WO2018235180A1

WO2018235180A1 - Distributed processing system

Info

Publication number: WO2018235180A1
Application number: PCT/JP2017/022798
Authority: WO
Inventors: 古庄　晋二
Original assignee: 株式会社ターボデータラボラトリー
Priority date: 2017-06-21
Filing date: 2017-06-21
Publication date: 2018-12-27
Also published as: JP6858378B2; JPWO2018235180A1

Abstract

The present invention achieves multitasking in which tasks are executed in a time-division manner with a simpler configuration in a distributed processing system. Provided is a distributed processing system in which reception data is relayed among nodes 1, thereby transferring data transmitted by each of the nodes 1 to all the other nodes 1, wherein in response to a task switch instruction, each of the nodes 1 saves data to be relayed, and switches a task of transmitting data to a task as a switch destination indicated by the task switch instruction. Further, the saved data of the task as the switch destination is recovered, and the relaying of the data is restarted. In each of the nodes 1, sub tasks 101 of a plurality of tasks are executed by multitasking. Each of the sub tasks 101 performs processing of corresponding task data received from another node 1. A sub task 101 which waits for task data enters a standby state, and an execution time therefor is cut, allowing a correspondingly longer execution time to be allocated to a sub task 101 having a process to be executed.

Description

Distributed processing system

The present invention relates to a technique of task switching in a distributed processing system.

As a distributed processing system, there is known a distributed processing system in which subtasks obtained by dividing tasks are allocated to a plurality of nodes, and the allocated subtasks in each node are processed in parallel (for example, Patent Document 1).

Also, as a distributed processing system, each node is connected in a ring shape with a bidirectional channel to form a ring network, and each node sequentially relays data along the ring. There is also known a distributed processing system that transfers data sent from a node to a ring to all other nodes (for example, Patent Document 2).

JP, 2013-23499, A International Publication No. 2005/073880

Operating the computer system in multitasking offers the following advantages.
(1) The use efficiency of the computer system can be improved by executing other tasks when waiting for processing such as I / O waiting.
(2) Since there is no need for queuing, short processing is completed in a short time even if short processing occurs after long processing has been input.
(3) Trouble tasks can be easily stopped.
On the other hand, in the case of realizing multitasking by executing tasks in a time division manner in the above-described distributed processing system, the following problems occur.
That is, in this case, since subtasks of tasks are executed asynchronously in each node, there is no inconsistency between processing of subtasks in each node, and to switch tasks so that tasks can be performed efficiently. In addition to complex control of centrally managing the state of subtasks of each node and switching the subtasks assigned to each node according to the state of the subtasks of each node, overhead for the control is generated.

Therefore, an object of the present invention is to realize, in a distributed processing system, a multitask that executes tasks in a time division manner with a simpler configuration.

In order to achieve the above object, the present invention performs control to simultaneously transmit a task switching instruction accompanied by an identifier of a task of a switching destination to each node to a distributed processing system which executes a plurality of tasks connecting a plurality of nodes by a network. A sub-task executing unit that executes, in each of the nodes, a sub-task that is a portion of each of a plurality of tasks executed by the distributed processing system in a multitasking process; Of the network and the task so that data of a task of which the subtask is a part propagates through the network to all other nodes other than the node where the subtask of the source of the data of the task is being executed. A data transmission / reception unit that transmits / receives data; and a data of a task received by the data transmission / reception unit from the network. Data transferred to a subtask of the task, and in response to the task switching instruction, data propagated through the network is switched to data of a task indicated by an identifier associated with the task switching instruction And a propagation data switching unit that controls the data transmission and reception operation of the task of the data transmission and reception unit.

