Workload control support device and workload control support method

Abstract

To provide a work load control support device that controls each work load executed by a renewable energy while considering a cost-benefit performance, and provide a work load control support method.SOLUTION: In a work load control system, a work load control support device executes: a parameter determination processing in which prediction values of an execution available period and a consumption power amount of each of a plurality of work loads are acquired, the work loads consuming a power and being scheduled to be executed in a future time zone, and a target value of the consumption power amount in the future time zone is calculated on the basis of each prediction value of the acquired execution available period and the consumption power amount so as to satisfy a condition of a use ratio of a renewable energy in a power consumption in the future time zone and a condition of a cost in accordance with a use of the renewable energy; and a work load control processing in which a timing of each work load executed in the future time zone is determined on the basis of the target value of a calculated consumption power amount, and each work load is executed at the determined timing.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present invention relates to a workload control support device and a workload control support method. [Background technology]

[0002] Fossil fuels are being used to prevent the emission of greenhouse gases such as carbon dioxide, which causes global warming. The so-called decarbonization, which aims to break away from the conventional approach, is attracting attention. A DC (Computer Handling Center) is equipped with a large number of information processing devices and communication devices to execute a given job. These are set to operate at a large amount of electricity. Attempts are being made to achieve decarbonization by relying on renewable energy.

[0003] In this case, the ratio of renewable energy to electricity consumption (renewable energy use) It is important to maintain the renewable energy rate, but it is necessary to change this renewable energy rate to a finer time granularity (e.g. For example, it is preferable to maintain the time on a per hour basis rather than a per day basis.

[0004] In this regard, Non-Patent Document 1 describes batch jobs that can change their execution timing and In a job system that includes interactive jobs whose scheduling cannot be changed, By shifting the execution timing of the project, the difference between the amount of electricity generated by renewable energy and the amount of electricity consumed can be reduced. It is stated that it can be made smaller. [Prior art documents] [Non-patent literature]

[0005] [Non-Patent Document 1] Li, Y.; Wang, X.; Luo, P.; Pan, Q. Thermal-Aware Hybrid Workload Management in a Green Datacenter towards Renewable Energy Utilization. Energies 2019, 12, 1494. https: / / doi.org / 10.3390 / en12081494 [Retrieved February 16, 2022] Summary of the Invention [Problem to be solved by the invention]

[0006] However, in Non-Patent Document 1, the cost of procuring renewable energy is not taken into consideration. Generally, the procurement of renewable energy sources often entails considerable costs, and non-patented Reference 1 does not take such cost into consideration. In particular, it does not take into consideration the cost-effectiveness aspect. In addition, Non-Patent Document 1 does not estimate the power consumption or workload that will be generated in the future due to the workload. It is assumed that the plans for the construction of the facility are known, but in reality, this is rarely the case. do not have.

[0007] The present invention has been made in view of the above background, and an object of the present invention is to provide a renewable energy It is a work environment that allows you to control each workload that is executed by the virtual machine cost-effectively. It is an object of the present invention to provide a workload control support device and a workload control support method. [Means for solving the problem]

[0008] One of the present inventions for solving the above problems is a processor and a memory. The available execution periods and execution times of multiple power-consuming workloads scheduled to run during a certain time period are and a predicted value of the executable period and the power consumption amount, Based on the target value of the power consumption amount in the future time period, Conditions for the utilization rate of renewable energy in the country and the costs associated with the utilization of renewable energy a parameter determination unit that calculates the power consumption amount so as to satisfy a target condition of the calculated power consumption amount; and determining a timing for each workload to be executed during said future time period based on said target value. and a workload control unit that executes each of the workloads at a determined timing. , a workload control support device. Effect of the Invention

[0009] According to the present invention, each workload that is run on renewable energy can be cost-effectively It can be controlled while taking into consideration. Problems, configurations and effects other than those described above will become apparent from the following description of the embodiments. [Brief description of the drawings]

