the next count.

After the task amount to be allocated is determined as described above, it is assumed that the task request from the latest slave C is transmitted to the master processor at time t2 (t2 = about 92). The task allocation management unit 10b of the master processor 10 counts the number of task requests from each slave transmitted from time t1 to time t2. Time t
During the predetermined time from 1 to t2, as shown in the figure, the task request is made twice from the slave A, twice from the slave B, and once from the slave C.
2, b = 2 and c = 1. Task assignment management unit 10
b determines the amount of tasks to be allocated to each slave again according to the number of task requests. Specifically, since the criterion (3) among the above criterions applies, the allocation to the slave C is reduced to 1/2 = 0.5 times, and the allocation amounts of the slaves A and B are maintained as they are. I do. Therefore, when a task request is transmitted from each slave processor after time t2, the task allocation amounts for slave A, slave B, and slave C are 12, 4, respectively.
4 × 0.5 = 2.

Similarly, when the task request from the slowest slave processor is received again, the assigned task amount is reviewed. In this case, the slowest slave processor is slave B, and when a task request from slave B is transmitted at time t3 (t3 = about 120),
The task assignment management unit 10b reviews the task amount to be assigned again and determines a new assigned amount. Within a predetermined time from time t2 to time t3, slave A, slave B,
Since all task requests from slave C are made once,
Since a = 1, b = 1, and c = 1, the criterion (4) among the above-mentioned criterion applies, and the assignment to each slave remains as it is.

Then, with the receipt of the task request from the slowest slave processor again as a trigger, the assigned task amount is reviewed. In this case as well, the slowest slave processor is slave B, and when a task request from slave B is made at time t4 (t4 = about 150), each slave processor within a predetermined time from time t3 to t4. The allocation amount is determined according to the number of task requests from. Since the number of task requests from each slave is all one as shown in the figure, a = 1, b = 1, c = 1, and
The criterion (4) among the above-mentioned criteria is applicable, and the assigned task amount remains as it is. As described above, the assignment of the total number of tasks 100 is completed, and the processing of all tasks is completed at time t5 (t5 = about 176).

As described above, in the present embodiment, the processing performance of each slave processor is evaluated based on the number of task requests from each slave processor within a predetermined time, that is, the time from when the slowest slave issues a task request. And the number of task requests is large (the processing capacity is large)
Since the amount of tasks assigned to a slave processor in response to a single task request is increased, tasks can be efficiently assigned to each slave processor while reducing the number of communications between the slave processor and the master processor.

That is, in the present embodiment, the processing capacity of the slave processor is not grasped by the master processor using a separate communication but is grasped by the number of task requests, so that the communication does not increase unnecessarily. further,
Since a large amount of tasks is assigned to a slave processor having a large processing capacity, task requests are not frequently transmitted from the slave processor to the master processor.

The applicant has set the total number of tasks to 100,
3 for communication between master processor and slave processor
(Time unit) Assuming that slave A, slave B, and slave C require 1 (hour unit), 3 (hour unit), and 5 (hour unit) to process one task, the task amount to be allocated is always constant. A computer simulation was performed for the case where the time was changed as in this embodiment and the case where the time was changed as in the present embodiment, and the following results were obtained for the processing time T of all tasks.

<When the assigned task amount is always one> T = 372 (time unit) <When the assigned task amount is always four> T = 210 (hour unit) <Allocated task amount is always 16 T = 204 (in units of time) <Case of the present embodiment> T = 176 (in units of time) As is clear from the simulation results, in the present embodiment, the task amount is optimally assigned to each slave processor. Since communication between the master processor and the slave processor is also suppressed, all tasks can be processed efficiently and quickly.

In this embodiment, the task processing capability of each slave processor is evaluated based on the number of task requests from each slave processor within a predetermined time, but each slave processor is evaluated based on the task request time interval of each slave processor. It is also possible to evaluate the task processing capability of the processor. That is, the master processor 10
Assigns four tasks to each slave processor as an initial value, measures these four tasks, and measures a time interval until a task request is transmitted from each slave processor. Then, the processing capacity is evaluated based on the difference in the time interval, and a large amount of tasks may be allocated to the slave processor having a large processing capacity at once from the next time. It will be clear to those skilled in the art that the number of task requests within a given time period and the time interval between task requests are physically equivalent.

<Second Embodiment> In the first embodiment described above, the task assignment management unit 10 of the master processor 10
b shows the case where the task amount assigned to a task request at one time is changed in accordance with the task processing capacity of each slave processor. In the present embodiment, each task amount is changed according to the remaining task amount not yet assigned. A case where the task amount assigned to the slave processor is dynamically changed will be described.

The configuration block diagram of the present embodiment is the same as the configuration block diagram shown in FIG. 1, and the task allocation management unit 10b in the master processor 10 assigns a task to be allocated to each of the slave processors A, B, and C. The amount is determined, and the task held in the task queue 10a is supplied to each slave processor. Task assignment management unit 1
When determining a task to be assigned to each of the slave processors A, B and C, 0b monitors the remaining task amount held in the task queue 10a and determines the task amount to be assigned according to the remaining task amount. Specifically, when the remaining task amount is large, the task amount assigned to each slave processor is increased, and as the remaining task amount decreases,
The amount of tasks to be allocated to each slave processor is leveled by reducing the granularity (unit) of the tasks to be allocated.