Further, in order to achieve the above task, the present invention performs control to simultaneously transmit a task switching instruction accompanied by an identifier of a task of a switching destination to each node to a distributed processing system having a plurality of nodes and executing a plurality of tasks. A subtask executing section for executing multitasks as subtasks each of which is a part of each of a plurality of tasks executed by the distributed processing system, and data originating from the task subtasks. A transmission unit that transmits data of a task whose subtask is a part to another node, a data capture unit that transfers data of a task received from another node to a subtask of the task, and a task received from another node Data to other nodes so that the data is sequentially transferred from the source node of the data to all other nodes. It is provided with a data transfer unit for splicing. Here, in response to the reception of the task switching instruction, the transmitting unit stops transmission of data of tasks other than the task indicated by the identifier associated with the task switching instruction to another node, and the relay unit is configured to In response to the reception of the task switching instruction, the relaying of the data of the task not relayed to the other node at that time received from the other node until the task switching instruction accompanied with the identifier of the task of the data is received put off.

Here, more specifically, for example, the sub task includes each of the transferred data in the data to be processed, performs predetermined processing on the data to be processed, and waits for the data to be processed. If the sub-task execution unit transitions to the standby state, the sub task execution unit decreases the execution time of the sub task that has transitioned to the standby state and increases the execution time of sub tasks not in other transition states. Good.

Further, in the distributed processing system, in the relay unit, in response to the reception of the task switching instruction, data of a task not relayed to another node at that time received from the other node is received from the other node. In addition to saving as save data, the save data of the task indicated by the identifier accompanying the task switching instruction may be relayed to the other node.

Also, in such a distributed processing system, in the data fetching unit, in response to the task switching instruction, transfer of data not transferred to the subtask at the point in time received from another node is transferred to the subtask. It may be configured to defer until receiving the task switching instruction accompanied by the data task identifier.

Further, in order to achieve the above task, the present invention performs control to simultaneously transmit a task switching instruction accompanied by an identifier of a task of a switching destination to each node to a distributed processing system having a plurality of nodes and executing a plurality of tasks. A subtask executing section for executing multitasks as subtasks each of which is a part of each of a plurality of tasks executed by the distributed processing system, and data originating from the task subtasks. While transmitting the data of the task received from the other node as a transmitter that transmits the data of the task whose subtask is a part to another node, and the task's data received from the other node to the task's subtask, A data transfer unit relaying to another node is provided so as to be sequentially transferred to all other nodes. Here, in response to the reception of the task switching instruction, the transmitting unit stops transmission of data of tasks other than the task indicated by the identifier associated with the task switching instruction to another node, and the data transfer unit And, in response to the reception of the task switching instruction, the data received from the other node at that time is stored as saving data of the task, and the stored identifier indicated by the task switching instruction is indicated Transfer of task save data to the subtask and relay to other nodes.

Further, each of the above distributed processing systems may have a plurality of rings in which n (where n is an arbitrary natural number of 2 or more) nodes are connected in a ring shape by channels. Here, the number of nodes included in each ring is equal, and each node of each ring is a channel whose transmission side is the node, and included in the same ring as the node of each ring other than the ring including the node. Each node is connected so as to be connected to a different node of another ring on a channel whose transmission side is the node. Then, in the data transfer means of each node, the node receives data of a task received from a channel connected to a node of the previous order of the node in connection order in the ring of the ring in which the node is included. Channels in which the order of connection in the ring included in the ring is connected to the node in the next order of the node, and channels in which the node is the transmitting side and in which the node is included in the ring other than the ring Relaying data of tasks received from the other node to another node by transmitting to a channel connected to Further, the transmitting unit is a channel for connecting data of a task whose source is a subtask of the task to a node next to the node in the order of connection on the ring of the ring including the node; The data of the task is transmitted to another node by transmitting to a channel having a node as a transmitting side and connected to a node of another ring other than the ring in which the node is included.

According to the distributed processing system as described above, data of a task whose part is the subtask of the task whose task is a subtask of each node is other than the node where the subtask of the source of the data of the task is executed. In the distributed processing system configured to propagate through the network to all other nodes, the control unit broadcasts a task switching instruction accompanied by an identifier of a task of a switching destination to each node. Data propagated through the network can be switched to data of a task of a switching destination of the task switching instruction.

