JPH07253951A - Distributed processing system - Google Patents

Abstract

PURPOSE:To provide the distributed processing system which is superior in reliability, maintainability, adaptivity and processing by removing problem points of conventional technology. CONSTITUTION:The distributed processing system which multiplies and executes plural tasks in task units is equipped with a multiplex level determining means 31 which determines the multiplex level of tasks in operation in a computer corresponding to resource information on the distributed processing system and broadcasts a multiplex level variation message showing the multiplex level, a multiplex level management means 32 which determines a method for implementing the determined multiplex level of tasks on the basis of a computer state message and generates a processing method message, and a task copying and deleting means 33 which receives the processing method message and realizes the determined implementation.

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 in which a plurality of computers are connected by a communication path.

[0002]

2. Description of the Related Art In a distributed processing system, when a plurality of tasks are multiplexed and performed simultaneously, a computer-based multiplexing method has been well known. There are the following two methods for this computer-based multiplexing method.

(1) A parallel multiplex system for each computer in which computers with the same task configuration are multiplexed and executed simultaneously.

(2) A standby redundancy system on a computer-by-computer basis in which only one of the multiplexed computers is operated and the computer is switched to the standby computer when a failure occurs.

In these computer-based multiplexing methods, the number of tasks in which a certain task is multiplexed and executed in parallel in the distributed processing system (hereinafter referred to as task multiplicity).
When 2 is set to 2, the number of computers required is twice as large as that when the multiplexing process is not performed. Similarly, if the multiplicity of tasks is set to 3, a computer which is three times as large as that in the case where no multiplexing processing is performed is required.

This will be specifically described. For example, in FIG.
In the schematic diagram of, four computers PU1, PU2, PU
3 and PU4 are shown as squares. Further, a state in which five types of tasks TA, TB, TC, TD, and TE are assigned to a computer is shown by a circle in which a task name is arranged in a rectangle representing the computer. Specifically, the task TA causes the computers PU1 and PU2 to send the task TB
To the computers PU1 and PU2, and the task TC to the computer PU1
, PU2, task TD is assigned to computers PU3 and PU4, and task TE is assigned to computers PU3 and PU4. That is, the computers PU1 and PU2
Has the same task configuration, the computers PU3 and PU4 have the same task configuration, and the multiplicity of tasks is 2.

In such a computer-based multiplexing system, all computers operate in the parallel multiplexing system.
On the other hand, in the standby redundancy method, for example, the computer PU1
And PU3 operate, and the computers PU2 and PU4 stand by. If it is desired to set the multiplicity of the task TA to 3 in these methods, (1) add another computer to the computers PU1 and PU
Use the same task configuration as 2.

(2) Assign task TA to computers PU3 and PU4
Add to and make it work.

(3) Add another computer and re-arrange the tasks.

There are methods such as

In the method (1), the cost is increased by adding a computer, and the multiplicity of the tasks TB and TC is reduced to 3, resulting in useless calculation.
In the method of (2), useless calculation occurs due to an increase in calculation load on the computers PU3 and PU4 and a task TA having a multiplicity of 4. The method (3) increases the equipment cost and is labor intensive.

In recent years, in consideration of the above-mentioned drawbacks of the computer-based multiplexing system, a task-based parallel multiplexing system has been proposed in which any task is multiplexed on a plurality of computers for parallel execution in order to improve reliability. Was done. In fact, if tasks are multiplexed, even if a computer in the distributed processing system fails, the task running on another normal computer can be continuously executed, so that the reliability is improved.

That is, in the task-based multiplexing method, the multiplicity of tasks can be changed on a task-by-task basis in accordance with the importance of the task. Therefore, the number of computers in the distributed processing system is simply doubled or tripled. It is not necessary to do so, and it is excellent in that it has the effect of suppressing the equipment cost. This will be specifically described. For example, in the schematic diagram of FIG. 28, four computers PU1, PU2, PU3, PU4 are each shown by a quadrangle. Furthermore, five types of tasks TA, TB, T
A state in which C, TD, and TE are assigned to a computer is shown by a circle in which a task name is placed in a rectangle representing the computer. Specifically, the task TA is the computers PU1 and PU2, and the task TB is the computers PU1 and PU2.
3, the task TC is assigned to the computers PU1, PU2 and PU3,
The task TD is assigned to the computers PU2 and PU3, and the task TE is assigned to the computer PU4 for execution.
In this case, the multiplicity of tasks TA, TB, and TD is 2, the multiplicity of task TC is 3, and the multiplicity of task TE is 1.

Thus, in the task-based multiplexing system,
The multiplicity of tasks can be freely specified for each task, and these multiplexed tasks are executed in parallel.
Here, when passing a message between tasks to another task, it is transmitted by a broadcast communication system (hereinafter referred to as broadcasting). Further, when a task receives a message, the first received message is adopted, or when a specified number of messages having the same content arrive, the task is adopted. Therefore, when executing a task, it is not necessary to know the computer to which the task is assigned or the multiplicity of the task. For example, when the task TA passes a message to TB, this message is broadcast from the computers PU1 and PU2 executing the task TA. This message is received by all the computers, but the computers actually used are the computers PU1 and PU3 executing the task TB. The task TB running on the computer PU1 receives two messages of the same content from the tasks TA of the computers PU1 and PU2, but one message is adopted by the above method. Further, the task TB being executed by the computer PU3 also adopts one message in the same manner.

In this method, the multiplicity of task TA is set to 3
If the task TA is to be set, the task TA may be operated by a computer having a low calculation load among the computers PU3, PU4 not operating. Therefore, if there is a margin in the calculation capacity of the entire distributed processing system, it is not necessary to add a computer, and the equipment cost can be suppressed.

[0016]

In the above-described task-based multiplexing system, the calculation load in the system can always change depending on the number of active tasks and the degree of calculation. Further, even when the task is being executed, the reliability required for the task may change depending on the calculation content. That is, the multiplicity required for a task may change depending on the calculation content.

However, in the conventional task-based multiplexing method, the multiplicity of tasks is fixed after the task is activated, and cannot be changed except by the user's designation. Therefore, when the computational load is high, the multiplicity of tasks can be reduced within the allowable range to reduce the computational load and improve the processing efficiency of the entire system. Since the multiplicity of is fixed, the processing efficiency of the entire system was reduced.

Furthermore, in the conventional task-based multiplexing method, it is necessary to determine the method of allocating tasks to computers before the task is activated, and the system status at task activation is reflected in the method of task allocation. Can not do it.
Moreover, in order to solve these problems, manual operation was required.

Further, when a computer fails, the multiplicity of tasks executed on the computer decreases and reliability decreases, so it is necessary to maintain the multiplicity of tasks when the computer fails. In addition, when a computer with a low computational load is added, such as when a failed computer is restored or when a new computer is added, tasks on the computer with a high computational load should be performed in order to utilize the capacity of the distributed processing system. It is necessary to move to the added computer to balance the load.

As described above, the conventional distributed processing system based on the task-based multiplexing method has the following problems.

(1) It is necessary to determine the multiplicity of tasks and the method of allocating the tasks to computers before starting the tasks. The status of the system in which the tasks are operating is determined by the multiplicity of tasks, and It cannot be reflected in the method of allocating tasks to computers, and the processing efficiency of the system cannot be made more desirable (the problem of adaptability).

(2) When a computer fails or is added, the task multiplicity must be changed or assigned to the computer manually, and the user needs to be aware of the task multiplicity and the computer to be assigned ( Maintainability problem).

(3) The task allocation means is installed in one or a plurality of predetermined service computers, and when those service computers fail, task allocation is not performed. Also, when a computer fails, the multiplicity of tasks is reduced. Furthermore, the computational load of task assignment processing is biased towards some computers (reliability,
Processability problem).

The distributed processing system of the present invention has been made in view of such problems, and an object thereof is to provide a distributed processing system excellent in reliability, maintainability, adaptability, and processability. To do.

[0025]

In order to achieve the above object, the present invention provides a distributed processing in which a plurality of tasks are multiplexed and executed in task units by a plurality of computers connected via a communication path. In the system, each of the plurality of computers determines the multiplicity of a task operating in the computer corresponding to the resource information of the distributed processing system, and broadcasts a multiplicity change message indicating this multiplicity. Determining means and a method for realizing the degree of multiplicity of the task decided by the multiplicity determining means, based on a computer status message, and a multiplicity managing means for generating a processing method message; and this multiplicity managing means And a task processing unit including a task copying / deleting unit that implements the determined implementation method.

Further, according to the present invention, in a distributed processing system in which a plurality of tasks are multiplexed in a task unit by a plurality of computers connected via a communication path and executed, each of the plurality of computers operates in its own computer. A task multiplicity decrease detection means for periodically broadcasting an operation status message including a task name list of active tasks, and periodically checking the multiplicity of running tasks to detect a decrease in the multiplicity of this task. When the decrease in the multiplicity is detected by the task multiplicity decrease detecting means, the computer which is the transfer source of the internal state of the task and the computer which is the transfer destination are determined, and the internal state of the task is transferred. Transfer source / destination determination means for determining the method, and the computer state when the transfer source / destination determination means determines the method of transferring the internal state of the task A calculation device status management means, the transfer source and
And a task processing device including a task internal state copying device for transferring the internal state of the task according to the transfer method determined by the transfer destination determining device.

