JP7367565B2 - Power control device and power control program - Google Patents

Description

The present invention relates to a power control device and the like.

A control method is disclosed for controlling the power consumption of the entire system to be below an upper limit value. For example, in one example, for a system composed of a plurality of computers, the plurality of computers are grouped into groups in order to keep the power value of the entire system below a certain value. Then, the management server caps the power for each group of computers to keep the power of the entire system below a certain level (for example, see Patent Document 1). Capping here means limiting power consumption.

In another example, when the power limit for the entire system is about to be exceeded, the surplus power obtained by lowering the operating level of the operating computers is allocated to other computers, thereby reducing the power consumption of the entire system. The power value is set below the upper limit (for example, see Patent Document 2).

By the way, a program that allocates computer resources to jobs and controls job execution is called a job scheduler. The job scheduler takes as input the number of requested nodes (computers) and the execution time of the submitted job, and assigns the job to the execution start time and the nodes (computers) to be used. Further, a computer center that manages a plurality of computers makes a contract with an electric power company regarding the upper limit of power that can be used. Therefore, the job scheduler needs to schedule not only the node allocation but also the power to be used for the job.

The job scheduler predicts the power to be used by the job before the job is executed, and schedules the power to be used for the job based on the predicted value so as to be below the upper limit power value of the system. FIG. 14 is a diagram showing a reference example of node and power allocation. In FIG. 14, the upper power value for the system is 500 watts (W). As shown in FIG. 14, for example, if job C and job D are executed at the same time, the upper limit power value of 500W will be exceeded, so the job scheduler will execute job D after job C is finished. assign. Note that, as a method for predicting the execution power of a job, for example, a method using past execution results is known (for example, Non-Patent Document 1).

Japanese Patent Application Publication No. 2015-106313 Japanese Patent Application Publication No. 2011-013764

“Job power consumption prediction using past execution results”, Yamamoto et al, IPSG SIG Technical Report, Vol. 2015-HPC-151 No.2 (2015)

Here, regarding the execution power of a job, the actual power may be larger than the predicted power. In such a case, capping occurs on the node that executes the job. When capping occurs, the job scheduler allocates the surplus power of the entire system to the node where capping occurs.

However, if the job scheduler allocates all the surplus power to the node where the capping occurs at the time when capping of power usage occurs, there is a problem that if capping occurs on another node, the job scheduler will not be able to respond to that node. be. This problem will be explained with reference to FIG. 15.

FIG. 15 is a diagram illustrating a reference example when capping of power usage occurs. In FIG. 15, the job scheduler allocates 100 W to job A and 300 W to job B as predicted power, and sets the upper limit to 500 W for the entire system. As shown in FIG. 15, it is assumed that the job scheduler allocates the entire surplus power of 100 W when job A attempts to exceed 100 W during execution. Then, when job B reaches the capping of 300W, the job scheduler cannot allocate surplus power to job B. That is, if the job scheduler allocates all the surplus power to the node where the capping has occurred at the time when the capping of power usage occurs, the job scheduler cannot secure the surplus power to be allocated to other nodes.

One aspect of the present invention is to be able to cope with situations where other nodes exceed the upper limit of power consumption even when surplus power is allocated to a node that is likely to exceed the upper limit of power consumption.

In one aspect, a power control device includes a management unit that manages surplus power of a power control system including a plurality of information processing devices and a power control device, and a management unit that manages surplus power of a power control system that includes a plurality of information processing devices and a power control device; a storage unit that stores a value of allocated power; and a storage unit that stores information stored in the storage unit when the value of power consumption of any of the information processing devices among the plurality of information processing devices exceeds the value of the allocated power; a specifying unit that specifies a job that the information processing device is executing; and a specifying unit that supplies surplus power to the information processing device that executes the identified job based on the information stored in the storage unit and the surplus power of the power control system. and an allocation unit that allocates in stages.

According to one embodiment, even if surplus power is allocated to a node that is likely to exceed the upper limit of power consumption, it is possible to cope with the case where other nodes exceed the upper limit of power consumption.

FIG. 1 is a block diagram showing the functional configuration of a power control system according to a first embodiment. FIG. 2 is a diagram illustrating an image of power control according to the first embodiment. FIG. 3 is a block diagram showing the functional configuration of the control device according to the first embodiment. FIG. 4 is a diagram illustrating an example of the data structure of a job table according to the first embodiment. FIG. 5 is a block diagram showing the functional configuration of a calculation node according to the first embodiment. FIG. 6 is a diagram showing an example of an updated job table. FIG. 7 is a diagram illustrating an example of a flowchart of the basic operation of the scheduler according to the first embodiment. FIG. 8 is a diagram illustrating an example of a flowchart of power control according to the first embodiment. FIG. 9 is a block diagram showing the functional configuration of the control device according to the second embodiment. FIG. 10 is a diagram illustrating power allocation according to the second embodiment. FIG. 11 is a diagram illustrating an example of a flowchart of the basic operation of the scheduler according to the second embodiment. FIG. 12 is a diagram illustrating an example of a flowchart of power control according to the second embodiment. FIG. 13 is a diagram illustrating an example of a computer that executes a power control program. FIG. 14 is a diagram showing a reference example of node and power allocation. FIG. 15 is a diagram illustrating a reference example when capping of power usage occurs.

