JPH1078938A - Dynamic load distribution method and its information processor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce the network traffic and to attain the fast processing for every information processor. SOLUTION: The resources necessary for execution of a task are reserved for each of plural information processors (S2), and one of these processors to which the task is assigned is decided (S4). Then the information processor of the transmitter side transmits the task to the decided information processor (S5), and each of those information processors detects the available period of the said reservation. Then the information processor having an expired available period cancels the reservation for its own resources (S6).

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a dynamic load distribution method for a parallel computer requiring high-speed processing and an information processing apparatus therefor.

[0002]

2. Description of the Related Art FIG. 15 shows a conventional parallel computer system. In FIG. 15, reference numeral 100 denotes a network, and 101 denotes a processing node connected to the network 100 for processing a task.

Conventionally, the dynamic load distribution of such a parallel computer system is described in the literature "Stankovic, JA, Ramaritham, K., Tu.
torial Hard Realtime Systems, IEEE, IEEE Catalog N
umber EH0276-6, ISBN 0-8186-0819-6, 1988 ", when each processing node 101 requests to generate a task under a heavy load of itself, another processing node 101 The amount of resources such as surplus calculation time of the processing node 101 is checked. As a result of the determination as to whether or not processing is possible in the processing node 101, if there is one or more processing nodes 101 determined to be possible, the task is assigned to one of the processing nodes 101. It decides to request, and assigns the task via the network 100.

As a dynamic load distribution method for a parallel computer capable of higher-speed processing, there is Japanese Patent Application No. Hei 7-315394 filed by the same applicant as the present invention. In this dynamic load distribution method, resources required for execution of a task are reserved in advance for each of a plurality of nodes in order to cause another node to process the task. Next, a node to which a task is assigned is determined from those nodes, and the task is transmitted to the node. On the other hand, a reservation release instruction is transmitted to other nodes that have not transmitted the task.

[0005]

In such a conventional dynamic load distribution system, a transmission delay of the network 100 is included until a task is actually allocated after a determination of processing availability and a task allocation determination. Delay occurs. The task is a processing node 101 determined with delay
, The resource status at that time may have changed from the resource status at the time of the determination. That is, the resources that were available at the time of the determination may not be available at the time of transmission due to the allocation of other tasks that are proceeding in parallel. In this case, the transmitted task is rejected in the processing node 101. In this case, another processing node 101 to which the task can be assigned must be searched again. In addition, the transmitted task becomes useless and causes an increase in traffic on the network.

In the dynamic load balancing method for reserving resources in advance in order to prevent retransmission of tasks as described above, it is necessary to transmit a reservation release instruction message in order to release reservations for unused resources. Becomes

As described above, in the conventional parallel computer system, the processing of the transmitted task may be rejected at the determined processing node due to the transmission delay time. Therefore, another processing node must be searched for and retransmitted, which causes a problem of increasing network traffic. In addition, a message must be transmitted to release the resource reservation, and since this reservation release message is transmitted, there is a problem that it is not suitable for higher-speed processing.

The present invention has been made to solve such a problem, and an object thereof is to provide a dynamic load distribution method capable of reducing traffic on a network and performing high-speed processing, and an information processing apparatus thereof. I have.

[0009]

In the dynamic load distribution method according to the present invention, a resource reserving step of reserving resources required for executing a task for each of a plurality of information processing apparatuses, and after the resource reserving step, A determining step that is executed and determines an information processing apparatus to which the task is to be assigned from the plurality of information processing apparatuses; and a transmitting side information processing apparatus having the task transmits the task to the information processing apparatus determined by the determining step. A task transmitting step of performing the above operation, wherein each of the plurality of information processing apparatuses detects an expiration date of the reservation, and the information processing apparatus having the reservation whose expiration date has passed releases the reservation for its own resource. And steps.

In the task transmitting step, the transmitting side information processing apparatus transmits the task to the information processing apparatus determined by the determining step before the expiration date of the reservation, and the reservation canceling step includes the step of releasing the plurality of information processing apparatuses. Each detects the expiration date of the reservation, and among the plurality of information processing devices, the information processing device to which the task has not been transmitted before the expiration date has expired cancels the reservation for its own resource. It is assumed that.

[0011] In the reservation canceling step, each of the plurality of information processing devices detects an expiration date of the reservation, and among the plurality of information processing devices, a task whose task has not been transmitted before the expiration date has passed. Cancels the reservation for its own resource step by step, and the information processing apparatus that has received the task transmitted in the task transmitting step executes a reservation recovery step for restoring the reservation for its own resource released by the reservation release step. It is provided.

The information processing apparatus further includes a determination signal transmitting step of transmitting a reservation determination signal to the information processing apparatus determined by the determination step. The reservation release step includes a step of detecting the expiration date of the reservation by each of the plurality of information processing apparatuses. The information processing apparatus which does not receive the reservation confirmation signal before the expiration date expires cancels the reservation for its own resource.

[0013] In addition, the reservation canceling step includes the step of detecting the expiration date of the reservation by each of the plurality of information processing devices and the information processing device that has not received the reservation confirmation signal before the expiration date of the plurality of information processing devices. A processing device that releases the reservation for its own resource step by step, and the information processing device that receives the reservation confirmation signal includes a reservation recovery step of recovering the reservation for its own resource released by the reservation release step. Things.