[0027]

EXAMPLES Next, the distributed processing system of the present invention will be described.
The preferred embodiments will be specifically described with reference to the accompanying drawings. Here, the distributed processing system of this embodiment uses a task-based multiplexing method, and its basic operation is as described above.

First, the configuration of a distributed processing system according to a first embodiment of the present invention will be described in detail with reference to FIGS.

In the distributed processing system of the first embodiment, a plurality of computers 1 are connected by a communication path 2 as shown in the block diagram of FIG. The computer 1 includes a task processing device 3 that executes a task, a storage device 4 that stores data handled by the task processing device 3, and a communication device 5 that processes message communication between the computers. Is communication device 5
Are connected through.

FIG. 2 is a diagram showing the configuration of the task processing device 3 shown in FIG. 1, which comprises a multiplicity determining means 31, a multiplicity managing means 32, and a task copying / deleting means 33. The multiplicity determining means 31 determines the multiplicity of a task operating in the task processing device 3 in accordance with the resource information of the distributed processing device, and outputs a multiplicity change message indicating the multiplicity in the distributed processing system. It is broadcast to the multiplicity determination means 31 and the multiplicity management means 32.

When the multiplicity managing means 32 receives the multiplicity changing message, the multiplicity managing means 32 in the distributed processing system sends a computer status message indicating the status of the computer 1.
While broadcasting to the destination, the computer status message in the distributed processing system is received and stored in the storage device 4. Then, a method for realizing the multiplicity of the task is determined by using these data, specifically, a method for copying and deleting the task is determined, and a processing method message stating the method is copied to the task copy in the computer. It transmits to the deletion means 33.

Upon receipt of the processing method message, the task copying / deleting means 33 realizes the determined copying / deletion method. Here, copying a task means transferring the internal state of a task, which is composed of the values of variables used by the task and the address indicating the position of the program, to the computer that newly operates the task.

Data M1 indicating that the message when the task processing devices 3 of the respective computers perform message communication with each other is one of the above-mentioned three means 31, 32, 33, for example, the multiplicity determining means 31. And the data showing the computer name and the message body. When broadcasting a message, the data indicating the broadcast mode is used instead of the data indicating the computer name. For example,
When the computer PU1 sends a message to the multiplicity determination means 31 of PU2, the content is (M1, PU2, message body). When the computer PU1 broadcasts a message to the multiplicity determining means 31 of all other computers in the distributed processing system, the content is (M1, broadcast, message body).

The processing operation procedure of the above-mentioned multiplicity determining means 31, multiplicity managing means 32, and task copying / deleting means 33 will be described in detail below.

FIG. 3 is a diagram showing an outline of a data flow for explaining the operation of the embodiment. The arrows in the figure show the flow of data transmitted between computers. Further, the required multiplicity table 41 and the computer status table 42 shown in FIG.
Is a table stored in the storage device 4 in the computer 1 shown in FIG. 1, and the configuration of the request multiplicity table 41 is shown in FIG.
The configuration of the computer status table 42 is shown in FIG.

First, the operation when a task is newly started and a task is copied to realize the multiplicity required by the task will be described. Here, it is assumed that the task is operating in the state of FIG. 28 and the task TE has just started in the computer PU4.

The following assumptions are made about the tasks that need to be multiplexed.

(1) At the time of system design, the system designer determines the upper limit (hereinafter referred to as the maximum multiplicity) and the lower limit (hereinafter referred to as the maximum multiplicity) of the task multiplicity according to the degree of task reliability required and the size of the system. , Called the minimum multiplicity).

(2) A request multiplicity message 51 (FIG. 3) describing the maximum multiplicity and the minimum multiplicity of a task is broadcast to the multiplicity determining means 31 (FIG. 3) in the distributed processing system when the task is activated. To describe the task. The requested multiplicity message 51 includes the name of the invoking task, the name of the computer on which the task is operating, the minimum multiplicity, and the maximum multiplicity. Note that if the computer name on which the task is operating is fixed and described in the task, maintainability will deteriorate, so the computer name on which the task is operating will be acquired at task startup through the functions that a distributed processing system normally has. And use it.

For example, the minimum multiplicity and the maximum multiplicity of the task TE are designed to be 2 and 3, respectively, and the computer P
The message body of the request multiplicity message 51 when activated in U4 is (TE, PU4,2,3). Upon receiving the request multiplicity message 51 (step 112 in FIG. 6), the multiplicity determining means 31 (FIG. 3) updates the request multiplicity table 41 as shown in FIG. 4 (step 113). Request multiplicity table 4
As shown in FIG. 4, each item 1 includes a task name described in the message, minimum multiplicity, and maximum multiplicity. If there is an item related to the task name described in the request multiplicity message 51, it is updated with the data described in the message. . If there is no item related to the task name, add it to the table. Next, if the computer name described in the message matches the own computer name, a multiplicity change message 52 (FIG. 3) is created, and the multiplicity determination means 31 and the multiplicity management means 32 in the distributed processing system are created.
Broadcast to the destination (steps 114, 115, 11 in FIG. 6).
6).

The multiplicity change message 52 includes a task name and multiplicity. This multiplicity takes a value between the minimum multiplicity and the maximum multiplicity described in the request multiplicity message 51. Here, assuming that the calculation load of the entire system is normal at the time of task activation, the maximum multiplicity is adopted as the multiplicity of the relevant task. Therefore, for example, the message body of the multiplicity change message 52 for the task TE is (TE, 3).

Upon receiving the multiplicity change message 52 from outside the computer, the multiplicity determining means 31 (step 11
1) It waits until it receives a processing end message 53 (FIG. 3) indicating that the current multiplicity changing processing has ended (step 117 in FIG. 6).

The multiplicity management means 32 (FIG. 3) receives the multiplicity change message 52 (step 2 in FIG. 7).
01), creates a computer status message 54 (FIG. 3) and broadcasts it to the multiplicity management means 32 in the distributed processing system (step 202 in FIG. 7). The computer status message 54 includes a computer name, a calculation load of the computer, and a task name list of tasks operating in the computer. For example, the message body of the computer status message 54 created by the multiplicity management means 32 in the computer PU1 is (PU1, 0.65, TA, TB, TC).

Upon receiving the computer status message 54, the multiplicity management means 32 updates the computer status table 42 as shown in FIG. 5 (step 203). As shown in FIG. 5, each item of the computer status table 42 includes a reception time of the computer status message 54, a computer name described in the message, a calculation load, and a task name list. here,
Check the reception time column and delete the items that have passed a certain time. This is because the computer status message 54 is received from all the computers in the distributed processing system, and it is considered that the computers described in the items after a certain time has stopped due to some reason.

Next, using the updated computer status table 42, the current multiplicity Od of the relevant task is calculated (step 204), and the magnitude is compared with the multiplicity Nd of the task after the change described in the message ( Step 205). Since the action when the task TE is started is considered here, the current multiplicity Od is 1. Since the multiplicity Nd of the task after the message description is changed is 3, Nd is larger. Therefore, the multiplicity management unit 32 performs a process of copying the relevant task to increase the multiplicity from 1 to 3, and this procedure will be described below with reference to the flowchart of FIG.

First, the multiplicity management means 32 of each computer
Check whether or not there is a corresponding task in the local computer (Fig. 8
Step 206). If there is a corresponding task, it becomes a candidate of a transfer source when copying the internal state of the task, and suspends the operation of the corresponding task and waits until a processing end message broadcast at the end of the transfer processing is received (step 2).
14). If there is no applicable task, it becomes a candidate for accepting the applicable task. Among these candidates, a computer with a low calculation load is preferable as a computer that accepts tasks from the viewpoint of load distribution. Therefore, a list is created from the computer status table 42 in which the computers that do not include the relevant task are arranged in order of increasing computational load (step 207). For example, in the computer status table 42, the computers that do not include the task TE are PU1, PU2, PU3, and when they are arranged in the ascending order of calculation load, (PU1, PU2, PU3) is obtained. Next, Nd-Od candidate computers are left from the head of the list and the others are removed (step 208).
For example, Nd from the beginning of the list for task TE
If only -Od = 3-1 = 2 computers are left, the list becomes (PU1, PU2). That is, the tasks TE are assigned to the computers PU1 and PU2.
Will be copied.

The above results are obtained in the same manner on all computers that do not include the corresponding task. This is because the computer status table 42 having the same contents, which has been updated using broadcast communication, is used. Therefore, even if the individual computer determines the copy processing method based on this result, no contradiction occurs. Here, the determination of the copy processing method is to determine the transfer source and the transfer destination when transferring the internal state of the task. Here, the copy destination computer determines the copy processing method. Therefore, if the computer name is in the list (step 209), the copy source computer determines the transfer source of the task internal state individually. If there is no own computer name, it waits until it receives a processing end message broadcast at the end of the transfer processing.