[0010] [Figure 1] 1 is a diagram illustrating an example of a configuration of a workload control system according to an embodiment of the present invention. [Diagram 2] FIG. 2 is a diagram illustrating an example of hardware and functions of a workload control assistance device. [Diagram 3] FIG. 13 is a diagram illustrating an example of a DC power prediction table. [Figure 4] FIG. 13 is a diagram illustrating an example of a time slot table. [Diagram 5] FIG. 13 is a diagram illustrating an example of a delay limit time prediction distribution table. [Figure 6] FIG. 13 illustrates an example of a user policy table. [Figure 7] FIG. 13 illustrates an example of a workload table. [Figure 8]FIG. 13 illustrates an example of a workload power consumption prediction distribution table. [Figure 9] FIG. 13 illustrates an example of a predicted workload table. [Figure 10] FIG. 11 is a flow diagram illustrating an overview of a workload control process. [Figure 11] FIG. 11 is a flow diagram illustrating details of a data update process. [Figure 12] FIG. 11 is a flow diagram illustrating a workload deployment process. [Figure 13] FIG. 11 is a flow diagram illustrating an example of a parameter determination process. [Figure 14] FIG. 11 is a flow diagram illustrating details of a parameter creation process. [Figure 15] FIG. 11 is a flow diagram illustrating details of a parameter creation process. [Figure 16] FIG. 11 is a flow diagram illustrating an example of a risk tolerance calculation process. [Figure 17] FIG. 11 is a flow diagram illustrating details of a process of calculating risk tolerance for each viewpoint. [Figure 18] FIG. 11 is a flow diagram illustrating an IT workload control process. [Figure 19] FIG. 11 is a flow diagram illustrating an IT workload control process. [Figure 20] FIG. 13 is a diagram illustrating an example of a workload migration information screen. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] FIG. 1 is a diagram showing an example of the configuration of a workload control system 1 according to this embodiment. The workload control system 1 includes one or more data centers 1000 (DC). The data centers 1000 are connected to each other via a wide area network 7000. are reliably connected.

[0012] The data center 1000 includes a management computer 2000 and an administrator or One or more server devices 3000 used by users and an administrator or The server device 300 includes one or more storage devices 4000 that are used by users. 0 and the storage device 4000 are connected to each other via a data network 6000 so that they can communicate with each other. In addition, a management computer 2000, a server device 3000, and a storage device 400 0, they are communicatively connected via a management network 5000.

[0013] The management network 5000, the data network 6000, and the wide area network The 7000 is used for connecting, for example, the Internet, LAN (Local Area Network), WAN (Wide A communications network is a wired or wireless communication network such as a Broadband Area Network (BRN), a Broadband Internetworking Area Network (BIWAN), or a dedicated line.

[0014] The server device 3000 and the storage device 4000 execute various types of processing. For example, the server device 3000 and the storage device 4000 may be used to In this case, the execution time period (hereafter, "time slot") is fixed. In addition to the interactive job, there is also a project on artificial intelligence (AI). In some cases, such as machine learning, the time slot is not necessarily fixed. Processing that is not scheduled but must be executed at least by a certain time period (hereafter referred to as batch processing) In the following, we will refer to batch jobs and interactive jobs. Together these are called jobs.

[0015] The management computer 2000 executes the operations performed by the server device 3000 and the storage device 4000. and to the system (data center 1000) by batch jobs scheduled to be executed in the future. The processing load (hereafter referred to as batch workload) is managed in units of time slots. Similarly, the management computer 2000 is executed by the server device 3000 and the storage device 4000. The system (data center, etc.) is being used by interactive jobs that have been executed and will be executed in the future. 1000) (hereafter referred to as batch interactive workload) It is managed on a slot basis.

[0016] In this specification, a workload may refer to a job (process) itself. .

[0017] By the way, the execution of each process in the server device 3000 and the storage device 4000 requires a certain A certain amount of power is required, but in the data center 1000, a certain percentage of this power is allocated to the power There is a demand to use electricity from renewable sources, not from the grid. That is, in this embodiment, this predetermined ratio (utilization rate as a minimum condition) is This is called the target value or target rate of the renewable energy utilization rate. The amount of electricity generated varies depending on the time of day.