Embodiments of a power control device and a power control program disclosed in the present application will be described in detail below based on the drawings. Note that the present invention is not limited to the examples.

[Configuration of power control system according to Example 1]
FIG. 1 is a block diagram showing the functional configuration of a power control system according to a first embodiment. The power control system 9 shown in FIG. 1 includes a control device 1, a plurality of calculation nodes 3, and a user terminal 5. The control device 1 is connected to a plurality of calculation nodes 3 and user terminals 5 via a network 7. The control device 1 predicts the power consumption of a job, and uses the predicted power to suppress the power consumption of the calculation node 3 that executes the job. When the power consumption exceeds the predicted upper limit of power, the control device 1 allocates the surplus power of the power control system 9 in stages to the calculation nodes 3 that execute the job. Note that in the embodiment, suppressing power consumption is referred to as "power capping", and power capping is implemented by hardware or software.

Here, an image of power control according to the first embodiment will be explained with reference to FIG. 2. FIG. 2 is a diagram illustrating an image of power control according to the first embodiment. In FIG. 2, the job scheduler allocates 100 W to job A and 300 W to job B as predicted power, and sets the upper limit to 500 W for the entire system.

As shown in FIG. 2, when the power consumption of job A is about to exceed the predicted upper limit of 100 W of power consumption, the control device ₁ generates a surplus Allocate power in stages. Here, it is assumed that a certain amount of 20W is allocated. As a result, even if the power consumption of job B exceeds the upper limit of power consumption during execution, the control device 1 can allocate surplus power to the calculation nodes 3 ₂ and 3 ₃ that execute job B. , it is possible to secure power to be allocated to the other calculation nodes 3 ₂ and 3 ₃ . Furthermore, the control device 1 can suppress a decline in job execution performance of the calculation node ₃₄ to which a certain amount of surplus power is allocated.

Returning to FIG. 1, the calculation node 3 executes the job assigned by the control device 1. The calculation node 3 executes the job assigned by the control device 1, and suppresses the power consumption of the job so as not to exceed the power consumption assigned to the job. Note that the functional configuration of the calculation node 3 will be described later.

The user terminal 5 is a terminal used when a user submits a job. When submitting a job, the user logs in using the user terminal 5 and submits the job. The user terminal 5 associates the submitted job with the number of nodes and requests the control device 1 to execute the job.

[Functional configuration of control device]
FIG. 3 is a block diagram showing the functional configuration of the control device according to the first embodiment. As shown in FIG. 3, the control device 1 includes a control section 10 and a storage section 20.

The control unit 10 corresponds to an electronic circuit such as a CPU (Central Processing Unit). The control unit 10 has an internal memory for storing programs and control data that define various processing procedures, and executes various processes using these. The control unit 10 includes a scheduler unit 11, a job execution instruction unit 12, and a power control unit 13. Note that the power control unit 13 is an example of a specifying unit and an allocation unit.

The storage unit 20 is, for example, a semiconductor memory element such as a RAM or a flash memory, or a storage device such as a hard disk or an optical disk. The storage unit 20 has a job table 21 and system power information 22.

The job table 21 is a table for managing jobs. The job table 21 stores, for each job, the calculation node 3 to be executed, the start time of execution, and the value of the usable allocated power at the time of execution. For each job, the calculation node 3 to execute, the execution start time, and the value of usable allocated power are determined by the scheduler unit 11, which will be described later. Note that an example of the data structure of the job table 21 will be described later.

FIG. 4 is a diagram illustrating an example of the data structure of a job table according to the first embodiment. As shown in FIG. 4, the job table 21 stores a job ID (IDentifier), expected job execution time, job start time, allocated power, and allocated node ID in association with each other. The job ID is an identifier that identifies the submitted job. The expected job execution time indicates the expected execution time from the start to the end of the job. The job start time indicates the start time of the job. Allocated power indicates the value of allocated power. The assigned node ID indicates the identifier of the assigned node. The job start time, allocated power, and allocated node ID are determined by the scheduler unit 11, which will be described later. The expected job execution time may be acquired by the scheduler unit 11, which will be described later, or may be specified by the user.