Next, the procedure for determining the copy source computer will be described. First, the computer state table 42 is used to find the number T of computers on which the task is operating. Next, the order S of the own computer when the top of the list is the first is calculated. Then, the remainder R when S is divided by T is calculated. At this time, when the computers on which the relevant tasks are operating are arranged in order from the one with the lowest calculation load, the R + 1th computer is set as the transfer source computer. For example, in the case of the computer PU1, the number T of computers on which the task TE is operating is 1 and the order S of its own computer in the list is 1. Therefore, the remainder R is 0, and when the computers on which the task TE is operating are arranged in order from the one with the lowest calculation load, R + 1 = 0 +
The 1st first computer PU4 is the transfer source computer.
Similarly for the computer PU2, T = 1, S = 2, R = 0
Therefore, PU4 is used as the transfer source computer.

Therefore, the processing method message 55 (FIG. 3) is created based on the above result and transmitted to the task copying / deleting means 33 (FIG. 3) in the own computer (step 210 in FIG. 8). Whether the processing method message 55 is a process of copying the corresponding task name, transfer source computer name, or task,
It is made up of data indicating whether a task can be deleted (hereinafter, the former is called a copy flag and the latter is called a delete flag). For example, the message body of the processing method message 55 created by the multiplicity management means 32 of the computer PU1 for the task TE is (TE, PU4, copy).

The task copying / deleting means 33 which has received the processing method message 55 (see step 30 in FIG. 10).
1) Check the processing method message and confirm that there is a copy flag in this case (step 302). If there is a copy flag, transfer request message 56 (FIG. 3)
Is created and sent to the task copy / deletion means 33 of the transfer source computer described in the message (step 30 in FIG. 10).
3). The transfer request message 56 includes a task name and a computer name. For example, for task TE, computer PU
The message body of the transfer request message 56 created by the task copy / deletion unit 33 of No. 1 is (TE, PU1), and the destination is the task copy / deletion unit 3 of the computer PU4.
It is 3.

Upon receiving the transfer request message 56, the task copying / deleting means 33 (step 307) transfers the transfer data 57 including the internal state of the task described in the message.
(FIG. 3) is transmitted to the task copy / deletion means 33 of the computer described in the message (step 308 in FIG. 10). The transfer data 57 has a structure such as (“DATA”, data body). Here, "DATA" is used to identify the message.

Upon receiving the transfer data 57, the task copying / deleting means 33 (step 304) executes the task copying process and, after the completion, creates a process end message 53 (FIG. 3) to perform the distributed process. Broadcast to the multiplicity determination means 31 and the multiplicity management means 32 in each computer of the system (step 305 in FIG. 10). Processing end message 5
Reference numeral 3 includes a self-computer name, a calculation load of the self-computer, and a task name list of tasks running on the self-computer. For example, the message body of the processing end message 53 created by the task copying / deleting means 33 in the computer PU1 after copying the task TE has a configuration such as (PU1, 0.73, TA, TB, TC, TE).

The multiplicity determining means 31 which has been waiting until now.
Receives the processing end message 53 (step 117 in FIG. 6), restarts the processing procedure for determining the multiplicity of tasks.

Further, when the multiplicity managing means 32, which has been on standby until now, receives Nd-Od processing end messages 53 (step 211 in FIG. 8), the operation of the task which has been interrupted as a copy target is executed. Resume, or the operation of the copied task is started (step 212).
Then, the computer status table 42 is updated using the data described in the processing end message 53 (step 21 of the same).
3). This is the same as the processing in step 203 of FIG.

In this way, the multiplicity required by a task can be realized in consideration of the calculation load in the distributed processing system when the task is activated.

Next, a method of reducing the multiplicity of tasks within the allowable range in consideration of reliability when the calculation load becomes high will be described. Here, a case where the task is operating in the state of the schematic diagram of FIG. 28 will be described. In the figure, the computer is shown as a rectangle and the tasks are shown as circles.

When the multiplicity determining means 31 in the computer determines that the computation load of a computer has exceeded a certain value and a certain time has passed (step 118 in FIG. 6), the multiplicity change message 52 (FIG. 3). ) Is created and broadcast to the multiplicity determination means 31 and the multiplicity management means 32 in the distributed processing system (steps 114, 115 and 11 in FIG. 6).
6). Here, the multiplicity change message 52 is composed of the task name of the task whose multiplicity is to be reduced and the multiplicity after the change. The task and the multiplicity are determined using the current multiplicity of the task obtained from the computer status table 42 and the allowable range of the multiplicity of the task obtained from the required multiplicity table 41.

Specifically, when a high-calculation load state is connected for a certain computer for a certain period of time, the multiplicity determining means 31 of the computer, in order from the task having the highest multiplicity among the tasks operating in its own computer, A search is made to see if it is larger than the minimum multiplicity described in the required multiplicity table 41, and when such a task is found, the search processing is terminated to reduce the multiplicity of the relevant task. In addition,
For tasks with the same multiplicity, for example, the tasks are checked in the order of task names. Here, the new multiplicity is the current multiplicity reduced by one.

If there is no task that meets the conditions, the multiplicity of the task cannot be further reduced, and therefore the multiplicity change message 52 is not created and the process returns to step 111 (step 115). For example, if the computer PU3 is in a high calculation load state, the multiplicity determination unit 31
Is a task TC with a multiplicity of 3, a task TB with a multiplicity of 2,
Check in order of TD. First, regarding the task TC, the current multiplicity is 3, and the minimum multiplicity is 2, which applies to the condition of the task being searched for. Therefore, the message body of the multiplicity change message 52 is (TC, 2). Here, the current multiplicity of 3 is reduced by 1
Is the new multiplicity. It should be noted that at this stage, it has not yet been decided to delete the corresponding task on the computer having a high calculation load that has made the above determination. However, as will be described below, since the computer to delete the corresponding task is determined based on the latest calculation load information of each computer in the distributed processing system, the corresponding task may be deleted from the above computer. Further, there is a possibility that the corresponding task whose calculation load suddenly becomes high for some reason may be deleted.

Here, the operation of the multiplicity determining means 31 which has received the multiplicity change message 52 is as described in the operation at the time of task activation (step 11 in FIG. 6).
1).

Upon receiving the multiplicity change message 52, the multiplicity management means 32 operates as described above up to step 114 in FIG. Here, since the multiplicity Nd of the relevant task after the change in the message description is smaller than the multiplicity Od of the current relevant task (step 205 of FIG. 7), the method of deleting the relevant task to reduce the multiplicity. Is determined as follows.

First, the multiplicity management means 32 of each computer
Check whether or not there is a corresponding task in the local computer (Fig. 9
215). If the corresponding task is not in its own computer, it waits until it receives a processing end message broadcast at the end of the deletion processing executed by another computer. on the other hand,
If there is a corresponding task, it becomes a candidate for the computer that deletes the corresponding task. Among these candidates, a computer with a high computational load is desirable as a computer for deleting tasks from the viewpoint of load distribution. Therefore, a list is created from the computer status table 42 in which the computers including the relevant tasks are arranged in descending order of computational load (step 216). For example, in the computer status table 42, the computer including the task TC is PU
1, PU2, PU3, which are arranged in the descending order of calculation load as (PU3, PU2, PU1). Next, Od-Nd candidate computers are left from the head of the list and the others are removed (step 217).
For example, Od from the top of the list for task TC
-Nd = 3-2 = 1 If only one computer is left, the list becomes (PU3). That is, the task TC is deleted in the computer PU3.

The above results are obtained in the same manner on all computers including the relevant task. This is because the computer status table 42 having the same contents, which has been updated using broadcast communication, is used. Therefore, even if each computer decides to delete the corresponding task based on this result, no contradiction occurs. Therefore, if the computer name is in the list (step 218), the processing method message 5
5 (FIG. 3) is created and sent to the task copy / deletion means 33 (FIG. 3) in the own computer (step 21 in FIG. 9).
9).

The processing method message 55 includes a task name, a computer name, and a deletion flag. For example, the message body of the processing method message 55 created by the multiplicity management means 32 of the computer PU3 for the task TC is (TC, PU3, delete).

The task copying / deleting means 33 which has received the processing method message 55 (see step 30 in FIG. 10).
1) Check the message and confirm that there is a deletion flag in this case (step 302). Therefore, the task described in the message is deleted in the own computer (step 306). Then, the processing end message 53 (FIG. 3) is created and broadcast to the multiplicity determination means 31 and the multiplicity management means 32 in each computer of the distributed processing system (step 305 in FIG. 10). Processing end message 5
Reference numeral 3 includes a self-computer name, a calculation load of the self-computer, and a task name list of tasks running on the self-computer. For example, the message body of the processing end message 53 created by the task copying / deleting means 33 in the computer PU3 after deleting the task TC is (PU3, 0.68, TB, TD).

The multiplicity determining means 31 which has been on standby until now.
Receives the processing end message 53 (step 117 in FIG. 6), restarts the processing procedure for determining the multiplicity of tasks.

When the multiplicity management means 32, which has been waiting until now, receives Od-Nd processing end messages 53 (step 220 in FIG. 8), it uses the data described in the processing end message 53 to create the computer status table. 42 is updated (step 221). This is the same as the processing in step 203 of FIG.

In this way, when the calculation load is high,
It is possible to reduce the multiplicity of a task within the allowable range of the multiplicity required by the task and to return the computational load of the computer in the distributed processing system to the normal level.