A first resource reservation step of reserving a resource required for executing a task in each of the plurality of first information processing devices as a first resource reservation; A first resource check step for checking the first resource information; and a task from among the plurality of first information processing apparatuses, based on the first resource information checked by the first resource check step. An allocation candidate determining step of determining a first information processing apparatus to be allocated as an allocation candidate; and a transmission information processing apparatus having the task, which is executed after the completion of the allocation candidate determining step, sends a first reservation extension signal to the allocation candidate. Transmitting a first reservation extension signal to be transmitted, each of the plurality of first information processing devices detecting an expiration date of the first resource reservation, and The first information processing device, to which the first reservation extension signal has not been transmitted before the expiration date has passed, is the first resource reservation for its own resource reserved in the first resource reservation step. A second resource reservation step of reserving, as a second resource reservation, resources required for the execution of the task in the second information processing apparatus; A second resource examination step of examining second resource information of the device; and, based on the first resource information and the second resource information of the assignment candidate, the assignment candidate and the second information processing device. An assignment destination determining step of determining an information processing apparatus to which a task is to be allocated as an allocation information processing apparatus, and a second reservation in which the transmission-side information processing apparatus transmits a second reservation extension signal to the allocation information processing apparatus A long signal transmitting step, wherein the second information processing device detects an expiration date of the second resource reservation, and the second reservation extension signal is transmitted until the expiration date of the second resource reservation expires. If not, a second reservation release step of releasing the second resource reservation, and until the time after the allocation candidate receives the first reservation extension signal elapses a predetermined time. A third reservation release step for releasing the first resource reservation for the own resource reserved in the first resource reservation step when the second reservation extension signal is not received; And transmitting the task to the assigned information processing apparatus after the reservation extension signal transmitting step is completed.

Further, in the information processing apparatus according to the present invention, a reservation means for receiving a reservation for its own resource from another information processing apparatus which requests execution of a task, and for reserving its own resource for the received reservation, A reservation canceling means for canceling the reservation when the task is not received before the expiration date of the reservation;
A resource allocating unit that allocates the resource reserved by the reservation unit to the task when the task is received from the other information processing apparatus.

The reservation canceling means cancels the reservation step by step when the expiration date of the reservation has passed, and when the task is received from another information processing apparatus, reserves the resource canceled by the reservation canceling means. Is provided with a reservation recovery means for recovering the data.

Also, when there is a request for a new reservation, a threshold changing means for changing a threshold of the resource amount that can be reserved, based on the number of reservations received from another information processing apparatus, When the resource amount of the new reservation and the received reservation does not exceed the threshold, a reservation unit that reserves its own resource for the new reservation, and when the expiration date of the new reservation has passed, A reservation canceling means for canceling the new reservation; and a resource allocating means for allocating a resource reserved by the reservation means to the task when a task is received from the information processing apparatus requesting the new reservation. Things.

[0018]

BEST MODE FOR CARRYING OUT THE INVENTION

Embodiment 1 FIG. FIG. 1 shows a parallel computer system which is a platform common to the first to ninth embodiments. In FIG. 1, reference numerals 1A to 1D denote processors for processing tasks, and 2A to D denote memories for storing programs and data necessary for processing the tasks.
A to D are processors 1A to 1D and memories 2A to 2D, respectively.
D, EDUs 4A to 4D, and timers 7A to 7D are interfaces for controlling input and output. Reference numerals 4A to 4D denote EDUs (External Distribution Units) for connecting the interfaces 3A to 3D and the network 5, respectively, and transmit data from the interfaces 3A to 3D to the network 5 and send the data to the own information processing apparatus flowing on the network 5. And outputs it to the interfaces 3A to 3D. 6A to 6D are memories 2A to
D is a table provided in each of the nodes D, and stores information on the total resource amount, the used resource amount, the free resource amount, and the reserved resource amount of the nodes 8A to 8D. 7A to 7D are timers for measuring time.

8A to 8D are processors 1A to 1D, respectively.
, Memory 2A-D, interface 3A-D, ED
The nodes are information processing devices having U4A to U4D and timers 7A to 7D, and each of the nodes 8A to 8D has the ability to process tasks independently. Here, the network 5 is used as a signal line for connecting each information processing device. However, any network capable of transmitting information, such as a bus, may be used.

For example, when a task occurs in the node 8A, the task is assigned to any one of the nodes 8 (A to D). The dynamic load distribution processing in this case is described below with reference to FIGS.
This will be described with reference to FIG.

FIG. 2 is a sequence diagram illustrating the flow of the dynamic load distribution processing according to the first embodiment. 2, the same reference numerals as those in FIG. 1 denote the same or corresponding parts. FIG. 3 is a flowchart illustrating a task transmission process performed by the task transmission-side node according to the first embodiment. 3, the same reference numerals as those in FIG. 2 denote the same or corresponding processes. FIG. 4 is a flowchart illustrating a task receiving process of the task receiving node according to the first embodiment. 4, the same reference numerals as those in FIG. 2 denote the same or corresponding processes.

First, a task is generated in the node 8A, and it is determined whether the task is to be processed by the own node 8A or one of the other nodes 8B to 8D. This processing will be described with reference to FIGS.

First, in step S1, the task transmitting node 8A transmits a task allocation request message to the other nodes 8B to 8D. This message is broadcast by the processor 1A over the network 5 via the EDU 4A. By this broadcast, a plurality of nodes 8B to 8D receive the allocation request message at a time. The reception is performed by the EDUs 4B to 4D, respectively, and is normally performed even while the processors 1B to 1D are processing other tasks. The assignment request message includes information on resources necessary for executing the task (hereinafter referred to as resource information). Each of the other nodes 8B to 8D determines whether or not the task can be accepted based on the resource information of the received assignment request message.