As an example, when the job ID is "1", the expected job execution time is "60", the job start time is "10:00", the allocated power is "100" W, and the allocated node ID is "node1". are doing.

Returning to FIG. 3, the system power information 22 includes information indicating the upper limit power allocated to the power control system 9. The upper limit of power allocated to the power control system 9 is defined by a contract with the power company. Furthermore, the system power information 22 is set with the surplus power of the system.

The scheduler unit 11 schedules jobs. For example, the scheduler unit 11 receives the job ID of the job to be executed and the number of nodes specified by the user. The scheduler unit 11 predicts the power consumption of the job indicated by the received job ID. The scheduler unit 11 obtains the expected execution time of the job indicated by the received job ID. Any conventional technique may be used to measure the power consumption of the job. For power consumption of a job, the technique mentioned in Non-Patent Document 1 may be used, as an example. The expected execution time of a job may be measured and stored in advance for each job. Then, the scheduler unit 11 allocates a node for execution and a start time for each received job ID based on the number of nodes, expected execution time of the job, and power consumption. The scheduler unit 11 then stores the expected job execution time, job start time, allocated power, and allocated node ID in the job table 21 for each job ID. The allocated power may be the power consumption predicted as an initial value.

The job execution instruction unit 12 instructs the calculation node 3 to execute a job. For example, the job execution instruction unit 12 refers to the job table 21, and when the start time of the job indicated by the job ID arrives, the job execution instruction unit 12 sends the calculation node 3 indicated by the node ID that executes the job along with the cap value. Instruct job execution. The cap value here is the value of the power allocated to the job, and refers to the limit value of power consumption when the specified calculation node 3 performs power capping.

The power control unit 13 controls the power of the plurality of calculation nodes 3. For example, when the power control unit 13 receives a notification from one of the plurality of calculation nodes 3 to the effect that an excess power has occurred, the power control unit 13 uses the job table 21 to identify the calculation node 3 in which the excess power has occurred. Identify the job ID of the job being executed. Then, based on the job table 21 and the system power information 22, the power control unit 13 allocates a fixed amount of system surplus power in stages to the calculation node 3 that executes the job indicated by the identified job ID. Then, the power control unit 13 notifies the calculation node 3 where the power excess has occurred of the new cap value for the job being executed. Then, the power control unit 13 updates the new cap value as the allocated power for the specified job ID in the job table 21. Further, when the power control unit 13 receives a notification that power excess has occurred again from the same calculation node 3 during execution of the job with the specified job ID, if the system surplus power is greater than a certain amount, the power control unit 13 It is sufficient to allocate a certain amount of surplus power to the calculation node 3. Note that the fixed amount of power to be allocated in stages is, for example, "10" W or "20" W, but is not limited to this, and may be determined in advance according to the upper limit of power allocated to the power control system 9. Good. Further, the fixed amount once determined may be updated.

[Functional configuration of calculation node]
FIG. 5 is a block diagram showing the functional configuration of a calculation node according to the first embodiment. As shown in FIG. 5, the calculation node 3 includes a power suppression section 31, a power monitoring section 32, a job execution section 33, and a power measurement section 34.

The power suppression unit 31 suppresses power consumption. For example, upon receiving a job execution instruction from the control device 1, the power suppression unit 31 limits power consumption so that the cap value included in the execution instruction becomes the upper limit. That is, the power suppression unit 31 performs power capping. Note that the power suppression unit 31 may be hardware or software.

The power monitoring unit 32 monitors power consumption. For example, the power monitoring unit 32 monitors the value of power consumption measured by a power measuring unit 34, which will be described later. When the power monitoring unit 32 detects that the power consumption value exceeds the cap value, it notifies the control device 1 that the power excess has occurred.

The job execution unit 33 executes a job. For example, the job execution unit 33 executes a job that is instructed to be executed by the control device 1.

The power measurement unit 34 measures the power consumption of the calculation node 3. As an example, the power measuring unit 34 corresponds to a power measuring device. The power measurement unit 34 may be connected to the calculation node 3.

[Example of updated job table]
FIG. 6 is a diagram showing an example of an updated job table. Note that the system power is assumed to be 500W.

Upon receiving the job ID to be executed and the number of nodes designated by the user, the scheduler unit 11 stores the expected job execution time, job start time, allocated power, and allocated node ID in the job table 21 for each job ID. The job table 21 here is assumed to be the table shown in FIG. 4.

Then, the job execution instruction unit 12 refers to the job table 21, and when the start time of the job with the job ID "1" reaches "10:00", the job execution instruction unit 12 selects the calculation node 3 with the node ID "node1" that executes the job. to execute the job. The execution instruction includes the job ID "1" and the cap value "100". As a result, the calculation node 3 of "node1" executes the job of job ID "1" so that the cap value "100" becomes the upper limit.