Then, according to general multitask control, in each node, subtasks of tasks whose data is not transferred from other nodes loses or decreases its execution time because there is no data to be processed, The subtasks of a task that is to transfer data from a node increase execution time because data to be processed occurs sequentially.

Therefore, in each node, the task to which the task switching instruction is switched, that is, the subtask of the task in which data is transferred from other nodes in each node is dominantly executed, and distributed processing The tasks to be performed in the system are switched.
Therefore, according to the distributed processing system as described above, it is possible to realize a multitask for executing tasks in a time division manner by a simple configuration that does not need to centrally manage the state of subtasks of each node.

As described above, according to the present invention, in a distributed processing system, it is possible to realize a multitask that executes tasks in a time division manner with a simpler configuration.

FIG. 1 illustrates a distributed system according to an embodiment of the present invention. It is a block diagram showing composition of a node concerning an embodiment of the present invention. It is a block diagram which shows the function of the node which concerns on embodiment of this invention. It is a block diagram showing composition of a task data transfer processing part concerning an embodiment of the present invention. FIG. 5 is a diagram showing data transfer of the distributed system according to the embodiment of the present invention. It is a flow chart which shows transfer data change processing concerning an embodiment of the present invention. It is a figure which shows the process example of the transfer data switching process which concerns on embodiment of this invention. It is a figure which shows the process example of the transfer data switching process which concerns on embodiment of this invention. It is a figure which shows the process example of the transfer data switching process which concerns on embodiment of this invention.

Hereinafter, embodiments of the distributed system according to the present invention will be described.
FIG. 1 shows the configuration of the distributed system according to the present embodiment.
As shown, the distributed system has one controller (CNT) and a plurality of rings (Ri), and each ring i includes the same number of nodes (Nij). Note that Ri represents the i-th ring, and Nij represents the j-th node of the i-th ring.

Here, FIG. 1a shows the case where the number of rings is two and the number of nodes in the ring is three, and FIG. 1b shows the case where the number of rings is three and the number of nodes in the ring is four.
Then, as illustrated, in each ring, a ring-type network is formed in which each node included in the ring is sequentially connected in a ring shape with a channel.
Also, the j-th node of the i-th ring is connected to the j-th node of each ring other than the i-th ring via a bidirectional channel.
On the other hand, the controller (CNT) is directly connected to each node (Nij) of each ring (Ri).
The connection between the controller (CNT) and each node (Nij) is a channel provided for each node (Nij), and the node (Nij) may be connected to the controller (CNT), respectively. The controller (CNT) and each node (Nij) may be connected by a channel.

Next, FIG. 2 shows the configuration of each node.
As illustrated, the node 1 includes a storage 11 and a processor 12. The processor 12 has a configuration as a computer including a CPU and its peripheral devices, and can perform data processing using the storage 11 independently of other nodes 1.

Further, the node 1 is provided with a channel interface 13 connected to the node 1 and provided for each channel to perform data transmission with another node 1 using the channel.

Next, FIG. 3 shows the configuration of functional blocks of node 1 (Nij) formed by execution of software of processor 12.
Subtasks 101 of a plurality of tasks are assigned to the node 1 as the task portion obtained by dividing the task performed by the distributed processing system is referred to as a subtask. Each subtask 101 is executed in a time-division multitasking manner under the control of a multitask control unit 1021 that performs scheduling of subtasks 101, management of resources, etc. on the operating system 102 to provide a multitasking environment. .

Next, each subtask 101 has its own data space 1011 and performs processing using its own data space 1011. Each data space 1011 is provided with a transmission queue SQ for storing data to be transmitted to other nodes, and a reception queue RQ for storing data received from other nodes.