Next, the process proceeds to step S2, and the above-described step S2 is performed.
If it is determined in step 1 that the task can be accepted, the resource described in the resource information is reserved so that other tasks cannot use the resource. If it is determined that the request cannot be accepted, no resource reservation is made.

Subsequently, in step S3, the nodes 8B and 8C that have determined that the task can be accepted transmit an assignment OK message to the effect that the assignment is accepted to the task transmitting node 8A. Here, the node 8D that has not been accepted does not transmit a message to reduce network traffic. The transmitting node 8A receives the allocation OK message until a predetermined time, and proceeds to the next step S4.

In step S4, the node to which the task is to be allocated is determined from the nodes 8B and 8C from which the transmitting node 8A has transmitted the allocation OK message. This determination may be made by selecting a node at random or by selecting a node having the most resource. The node 8B determined as the task assignment destination in step S4 is
It becomes an assignment information processing device.

Next, the transmitting node 8A transmits the task body to the network 5 in step S5. The transmission is performed by designating the node 8B as the assigned information processing apparatus as the transmission destination. The node 8B receives the task via the network 5, and allocates its own resource reserved in step S2 to the received task. Then, at node 8B,
The task is performed.

On the other hand, the node 8C that has not received the task
Determines in step S6 whether the expiration date of the reservation has passed, and if it has passed, cancels the reservation. Here, the resource reservation performed in step S2 is released after a predetermined time T (hereinafter, referred to as a reservation release time T) has elapsed since the reception of the task allocation request message. The measurement of the reservation release time T is performed using the timer 7C. The release of the reservation allows other tasks to use the resource. The expiration date of the reservation only needs to be based on the reception of the task assignment request message,
The time may be set at a predetermined time after the resource is reserved, or a predetermined time may be set after transmitting the allocation OK message.

At this time, the reservation is automatically released after the reservation release time T has elapsed, so that the transmission of the reservation release message via the network 5 is not performed. Therefore, the number of messages transmitted and received on the network 5 can be reduced. That is, network traffic can be reduced. Further, in a parallel processing device that performs dynamic load distribution, data is frequently exchanged between the nodes 8A to 8D, and many nodes 8A to 8D use the network 5. Therefore, when traffic on the network 5 increases, data The time required to wait for transmission / reception increases, and the processing speed decreases. Conversely, by reducing the traffic of the network 5, the processing efficiency of the entire parallel processing device is improved,
High-speed processing becomes possible.

{Processing of Task Sending Node} Next, the above-mentioned dynamic load distribution processing is performed by the task sending node 8A.
The following description focuses on the processing of FIG. First, in step S101 of FIG. 3, the node 8A generates a task. This task may have been received from outside the node. Next, the process proceeds to step S102, where it is determined whether the task can be processed by the own node. The method of determining whether or not processing is possible is performed, for example, by determining whether the surplus calculation time and free memory are sufficient to execute the task. If it is determined that the processing can be performed by the own node, the process proceeds to step S103. If it is determined that the process cannot be performed, the process proceeds to step S1-1.

In step S103, the task is executed on the own node. When the execution is completed, step S10
Return to 1. However, this node has multiple processes,
Threads can be processed in parallel. Therefore, even if the processing in step S103 is not completed, step S10
1 is performed, and tasks are generated one after another.

On the other hand, when it is determined in step S102 that the other nodes 8B to 8D process the task,
As shown in step S1, the task assignment request message is transmitted and received. In the node 8A on the task transmission side, the step S1-1 in FIG.
A task assignment request signal is transmitted to B to D. This allocation request message is, as shown by M1 in FIG.
D, message type, source, destination, required resources, and other data. Here, information indicating "assignment request" is entered in the message type, and information indicating that the source is the node 8A on the task transmitting side and the destination is broadcast to the nodes 8B to 8D. In the data part,
A task number for identifying the task, a calculation time required for executing the task, and necessary resource information such as a memory are described.

Next, in step 3-2, the control waits for an allocation OK message for a predetermined time. The reception waiting time may be changed according to a task or a condition such as the number of nodes. The allocation OK message includes the task number as shown in M2 of FIG. By identifying the task number, it is possible to determine whether the assigned OK message is for the task.

Subsequently, in step S4, among the nodes 8B and 8C that have transmitted the allocation OK message, the node 8 that transmits the task is determined as an allocation information processing apparatus (hereinafter, referred to as an allocation node). Next, step S5-1
And sends a task transmission message to the assignment node. This task transmission message has a structure as shown in M3 of FIG. 5, the identifier of the assigned node is specified as the transmission destination, and the data portion includes the task body.

Thus, the dynamic load distribution processing is completed. Then, a new task is generated again in step S101,
Similar processing is performed. In the processing of the task transmission side node 8A, since the reservation release message is not transmitted to a node other than the allocation node, the processing load of the task transmission side node 8A is light.

(3) Processing of Another Node (Task Receiving Node) Next, the above dynamic load distribution processing is performed by the node 8 on the task receiving side.
The following description focuses on processes B to D.

First, in step S1-2 in FIG. 4, a task allocation request message is received, and it is determined whether the request can be accepted. If it is determined that the request cannot be accepted, the subsequent processing is not performed, and a new request arrives. If it is determined that it can be accepted, the process proceeds to the next step S1-3. In step S1-3, measurement of the time t is started using each of the timers 7B to 7D. That is, the measurement of the time length is started. This step S1-
Step 3 may be performed after the next step S2. continue,
In step S2, resources required for executing the task are reserved.
The resources to be reserved determine the required amount of resources from the resource information included in the task allocation request message, and reserve the resources managed by the own node by the determined amount.