Here, when the power control unit 13 receives a notification that power excess has occurred from the calculation node 3 of “node 1”, the power control unit 13 uses the job table 21 to check the job being executed by the calculation node 3 where the power excess has occurred. Specify job ID "1".

Then, based on the job table 21 and the system power information 22, the power control unit 13 allocates a certain amount of system surplus power in stages to the calculation node 3 that executes the job with the specified job ID “1”.

As an example, assume that the current time is "10:10". Then, the power control unit 13 calculates the remaining time of job ID "1". Here, since the expected job execution time for job ID "1" is "60" minutes, the remaining time is calculated as "40" minutes. Then, the power control unit 13 calculates newly allocable power. Here, the power already allocated at the current time is "300" W, which indicates the power corresponding to job IDs "1" and "2". The surplus power of the system power is "200" W. Such surplus power may be obtained from the system power information 22. Ten minutes have passed since the job with job ID "1" started execution, and the remaining time is 50 minutes. The job with job ID "3" is scheduled to use "100" W from "10:30" during the remaining time of job ID "1". Therefore, the allocatable power that can be used for the remaining "50" minutes of job ID "1" is calculated to be "100" W. Then, if the fixed amount to be allocated in stages is "10" W, "10" W is within the allocatable power "100" W, so the power control unit 13 considers "10" W as an additional amount of power. decide. Then, the power control unit 13 notifies the calculation node 3 of “node1” of the new cap value “110”W. Then, the power control unit 13 updates the allocated power to the new cap value “110” for the specified job ID “1” in the job table 21. The job table 21 shown in FIG. 6 is an updated job table updated by the power control unit 13. As a result, the calculation node 3 of "node1" executes the job of job ID "1" so that the new cap value "110" becomes the upper limit. That is, the calculation node 3 of "node1" performs power capping with a new cap value of "110".

[Basic operation of scheduler]
FIG. 7 is a diagram illustrating an example of a flowchart of the basic operation of the scheduler according to the first embodiment. It is assumed that the user terminal 5 transmits to the control device 1 a job execution request that includes a job ID and the number of nodes specified by the user.

As shown in FIG. 7, the scheduler unit 11 receives a job ID of a job to be executed (step S11). The scheduler unit 11 predicts the power consumption of the job indicated by the job ID (step S12). Any conventional technique may be used to measure the power consumption of the job. For power consumption of a job, the technique mentioned in Non-Patent Document 1 may be used, as an example.

Then, the scheduler unit 11 schedules the job indicated by the job ID based on the predicted power and the number of requested nodes (step S13). For example, the scheduler unit 11 obtains the expected execution time of the job indicated by the received job ID. The expected execution time of a job may be measured and stored in advance for each job. Then, the scheduler unit 11 allocates a node for execution and a start time for each received job ID based on the number of nodes, expected execution time of the job, and power consumption. The scheduler unit 11 then creates a job table 21 that includes the expected job execution time, job start time, allocated power (predicted power), and allocated node ID for each job ID.

Then, when the job start time arrives, the scheduler unit 11 subtracts the predicted power at this time from the system power in the system power information 22 as the allocated power (step S14). Then, the job execution instruction unit 12 instructs the calculation node 3 that executes the job whose start time has arrived to execute the job with the cap value allocated to the power, and executes the job (step S15). .

When the scheduler unit 11 detects that the job ends (step S16), it adds the allocated power to the system power in the system power information 22, and returns the allocated power to the system (step S17).

[Power control operation]
FIG. 8 is a diagram illustrating an example of a flowchart of power control according to the first embodiment.

The power control unit 13 determines whether or not a notification to the effect that power exceedance has occurred is received (step S21). If it is determined that the notification that the power excess has occurred has not been received (step S21; No), the power control unit 13 repeats the determination process until the notification is received.

On the other hand, if it is determined that a notification to the effect that power excess has occurred (step S21; Yes), the power control unit 13 performs the following processing. The power control unit 13 refers to the job table 21 and identifies the job ID of the job assigned to the calculation node 3 from the node ID of the calculation node 3 where the power excess has occurred (step S22).

Then, the power control unit 13 acquires the remaining time of the job indicated by the specified job ID (step S23). For example, the power control unit 13 refers to the job table 21, obtains the job start time and expected job execution time corresponding to the specified job ID, and stores the obtained job start time, expected job execution time, and current time. to calculate the remaining time for the job.

Then, the power control unit 13 acquires the allocatable power (step S24). For example, the power control unit 13 obtains the surplus power of the system from the system power information 22 at the current time. The power control unit 13 calculates the allocatable power using the system's surplus power and the power of other jobs scheduled to be used during the remaining time of the job.