[0018] Therefore, the management computer 2000 (workload control support device) of this embodiment is For each batch job execution in the time slot, the amount of renewable energy generated is forecast. Based on the measured values, a power consumption target value is set while taking into account the renewable energy rate and cost. Control the execution timing of each batch job so that the power consumption target value can be maximized ( More specifically, the batch jobs to be executed in each time slot are scheduled at the start of each time slot. By determining the renewable energy rate vs. cost in data center 1000, In the following, "renewable energy" will be referred to as "renewable energy" It is sometimes abbreviated as "ne".

[0019] Next, FIG. 2 shows the hardware of the management computer 2000 (workload control support device). FIG. 1 is a diagram illustrating an example of the functions.

[0020] The management computer 2000 includes a CPU (Central Processing Unit), a DSP (Digital Signal Processor), or), GPU (Graphics Processing Unit), FPGA (Field-Programmable Gate Array), etc. Device 11000 (processor), ROM (Read Only Memory), RAM (Random Access Memory) Main memory 12000 (memory), HDD (Hard Disk Drive), SSD (Solid State Drive) e) A storage device 8000 such as a NIC (Network Interface Card), a wireless communication module , USB (Universal Serial Interface) module, or serial communication module, etc. A communication device 16000 configured as a communication device, and an input device 1400 configured as a mouse, keyboard, etc. 0 and a liquid crystal display or organic EL (Electro-Luminescence) display, etc. and an output device 15000.

[0021] In addition, the management computer 2000 includes a parameter determination program 8700, a risk tolerance calculation program Program 8800, IT Workload Control Program 8900, and Power Consumption Price Forecasting Each program of program 9000 is stored.

[0022] The parameter determination program 8700 determines the number of workpieces to be executed in future time slots. The predictions of the available period and power consumption of each load are obtained. The meter determination program 8700 determines the future based on the execution period and the predicted value of the power consumption. The target value of the power consumption in the current time slot is set to the target value of the power consumption in the future time slot. The calculation will be made so as to meet the target rate and the cost conditions for the use of renewable energy. .

[0023] In this embodiment, the delay limit time is used as the executable time. The delay limit time is expressed as follows: This is the latest time that can be set as the execution timing of a batch job.

[0024] The parameter determination program 8700 determines whether the condition is the renewable energy rate or the cost. When the control policy that prioritizes the renewable energy rate is specified, The timing of batch workloads to optimize the utilization of renewable energy is also important. We identify a parameter α that indicates the pattern of future events, and predict future events based on the identified parameter α. The target value of the power consumption in the real slot is calculated. 8700 is a renewable energy system that can be used in cases where a cost-conscious control policy is specified. The timing patterns of batch workloads are optimized to optimize the cost of using A parameter α that indicates the time slot is determined, and the future time slot is calculated based on the determined parameter α. Calculate the target value for the amount of power consumption in the facility.

[0025] In this embodiment, the parameter α is the number of executions scheduled to be performed in a certain time slot. The batch workload that is actually executed in that time slot is In this embodiment, the parameter α is a parameter that indicates the ratio of the time slot. Each time slot has a value between 0 and 1. Note that this is just an example. This reflects the percentage of the batch workload that is actually executed out of the batch workload that is scheduled to be executed. Other values ​​may be used if desired.

[0026] The IT Workload Control Program 8900 calculates the power consumption in future time slots. The risk tolerance, which indicates the risk due to the uncertainty of the predicted value of the quantity, is calculated using a specified algorithm. Based on the calculated risk tolerance and the target value of the power consumption, the system determines the future time slot. Determines when each workload should run on the network and executes each workload at the determined timing. Run the command.

[0027] The risk tolerance calculation program 8800 calculates the risk tolerance in future time slots. The predicted power generation amount related to renewable energy in the future and the power consumption amount in the future time slots are It is calculated based on the difference between the actual and target strength values.

[0028] The power consumption price prediction program 9000 predicts the workload to be executed in future time slots. The execution period and power consumption forecast for the code, and the power generation forecast for future time slots , the predicted value of the price of electricity in future time slots, etc. are calculated.