Next, description will be given of the operation in the case where the multiplicity of the tasks which are lacking is increased and the degree of reliability is recovered when the calculation load becomes low. The case where the task is operating in the state of the schematic diagram of FIG. 28 will be described below. In this figure, the computer is shown as a rectangle and the tasks are shown as circles.

When the multiplicity determining means 31 in the computer determines that the calculation load of a computer has become less than a certain value and a certain time has passed (step 119 in FIG. 6), the multiplicity change message 52 (FIG. 3). ) Is created and broadcast to the multiplicity determination means 31 and the multiplicity management means 32 in the distributed processing system (steps 114, 115 and 11 in FIG. 6).
6). The multiplicity change message 52 includes the task name of the task whose multiplicity is increased and the multiplicity after the change. The task and the multiplicity are determined using the current multiplicity of the task obtained from the computer status table 42 and the allowable range of the multiplicity of the task obtained from the required multiplicity table 41.

Specifically, when a low calculation load state continues for a certain computer for a certain period of time, the multiplicity determining means 31 of the computer uses the computer state table 42 to operate in each computer in the distributed processing system. Of the tasks, check the current multiplicity of tasks that are not running on your computer,
The multiplicity will be increased for a task having a large difference from the maximum multiplicity described in the required multiplicity table 41. The purpose of excluding the tasks running on the local computer is that it is impossible to copy such tasks to the local computer.
This is because the load on the computer will not be increased in such a case.

For tasks having the same multiplicity, for example, the tasks are checked in the order of task names. Here, the new multiplicity is increased by one from the current multiplicity. If there is no task that meets the conditions, the multiplicity of tasks in the distributed processing system satisfies all the requirements, and therefore the multiplicity change message 52 is not created and the process returns to step 111 (step 115).

For example, when the computer PU4 is in a low calculation load state, the multiplicity determining means 31 causes the computer state table 42 to be used.
Using the required multiplicity table 41, the current multiplicity and the maximum multiplicity required by the task are compared for the tasks that are not operating in the computer PU4. For example task TA
Since the present multiplicity is 2 and the maximum multiplicity is 2, it is not necessary to increase the task any more. Also, the task TB
The current multiplicity is 2, and the maximum multiplicity is 3, so the difference is 1. Similarly, the current multiplicity of task TC is 3, the maximum multiplicity is 3, and the difference is 0. The current multiplicity of task TD is 2, the maximum multiplicity is 2, and the difference is 0. Task T
E is not a target because it is running on its own computer. Therefore, the task TB has the largest difference.
Therefore, the message body of the multiplicity change message 52 is (TB, 3). Here, a new multiplicity of 3 which is one increase from the current multiplicity of 2 is used. At this stage, it has not yet been decided to copy the corresponding task to the computer that has made the above determination and has a low calculation load. However, as described below, since the computer to which the relevant task is copied is determined based on the latest calculation load information of each computer in the distributed processing system, there is a possibility that the relevant task will be copied to the above computer. Further, there is a possibility that the corresponding task may be copied to another computer whose calculation load suddenly becomes low for some reason.

Here, the operation of the multiplicity determining means 31 which has received the multiplicity change message 52 is as described in the operation at the time of task activation (step 11 in FIG. 6).
1).

Further, the operation of the multiplicity management means 32 in this case coincides with the operation at the time of task activation in that the method of copying the task is determined in order to increase the multiplicity of the task. Therefore, the operation of the multiplicity management means 32 that has received the multiplicity change message 52 is as described in the operation at the time of task activation (step 201 and subsequent steps in FIG. 7 and step 206 and subsequent steps in FIG. 8). Eventually,
For example, the task TB is copied to the computer PU4 having the lowest calculation load among the computers in which the task TB is not operating, and the task TB in the distributed processing system restarts its operation to complete all operations. At this time, the task TB is operating on the computers PU1, PU3, PU4.

In this way, when the calculation load is low,
It is possible to increase the multiplicity of tasks within an allowable range of multiplicity required by the tasks and improve the reliability in consideration of the calculation load in the distributed processing system.

As described above, the first embodiment described above
(1) It is not necessary to determine the multiplicity of tasks and the method of allocating tasks to computers before starting the tasks.
Since the status of the system in which the task is operating can be reflected in the multiplicity of the task and the method of allocating the task to the computer, the adaptability effect that the processing efficiency of the system can be made more desirable. (2) Since it is possible to change the multiplicity of tasks and assign them to computers without human labor, the effect of maintainability that the user does not need to be aware of the multiplicity of tasks and the computers to be assigned,
(3) Since the means for allocating tasks is distributed among the computers in the distributed processing system, the reliability effect that the tasks can be allocated even if some computers fail, and the task allocation process Since the calculation load of the above is distributed to each computer in the system, the processing efficiency of the entire system is improved, and the distributed processing system has excellent reliability, maintainability, adaptability, and processing performance. Can be provided.

Further, task copying and deletion are executed for the purpose of maintaining reliability or load distribution. By executing the tasks in a distributed manner between normal task executions, computations caused by task multiplicity change processing are performed. An increase in load and an increase in communication volume can be suppressed.

Further, since the multiplicity determining means waits until the current multiplicity changing processing is completed, it does not broadcast an unnecessarily large multiplicity changing message. Therefore,
It is possible to suppress the increase of the calculation load and the increase of the communication volume caused by the task multiplicity change processing. Also, when acquiring the status in the system, the processing method is executed by each computer to realize the required multiplicity. The information used for the determination is matched in the distributed processing system, but the determination of the multiplicity and the processing method for realizing the multiplicity can be performed by any computer currently operating. Therefore, in the first embodiment, the reliability of the multiplicity change processing is maintained, while the reliability of the processing is balanced by the load distribution of the processing.

A modification of the above-described first embodiment will be described below.

In the first embodiment, the multiplicity change message is broadcast by the multiplicity determining means at the time of task activation, high calculation load, and low calculation load. When it is known that the degree of importance changes, the multiplicity of the task can be changed according to the status of the system by broadcasting the multiplicity change message of the relevant task at each time zone delimiter. Alternatively, the task created by the system designer can create and broadcast the multiplicity change message based on the calculation status in the framework of the above embodiment.

Further, in the first embodiment, the multiplicity described in the multiplicity change message at the time of task activation, high computation load, and low computation load is the maximum multiplicity and the current multiplicity, -1. Although the current multiplicity +1 is used, a method of adopting a value between the minimum multiplicity and the maximum multiplicity can be incorporated in consideration of the situation of the distributed processing system. For example, in a system in which the maximum multiplicity is at most 2, the maximum multiplicity may be used as the multiplicity described in the multiplicity change message at the time of high calculation load, and the minimum multiplicity at the time of low calculation load. Can be used.

Furthermore, the upper limit value or the lower limit value of the calculation load when determining the high calculation load or the low calculation load is provided with a plurality of stages, and the calculation load is increased depending on the stage. Severe decrease or increase can be determined. As a result, a more detailed response can be made according to the system status. Also, a method of changing the upper limit value or the lower limit value of the calculation load can be incorporated. For example, when the number of tasks that operate in each time zone of the day is known at the time of system design, changing the upper limit value or the lower limit value depending on the time zone reduces unnecessary copying or deletion of tasks. The effect is that you can.

The method of determining the multiplicity and the upper limit value and the lower limit value of the calculation load can be determined depending on the scale of the actual distributed processing system and the number and characteristics of operating tasks. Further, although the multiplexed tasks are executed in parallel in the above-described embodiment, the present invention can be applied to the case where the task-unit standby redundancy system is adopted. Here, the standby redundancy method on a task-by-task basis is one task that is actually running among the multiplexed tasks, and the remaining tasks are either waiting and sending periodically, or from some task that is running. This is a method in which one of the waiting tasks starts operating when it receives the internal state of the task and detects that the running task has stopped.

Next, the configuration of the second embodiment of the distributed processing system of the present invention will be described in detail with reference to FIGS. 11 and 12.

The distributed processing system of the second embodiment is shown in FIG.
As shown in the configuration diagram of FIG.
(Corresponding to 2) in FIG. 28 is connected via a communication path 102 (corresponding to 2 in FIG. 28). The computer 101 includes a task processing device 103 that executes a task, an allocation device 104 that determines and allocates a computer that executes a task when a computer fails or is added, and data handled by the task processing device 103 and the task allocation device 104. The storage device 105 stores the data, and the communication device 106 processes message communication between computers. The computer 101 and the communication path 102 are connected to each other through the communication device 106.

FIG. 12 is a task assignment device 1 shown in FIG.
It is a figure which shows the structure of 04. In the figure, the task multiplicity decrease detecting means 141 periodically sends an operation state message indicating the operation state of a task to the task operating state collecting means 14
2 and the task processing device 10 is periodically broadcasted.
In step 3, the multiplicity of the task being executed is checked, and if the multiplicity of the task is reduced, the multiplicity-reduced task name list describing the corresponding task is used as the transfer source / transfer destination determination means 143 in the same computer 101. Pass to.

The task operating status collecting means 142 receives the operating status messages periodically broadcast by the task multiplicity decrease detecting means 141 in all the computers 101 in the distributed processing system and stores them in the storage device 105. Store.