Then, the power control unit 13 calculates the additional power to be allocated (step S25). For example, the power control unit 13 calculates a certain amount of allocable power as the allocated power.

Then, the power control unit 13 determines whether the allocated power is greater than 0 (step S26). If it is determined that the allocated power is 0 or less (step S26; No), the power control unit 13 ends the power control process because no additional allocation can be made.

On the other hand, if it is determined that the allocated power is greater than 0 (step S26; Yes), the power control unit 13 subtracts the allocated power from the system surplus power in the system power information 22 (step S27).

Then, the power control unit 13 notifies the calculation node 3 of the new cap value to the calculation node 3 in which power excess occurred during execution of the specified job (step S28). The new cap value indicates a power value obtained by adding additional allocated power to the already allocated allocated power. Then, the power control unit 13 updates the allocated power to the new cap value for the job specified in the job table 21. Then, the power control unit 13 ends the power control process.

[Effects of Example 1]
According to the first embodiment, the power control system 9 includes a plurality of calculation nodes 3 and a control device 1. The control device 1 has a job table 21 that stores, for each job, the value of usable allocated power when executed by the calculation node 3. The control device 1 manages surplus power of the power control system 9. When the power consumption value of any one of the plurality of calculation nodes 3 exceeds the allocated power value, the control device 1 determines whether the calculation node 3 is currently being executed based on information stored in the job table 21. Identify jobs. Based on the information stored in the job table 21 and the surplus power of the power control system 9, the control device 1 allocates surplus power in stages to the calculation nodes 3 that execute the specified job. According to this configuration, when the power consumption exceeds the allocated power, the control device 1 allocates the surplus power in stages to secure the surplus power to be allocated to the calculation nodes 3 that execute other jobs. becomes possible. In addition, the control device 1 can suppress a decline in job execution performance of the calculation node 3 to which surplus power is allocated.

Further, according to the first embodiment, the control device 1 allocates the preset allocated power out of the surplus power in stages. According to this configuration, the control device 1 can secure surplus power to be allocated to the calculation nodes 3 that execute other jobs by allocating the preset allocated power out of the surplus power in stages. .

Further, according to the first embodiment, the job table 21 stores, for each job, the computing node 3 to be executed, the expected execution time, the execution start time, and the value of the usable allocated power in association with each other. Based on the remaining time of the identified job, the execution start time of other jobs, the available allocated power, and the surplus power of the power control system 9, the control device 1 determines whether the computing node 3 executing the identified job has surplus power. Allocate power in stages. According to this configuration, the control device 1 can reliably secure surplus power to be allocated to the calculation nodes 3 that execute other jobs.

By the way, in the first embodiment, when the control device 1 receives a notification from the calculation node 3 that executes a job that power excess has occurred, it is explained that the system allocates a certain amount of surplus power to the calculation node 3 in stages. did. That is, when the control device 1 receives a notification from the calculation node 3 that executes the job that power excess has occurred, the control device 1 fixes the additional allocated power. However, the control device 1 is not limited to this, and if a power excess has occurred in the past from the calculation node 3 that executes the job, the control device 1 can generate a new additional power based on the amount of power exceeded in the past. The allocated power may be changed dynamically.

Therefore, if a power excess has occurred in the past from the calculation node 3 that executes the job, the control device 1 dynamically allocates new additional power based on the amount of excess power that has exceeded in the past. Explain when to change.

[Functional configuration of control device]
FIG. 9 is a block diagram showing the functional configuration of the control device according to the second embodiment. Note that the same components as those of the control device 1 shown in FIG. 2 are denoted by the same reference numerals, and descriptions of the overlapping components and operations will be omitted. The difference between the first embodiment and the second embodiment is that the scheduler section 11 and the power control section 13 of the control section 10 are changed to a scheduler section 11A and a power control section 13A, respectively. Further, the difference between the first embodiment and the second embodiment is that power history information 23A is added to the storage unit 20.

The power history information 23A is information that stores, as a history, the final additionally allocated power for each job in which power exceedance has occurred.

The scheduler unit 11A schedules jobs.

For example, the scheduler unit 11A receives the job ID and number of nodes of the job to be executed specified by the user. The scheduler unit 11A predicts the power consumption of the job indicated by the received job ID. As an example, if the job indicated by the job ID has experienced excess power in the past, the scheduler unit 11A uses the last final allocated power as the predicted value. The scheduler unit 11A sets the predicted power consumption as a predicted value if power excess has not occurred in the past. Any conventional technique may be used to measure the power consumption of the job.