FIG. 3 shows the change over time of the task amount assigned to each slave processor. In the figure, the horizontal axis represents the task ratio that has not been assigned, that is, the ratio of the remaining task amount to the total task amount (remaining task amount / total task amount), and the vertical axis represents the task request to each slave processor once. This is the task amount to be assigned to the task. Note that since the remaining task amount decreases with time, the horizontal axis also functions as a time axis (that is, FIG. 3 shows a time change of the assigned task amount). In the figure, the task amount to be assigned represents the minimum assignment unit as n.

As shown in the figure, until the remaining task amount / the total task amount (remaining task ratio) becomes 1/4, the task allocation management unit 10b of the master processor 10 uniformly assigns 8n tasks to each slave processor. Assign. And if the task remaining rate is less than 1/4,
The task amount assigned to each slave processor for one task request is reduced from 8n to 4n. Further, when the task remaining rate becomes 1/8 or less, the task amount to be allocated is reduced from 4n to 2n, and the task remaining rate becomes 1/1.
When the number of tasks becomes 6 or less, the granularity of the task amount to be allocated is further reduced to 2n, which is the minimum unit.

As described above, when the amount of tasks not yet allocated and held in the task queue is large (that is, in the case of initial time), the amount of tasks allocated at a time is increased to increase the The task amount to be assigned to each slave processor by reducing the number of task requests from the slave processor (particularly, a slave processor having a large processing capacity) and reducing the granularity of the task amount to be assigned as the remaining task amount decreases (that is, when time elapses). And all tasks can be processed efficiently.

In this embodiment, a computer simulation is performed under the same conditions as those in the first embodiment.
The time T until the processing of all tasks is completed is calculated. As a result, the following result was obtained in the case of the present embodiment: T = 192 (time unit).

Therefore, as in the present embodiment, when the task amount to be assigned is temporally changed according to the remaining task amount held in the task queue 10a, the task can be uniformly assigned to each slave processor. compared,
The number of communications can be reduced, and all tasks can be processed more efficiently.

<Third Embodiment> In the above-described second embodiment, the task amount assigned to each slave processor is temporally changed according to the remaining task amount that has not been assigned. In the following, a distributed processing method that combines the above-described distributed processing method of the first embodiment and the distributed processing method of the second embodiment will be described.

The configuration block diagram of this embodiment is also the same as the configuration block diagram shown in FIG. 1, and the task allocation management unit 10b in the master processor 10 evaluates the task processing capability of each slave processor, and executes the task processing. The amount of tasks to be assigned at one time is temporally changed according to the capability, and the amount of tasks to be assigned is also changed according to the amount of remaining tasks held in the task queue 10a.

FIG. 4 shows the amount of tasks allocated from the master processor 10 to each slave processor and the task requests from each slave processor in chronological order. (A) is the assigned task amount from time t = 0 to 100, and (B) is the assigned task amount from time t = 100 to 200.

In this embodiment, as in the first embodiment, the performance ratio of each slave processor is the slave A:
B: C = 15: 5: 3 (if there is a difference in performance itself,
And the performance itself is uniform, but there is a difference in the processing capability of the assigned task because it is processing another task), and the total number of tasks to be assigned by the master processor 10 is 100. .

At time t = 0, master processor 1
The task allocation management unit 10b in 0 sets the initial value of the task amount allocated to each slave processor to 8, and allocates eight tasks to each slave processor (slaves A, B, and C). Then, the number of task requests from each slave processor within a predetermined time until time t1 (t1 = about 86) at which the task request from the latest slave C is made is counted. In this case, a task request is made three times in total from the slave A as shown in FIG.
, The task request is made twice in total, and the slave C makes one task request. Therefore, a = 3,
Since b = 2 and c = 1, the task assignment management unit 10
b is (2) out of the four criteria in the first embodiment.
According to the criterion, the assignment to the slave A is 3/2 =
Increase by 1.5 times, and assign the
/2=0.5 times and keep the assignment to slave B as it is. Therefore, for a task request after time t1, 8 × 1.
5 = 12, 8 for slave B, and 8 × 0.5 = 4 for slave C.

After determining the assigned task amount, in the first embodiment described above, the assigned task amount is reviewed when the task request from the slowest slave processor is made again. The assignment management unit 10b not only monitors task requests, but also monitors the remaining task amount held in the task queue 10a. Then, when the remaining task rate is 1/4, that is, when the remaining task amount is 100 × １ ／ = 25 or less, the reference value 8 of the assigned task is reduced to 4 (fine granularity), The task amount assigned to each slave is reviewed with respect to the reference value 4, and a new task amount is determined.

More specifically, the assigned task amount determined last time is 1.5 times the reference value for the slave A, the reference value is unchanged for the slave B, and the reference value is 0 for the slave C. 0.5 times, the time t2 (t2 = about 1
08), if the remaining task amount becomes 25, the reference value is reduced from 8 to 4, and 4 ×
1.5 = 6 tasks are allocated, 4 tasks are allocated to slave B, and 4 ×
0.5 = 2 tasks are assigned.