When the task is copied or moved, the transfer source / destination determining means 143 provides a transfer source computer and transfer destination of the internal state of the task, such as a variable value used by the task and an address indicating the program position. Determine which computer will be. Also, when moving a task, a computer for deleting the task is determined. At this time, in order to reflect a more accurate state of the distributed processing system in the judgment processing and obtain a more preferable load distribution result, the method of allocating one task at a time is judged. For that purpose, a decision right request message for adjusting the decision order is broadcast to the decision order management means 144.

Further, in order to collect information on the state of the computer in the distributed processing system used as the criterion for the determination process, the computer state management means 1 sends a computer state request message.
Broadcast to 45 and receive the result. Further, when the computer as the transfer source is selected, an inquiry message for confirming whether the computer can be accepted is transmitted to the task acceptance determining means 146 in the relevant computer and the result is received.
A transfer method message describing the determination result determined based on the information thus obtained is broadcast to the task internal state copying means 147.

The judgment order management means 144 receives the judgment right request message from the transfer source / transfer destination judgment means 143, determines the judgment order, and stores it in the storage device 105.
The transfer source / transfer destination determination means 143 starts the determination processing in this order.

When the computer status management means 145 receives the computer condition request message from the transfer source / transfer destination determination means 143, the task processing device 103 within the same computer 101.
The calculation load and the list of task names of operating tasks are returned to the transfer source / transfer destination determination means 143 of the transmission source computer.

The task acceptance determination means 146 determines the transfer source
When receiving the inquiry message from the transfer destination judging means 143, it is judged whether or not the task described in the message can be accepted, and the result is returned to the transfer source / transfer destination judging means 143 of the transmission source computer. .

If the task described in the received message is operating in the task processing device 103 in the same computer 101, the task internal state copying means 147 interrupts the task operation. In the case of the transfer source computer described in the message, the internal state of the task is copied to the task internal state copying means 1 in the transfer destination computer.
Send to 47. If it is a transfer destination computer, after receiving the internal state, a transfer end message for notifying the end of transfer is broadcast to the transfer source / transfer destination determination means 143 and the task internal state copying means 147. If the computer deletes the task, the corresponding task is deleted. When the transfer end message is received, the operation of the relevant task is restarted.

The task allocation requesting means 148 is the computer 10
When 1 is added, a task assignment request message indicating that a computer has been added as a trigger for moving a task from a computer with a high computational load to a computer with a lower computational load is a transfer source / transfer destination determination means 143. Broadcast to. The task allocation request message is received by the transfer source / transfer destination determining means 143 in all the computers 101 in the distributed processing system, and the determination process is started. When the task allocation devices 104 of the respective computers perform message communication with each other, the message is data M2 and computer name indicating that the means is one of the task allocation means 104, for example, the task allocation request means 148. Data and a message body. When transmitting by the broadcast communication system, the data indicating the broadcast mode is used instead of the data indicating the computer name. For example,
When the computer PU1 sends a message to the task assignment requesting means 148 of PU2, the content is (M2, PU2, message body). When the computer PU1 broadcasts a message to all task assignment requesting means 148 in the distributed processing system, the content is (M2, broadcast, message body).

Below, the task multiplicity decrease detecting means 141, the task operation status collecting means 142, the transfer source / transfer destination judging means 143, the judgment order managing means 144, the computer status managing means 145, the task acceptance judging means 146,
The processing operation procedure of the task internal state copying means 147 and the task allocation requesting means 148 will be described in detail.

FIG. 13 is a diagram showing the outline of the flow of data for explaining the operation of the embodiment, and the arrows in the figure show the flow of data. The operation status message 151, the task status table 152, the judgment order table 153, the computer status table 154 and the insufficient multiplicity task table 155 in FIG. 13 are the storage device 1 in the computer 101.
05 (FIG. 11), the operation status message table 151 has a structure shown in FIG. 14, the task status table 152 has a structure shown in FIG. 15, and the determination order table 153 has a structure shown in FIG. Computer status table 15
The configuration of No. 4 is shown in FIG.
The configuration of is shown in FIG.

First, the normal operation when there is no change in the configuration of the computer in the distributed processing system will be described. here,
It is assumed that the task is operating in the state of the schematic diagram of FIG.
In the figure, the computer is shown as a rectangle and the tasks are shown as circles.

First, the task multiplicity decrease detecting means 141
(FIG. 13) examines a task running in the task processing device in the computer, creates an operating state message 161 (FIG. 13), and broadcasts it to the task operating state collecting means 142 in the distributed processing system. (Step 401 in FIG. 19). The operating status message 161 includes a computer name and a task name list of tasks operating in the computer. For example, in the computer PU1, the message body is (PU1, TA, TB, TC).

When the task operating status collecting means 142 receives the operating status message 161 in the distributed processing system (step 501 in FIG. 20), the task operating status collecting means 142 outputs the computer name of the operating status message table 151 as shown in FIG. Check the column (step 502), and if there is a computer name described in the message, delete that item from the table (step 5).
03). After that, the reception time of the operating status message 161 and the operating status message 161 are displayed as the operating status message.
Write to the bull 151 (step 504). here,
The column of the reception time is checked, and the item after a certain period of time is deleted (step 505). As will be described later, since the task multiplicity decrease detecting means 141 periodically broadcasts the operating state message 161, it is considered that the computers described in such items are stopped for some reason.

Next, the task multiplicity decrease detecting means 141
Updates the task status table 152 (FIG. 15) using the operation status message table 151 which the task operation status collection means 142 updates at any time (step 402 in FIG. 19). The task status table 152 is a task name of a task running in the computer, the current multiplicity of the task,
This is a record of the multiplicity at the time of the previous update (called the old multiplicity). Here, updating the table means the data in the column of multiplicity.
This means that the data is copied to the old multiplicity column, the task multiplicity is calculated using the task state table 152, and the task multiplicity is entered in the multiplicity column.

After updating the table, step 403 in FIG.
After that, the difference between the task multiplicity and the old multiplicity is calculated in order from the beginning of the table in order to check whether the task multiplicity has decreased. Normally, since the multiplicity of tasks does not change, other processing is not executed, the process reaches step 412, waits for a fixed time, and then executes step 401 again. That is, the task multiplicity decrease detecting means 141 periodically broadcasts the operation status message 161 to check the multiplicity of tasks. The task status table 152
Since only the tasks running in the computer are recorded in, the load of the computing of the multiplicity of tasks and the inspection process is distributed to the computers in the distributed processing system, which is efficient.

Next, the operation in the case where the computer fails and the multiplicity of tasks decreases will be described.

For example, consider a case where a task is operating in the state of FIG. 28 and the computer PU2 fails and is stopped. Then, in the processing operation procedure of the task multiplicity decrease detection means 141 (FIG. 13), the multiplicity decrease of the tasks TA, TC, and TD is detected after step 403 (FIG. 19) described above.
For example, since the tasks TA, TB, and TC are running on the computer PU1, the item of the task TA of the task status table 152 (FIG. 15) is read (step 40 in FIG. 19).
3) Since multiplicity-old multiplicity = 1-2 <0 (step 406), the task name TA is added to the multiplicity decrease task name list 162 (FIG. 13) (step 407 in FIG. 19). Next task status table 15
Although the item of task TB of 2 is read (step 40
8, step 409), and multiplicity-old multiplicity = 2-2 = 0 (step 406), so nothing is done. Furthermore, the item of the task TC of the task status table 152 is read (steps 408 and 409).

Since multiplicity-old multiplicity = 2-3 <0 (step 406), the task name TC is added to the multiplicity decrease task name list 162. In this case, the end of the task state table 152 as shown in FIG. 15 has been reached. If the multiplicity becomes 0 (step 404), there is an active task regardless of whether the cause is the normal end of the task or the failure of the computer. Since it does not, it is impossible to copy the task. Therefore, the item of the corresponding task is deleted from the task status table 152 (step 405).

Next, since the multiplicity decreasing task name list 162 is not empty (step 410), the multiplicity decreasing task name list 162 is output to the transfer source / transfer destination determining means 143 (FIG. 13) in the computer PU1. (Step 41
1). For example, the message body of the multiplicity reduction task name list 162 is (PU1, TA, TC). Although the multiplicity of the task TD is also reduced, this is detected by the computer PU3 in operation of the task TD.

When the transfer source / transfer destination judging means 143 in the computer PU1 receives the above-mentioned multiplicity decreasing task name list 162 (step 601 in FIG. 21), it creates a judgment right request message 163 (FIG. 13). And broadcast (step 603 in FIG. 21). The destination is the judgment order management means 144 (see FIG.
3). This broadcast determines the computer that acquires the right to judge the transfer source computer and the transfer destination computer when transferring the internal state of the task (hereinafter referred to as the judgment right),
It is implemented as a trigger for determining the order of judgment.

Actually, the decrease in the multiplicity of the task TC is calculated by the computer P.
Since U1 and PU3 are detected, PU1 and PU3 compete with each other for acquisition of the decision right, and this conflict needs to be resolved. Further, the order of judgment needs to be determined consistently within the distributed processing system. Further, if the determinations can be executed in order, the calculation load after each transfer process is accurately reflected. That is, a more preferable load distribution result is obtained.