The scheduler unit 11A also obtains the expected execution time of the job indicated by the received job ID. The scheduler unit 11A then assigns a node to execute and a start time for each received job ID based on the number of nodes, expected job execution time, and power consumption. The scheduler unit 11A then stores the expected job execution time, job start time, allocated power, and allocated node ID in the job table 21 for each job ID. The allocated power may be the power consumption predicted as an initial value.

Furthermore, the scheduler unit 11A records the excess amount and the final allocated power value to which the excess amount has been added, in the power history information 23A, for a job that exceeds the allocated power.

The power control unit 13A controls the power of the plurality of calculation nodes 3. For example, when the power control unit 13A receives a notification from one of the plurality of calculation nodes 3 to the effect that an excess power has occurred, the power control unit 13A uses the job table 21 to identify the calculation node 3 in which the excess power has occurred. Identify the job ID of the job being executed. Then, the power control unit 13A refers to the power history information 23A and determines whether or not the job indicated by the job ID has experienced power overrun in the past. If the job indicated by the job ID has caused excess power in the past, the power control unit 13A transmits the previous excess amount to the calculation node 3 that executes the job if the surplus power of the system is larger than the previous excess amount. Assign additionally. Then, the power control unit 13A notifies the calculation node 3 of the new cap value of the job being executed. Then, the power control unit 13A updates the allocated power to the new cap value for the specified job ID in the job table 21. Further, when the power control unit 13A receives a notification that power excess has occurred again from the same calculation node 3 while executing the job with the specified job ID, the power control unit 13A further adds and allocates the previous excess amount. good.

Note that if the job indicated by the job ID has not experienced excess power in the past, the power control unit 13A may allocate a fixed amount of surplus power in stages to the calculation node 3 that executes the job.

[Power allocation according to Example 2]
FIG. 10 is a diagram illustrating power allocation according to the second embodiment. As shown in the upper part of FIG. 10, it is assumed that job The scheduler unit 11A predicts the power consumption of job X to be "40" W when job X is submitted for the nth time. The job execution instruction unit 12 instructs execution of the job along with the cap value "40" W. When the power control unit 13A receives a notification that a power excess has occurred from the calculation node 3 that executes the job X, it additionally allocates a certain amount of "10" W since a power excess has not occurred in the past. Job X has been additionally allocated twice, so the final allocated power amount is "60" W. Therefore, the scheduler unit 11A records the excess amount "20" W and the final allocated power value "60" W for the job X that exceeded the allocated power in the power history information 23A.

Then, the scheduler unit 11A predicts that the power consumption when the job X is submitted for the (n+1) time will be "60" W, which is the value of the previous allocated power. Then, this job X is executed without exceeding "60" W until the end of execution.

As shown in the lower part of FIG. 10, it is assumed that job Z has not experienced an excess power in the past until the (n-1)th input. The scheduler unit 11A predicts the power consumption of job Z to be "40" W when job Z is submitted for the nth time. The job execution instruction unit 12 instructs execution of the job along with the cap value "40" W. When the power control unit 13A receives a notification that a power excess has occurred from the calculation node 3 that executes the job Z, it additionally allocates a certain amount of "10" W since a power excess has not occurred in the past. Job Therefore, the scheduler unit 11A records the excess amount "20" W and the final allocated power value "60" W for the job X that exceeded the allocated power in the power history information 23A.

Then, the scheduler unit 11A predicts that the power consumption when the job Z is submitted for the (n+1) time will be "60" W, which is the previously allocated power. Then, when the power control unit 13A receives a notification from the calculation node 3 that executes job Z that power has exceeded, it additionally allocates the previous excess amount "20" W.

As a result, for example, in the case of a job with large power fluctuations, the power control unit 13A dynamically allocates additional power using the excess amount given in the past, thereby allocating power according to the characteristics of the job. becomes possible. Moreover, the power control unit 13A can respond when another node exceeds the upper limit of power consumption, compared to a case where all surplus power is allocated.

[Basic operation of scheduler]
FIG. 11 is a diagram illustrating an example of a flowchart of the basic operation of the scheduler according to the second embodiment. It is assumed that the user terminal 5 transmits to the control device 1 a job execution request that includes a job ID and the number of nodes specified by the user.

As shown in FIG. 11, the scheduler unit 11A receives a job ID of a job to be executed (step S31). The scheduler unit 11A predicts the power consumption of the job indicated by the job ID (step S32).

The scheduler unit 11A determines whether there is a record of excess power in the past (step S32A). For example, the scheduler unit 11A refers to the power history information 23A and determines whether or not there is a history of occurrence of power overrun for the job indicated by the job ID. If it is determined that there is no record of excess power in the past (step S32A; No), the scheduler unit 11A moves to step S33.