Subsequently, the flow shifts to step S3-1, where the assignment O
Send K message. Next, the process proceeds to step S5-2, where it is checked whether the EDUs 4B to 4D have received the task. If it has been received, the process proceeds to step S5-4 and proceeds to step S2.
Allocate the reserved resource to the received task. Subsequently, the task is executed in step S111.

On the other hand, if it is determined in step S5-2 that the task has not been received, the process proceeds to step S5-3, and the time t at which the measurement is started from step S1c exceeds the reservation release time T. Is detected. If the time t has not passed the reservation release time T, step S5-
Return to 2. If the reservation release time T has passed, the process proceeds to step S6. In step S6, the reservation of the resource made in step S2 is released, and the resource can be used for another task.

As described above, since the reservation is released without receiving the message for releasing the resource reservation, there is an effect that the network traffic is reduced and the processing speed of the entire system is improved.

Embodiment 2 In the first embodiment, the task assignment request message is broadcast (broadcast). However, the task assignment request message may be sequentially transmitted to each node. The operation of the second embodiment will be described below with reference to FIG.
FIG. 6 is a sequence diagram illustrating the flow of the dynamic load distribution processing according to the first embodiment. 6, the same reference numerals as those in FIG. 2 denote the same or corresponding parts.

First, in step S1a, the task transmitting node 8A transmits a task allocation request message. In this task allocation request message, one node 8B is designated and transmitted. Next, in step S2a, the node 8B makes a resource reservation in the same manner as described in step S2 of FIG. Then, step S3
At a, the allocation OK message is transmitted from the node 8B to the task transmitting node 8A. When task assignment is impossible, this assignment OK message is not transmitted.

Next, steps S1b, S2b, and S3
At b, the task transmitting node 8A transmits a task allocation request message to another node 8C, and checks whether the task can be allocated. This investigation is performed in step S1 described above.
a, S2a, and S3a. Similarly, it is checked whether the other nodes 8D can be assigned tasks. This check is performed in the same manner as in steps S1c, S2c, and S3c described above.

Next, the process proceeds to step S4, where the assignment destination is determined, and in step S5, the task is transmitted to the assignment node 8B as the assignment information processing device. Each of the nodes 8B to 8D measures the time t after receiving the task assignment request message, and the task transmission in step S5 is performed before the measured time t elapses the reservation release time T. The nodes 8C and 8D to which the tasks have not been transmitted by the reservation release time T release the resource reservation in step S6.

According to the second embodiment, the reservation is released without receiving the message for releasing the resource reservation, so that there is an effect that the network traffic is reduced and the processing speed of the entire system is improved. .

The allocation destination determining step S3 may be performed each time a new allocation OK message arrives. Then, when an information processing device (that is, a node) that satisfies a predetermined resource condition is found, transmission of a task assignment request message to another node may not be performed. In this case, the task request message and the assignment O
There is an effect that traffic generated by the K message can be suppressed low. The above resource conditions are:
This is a resource condition that allows the task to be executed with a margin, and is determined based on the amount of resources necessary for executing the task. For example, the surplus calculation time or memory of the node is calculated as the calculation time required for executing the task or 20 times of the memory.
This is a condition that 0% is present.

Embodiment 3 The third embodiment is an embodiment in which a reservation confirmation message for confirming a reservation is transmitted so that the reservation is not released even if the task is not transmitted before the reservation release time T.

FIG. 7 is a sequence diagram showing a dynamic load distribution process according to the third embodiment. 7, the same reference numerals as those in FIG. 2 indicate the same or corresponding parts.

Since the basic operation of the third embodiment is the same as that of the first embodiment, the characteristic operation of the third embodiment will be described below. When the processes in steps S1 to S4 are completed and the assignment node is determined, in step S41, the task transmitting node 8A transmits a reservation confirmation message to the assignment node. The node 8B that has received the reservation confirmation message determines the resource reservation performed in step S2, and prevents the reservation from being released in step S6.

In nodes 8C and 8D to which the reservation confirmation message has not been transmitted, the reservation release processing is performed in step S6.

After the end of step S41, task transmission is performed in step S5. At the allocation node 8B,
In step S41, the resource whose reservation has been confirmed is allocated to the received task.

As described above, in the third embodiment, even when a task cannot be transmitted within the reservation release time T, there is an effect that the network traffic is reduced and the processing speed of the entire system is improved.

The reservation confirmation message according to the third embodiment can be applied to the second embodiment. FIG. 8 is a sequence diagram illustrating the dynamic load distribution processing in this case.
8, the same reference numerals as those in FIG. 6 denote the same or corresponding parts. Step S41 is the same as step S41 in FIG.
The above-mentioned effect can be obtained by this reservation confirmation message.

Embodiment 4 Embodiment 4 is an embodiment in which the reservation of the current assignment node candidate is extended until a good assignment node is found.

FIG. 9 is a sequence diagram showing a dynamic load distribution process according to the fourth embodiment. 9, the same reference numerals as those in FIG. 6 denote the same or corresponding parts.

Since the basic operation of the fourth embodiment is the same as that of the second embodiment, the characteristic operation of the fourth embodiment will be described below. Steps S1a to S3b
Up to this point, as a first resource checking step, the resource statuses of the plurality of nodes 8B and C as the first information processing device are checked, and the first resource reservation step is executed. Next, the process proceeds to step S4a, where allocation destination determination processing, which is an allocation candidate determination step, is performed. In this assignment destination determination processing, a node having the best resource status is determined as an allocation candidate (hereinafter, referred to as an allocation node candidate) from the nodes 8B and 8C that have transmitted the allocation OK message. Whether the resource situation is excellent or not
For example, how to determine whether there is a lot of surplus resources,
There is a method in which the task is determined based on conditions such as execution of the task can be performed faster than other nodes.