The decision right request message 163 is data indicating the task name, the computer name, and the process for copying the task or the process for moving the task (in the following, in the case of the former case). Is called a copy flag, and in the latter case it is called a move flag). In this case, the message body of the decision right request message 163 relating to the task TA is (TA, PU1, copy), and the decision right request message 16 relating to the task TC is shown.
The body of message 3 is (TC, PU1, copy). As described above, since the computer PU3 also detects the decrease in the multiplicity of the task TC, the computer PU3 also sends a decision right request message 163 such that the message body relating to the task TC is (TC, PU3, copy). Trying to broadcast. However, if any one of the decision right request messages 163 relating to the same task is adopted, if the first-come-first-served method is used, the one broadcasted later is wasted. Therefore, as will be described later, it is possible to confirm whether or not the decision right request message 163 regarding the task is already received. This reduces the amount of communication and processing for communication, and reduces communication delay. Furthermore, the earliest judgment request message 163
It is considered that the computer that was able to broadcast the above has a lower calculation load, and therefore a computer with a lower load executes the determination processing, and the processing efficiency of the entire distributed processing system improves.

The means for making it possible to confirm whether or not the decision right request message 163 regarding the task is received is the decision order management means 144 (FIG. 13). When the judgment order management means 144 receives the judgment right request message 163 (step 701 in FIG. 23), the judgment order table 153 shown in FIG. 16 is updated. Each item of the judgment order table 153 is the same as the judgment right request message 163. That is, it consists of a task name, a computer name, a copy flag or a move flag. In this case, since the received judgment right request message 163 has a copy flag (step 702), the task name described in the message is compared with the task name of each item in the judgment order table 153. (Step 703), if there is an item with the same task name and a copy flag, there is a first-arrival message, so the received message
Ji throws away. If the task name is the same and there is no item with the copy flag, the message is judged in the order table 1
It is added to the end of 53 (step 704).

Returning to the explanation of the processing operation procedure of the transfer source / transfer destination determining means 143 again. Transfer source / transfer destination determination means 143
In order to reduce the communication delay as described above, when the judgment right request message 163 is broadcast, the judgment order table 15
3, it is checked whether the decision right request message 163 regarding the task is already received and it is decided whether or not to broadcast (step 603 in FIG. 21). Then, it stops for a certain period of time (step 605). This is to wait for the update of the determination order table 153. Next, in order to confirm whether the judgment right has been acquired, the judgment order table
The bull 153 is referenced again (step 606). If there is no own computer name in the judgment order table 153, the judgment right cannot be acquired, and the process ends. Judgment order table 15
If there is a computer name at the beginning of 3 (step 607),
The determination process regarding the task described at the head of the determination order table 153 is started (step 612 and thereafter).

The computer name is 2 in the determination order table 153.
In the case of the second and subsequent times, a certain period of time is waited until the transfer end message 164 (FIG. 13) transmitted when the transfer process executed as a result of the judgment process is completed (step 608 of FIG. 21). ), And the confirmation of the presence or absence of reception is repeated (step 609). The transfer end message 164 includes the task name of the transfer processing target. For example, a transfer end message when the transfer process of the task TA is completed
The message body of J164 is (TA). When the transfer end message 164 is received, the item including the task name described in the message is judged in the order table 15
3 is deleted (step 610). As a result, if the computer name is at the beginning of the determination order table 153 (step 611), the determination process is executed (step 612 onward). When the own computer name is in the second and subsequent positions in the judgment order table 153, the process returns to step 608, and the above processing is repeated until the own computer name comes to the head of the judgment order table 153.

The determination processing in the transfer source / transfer destination determination means 143 starts by broadcasting the computer status request message 165 (FIG. 13) (step 612 in FIG. 21).
The computer status request message 165 includes the name of the computer itself. The destination is the computer status management means 145 (FIG. 13). For example, the transfer source / transfer destination determining means 1 in the computer PU1
The message body of the computer status request message 165 broadcast by 43 is (PU1).

The computer status request message 165 is received by all the computer status management means 145 (FIG. 13) in the distributed processing system. Computer status management means 145
Receives the computer status request message 165 (see FIG. 2).
4, step 801), and the task being operated by the task load and the computation load is checked (step 802). Next, a computer status message 166 (Fig. 13) is created (Fig. 24).
Step 803). Computer status message 166
It consists of a computer name, a calculation load, and a task name list of tasks running on the task processor. For example, a computer status message created by the computer status management means 145 in the computer PU3.
The message body of J166 is (PU3, 0.72, TB, TC, TD). Then, the computer status message 166 is transmitted to the transfer source / transfer destination determination means 143 of the computer name described in the received message (step 804). The transfer source / transfer destination determination means 143 uses the computer status message 1 described above.
When 66 is received (step 613 in FIG. 22), the contents of the message indicate the computer status table 154 as shown in FIG.
Is updated (step 614). Each item of the computer status table 154 is the same as the computer status message 166. That is, the computer name, the calculation load of the corresponding computer, and the task name list of tasks running on the task processing device of the corresponding computer.

Next, from the information written in the computer status table 154, the computer to be the transfer destination when transferring the internal state of the task is determined. The task to be transferred must not be running at the transfer destination. Further, it is desirable that the calculation load is low from the viewpoint of load distribution. Then, the computer in which the corresponding task is not operating is found from the computer status table 154 (step 615). If it cannot be found, it cannot be copied. Therefore, the multiplicity of the corresponding task cannot be maintained. Regarding such a task, in order to recover the multiplicity when a computer is added, the fact that the multiplicity is insufficient is indicated in the multiplicity insufficient task table 155 as shown in FIG. It is recorded (step 616). When a computer in which the relevant task is not running is found, an inquiry message 167 (Fig. 1) for confirming whether or not the relevant task can be accepted is sent to the computer with the lowest computational load.
3 is transmitted (step 617 in FIG. 22). The inquiry message 167 is made up of the self-computer name and the corresponding task name. For example, the transfer source / transfer destination determination means 14 of the computer PU1
17 to find the transfer destination of the task TA in FIG.
And the computers PU3 and PU4 in which the task TA is not operating are candidates for the destination computer. Then, an inquiry message 167 whose message body is (PU1, TA) is transmitted to the task acceptance determining means 146 (FIG. 13) of the computer PU3 having the lowest calculation load.

The inquiry acceptance message 167 is received by the task acceptance determining means 146 (FIG. 13) in the computer designated as the destination. When the task acceptance determination means 146 receives the inquiry message 167 (step 901 in FIG. 25), it determines the acceptability of the task described in the message (step 902). For example, the upper limit of the calculation load is set to be acceptable, and if the current calculation load is below that value, it is acceptable, and if it exceeds that value, it is not acceptable. There is a method to determine that there is.
Alternatively, it is possible to use means for determining whether or not the task group can be executed within the time limit based on information about the influence of the task on the calculation load. If the result of the judgment (step 903) is that the message is acceptable, an acceptable message 168 (FIG. 13) is sent to the transfer source / transfer destination determination means 143 of the computer described in the received inquiry message 167. It is returned (step 904 in FIG. 25). If it is acceptable, the unacceptable message 16 addressed to the transfer source / destination determination means 143 of the computer described in the received inquiry message 167.
9 (return FIG. 13 (step 905 in FIG. 25). Acceptable message 168 and unacceptable message)
The message body of page 169 is like (“OK”) (“NG”), respectively. "OK" and "NG" are used to identify the message.

When the transfer source / transfer destination determining means 143 receives the acceptable message 168 within a fixed time (step 618 in FIG. 22), it sets the corresponding computer as the transfer destination computer (step 621). If the unacceptable message 169 is received within a certain period of time, or if the message cannot be received within a certain period of time, the computer with the lowest computational load as a result of removing the relevant computer from the transfer destination candidates Inquiry message 167 is transmitted to (step 619, step 620).

Such a process can be accepted by a message
Message 168, the transfer destination computer is determined, or the transfer destination candidate is exhausted and it is determined that there is no transfer destination computer. For example, in the computer PU1, the candidates of the transfer destination of the task TA are the computers PU3 and PU4 as described above,
If the acceptable message 168 does not arrive within a fixed time after transmitting the inquiry message 167 to the computer PU3 having the lowest calculation load, the inquiry message 167 is transmitted to the computer PU4. In some cases, unacceptable message 169 from computer PU4
May arrive, or the message may not arrive in time. In that case, since there is no computer as the transfer destination, the multiplicity of the relevant task remains reduced. Therefore, in such a case, the relevant task name is registered in the multiplicity-deficient task table 155 as in the previous case. If the acceptable message 168 arrives from the computer PU4, the computer PU4 is set as the transfer destination computer.

When the transfer destination computer is determined, the transfer source computer is further determined. The transfer source computer must be running the task to be transferred. Further, since the transfer work is executed after the normal task processing is completed, it is more efficient that the transfer source is a computer with a low calculation load. Therefore, the computer on which the relevant task is running is set to the computer status table.
It is found from the bull 154 (step 622), and the computer with the lowest calculation load is set as the transfer source (step 623). In the example considered here, the only computer PU1 on which the task TA is operating is the transfer source computer. Note that at least the computer that has acquired the decision right is running the corresponding task, so the transfer source computer can always be found.