[0029] Furthermore, the management computer 2000 stores a DC power prediction table 8100, a time slot table 8200, delay limit time prediction distribution table 8300, user policy table 8400 , a workload table 8500, a workload power consumption forecast distribution table 8600, and and a predicted workload table 8650.

[0030] The DC power forecast table 8100 is a recalculation table for calculating the power consumption price forecasted by the power consumption price forecast program 9000. The amount and price of renewable energy generation, the forecast price of electricity provided by the power grid, and These actual values ​​are stored.

[0031] The time slot table 8200 stores the predicted power consumption value and For each time slot, the actual value, the target value of power consumption, the parameter α, and the risk tolerance, etc. It stores information on.

[0032] The delay limit time forecast distribution table 8300 is a table showing the workload distribution in each future time slot. It stores information on the distribution of predicted delay limits for each node.

[0033] The user policy table 8400 includes a target value of the renewable energy rate (hereinafter referred to as the target rate) and The policy of the user regarding the use of renewable energy, such as the cloud control policy, In this embodiment, the data of the renewable energy utilization rate (renewable energy utilization policy) is stored. The renewable energy rate is the percentage of electricity consumed by renewable energy sources in a given period of time. However, other definitions may be used.

[0034] The workload table 8500 stores information about each workload (batch workload and interactive workload). The server stores and accumulates information about the execution schedule of the active workload. The device 3000 and the storage device 4000 perform the following operations according to the workload table 8500. Each workload is executed using

[0035] The workload power consumption forecast distribution table 8600 shows the forecast of the power consumption of each workload. It stores information about the distribution of values.

[0036] The predicted workload table 8650 stores the predicted workload for each future time slot. Stores information. Next, specific examples of each database will be described.

[0037] (DC Power Estimation Table) 3 is a diagram showing an example of the DC power prediction table 8100. DC power prediction table 8100 is a time slot ID 8110 in which identification information of a time slot is set, a prediction Or the time 8120 at which the target time of the actual measurement is set, the renewable energy at that target time The predicted value of the amount of power generated by the renewable energy source (the amount of power that can be provided to the data center 1000) is set. Energy generation forecast 8130, actual measured renewable energy at the target time Renewable energy power generation actual measurement 8140, where the actual value of the power generation is set, at the target time Renewable energy prices, which set the forecast price per unit of renewable energy electricity Price forecast 8150, the actual renewable energy price set at the target time is set Actual renewable energy prices measured 8160, for a given power system (e.g., a commercial power system) A system in which the predicted price of electricity per unit amount (for example, 1 kW) at a target time is set Price forecast 8170 and the unit price actually set in the above power system at the target time Each data of 8180 grid price measurements where the price of electricity per unit (e.g. 1kW) is set It consists of one or more records containing items.

[0038] The actual measured values ​​and performance values ​​in the DC power prediction table 8100 are input by the user. Alternatively, the information may be automatically acquired from a predetermined database.

[0039] (Time slot table) FIG. 4 is a diagram showing an example of the time slot table 8200. The bus 8200 includes a timeslot ID 8210 in which identification information of a timeslot is set. The time 8220 at which the start time of the time slot is set, Power consumption of the data center 1000 (server device 3000, storage device 4000, The predicted value of the power consumption for all facilities or equipment in the data center, including air conditioning equipment (not shown) The power consumption forecast 8230 that is set, the data actually measured in that time slot The actual power consumption value of the center 1000 is set as the power consumption actual measurement 8240, and the time slot The predicted power consumption of batch workloads (hereafter referred to as batch power consumption) in the The batch power consumption forecast 8250 in which the batch work load in the time slot is set The actual power consumption of the batch is set to 8260, and the time slot A power consumption target value 8270 in which a target value for power consumption of the data center 1000 in , a parameter α 8280 in which the parameter α in that time slot is set, and Each data of risk tolerance 8290 where the risk tolerance for that time slot is set It consists of one or more records containing items.

[0040] (Delay limit time forecast distribution table) 5 is a diagram showing an example of a delay limit time prediction distribution table 8300. The predictive distribution table 8300 is a time slot table in which the identification information of future time slots is set. The delay limit time of the workload in the time slot is set. The delay limit time 8320 to be set and the time slot having the delay limit time A prediction value of the number of workloads in the workload is set to 8330, each of which has one or more data items. It consists of the above records.