Then, the operating system 102 transfers task data stored in the transmission queue SQ of each subtask 101 to another node 1 and stores data received from another node 1 in the reception queue RQ, task data transfer A processing unit 1022 is provided.

Next, FIG. 4 shows the configuration of the task data transfer processing unit 1022 of the operating system 102.
As illustrated, the task data transfer processing unit 1022 controls the reception buffer 41, the transmission buffer 42, the save buffer 43, the duplicate transfer suppression unit CHK 44, the reception data transfer unit 45, the transmission data transfer unit 46, and the above-described units. A unit 47 is provided.

The reception buffer 41 is provided with a buffer RS (r) in which data received by the channel interface 13 from the node 1 immediately preceding the node on the ring included in the same ring as the own node 1 is stored. There is.

Here, the order on the ring is the order of the node 1 that passes along the ring in the data transfer direction of the channel.
Also, the reception buffer 41 stores the data received by the channel interface 13 from the node 1 having the same order in the ring of the other rings different from the ring in which the own node 1 is included, and the same number as the number of the other rings A buffer Ri (r) is provided.

Next, the transmission buffer 42 stores a buffer RS (s) for storing data to be transmitted via the channel interface 13 to the node 1 one order later in the ring included in the same ring as the own node 1 Have. In addition, the transmission buffer 42 stores data to be transmitted via the channel interface 13 to the node 1 having the same order in the rings of the other rings different from the ring in which the own node 1 is included, the same number as the number of the other rings Buffer Ri (s).

The transfer control unit 47 normally controls the respective units so that the following operation is performed.
That is, the transfer control unit 47 sets, in the transmission data transfer unit 46, the task that is the execution target task as the transmission target task. The execution target task will be described later.
The transmission data transfer unit 46 in which the transmission target task is set includes the data stored in the transmission queue SQ of the data space 1011 of the sub task 101 of the transmission target task, the buffer RS (s) of the transmission buffer 42 and each buffer Ri (s). ), And the order in the ring which is included in the same ring as the own node 1 is the same in the ring in each of the other rings different from the ring in which the own node 1 is included, and the same in the ring Send to each of node 1

Note that, to the data stored in the transmission queue SQ of the data space 1011 of the subtask 101, a task identifier indicating the identification of the task to which the subtask 101 belongs is added.
Also, the transfer control unit 47 receives the data stored in the buffer RS (r) of the reception buffer 41, received from the node 1 that is included in the same ring as the own node 1 and the order on the ring is one before. , Transfer to the duplicate transfer inhibiting unit CHK44. The duplicate transfer inhibiting unit CHK 44 checks whether the transferred data is data of the origin of the own node 1 and outputs the data only when the own node 1 is not the origination of data, and the own node 1 outputs the data. If it is data of the originator, the data is discarded.

Then, the data output from the duplicate transfer inhibition unit CHK 44 is transferred to the buffer RS (s) of the transmission buffer 42 and each buffer Ri (s), and the data is included in the same ring as the own node 1 The order in the ring on the ring and the order in the ring of each other ring different from the ring in which the own node 1 is included are transmitted to each of the same node 1.

The data output from the duplicate transfer inhibition unit CHK44 is also transferred to the received data transfer unit 45, and the received data transfer unit 45 transfers the transferred data to the subtask 101 of the task with the task identifier added to the data. The data is stored in the reception queue RQ of the data space 1011.

Further, data received in each of the nodes 1 having the same order in the ring of the other rings different from the ring including the own node 1 and stored in each Ri (r) is transferred to the received data transfer unit 45, The reception data transfer unit 45 stores the transferred data in the reception queue RQ of the data space 1011 of the subtask 101 of the task with the task identifier added to the data.

Hereinafter, the data transfer operation between the nodes 1 realized by the operation of the task data transfer processing unit 1022 will be described by taking a distributed system in which the number of rings is 2 and the number of nodes in the ring is 3 as an example.