Subsequently, a predetermined reservation release time T1
Before elapses, a reservation extension message, which is a first reservation extension signal, is transmitted to the allocation node candidate in step S42a. Upon receiving the reservation extension message, the allocation node candidate 8B clears the time measured from step S1a and starts a new time measurement.

Next, after transmitting the reservation extension message, the second
In order to search for a node to be an information processing device of the above, that is, a node to which a task can be newly assigned, steps S1c and S3c as a second resource investigation step and resource reservation S2c as a second resource reservation step are executed. Next, the process proceeds to step S4b, which is an assignment destination determination step, and after transmitting the reservation extension message, the newly discoverable assignable node 8
D, and compares the resource status of the assigned node candidate 8B determined in step S4a with those nodes 8C,
D is searched for a node with an excellent resource status, and an allocation node candidate 8D is newly determined.

The task transmitting node 8A determines that the time from the arrival of the reservation extension message to the assigned node candidate 8B is
Before the predetermined reservation release time T2 elapses, the second
Is transmitted as a second reservation extension signal transmission step. As described above, the reservation of the assigned node is extended while checking the resource status of other nodes one after another. The search for the assigned node and the extension of the reservation are repeated as long as time permits, and the optimum assigned node is found.

Here, it is assumed that there is no time after one reservation extension, and the allocation node candidate 8D determined in the final allocation destination determination step S4b is determined as the allocation node. In this case, the task is transmitted in step S5 instead of the reservation extension message described above. The task transmission is performed by the task transmission side node 8A by the assignment destination determining step S
This is a process of transmitting a task to the allocation node 8D determined in 4b.

On the other hand, the resource reservation that has passed the reservation release time T1 is released in step S6b, which is the first reservation release step. Similarly, the resource reservation after the reservation release time T2 is released in step S6a, which is the third reservation release step. The case where the task is transmitted to the allocation node 8D as the second information processing apparatus has been described above. However, when the task is transmitted to the allocation candidate node 8B, the second information processing is performed as a second reservation release step. At the allocation node 8D as a device, the reservation is released as in step S6b. This eliminates the need for a reservation release message as in the above-described embodiment.

As described above, according to the fourth embodiment, the reservation is released without receiving the message for releasing the resource reservation, so that the network traffic is reduced and the processing speed is improved as a whole system. There is an effect of doing.

In the first embodiment, the reservation release time T
Although it is necessary to set a time sufficient to receive the allocation OK message from many nodes, in the fourth embodiment, the reservation release time T1 or T2 can be set shorter. The reason is, for example, T
This is because in 1 it is sufficient to set a time sufficient to receive the allocation OK message from at least two nodes, and in T2 it is sufficient to set a time sufficient to receive the allocation OK message from at least one node. However, since the reservation extension message causes an increase in traffic, at this point, the number of nodes that can be investigated in one reservation extension and the lengths of the reservation release times T1 and T2 are adjusted. Note that the same time may be set for T1 and T2.

As described above, since the reservation release times T1 and T2 can be set short, the system as a whole
The effect is that the amount of reserved resources can be reduced and many tasks can be executed.

The task assignment request message is 1
Instead of individually transmitting to each of the nodes 8B to 8D, one assignment request message may be received by a plurality of nodes as shown in FIG. 10, the same reference numerals as those in FIG. 9 represent the same or corresponding parts. The task assignment request message in step S1 is the same as the task assignment request message in step S1 shown in FIG. Therefore, one task allocation request message in step S1 is used to execute step S1 in FIG.
It plays the role of request messages for the tasks a, S1b and S1c.

Embodiment 5 Embodiment 5 is an embodiment in which reserved resources are deleted step by step. FIG. 11 is a sequence diagram showing a dynamic load distribution process according to the fifth embodiment. 11, the same reference numerals as those in FIG. 2 denote the same or corresponding parts.

Since the basic operation of the fifth embodiment is the same as that of the first embodiment, the characteristic operation of the fifth embodiment will be described below. In steps S1 to S4, an allocation node is determined from the plurality of nodes 8B to 8D. Next, the task transmission is executed. Here, it is assumed that the task transmission step S5 cannot transmit the task by the reservation release time T in preparation for the task transmission.

In steps S9a, each of the nodes 8B to 8D that have not received the task by the reservation release time T releases a part of the resource reservation made in step S2.
In this case, 20% of the reserved resource amount is released. Next, after a lapse of a predetermined time from step S9a, each of the nodes 8B to 8D performs step S9.
9b is executed. In step S9b, step S2
Releases part of the resource reservation made in. Here, 50% of the reserved resource amount is released. Subsequently, after a predetermined time has elapsed from step S9b, the node 8
Step S9c is performed in each of B to D. In step S9c, all the resource reservations made in step S2 are released. However, at the node where the reservation has been confirmed or the task has been transmitted, the reservation cancellation at these steps S9a to S9c is not performed.

The stepwise cancellation of the reservation in steps S9a to S9c is not only divided into three as described above, but also a predetermined number of times is appropriately determined in advance, and the reservation is deleted stepwise according to the predetermined number of times. be able to. Also, the amount of one deletion is appropriately set. Here, "gradual release"
This means that a part of the resource reservation reserved for the own node is gradually released, and finally all the reservations for the resource of the own node are released.