Since we are considering copying tasks here,
When the transfer destination and the transfer source are determined, the transfer method is determined (step 624). Therefore, a transfer method message 170 (FIG. 13) is created and broadcast (step 626 in FIG. 22). The transfer method message 170 includes a corresponding task name, a transfer source computer name, a transfer destination computer name, a computer name for deleting the corresponding task when moving, a copy flag or a move flag. Since the case of copying is considered here, a message to be deleted is not specified and a message is created using data that cannot be the computer name, for example, -1.
The destination is the task internal state copying means 147 (FIG. 13). For example, when copying the task TA, if the source computer is PU1 and the destination computer is PU4,
Transfer method The message body of the message 170 is as follows: (TA, PU1, PU4, -1, copy).

The transfer method message 170 is received by all task internal state copying means 147 (FIG. 13) in the distributed processing system. Task internal state copying means 14
When receiving the transfer method message 170 (step 911 in FIG. 26), the server 7 checks whether or not the task described in the message received by the task processing device in its own computer is operating (step 914). If the task in question is not running, the process is terminated. If the relevant task is operating, the operation of the task is interrupted (step 91 of the same).
5). If the computer name of the own computer matches the transfer source computer name described in the received message (step 916),
The data of the internal state of the task is transferred to the transfer destination computer described in the received message as transfer data 171 (FIG. 13) (step 917 in FIG. 26). For example, the transfer data 171 has a configuration such as ("DATA", data body). "DATA" is used to identify the message. When the task internal state copying means 147 of the transfer destination computer receives the transfer data 171 (step 912), it copies the task internal state data (step 920). After the copying process is completed, the transfer completion message 164 is broadcast (step 921). The transfer end message 164 includes the task name of the transfer processing target, as described above. The destinations are the transfer source / transfer destination determining means 143 and the task internal state copying means 147.
For example, the main message body of the transfer end message 164 when the transfer process of the task TA is completed is (TA).

The transfer end message 164 is received by all task internal state copying means 147 and transfer source / transfer destination determining means 143 in the distributed processing system. Since the processing when the transfer source / destination determining means 143 has received has already been described, the processing when the task internal state copying means 147 has received will be described here. Upon receiving the transfer end message 164 (step 913), the task internal state copying means 147 restarts the operation of the task described in the received message (step 922).

In this way, the multiplicity of tasks can be maintained even if the computer fails. Also, if the multiplicity of tasks cannot be maintained due to the computational load in the distributed processing system or the balance between the number of computers and the multiplicity, record the relevant task name so that the multiplicity can be restored when a computer is added. To

The operation when a load is required to be distributed by adding a computer will be described below. Since the contents are almost the same as the operation when the computer fails, the different parts will be mainly described.

When a new computer is added to the distributed processing system, the task allocation requesting means 148 (FIG. 13) of that computer operates. This can be realized by the user inputting an operation start command when adding a computer. The task assignment requesting means 148 uses the task assignment request message 172.
(FIG. 13) is created and broadcast (step 9 in FIG. 27).
31, step 932). This is a trigger for moving a task from a computer with a high computational load to a computer with a lower computational load. For example, the task allocation request message 172 is something like ("REQ"). "REQ" is used to identify the message.

It is the transfer source / transfer destination determining means 143 that receives the task allocation request message 172. When the transfer source / transfer destination determining means 143 receives the task assignment request message 172 (step 602 in FIG. 21), it creates a determination right request message 163 to acquire the right to determine the task assignment method. Broadcast (Step 6)
04). At that time, the multiplicity-deficient task table 155 (FIG. 18) and the judgment order table 153 (FIG. 16) are referred to. A task recorded in the multiplicity-deficient task table 155, which is still operating, has a lack of multiplicity and should be copied. Whether or not the corresponding task is operating depends on whether or not the task is operating in the own computer. This is because the computer recorded in the insufficient multiplicity task table 155 is a computer that has acquired the right to determine the task allocation method, and in such a computer, the corresponding task is originally operating.

Therefore, a decision right request message 163 is created for that task and is broadcast to the decision order management means 144, and the task name is assigned to the task number lacking multiplicity.
Bull 155 to delete. For example, if the task TD is recorded in the task table 155 with insufficient multiplicity of the computer PU3, the message body of the decision right request message 163 is (TD, PU3, copy). In addition, a computer with a computational load above a certain value may move a running task to a computer with a lower computational load in order to send a decision right request message with a move flag.
163 is created. For example, if the calculation load of the computer PU3 is a certain value or more and the tasks TB, TC, and TD are in operation, the message body of the decision right request message 163 is (TB, PU3, move) (TC, PU3, move) (TD, PU3, move). In this case, the task does not necessarily move to the added computer, but the task can be moved to the computer with the lowest load at that time, so that more desirable load distribution is realized in the entire distributed processing system. When broadcasting the decision right request message 163, refer to the decision order table 153 to check whether the decision right request message 163 for the relevant task has already been received, and whether to broadcast. Decide. If no message related to the task is received, broadcast it. When the message with the movement flag is received for the task, the message with the copy flag is broadcast, and the message with the movement flag is not broadcast. When the message with the copy flag is received for the relevant task, the message with the move flag is broadcast, and the message with the copy flag is not broadcast.

By the way, as described above, the judgment right request message
The judgment order management means 144 receives the judgment order table 144 so that the task names do not overlap.
3 is updated. Here, since there is a message with a movement flag (step 702 in FIG. 23), the task name described in the message is compared with the task name of each item of the judgment order table 153 (step 705), If the task name is the same and the item has a move flag, the received message is discarded because there is a first-arrival message. If the task name is the same and there is no item with the move flag, the message is inserted before the item with the copy flag (step 706). By moving tasks first to balance the load and then addressing the reliability issue of maintaining multiplicity of tasks, the overall efficiency of task migration and copying is improved.

The transfer source / transfer destination determining means 143 operates as described above except for the branch relating to the presence / absence of the movement flag (step 624 in FIG. 22). If there is a migration flag (step 624), it is necessary to copy the relevant task and then delete the relevant task from the computer with a high computational load. Therefore, the computer status table 154 (see FIG.
Referring to 7), as a computer that deletes the computer task with the highest computational load among the computers running the relevant task (step 625 in FIG. 22), transfer method message 1
70 is created and broadcast (step 626). For example, when moving about task TB, if the source computer is PU1, the destination computer is PU4, and the computer that deletes the task is PU3, the message body is (TB, PU1, PU4, PU3, move). ) become that way. In this way, when moving tasks,
The source computer and the computer that deletes the task do not necessarily match. This is an important point for achieving a more desirable load balancing result and realizing a more efficient process of transferring the internal state of a task.

When the task internal state copying means 147 receives the transfer method message 170 (step 911 in FIG. 26), it executes the processing operation procedure already described. In particular, if the message has a move flag and the computer name for deleting the task described in the message is the own computer name (step 918), the task is deleted (step 919).

With this, it is possible to realize load distribution when a computer is added and to recover the multiplicity of tasks that were deficient when the computer failed.

As described above, in the second embodiment, (1) the task multiplicity is maintained because the decrease in the multiplicity of tasks is detected when a computer fails and the tasks are copied. And (2) the effect of maintainability that it is not necessary to make the user aware of the task multiplicity and the computer that allocates the task, since the task allocation and the maintenance of the multiplicity can be performed without human labor. 3) It is not necessary to decide the task allocation method in advance, and even when the task to be executed in the distributed processing system or the computer is changed, the allocation method can be changed without human intervention, so a more desirable task. The effect of adaptability and maintainability that tasks are allocated, and (4) task allocation means are distributed to each computer in the distributed processing system, so when some computers fail. However, it has the reliability effect that task allocation can be performed and the processing efficiency that the overall system processing efficiency is improved because the calculation load of task allocation processing is distributed to each computer in the system. However, it is possible to provide a distributed processing system having excellent reliability, maintainability, adaptability, and processability.

Further, the copying and moving of tasks are executed for maintaining reliability or load distribution. By executing the tasks in a distributed manner between normal task executions, the calculation load caused by the task allocation processing increases. Also, it is possible to suppress an increase in communication volume.

Further, since the method by which the transfer source / transfer destination judging means 143 acquires the judgment right is first-come-first-served, it is highly possible that the computer acquiring the judgment right has a relatively low calculation load. Since the transfer source computer and the transfer destination computer are likely to have relatively low calculation loads, the processing efficiency of the entire distributed processing system is improved.

Further, since the computer to be deleted when the task is moved is not a transfer source but a computer having a high calculation load, more desirable load distribution can be realized.

Further, the transfer source / transfer destination judging means 143 uses the data provided by the judgment order managing means 144 to judge whether the competing judgment right request message has already been received. Since the broadcasting of messages is canceled, the amount of communication and the processing for communication are reduced, and the communication delay can be reduced.

Further, since the task multiplicity decrease detecting means 141 inspects the change in multiplicity only for the task operating in its own computer, the inspection processing is performed for each task by the computer in the distributed processing system to balance the load. It will be efficient.

Further, since the task operating state collecting means 142 checks the message reception time and deletes the item relating to the task which has passed a certain time, it is necessary to notify the fact that the task is finished when the task is finished. As a result, communication volume can be suppressed.

Further, in the judgment order management means 144,
Since the order is determined so that the task movement is executed with priority over the copying, the load balancing can be realized by the task movement earlier and the efficiency of the entire processing is improved.