Now, as shown in FIG. 5a, when the task A is a task to be executed in the node N11 of the ring R1, the transmission data transmitted by the subtask 101 of the task A is transferred from the node N11 to the nodes N12 and N21, It is sent to subtask 101 of task A of nodes N12 and N21. The data sent to the node N12 is relayed by the node N12 to the nodes N13 and N22, and sent to the subtask 101 of the task A of the nodes N13 and N22. The data sent to the node N13 is relayed by the node N13 to the node N23, and sent to the subtask 101 of the task A of the node N23.
As a result, when the task A is an execution target task, the transmission data transmitted by the subtask 101 of the task A of the node N11 sequentially propagates the node 1 one by one as described above, and all nodes other than the own node N11 are transmitted. Is transferred to sub task 101 of task A of FIG.

Similarly, as shown in FIG. 5b, when the task B is an execution target task at the node N12 of the ring R1, the transmission data transmitted by the subtask 101 of the task B is, as illustrated, node N12 to node N13. And the node N22 is sent to the subtask 101 of the task B of the node N13 and the node N22. The data sent to the node N13 is relayed by the node N13 to the nodes N11 and N23, and sent to the subtask 101 of the task B of the nodes N11 and N23. The data sent to the node N11 is relayed by the node N11 to the node N21 and sent to the subtask 101 of the task B of the node N21.
As a result, when the task B is the execution target task, the transmission data of the subtask 101 of the task B of the node N12 sequentially propagates the node 1 one by one as described above, and the transmission data of all the nodes 1 other than the own node N12 It is transferred to sub task 101 of task B.

Likewise, similarly, transmission data transmitted by the subtask 101 of the arbitrary task X of the arbitrary node 1 (Nij) of the arbitrary ring is not the own node (Nij) when the task X is an execution target task It is transferred to the subtask 101 of task X of all nodes 1.

Therefore, when sub task 101 of task X of each node 1 of each ring transmits data when task X is an execution target task, the sub task 101 of task X of each node 1 of each ring receives the other data. The data transmitted by the subtask 101 of the task X of all the nodes 1 of the node 1 is transferred.

The operation of switching tasks in such a distributed processing system will be described below.
First, the subtask 101 of each task of each node 1 according to the present embodiment performs the following processes P1 and P2.
Process P1: A predetermined operation is performed, data representing the operation result is stored in the transmission queue SQ, and is transmitted to the subtask 101 of the same task in another node.
Process P2: Take out the data received from the subtask 101 of the same task of another node stored in the reception queue RQ, and carry out an operation of integrating the operation result represented by the data into the operation result of process P1.

However, the operation of the process P2 is performed on the data received from the subtask 101 of the same task of all the other nodes. Furthermore, there is no order dependency in at least the operation P2, and the same result can be finally obtained by processing the data received from the subtask 101 of the same task in another node 1 in any order. It is.

Further, when there is no process to be performed by each subtask 101, that is, the process P1 is ended, and data received from the subtask 101 of the same task of another node to be processed next in process P2 is the reception queue RQ If there is not, then it transitions to the standby state until the data stored in the reception queue RQ is received.

Then, the multitask control unit 1021 of the operating system 102 of each node 1 does not give the execution time (CPU time) to the subtask 101 in the standby state, or reduces the execution time, and is not in the standby state by that amount. The execution time (CPU time) of another subtask 101 is increased.

Referring back to FIG. 1, the switching of tasks in the distributed processing system is performed by the controller (CNT) in accordance with a predetermined schedule.
Further, task switching is performed by the controller (CNT) simultaneously transmitting, to each node 1 (Nij) of each ring, a task switching instruction specifying a task identifier of a task to be switched at the time of task switching.