On the other hand, if it is determined that there is a record of excess power in the past (step S32A; Yes), the scheduler unit 11A sets the previously allocated power as the predicted power (step S32B). For example, the scheduler unit 11A obtains the allocated power corresponding to the job ID of the power history information 23A and uses it as the predicted power. The scheduler unit 11A then proceeds to step S33.

In step S33, the scheduler unit 11A schedules the job indicated by the job ID based on the predicted power and the number of requested nodes (step S33). The scheduler unit 11A then creates a job table 21 that includes the expected job execution time, job start time, allocated power (predicted power), and allocated node ID for each job ID.

Then, when the job start time arrives, the scheduler unit 11A subtracts the predicted power at this time from the system power in the system power information 22 as the allocated power (step S34). Then, the job execution instruction unit 12 instructs the calculation node 3 that executes the job whose start time has arrived to execute the job with the cap value allocated to the power, and executes the job (step S35). .

When the scheduler unit 11A detects that the job ends (step S36), it records the excess allocation amount (step S37). For example, the scheduler unit 11A records the excess amount and the final allocated power value to which the excess amount has been added, in the power history information 23A, for a job that exceeds the allocated power.

Then, the scheduler unit 11 adds the allocated power to the system power in the system power information 22, and returns the allocated power to the system (step S38).

[Power control operation]
FIG. 12 is a diagram illustrating an example of a flowchart of power control according to the second embodiment.

The power control unit 13A determines whether or not a notification to the effect that power excess has occurred has been received (step S41). If it is determined that the notification that the power excess has occurred has not been received (step S41; No), the power control unit 13A repeats the determination process until the notification is received.

On the other hand, if it is determined that a notification to the effect that power exceedance has occurred (step S41; Yes), the power control unit 13A performs the following processing. The power control unit 13A refers to the job table 21 and identifies the job ID of the job assigned to the calculation node 3 from the node ID of the calculation node 3 where the power excess has occurred (step S42).

Then, the power control unit 13A obtains the remaining time of the job indicated by the specified job ID (step S43). For example, the power control unit 13A refers to the job table 21 and obtains the job start time and job expected execution time corresponding to the specified job ID. The power control unit 13A calculates the remaining time of the job using the acquired job start time, expected job execution time, and current time.

Then, the power control unit 13A obtains the allocatable power (step S44). For example, the power control unit 13A obtains the system surplus power from the system power information 22 at the current time. The power control unit 13A calculates the allocatable power using the system's surplus power and the power of other jobs scheduled to be used during the remaining time of the job.

Then, the power control unit 13A determines whether there is a record of excess power in the past for the specified job ID (step S45). For example, the power control unit 13A refers to the power history information 23A and determines whether or not the job indicated by the job ID has experienced power overrun in the past. If it is determined that there is a record of excess in the past (step S45; Yes), the power control unit 13A calculates the allocated power (step S47). For example, the power control unit 13A calculates the previous excess amount as additional allocated power to the calculation node 3 that executes the job indicated by the job ID. The power control unit 13A then proceeds to step S48.

On the other hand, if it is determined that there is no record of excess in the past (step S45; No), the power control unit 13A calculates the fixed allocated power as additional allocated power (step S46). For example, the power control unit 13A calculates a certain amount of allocable power as the allocated power. The power control unit 13A then proceeds to step S48.

Then, the power control unit 13A determines whether the allocated power is greater than 0 (step S48). If it is determined that the allocated power is less than or equal to 0 (step S48; No), the power control unit 13A ends the power control process because additional allocation is not possible.

On the other hand, if it is determined that the allocated power is greater than 0 (step S48; Yes), the power control unit 13A subtracts the allocated power from the system surplus power in the system power information 22 (step S49).

Then, the power control unit 13A notifies the calculation node 3 of the new cap value to the calculation node 3 in which power excess occurred during execution of the specified job (step S50). The new cap value indicates a power value obtained by adding additional allocated power to the already allocated allocated power. Then, the power control unit 13A updates the allocated power to the new cap value for the specified job in the job table 21. Then, the power control unit 13A ends the power control process.

Note that the power control unit 13A can add the same amount of power as the previous excess amount to the calculation node 3 that executes the job if a power excess has occurred in the past for a job in which a power excess has occurred. Assign explained. However, the power control unit 13A is not limited to this, and may additionally allocate 1/2 or 1/3 of the previous excess amount of power.

[Effects of Example 2]
According to the second embodiment, if the job being executed on the calculation node 3 whose power consumption value exceeds the allocated power value has exceeded the allocated power in the past, the control device 1 can Calculate the new value to be assigned. The control device 1 allocates the calculated surplus power in stages. According to this configuration, the control device 1 is able to allocate power according to the characteristics of the job by calculating the value to be allocated using past exceeded values. Moreover, the control device 1 can respond when the other calculation nodes 3 exceed the upper limit of power consumption, compared to the case where all the surplus power is allocated.