On the other hand, when step S5 is executed at the timing shown in FIG. 11, the node 8B that has received the task recovers the deleted reserved resource in step S10, and sets the resources necessary for executing the task. Return the quantity to a fully reserved state. After step S10, the recovered reservation resource is allocated to the received task and the task is executed.

In the fifth embodiment, even if the transmission of a task has passed the reservation release time T, resources are deleted step by step, so that there is an effect that the possibility of assignment of a task is high. This is because it is more likely that a small resource can be secured than a large resource.

Further, similarly to the third embodiment, a reservation confirmation message may be transmitted, and the resource reservation deleted by the node receiving the reservation confirmation message may be recovered. FIG. 12 is a sequence diagram illustrating the dynamic load distribution processing in this case. 12, the same reference numerals as those in FIG. 11 denote the same or corresponding parts. When the reservation release time T1 has passed, as described in steps S9a to S9c above,
Release reservation resources step by step. On the other hand, the node that has received the reservation confirmation message does not perform the stepwise release of the reservation, or stops the stepwise release, and performs the reservation recovery in step S10.

Further, similarly to the fourth embodiment, a reservation extension message may be transmitted to recover the resource reservation deleted by the node which has received the reservation extension message. FIG. 13 is a sequence diagram illustrating the dynamic load distribution processing in this case. 13, the same reference numerals as those in FIG. 11 denote the same or corresponding parts. When the reservation release time T1 has passed, as described in steps S9a to S9c above,
Release a reserved resource. On the other hand, when the reservation extension message is transmitted after the reservation release time T1 has passed,
A new reservation release time T2 is set, and the above-described reservation recovery step S10 is executed. Similarly, when there is no reservation extension message or task transmission before the reservation release time T2 elapses, the reservation resource is also released in stages.

Embodiment 6 FIG. The sixth embodiment is an embodiment in which a reservation is received beyond the amount of free resources held by the node. FIG. 14 is a flowchart of a task receiving process of the receiver-side system according to the sixth embodiment. This flowchart describes step S1-2 in FIG. 4 in detail.

First, in step S1-21, the threshold value Th
Is calculated. This threshold value Th is set based on the number of reservations received from another information processing device. For example,
When the number of accepted reservations is 1, the resource amount is 100% of the free resource amount, and when the number of reservations is 5,
When 150% of the free resource amount and the number of reservations are 10, 200% of the free resource amount is the threshold Th.
Is set as

Next, in step S1-22, a task allocation request message is received. Subsequently, step S1-2
In step 3, it is determined whether the resource amount requested by the task allocation request received in step S1-22 is larger than a threshold Th. If it is larger, it is determined that the task cannot be assigned,
It returns to step S1-21. Returning to step S1-21, a new threshold value Th is set based on the latest available resource amount. On the other hand, if it is smaller, it is determined that task assignment is possible, and the routine goes to step S1-3 in FIG.

Here, the threshold value T is set according to the number of resource reservations.
Since h is changed, there is a high possibility that a task will be accepted even if a reservation is accepted beyond the amount of free resources. That is, the resource reservation may be canceled in time, and the probability increases as the number of nodes that can accept the task increases. Therefore, many resources that are reserved but not used occur, and the number of tasks that can be executed in the entire system is limited. On the other hand, when a plurality of reservations are received, the more the number of reservations increases, the higher the possibility that one of the reservations will be released will be increased. For the above reasons, in the sixth embodiment, the threshold value Th is changed according to the number of resource reservations, so that the reservation of the resource amount exceeding the free resource amount can be accepted and many tasks can be executed. I have.

[0078]

Since the present invention is configured as described above, it has the following effects.

In the dynamic load distribution method according to the present invention, a resource reservation step for reserving resources required for executing a task in each of a plurality of information processing apparatuses, and the plurality of information processing apparatuses are executed after the resource reservation step. A determining step of determining an information processing apparatus to which the task is to be assigned from among the information processing apparatuses, and a task transmitting step of transmitting the task to the information processing apparatus in which the determining step is determined by the transmitting-side information processing apparatus having the task; Each of the plurality of information processing devices detects an expiration date of the reservation,
The information processing device having the reservation after the expiration date,
A reservation canceling step of canceling the reservation for its own resource, and canceling the time limit of the reservation, thereby reducing network traffic and enabling high-speed processing.

In the task transmitting step, the transmitting side information processing apparatus transmits the task to the information processing apparatus determined by the determining step before the expiration date of the reservation. Each detects the expiration date of the reservation, and among the plurality of information processing devices, the information processing device to which the task has not been transmitted before the expiration date has expired releases the reservation for its own resource and cancels the reservation. Since the time limit is canceled, network traffic is reduced, and high-speed processing can be performed.

In the reservation canceling step, each of the plurality of information processing devices detects the expiration date of the reservation, and among the plurality of information processing devices, the information processing device whose task has not been transmitted before the expiration date has passed. Cancels the reservation for its own resource step by step, and the information processing apparatus that has received the task transmitted in the task transmitting step executes a reservation recovery step for restoring the reservation for its own resource released by the reservation release step. In order to cancel the reservation, the network traffic is reduced, high-speed processing is possible, and even if a task is assigned after a predetermined time has elapsed, resources can be recovered, and the task can be recovered. More likely to be able to do it.

The information processing apparatus further includes a determination signal transmitting step of transmitting a reservation determination signal to the information processing apparatus determined by the determination step. The reservation canceling step includes the steps of: detecting the expiration date of the reservation by each of the plurality of information processing apparatuses; Of the information processing devices, the information processing device that did not receive the reservation confirmation signal before the expiration date has expired releases the reservation for its own resource, and cancels the reservation in a time-limited manner, thereby reducing network traffic, Since high-speed processing is possible and the reservation confirmation can be notified in advance, the reservation is not released even if the task transmission takes time.