On the other hand, the transfer control unit 47 of the task data transfer processing unit 1022 of each node 1 (Nij) performs the transfer data switching process shown in FIG. 6 in order to switch tasks.
As illustrated, the transfer control unit 47 monitors the occurrence of reception of the task switching instruction sent by the controller (CNT) (step 602), and if reception of the task switching instruction occurs, the task to be executed is The task is switched to the task of the switching destination indicated by the task identifier of the task switching instruction, and the task that is the execution target task is set as the transmission target task in the transmission data transfer unit 46 (step 604).

Note that when the transmission target task is switched in step 604, as described above, the data stored in the transmission queue SQ of the data space 1011 of the subtask 101 of the transmission target task after switching is in the same ring as the own node 1. The order in the ring of the ring 1 which is included in the ring subsequent to the ring 1 and the ring other than the ring in which the own node 1 is included is transmitted to each of the same node 1; The data stored in the transmission queue SQ of the data space 1011 of the subtask 101 of the transmission target task before switching is not transmitted to other nodes.

Next, the transfer destination of the data stored in the buffer RS (r) of the reception buffer 41 is changed from the duplicate transfer inhibition unit CHK44 to the save buffer 43, and the save buffer 43 receives the data from the buffer RS (r). The transferred data is stored as the save data of the task indicated by the task identifier added to the data, and the save data of the task of the execution target task after switching in step 604 is stored in the save buffer 43 , Transfer to the duplicate transfer inhibiting unit CHK 44 (step 606).

The duplicate transfer inhibiting unit CHK 44 processes the data transferred from the save buffer 43 in the same manner as the data transferred from the buffer RS (r) of the reception buffer 41.
That is, the duplicate transfer inhibiting unit CHK 44 checks whether the data transferred from the save buffer 43 is the data of the origin of the own node, and transmits the data only when the own node is not the data of the origin. The ring that is transferred to the 42 buffers RS (s) and each buffer Ri (s) and is included in the same ring as the own node 1 and is the node 1 after the order on the corresponding ring and the own node 1 And transmit the data to the reception data transfer unit 45 and transmit the data to the reception data transfer unit 45, and the sub-task 101 of the task with the task identifier added to the data. It is stored in the reception queue RQ of the data space 1011.

Then, the transfer destination of the data stored in the buffer RS (r) of the reception buffer 41 is returned to the duplicate transfer inhibition unit CHK 44 (step 608), and the process returns to the monitoring of step 602.
The transfer data transfer process performed by the transfer control unit 47 of the task data transfer processing unit 1022 has been described above.
Here, processing examples of such transfer data switching processing are shown in FIG. 7 to FIG.
In each diagram, Dk (Nij) represents data originating from the subtask 101 of the task k of the j-th node 1 of the i-th ring.
Now, as shown in FIG. 7a, task A is set as an execution target task, and data of task A originating from subtask 101 of task A of each node 1 is node 1 as described above. As shown in FIG. 7b, when a task switching instruction for setting task B as a switching destination is transmitted from the controller (CNT) to each node 1 (Nij) during data propagation of task A while propagating one by one. Thus, in each node 1, data of task A not transferred to other nodes and subtasks 101 received at that time is saved in the save buffer 43 as save data of task A, and data propagation of task A is propagated. Stop.

Also, in each node 1, task B is set as an execution target task, data of task B to which the subtask 101 of task B of each node 1 originates is transmitted, and data of task B is shown in FIGS. 7b and 8a. As shown, node 1 is propagated one by one, and data of task B is transferred to subtask 101 of task B of each node 1.

Then, as shown in FIG. 8b, when a task switching instruction for setting task A as a switching destination is transmitted from the controller (CNT) to each node 1 (Nij) during propagation of data of task B shown in FIG. 8a. In each node 1, data of task B not transferred to other nodes and subtasks 101 received at that time are saved in the save buffer 43 as save data of task B, and the propagation of data of task B is Stop.