[others]
In the first and second embodiments, a case has been described in which the power control units 13 and 13A allocate additional power so that the power consumption of the job becomes the upper limit. However, the power control units 13 and 13A are not limited to this, and when the power consumption of the job is less than the additionally allocated power, the power controllers 13 and 13A may convert the additionally allocated power into system power.

Further, each component of the illustrated control device 1 does not necessarily need to be physically configured as illustrated. In other words, the specific manner of distributing and integrating the control device 1 is not limited to what is shown in the diagram, and all or part of it may be functionally or physically distributed in arbitrary units depending on various loads and usage conditions. -Can be integrated and configured. For example, the scheduler section 11 and job execution instruction section 12 may be integrated as one section. Further, the power control unit 13 may be separated into a specifying unit that specifies a job with excess power and an allocation unit that allocates power to the calculation node 3 that executes the specified job. Further, the storage unit 20 may be connected as an external device to the control device 1 via a network.

Moreover, the various processes described in the above embodiments can be realized by executing a program prepared in advance on a computer such as a personal computer or a workstation. Therefore, an example of a computer that executes a power control program that implements the same functions as the control device 1 shown in FIG. 2 will be described below. FIG. 13 is a diagram illustrating an example of a computer that executes a power control program.

As shown in FIG. 13, the computer 200 includes a CPU 203 that executes various calculation processes, an input device 215 that receives data input from a user, and a display control unit 207 that controls a display device 209. The computer 200 also includes a drive device 213 that reads programs and the like from a storage medium, and a communication control unit 217 that exchanges data with other computers via a network. Further, the computer 200 includes a memory 201 that temporarily stores various information, and an HDD (Hard Disk Drive) 205. The memory 201, CPU 203, HDD 205, display control section 207, drive device 213, input device 215, and communication control section 217 are connected via a bus 219.

The drive device 213 is, for example, a device for the removable disk 210. The HDD 205 stores a power control program 205a and power control processing related information 205b.

The CPU 203 reads the power control program 205a, expands it into the memory 201, and executes it as a process. This process corresponds to each functional section of the control device 1. The power control processing related information 205b corresponds to the job table 21 and the system power information 22. For example, the removable disk 210 stores information such as the power control program 205a.

Note that the power control program 205a does not necessarily need to be stored in the HDD 205 from the beginning. For example, a "portable physical medium" such as a flexible disk (FD), CD-ROM (Compact Disk Read Only Memory), DVD (Digital Versatile Disk), magneto-optical disk, or IC (Integrated Circuit) card that is inserted into the computer 200 is used. The program is stored in ``. Then, the computer 200 may read out and execute the power control program 205a from these programs.

1 Control device 10 Control unit 11, 11A Scheduler unit 12 Job execution instruction unit 13, 13A Power control unit 20 Storage unit 21 Job table 22 System power information 23A Power history information 3 Computation node 31 Power suppression unit 32 Power monitoring unit 33 Job execution Section 34 Power measurement section 5 User terminal 7 Network 9 Power control system

Claims (5)

a management unit that manages surplus power of a power control system including a plurality of information processing devices and a power control device;
a storage unit that stores, for each job, a value of usable allocated power when executed by the information processing device;
When the power consumption value of any one of the plurality of information processing devices exceeds the allocated power value, identifying a job being executed by the information processing device from information stored in the storage unit. A specific part to
an allocation unit that allocates surplus power in stages to an information processing device that executes a specified job based on information stored in the storage unit and surplus power of the power control system;
A power control device comprising: The power control device according to claim 1, wherein the allocation unit allocates preset allocated power out of the surplus power in stages. The storage unit stores, for each job, an information processing device to be executed, an expected execution time, an execution start time, and a value of available allocated power in association with each other;
The allocation unit is configured to allocate surplus power to the information processing device that executes the specified job based on the remaining time of the specified job, the execution start time of other jobs, available allocated power, and surplus power of the power control system. The power control device according to claim 1, wherein power is allocated in stages. The allocation unit is characterized in that, when the identified job exceeds the allocated power in the past, a new value to be allocated is calculated based on the exceeded value, and the surplus power is allocated in stages by the calculated value. The power control device according to claim 3. Manages surplus power in a power control system that includes multiple information processing devices and power control devices,
If the power consumption value of any of the information processing devices among the plurality of information processing devices exceeds the allocated power value, for each job, the available allocated power value when executed by the information processing device is determined. Identifying the job being executed by the information processing device from the information stored in the storage unit,
A power supply system that causes a computer to allocate surplus power in stages to an information processing device that executes a specified job based on information stored in the storage unit and surplus power of the power control system. control program.

Images