In the reservation canceling step, each of the plurality of information processing devices detects the expiration date of the reservation, and information of the plurality of information processing devices for which the reservation confirmation signal has not been received before the expiration date has passed. A processing device that gradually releases the reservation for its own resource, the information processing device that receives the reservation confirmation signal includes a reservation recovery step for recovering the reservation for its own resource released by the reservation release step, Since the reservation is canceled, the network traffic can be reduced, high-speed processing can be performed, and even if a task is assigned after a predetermined time has elapsed, there is a high possibility that resources can be recovered.

A first resource reservation step of reserving a resource required for executing a task as a first resource reservation in each of the plurality of first information processing apparatuses; A first resource check step for checking the first resource information; and a task from among the plurality of first information processing apparatuses, based on the first resource information checked by the first resource check step. An allocation candidate determining step of determining a first information processing apparatus to be allocated as an allocation candidate; and a transmission information processing apparatus having the task, which is executed after the completion of the allocation candidate determining step, sends a first reservation extension signal to the allocation candidate. Transmitting a first reservation extension signal to be transmitted, each of the plurality of first information processing devices detecting an expiration date of the first resource reservation, and The first information processing device, to which the first reservation extension signal has not been transmitted before the expiration date has passed, is the first resource reservation for its own resource reserved in the first resource reservation step. A second resource reservation step of reserving, as a second resource reservation, resources required for the execution of the task in the second information processing apparatus; A second resource examination step of examining second resource information of the device; and, based on the first resource information and the second resource information of the assignment candidate, the assignment candidate and the second information processing device. An assignment destination determining step of determining an information processing apparatus to which a task is to be allocated as an allocation information processing apparatus, and a second reservation in which the transmission-side information processing apparatus transmits a second reservation extension signal to the allocation information processing apparatus A long signal transmitting step, wherein the second information processing device detects an expiration date of the second resource reservation, and the second reservation extension signal is transmitted until the expiration date of the second resource reservation expires. If not, a second reservation release step of releasing the second resource reservation, and until the time after the allocation candidate receives the first reservation extension signal elapses a predetermined time. A third reservation release step for releasing the first resource reservation for the own resource reserved in the first resource reservation step when the second reservation extension signal is not received; A transmitting step in which the transmitting side information processing apparatus transmits the task to the assigned information processing apparatus, and the network traffic is transmitted in order to cancel the time limit of the reservation. In addition to reducing the number of resources and enabling high-speed processing, the resources are reserved and released separately in the first resource reservation and the second resource reservation, so that the amount of resources reserved at a time is reduced as a whole system. Can be.

Further, in the information processing apparatus according to the present invention, a reservation means for receiving a reservation for its own resource from another information processing apparatus which requests execution of a task, and for reserving its own resource for the received reservation, A reservation canceling means for canceling the reservation when the task is not received before the expiration date of the reservation;
Resource receiving means for allocating the resource reserved by the reservation means to the task when the task is received from the other information processing apparatus. Processing becomes possible.

The reservation canceling means cancels the reservation step by step when the expiration date of the reservation has passed, and when the task is received from another information processing apparatus, reserves the resource canceled by the reservation canceling means. It has a reservation recovery means to recover the time limit, cancels the reservation time limit, reduces network traffic, enables high-speed processing, and even if a task is assigned after a predetermined time has elapsed It is highly possible to recover resources.

Further, when there is a request for a new reservation, a threshold changing means for changing the threshold of the resource amount which can be received based on the number of reservations received from another information processing apparatus. When the resource amount of the new reservation and the received reservation does not exceed the threshold, a reservation unit that reserves its own resource for the new reservation, and when the expiration date of the new reservation has passed, A reservation canceling means for canceling the new reservation, and a resource allocating means for allocating a resource reserved by the reservation means to the task when a task is received from the information processing apparatus which has requested the new reservation, In order to cancel the reservation, the network traffic is reduced, high-speed processing is enabled, and the threshold for the amount of resources that can be reserved is changed based on the number of reservations. The number of reservations to be kicked is increased, and it is possible to perform many tasks while suppressing the phenomenon can not execute the transmission task.

[Brief description of the drawings]

FIG. 1 is a functional block diagram showing a parallel computer system according to Embodiment 1 of the present invention.

FIG. 2 is a sequence diagram showing dynamic load distribution processing according to the first embodiment of the present invention.

FIG. 3 is a flowchart of a task assignment process of the center-side system according to the first embodiment of the present invention.

FIG. 4 is a flowchart of a task receiving process of the receiver-side system according to the first embodiment of the present invention.

FIG. 5 is a diagram showing the contents of various messages according to the first embodiment of the present invention.

FIG. 6 is a sequence diagram showing a dynamic load distribution process according to the second embodiment of the present invention.

FIG. 7 is a sequence diagram showing dynamic load distribution processing according to Embodiment 3 of the present invention.

FIG. 8 is a sequence diagram showing another dynamic load distribution process according to the third embodiment of the present invention.

FIG. 9 is a sequence diagram showing dynamic load distribution processing according to Embodiment 4 of the present invention.

FIG. 10 is a sequence diagram showing another dynamic load distribution processing according to the fourth embodiment of the present invention.

FIG. 11 is a sequence diagram showing dynamic load distribution processing according to Embodiment 5 of the present invention.

FIG. 12 is a sequence diagram showing another dynamic load distribution process according to the fifth embodiment of the present invention.