Also, the save data of task A that has been saved is transmitted from each node 1, and as a result, as shown in FIG. 8b and FIG. 9, the propagation of data of task A is resumed, and the data of task A is It is transferred to the subtask 101 of task A of each node 1.

As described above, according to the present embodiment, the controller (CNT) transmits a task switching instruction to each node 1 (Nij) to transfer data between each node 1 (Nij) of the distributed processing system. Can switch tasks.

Then, in each node 1, the subtask 101 of the task whose data is no longer transferred from the other nodes 1 is in a standby state because there is no data to be processed in the above-described processing P2, while data is transferred from the other nodes 1 The subtask 101 of the task which has come to be executed comes to generate data to be processed in the above-described process P2.

Therefore, in each node 1, execution of the subtask 101 of the task in which data is transferred from another node 1 in each node 1 to a task switching instruction of the task switching instruction transmitted by the controller (CNT) It becomes dominant and switches tasks to be performed in the distributed processing system.

That is, task switching in the distributed processing system is realized by transmission of a task switching instruction of the controller (CNT).
Therefore, according to the present embodiment, it is possible to realize a multitask for executing tasks in a time division manner by a simple configuration in which there is no need to centrally manage the state of the subtasks of each node.
Although FIGS. 7, 8 and 9 show the case where the nodes 1 operate in synchronization with each other, in actuality, the operation of each node 1 may be asynchronous. Even in this case, when the controller (CNT) transmits a task switching instruction to each node 1 (Nij), the data of the task of the switching destination specified by the task switching instruction is dominant among the nodes 1 (Nij) promptly. Will be transferred to Therefore, in each node 1, the execution of the subtask 101 of the task of the switching destination designated by the task switching instruction is rapidly performed dominantly, and the task to be executed in the distributed processing system is switched.

In the above embodiment, the controller (CNT) transmits, to each node 1 (Nij), a pend instruction specifying a task identifier and a stop instruction specifying a task identifier, in addition to the task switching instruction. May be

The pend instruction is an instruction to temporarily stop the task, and when each node 1 receives the pend instruction, it suspends the transfer of the data to which the task identifier specified by the pend instruction is added, and transfers the data. Save in a resumable form.

The stop instruction is an instruction to stop the task, and each node 1 clears the transmission queue SQ and the reception queue RQ of the subtask 101 of the task indicated by the task identifier designated by the stop instruction when the node 1 receives the stop instruction. When the data to which the task identifier specified by the stop instruction is added is received, the data is discarded.

Reference Signs List 1 node 11 storage 12 processor 13 channel interface 41 reception buffer 42 transmission buffer 43 save buffer 44 duplication transfer inhibition unit CHK 45 reception data transfer unit 46 transmission Data transfer unit 47 Transfer control unit 101 Subtask 102 Operating system 1011 Data space 1021 Multitask control unit 1022 Task data transfer processing unit