When the task request is transmitted from the slowest slave processor (slave C in this case) at time t3, the task allocation management unit 10b sets the time from time t1 to t3 as in the first embodiment. The task amount to be assigned is determined according to the number of task requests within a predetermined time up to the specified time. Within a predetermined time from time t1 to t3, there are a total of two task requests from slave A, a total of one task from slave B, and a total of one task request from slave C. Therefore, a = 2, b =
1, c = 1, and 6 × 2 for the slave A for the task request after the time t3 according to the criterion (1).
/ 1 = 12 tasks are assigned, 4 tasks are assigned to slave B, and 2 tasks are assigned to slave C. As described above, the assigned task amount is changed according to the processing capacity and the remaining task amount of each slave processor, and all the tasks are processed by the slave processors A, B, and C.

Also in the present embodiment, the present applicant has calculated the time T required to process all tasks using computer simulation. <In the case of the present embodiment> T = 166 (time unit) I got it.

As described above, by dynamically changing the task amount to be allocated over time based on the processing capacity of each slave processor and the remaining task amount, compared to a case where tasks are uniformly assigned to each slave processor, All tasks can be processed more efficiently and quickly.

The case where image data is processed by a plurality of processors has been described in the embodiment of the present invention. However, it goes without saying that the present invention is not limited to this and can be used for arbitrary data processing. Also, the number of slave processors is not limited to three, and can be applied to a system having four or more slave processors.

Further, the number of master processors is not limited to one, but may be two or more in some cases. Further, the present invention can be applied to a case where the master processor also has a slave processor function and processes a task. Furthermore, the present invention can be applied to a case where the slave processor appropriately takes the form of functioning as a master processor. In this case, the slave processor and the master processor are replaced as appropriate.

[0065]

According to the present invention, the amount of tasks to be allocated to a plurality of slave processors is dynamically changed over time, so that even if there is a difference in processing capacity among the plurality of slave processors, task allocation can be performed efficiently. By doing so, the time required for all task processing can be reduced.

Further, tasks can be efficiently allocated without increasing the communication between the master processor and the slave processors.

[Brief description of the drawings]

FIG. 1 is a configuration block diagram of an embodiment of the present invention.

FIG. 2 is an explanatory diagram showing a time change of an assigned task amount according to the first embodiment of this invention.

FIG. 3 is a graph showing a time change of an assigned task amount according to a second embodiment of the present invention.

FIG. 4 is an explanatory diagram showing a time change of an assigned task amount according to a third embodiment of the present invention.

[Explanation of symbols]

10 Master processor, 10a Task queue, 10
b Task allocation management unit, 12 slave processors A (slave A), 14 slave processors B (slave B), 16 slave processors C (slave C), 100 networks.

Claims (12)

[Claims] 1. A distributed processing system in which a plurality of processors are connected via a network, and a master processor among the plurality of processors assigns a task to be processed by another plurality of slave processors. And a allocating means for changing a task amount allocated to the slave processor in accordance with a status of the slave processor. 2. The distributed processing system according to claim 1, wherein said allocation management means changes a task amount to be allocated according to a task processing capability of said slave processor. 3. The distributed processing system according to claim 2, wherein said assignment management unit evaluates said task processing capability based on a time interval of a task request from said slave processor. 4. The distributed processing system according to claim 2, wherein said assignment management means evaluates said task processing capability based on the number of task requests within a predetermined time from said slave processor. 5. The master processor further includes task queue means for holding a task to be assigned to the slave processor, and the assignment management means stores the task amount to be assigned to the remaining task amount held in the task queue means. The distributed processing system according to any one of claims 1 to 4, wherein the value is changed according to a task amount. 6. A distributed processing system in which a plurality of processors are connected via a network, and a master processor among the plurality of processors assigns a task to be processed by another plurality of slave processors, wherein the master processor comprises: A task queue unit that holds tasks to be assigned to the slave processors; and an assignment management unit that changes a task amount to be assigned to the slave processors according to a remaining task amount held in the task queue unit. A distributed processing system characterized by the following. 7. A distributed processing method in which a plurality of processors are connected via a network, and a master processor among the plurality of processors assigns a task to be processed by another plurality of slave processors. A distributed processing method, wherein the task amount to be allocated is changed according to the status of the slave processor. 8. The distributed processing method according to claim 7, wherein the task amount to be allocated is determined according to a task processing capability of the slave processor. 9. The system according to claim 8, wherein the task processing capability is determined based on a time interval between task requests.
The distributed processing method as described above. 10. The distributed processing method according to claim 8, wherein the task processing capacity is determined based on the number of task requests within a predetermined time. 11. The distributed processing method according to claim 7, wherein the task amount to be assigned is determined according to a remaining task amount that has not been assigned yet. 12. A distributed processing method in which a plurality of processors are connected via a network, and a master processor among the plurality of processors assigns a task to be processed by another plurality of slave processors. A distributed processing method characterized by changing an assigned task amount according to a remaining task amount that has not been assigned yet.