FIG. 13 is a sequence diagram showing another dynamic load distribution process according to the fifth embodiment of the present invention.

FIG. 14 is a flowchart of a task receiving process of a receiver-side system according to Embodiment 6 of the present invention.

FIG. 15 is a functional block diagram showing a parallel computer system according to a conventional technique.

[Explanation of symbols]

1A-D processor, 2A-D memory, 3A-D
Interface, 4 EDU, 5 networks,
6A-D table, 7A-D timer, 8A-
D node

Claims (9)

[Claims] A resource reserving step of reserving resources required for executing a task for each of a plurality of information processing apparatuses; and a task executed after the resource reserving step, and allocating the task from the plurality of information processing apparatuses. A determining step of determining an information processing device, a task transmitting step in which a transmitting-side information processing device having the task transmits the task to the information processing device determined by the determining step, and each of the plurality of information processing devices A dynamic load balancing method comprising: a reservation release step of detecting an expiration date of a reservation and causing an information processing apparatus having the reservation whose expiration date has passed to release the reservation for its own resource. 2. The task transmitting step, wherein the transmitting side information processing apparatus transmits a task to the information processing apparatus determined by the determining step until the expiration date of the reservation has passed, and the reservation releasing step includes: Each detects the expiration date of the reservation, and among the plurality of information processing devices, the information processing device to which the task has not been transmitted before the expiration date has expired cancels the reservation for its own resource. The dynamic load distribution method according to claim 1, wherein 3. The reservation canceling step includes the steps of: detecting the expiration date of the reservation by each of the plurality of information processing devices; and transmitting the task of the plurality of information processing devices to which the task has not been transmitted before the expiration date has passed. Cancels the reservation for its own resource step by step, and the information processing apparatus that has received the task transmitted in the task transmitting step performs a reservation recovery step of restoring the reservation for its own resource released in the reservation release step. 3. The dynamic load distribution method according to claim 2, further comprising: 4. A reservation signal transmitting step of transmitting a reservation confirmation signal to the information processing device determined by the determining step, wherein the reservation releasing step includes detecting each of the plurality of information processing devices to detect the expiration date of the reservation. 2. The dynamic load according to claim 1, wherein the information processing device that has not received the reservation confirmation signal before the expiration date expires cancels the reservation for its own resource. Dispersion method. 5. A reservation canceling step, wherein each of the plurality of information processing devices detects an expiration date of a reservation, and information of the plurality of information processing devices not receiving a reservation confirmation signal before the expiration date has passed. A processing device that gradually releases the reservation for its own resource, and the information processing device that receives the reservation confirmation signal includes a reservation recovery step for recovering the reservation for its own resource released by the reservation release step. 5. The dynamic load distribution method according to claim 4, wherein: 6. A first resource reservation step of reserving, as a first resource reservation, a resource required for executing a task in each of the plurality of first information processing devices; and for each of the plurality of first information processing devices. A first resource check step for checking the first resource information; and a task from among the plurality of first information processing apparatuses, based on the first resource information checked by the first resource check step. An allocation candidate determining step of determining a first information processing apparatus to be allocated as an allocation candidate; and a transmitting side information processing apparatus which is executed after the completion of the allocation candidate determining step and has the task, transmits a first reservation extension signal to the allocation candidate. Transmitting a first reservation extension signal to be transmitted; each of the plurality of first information processing devices detecting an expiration date of the first resource reservation; The first information processing apparatus that until the expiration date has passed out of the physical device the first reservation extension signal is not transmitted,
A first reservation release step for releasing the first resource reservation for the own resource reserved in the first resource reservation step; A second resource reserving step for reserving as resource reservation, a second resource retrieving step for retrieving second resource information of the second information processing apparatus, a first resource information of the allocation candidate, and the second resource An assignment destination determining step of determining an information processing device to which a task is to be assigned from among the allocation candidates and the second information processing device as an allocation information processing device based on resource information; Second to processing equipment
A second reservation extension signal transmitting step of transmitting the reservation extension signal of the above, and the second information processing device detects the expiration date of the second resource reservation, and until the expiration date of the second resource reservation expires. A second reservation release step of releasing the second resource reservation when the second reservation extension signal has not been transmitted to the communication device; If the second reservation extension signal is not received before the time elapses a predetermined time, the first resource reservation for the own resource reserved in the first resource reservation step is released. A dynamic load comprising: a third reservation release step; and a transmission step executed after the second reservation extension signal transmission step ends, wherein the transmission side information processing apparatus transmits the task to the allocation information processing apparatus. Minute Method. 7. Receiving a reservation for its own resource from another information processing device requesting execution of a task,
A reservation means for reserving its own resource for the received reservation; a reservation release means for releasing the reservation if the task is not received before the expiration date of the reservation; and An information processing apparatus comprising: a resource allocating unit that allocates a resource reserved by the reservation unit to the task when the task is received. 8. The reservation canceling means cancels the reservation step by step when the expiration date of the reservation has passed, and reserves the resource canceled by the reservation canceling means when a task is received from another information processing apparatus. The information processing apparatus according to claim 7, further comprising a reservation recovery unit that recovers the information. 9. A threshold changing means for changing a threshold of a resource amount that can be reserved based on the number of reservations received from another information processing apparatus; When the resource amount of the new reservation and the received reservation does not exceed the threshold value, a reservation unit that reserves its own resource for the new reservation, and when the expiration date of the new reservation has passed, A reservation canceling means for canceling the new reservation; and a resource allocating means for allocating a resource reserved by the reservation means to the task when a task is received from the information processing apparatus requesting the new reservation. Information processing device.