Claims

A distributed processing system that executes a plurality of tasks, in which a plurality of nodes are connected by a network,
A control unit that simultaneously transmits to each node a task switching instruction accompanied by an identifier of a task of a switching destination;
Each node is
A subtask execution unit that executes, in multitasking, a subtask that is a part of each of a plurality of tasks executed by the distributed processing system;
The task's subtask of each node originates, and the data of the task of which the subtask is a part passes through the network to all other nodes other than the node where the subtask of the data origin of the task is executed. A data transmission / reception unit that transmits / receives data of the task and the network so as to propagate
A data fetching unit that transfers task data received from the network by the data transmission / reception unit to a subtask of the task;
In response to the task switching instruction, the data transmitting / receiving operation of data of the task of the data transmitting / receiving unit is switched so that data propagated through the network is switched to data of a task indicated by an identifier accompanying the task switching instruction. A distributed processing system comprising: a propagation data switching unit to control.
A distributed processing system for executing a plurality of tasks, comprising a plurality of nodes, comprising:
A control unit that simultaneously transmits to each node a task switching instruction accompanied by an identifier of a task of a switching destination;
Each node is
A subtask execution unit that executes, in multitasking, a subtask that is a part of each of a plurality of tasks executed by the distributed processing system;
A transmitting unit that transmits data originating from a subtask of a task to another node as data of a task of which the subtask is a part;
A data acquisition unit that transfers task data received from another node to a subtask of the task;
A data transfer unit relaying data of a task received from another node to another node such that the data is sequentially transferred from the source node of the data to all other nodes;
In response to the reception of the task switching instruction, the transmitting unit stops transmission of data of tasks other than the task indicated by the identifier associated with the task switching instruction to another node.
The relay unit, in response to the reception of the task switching instruction, relays data of a task not relayed to another node at the point in time received from the other node, the task switching accompanied by an identifier of the task of the data A distributed processing system characterized by deferring until an instruction is received.
The distributed processing system according to claim 2, wherein
The subtask includes each of the transferred data in the data to be processed, performs predetermined processing on the data to be processed, and is in a standby state when the data to be processed is in a waiting state. Transition,
The distributed processing system, wherein the subtask execution unit decreases the execution time of a subtask that has transitioned to a standby state, and increases the execution time of a subtask not in another transition state.
The distributed processing system according to claim 2 or 3, wherein
The relay unit, in response to the reception of the task switching instruction, stores data of a task not relayed to another node at that time received from another node as save data of the task and is stored A distributed processing system, comprising: relaying save data of a task indicated by an identifier associated with the task switching instruction to the other node.
The distributed processing system according to claim 2, 3 or 4, wherein
The data fetching unit is responsive to the task switching instruction to transfer data not transferred to the subtask at that time received from another node to the subtask, the task switching accompanied by an identifier of the task of the data A distributed processing system characterized by deferring until an instruction is received.
A distributed processing system for executing a plurality of tasks, comprising a plurality of nodes, comprising:
A control unit that simultaneously transmits to each node a task switching instruction accompanied by an identifier of a task of a switching destination;
Each node is
A subtask execution unit that executes, in multitasking, a subtask that is a part of each of a plurality of tasks executed by the distributed processing system;
A transmitting unit that transmits data originating from a subtask of a task to another node as data of a task of which the subtask is a part;
While transferring data of a task received from another node to a subtask of the task, it is relayed to the other node so that the data is sequentially transferred from the source node of the data to all other nodes. Data transfer unit, and
In response to the reception of the task switching instruction, the transmitting unit stops transmission of data of tasks other than the task indicated by the identifier associated with the task switching instruction to another node.
The data transfer unit, in response to the reception of the task switching instruction, stores data received from another node at that time as save data of the task, and is stored. 4. A distributed processing system for transferring to the subtask the save data of the task indicated by the identifier accompanied by and relaying to the other node.
A distributed processing system according to claim 2, 3, 4, 5 or 6, wherein
A plurality of rings in which n (where n is an arbitrary natural number of 2 or more) nodes are connected in a ring shape by channels,
The number of nodes included in each ring is equal,
Each node in each ring is a channel whose transmission side is the node, and each node in each ring other than the ring including the node and each node included in the same ring are channels whose transmission side is the node. Connected so as to be connected to different nodes of other rings respectively,
The data transfer means of each of the nodes includes the data of the task received from the channel connected to the node of the previous order of the node on the connection in the ring of the ring including the node. A channel connected to the next node of the node in the order of connection in the ring of the ring and a channel having the node as a transmitting side and connected to a node of another ring other than the ring including the node Relay the task data received from the other node to the other node by transmitting to the
The transmission unit is configured to transmit data of a task whose source is a subtask of the task, a channel connected to a node next to the node in the order of connection on the ring of the ring including the node, and the node A transmission characterized in that data of the task is transmitted to another node by transmitting to a channel that is a transmitting side and is connected to a node of another ring other than the ring that includes the node. Processing system.