A modification of the second embodiment will be described below.

In the second embodiment described above, the multiplexed tasks are executed in parallel, but the present invention can be applied to the case where the above-mentioned standby redundancy system of task units is adopted. In addition, as a method of actually mounting in a computer, there are a method of mounting a task processing device and a task allocation device in the same processor, and a method of mounting in a different processor.
The latter is superior in treatment efficiency, but the equipment cost is lower in the former.

In the second embodiment, the task allocation request message is broadcast when a computer is added. However, by broadcasting this message from a computer in which the computational load has decreased for a certain period of time, load distribution and restoration of multiplicity can be realized. it can.

[0143]

As described above, according to the present invention, (1) it is not necessary to determine the multiplicity of tasks and the method of allocating tasks to computers before starting the tasks, Since the system status can be reflected in the multiplicity of tasks and the method of allocating tasks to computers, the adaptability effect that the processing efficiency of the system can be made more desirable, and (2) tasks Since the multiplicity of tasks can be changed and assigned to computers without human intervention, the effect of maintainability that the user does not need to be aware of the multiplicity of tasks and the computers to be assigned, and (3)
Since the means of task allocation is distributed to each computer in the distributed processing system, task allocation can be performed even if some computers fail, and task decrease can be detected by detecting decrease in the multiplicity of tasks. Since the task is duplicated, the multiplicity of tasks can be maintained, and the computational load of task allocation processing is distributed to each computer in the system, which improves the processing efficiency of the entire system. It is possible to provide a distributed processing system that has the effects and is excellent in reliability, maintainability, adaptability, and processability.

[Brief description of drawings]

FIG. 1 is a system configuration diagram of a first embodiment of the present invention.

FIG. 2 is a block diagram of a task processing device according to a first embodiment.

FIG. 3 is a diagram showing an outline of a data flow for explaining the operation of the first embodiment.

FIG. 4 is an example of a required multiplicity table used in the first embodiment.

FIG. 5 is an example of a computer status table used in the first embodiment.

FIG. 6 is a flowchart showing a processing operation procedure of a multiplicity determining unit in the first embodiment.

FIG. 7 is a first flowchart showing a processing operation procedure of a multiplicity management unit in the first embodiment.

FIG. 8 is a second flowchart showing the processing operation procedure of the multiplicity management means in the first embodiment.

FIG. 9 is a third flowchart showing the processing operation procedure of the multiplicity management means in the first embodiment.

FIG. 10 is a flowchart showing a processing operation procedure of task copying / deleting means in the first embodiment.

FIG. 11 is a system configuration diagram of a second embodiment of the present invention.

FIG. 12 is a block diagram of a task allocation device according to a second embodiment.

FIG. 13 is a diagram showing an outline of the flow of data for explaining the operation of the second embodiment.

FIG. 14 is an example of an operation status message table processed by the task assignment device in the second embodiment.

FIG. 15 is an example of a task status table processed by the task allocation device in the second embodiment.

FIG. 16 is an example of a determination order table processed by the task assignment device in the second embodiment.

FIG. 17 is an example of a computer status table processed by the task allocation device in the second embodiment.

FIG. 18 is an example of a multiplicity-deficient task table processed by the task allocation device in the second embodiment.

FIG. 19 is a flowchart showing the processing operation procedure of the task multiplicity decrease detecting means in the second embodiment.

FIG. 20 is a flowchart showing the processing operation procedure of the task operation status collecting means in the second embodiment.

FIG. 21 is a first flowchart showing a processing operation procedure of a transfer source / transfer destination determining means in the second embodiment.

FIG. 22 is a second flowchart showing the processing operation procedure of the transfer source / transfer destination determining means in the second embodiment.

FIG. 23 is a flowchart showing the processing operation procedure of the judgment order management means in the second embodiment.

FIG. 24 is a flowchart showing the processing operation procedure of the computer status management means in the second embodiment.

FIG. 25 is a flowchart showing the processing operation procedure of the task acceptance determining means in the second embodiment.

FIG. 26 is a flowchart showing the processing operation procedure of the task internal state copying means in the second embodiment.

FIG. 27 is a flowchart showing the processing operation procedure of the task allocation requesting means in the second embodiment.

FIG. 28 is a schematic diagram for explaining a task-based multiplexing method.

FIG. 29 is a schematic diagram for explaining a computer-based multiplexing method.

[Explanation of symbols]

1, 101 ... Computer, 2, 102 ... Communication path, 3, 103
... task processing device, 4, 105 ... storage device, 5, 106
... communication device, 31 ... multiplicity determining means, 32 ... multiplicity managing means, 33 ... task copying / deleting means, 104 ... task assigning device, 141 ... task multiplicity decrease detecting means, 142
... task operation status collection means, 143 ... transfer source / transfer destination determination means, 144 ... determination order management means, 145 ... computer status management means 146 ... task acceptance determination means, 147 ...
Task internal state copying means, 148 ... Task allocation requesting means.

Claims (11)

[Claims] 1. A distributed processing system in which a plurality of tasks are multiplexed and executed in task units by a plurality of computers connected via a communication path, wherein each of the plurality of computers corresponds to resource information of the distributed processing system. Then, the multiplicity of the task operating in the computer is determined, and the multiplicity determining means for broadcasting the multiplicity change message indicating this multiplicity and the multiplicity of the task determined by this multiplicity determining means are realized. Determining the method to be performed based on the computer status message and generating a processing method message, and a task copy for receiving the processing method message from the multiplicity managing means and realizing the determined realization method. -Distributed processing system characterized by comprising task processing means including deletion means. 2. Each of the plurality of computers further comprises a request multiplicity table including a task name and minimum multiplicity and maximum multiplicity information corresponding to each task, and the multiplicity determining means 2. The distributed processing system according to claim 1, wherein, when a multiplicity message is received, the request multiplicity table is changed according to the content thereof and the multiplicity change message is created. 3. Each of the plurality of computers further comprises a computer status table consisting of a computer name in the distributed processing system, a computation load of the computer, and task name list information of tasks running on the computer, The multiplicity management means, when receiving the computer status message, changes the computer status table according to the contents of the message, and obtains the current multiplicity of the task using the changed computer status table, 3. The distributed processing system according to claim 1, wherein this multiplicity is compared with the multiplicity of the changed task described in the multiplicity change message. 4. As a result of the comparison, when the multiplicity of the changed task is higher, the multiplicity management means judges whether or not there is a corresponding task in its own computer, and if not, the computer status. From the table, create a list in which the computers that do not include the relevant task are listed from the one with the lowest load, and then delete the list other than the list corresponding to the difference between the changed multiplicity of the task and the current multiplicity of the relevant task. 4. The distributed processing system according to claim 3, wherein the processing method message is created if the computer name is present in the list at this time. 5. As a result of the comparison, when the current multiplicity of the relevant task is higher, the multiplicity managing means judges whether or not the relevant task exists in the own computer, and if there is, the computer. From the status table, create a list of computers that include the task in order from the one with the highest load, and then delete the list that corresponds to the difference between the current multiplicity of the task and the multiplicity of the changed task. 4. The distributed processing system according to claim 3, wherein the processing method message is created when the own computer name is present in the list. 6. The task copying / deleting means, when receiving the processing method message from the multiplicity managing means, judges whether or not a copying flag is present in the processing method message, and if there is, a transfer flag is sent. When a request message is created and transmitted to the computer described in the processing method message, and when a transfer request message is received from the computer described in the processing method message, the internal state of the task described in the processing method message is included. The transfer data is transmitted to the task copying / deleting means of the computer described in the processing method message.
The distributed processing system described. 7. A distributed processing system in which a plurality of computers are connected to each other via a communication path to multiplex and execute a plurality of tasks in task units, wherein each of the plurality of computers is a task operating in its own computer. And a task multiplicity decrease detecting means for periodically examining the multiplicity of the running task and detecting the decrease in the multiplicity of this task, while periodically broadcasting the operation status message including the task name list of When the decrease in the multiplicity is detected by the multiplicity decrease detecting means, the computer which is the transfer source of the internal state of the task and the computer which is the transfer destination are determined, and the method of transferring the internal state of the task is also determined. Transfer source / transfer destination determining means, and a computer status pipe that gives the state of the computer when the transfer source / transfer destination determining means determines the method of transferring the internal state of the task. And a task internal state copying means for transferring the internal state of the task according to the transfer method determined by the transfer source / transfer destination determining means, and a task assigning device or a task processing device. Distributed processing system. 8. The task allocation device or task processing device further comprises task allocation request means for broadcasting a message requesting task allocation when a computer is added to the distributed processing system. The distributed processing system described. 9. The task allocating device or the task processing device further comprises a task operating state collecting means for receiving and recording the operating state message from the task multiplicity decrease detecting means. Alternatively, the distributed processing system according to item 8. 10. The task allocation device or the task processing device further includes a judgment order management unit for managing an order in which the transfer source / destination judgment unit determines a method of transferring an internal state of a task. Claim 7, characterized in that
8. The distributed processing system according to any one of 8 and 9. 11. A task for determining whether or not the task assigning device or the task processing device can accept a task when the transfer source / transfer destination determining means determines the method of transferring the internal state of the task. 11. The distributed processing system according to claim 7, further comprising acceptance determination means.