[0041] (User Policy Table) 6 is a diagram showing an example of the user policy table 8400. User Policy Table 8400 is a policy ID 8410 for which identification information of a user policy is set, Renewable Energy Target Rate 8420, in which the target value (target rate) of the renewable energy rate in Si is set , the start date and time 8430 at which the application start date and time of the user policy is set, A goal achievement date 8440 for which a deadline for achievement is set, and a reproducible status in the user policy One or more data items of the control policy 8450 for setting the control policy for energy It consists of the above records.

[0042] In addition, in the Control Policy 8450, "COST" refers to the use of renewable energy as well as The actual renewable energy utilization rate exceeds the target rate, with an emphasis on the cost of electricity utilization. In cases where renewable energy sources are insufficient, the use of renewable energy will be restricted (and the shortfall will be made up by using electricity from the power grid). "RE" stands for renewable energy. The actual utilization rate of renewable energy has exceeded the target rate. This means that there are no particular restrictions on renewable energy sources (maximum use of renewable energy). The content of the control policy 8450 shown here is an example, and the cost and utilization of renewable energy are also included. It is also possible to set other policy information in terms of the balance of

[0043] In this embodiment, the data in the user policy table 8400 is input in advance by the user. However, it may be set or changed automatically.

[0044] (Workload table) 7 is a diagram showing an example of a workload table 8500. Workload Table 8500 is a workload ID 8510 in which the identification information of the workload is set, A power consumption prediction 8520 is a system for predicting the power consumption of a workload. The actual power consumption measurement value of the workload is set by the power consumption measurement value 8530. The time when the information is initially set in the management computer 2000 as an execution schedule (the time when the information is input) The execution time 8540 is set to the execution timing of the workload. Line Schedule 8550, for that workload, IT workload control program The execution schedule after the change is set by the 8900. The latency limit for the workload is set by the user. The delay limit 8570 that is set for that workload, and the queue flag is set for that workload. Queue flag 8580 data items each having information indicating whether or not It consists of the following records.

[0045] If the workload is an interactive workload, the execution schedule is The value of 50 is automatically set to the same value as the input time 8540. Also, the queue flag 85 80 indicates that the execution timing indicated by the execution schedule 8550 will be delayed. If the queue flag is confirmed, it will be automatically set to "Y". State.

[0046] (Workload Power Consumption Forecast Distribution Table) FIG. 8 is a diagram showing an example of a workload power consumption prediction distribution table 8600. The workload power consumption forecast distribution table 8600 is a table for which the identification information of the workload is set. The workload ID 8610, the range of predicted power consumption for that workload is set The power 8620 and the probability 8630 that the predicted value of the power consumption is realized are set. It consists of one or more records with data items. Workload Power Consumption Prediction Distribution Table The Bull 8600 generates data as historical power consumption data for each workload is acquired. (The distribution of predicted power consumption values ​​is statistically calculated) and updated.

[0047] (Predicted Workload Table) FIG. 9 is a diagram showing an example of the predicted workload table 8650. The Bull 8650 is configured with information identifying the predicted future timeslot workload. The predicted workload ID 8655, the time when the prediction for that workload was made, 8660, and a time slot 866 indicating the time slot for which that workload was predicted. 5. Power Consumption Prediction 867, where the predicted power consumption value for the predicted workload is set. 0, the predicted value of the delay limit time in the predicted workload is set. Measurement 8675 and the queue flag is set for that predicted workload. The queue flag 8680 has one or more data items each having information indicating whether It consists of code.

[0048] Each of the programs described above is executed by the processing unit 11000 (the main memory unit 12000 or the memory The program is executed by reading out the program stored in the memory device 8000. The program can be distributed, for example, by recording it on a recording medium. 000 is provided in whole or in part as a virtual server provided by, for example, a cloud system. Virtual information processing provided by using virtualization technology and process space isolation technology, such as a The management computer 2000 may provide the following: All or part of the functions are provided by the cloud system through API (Application Programming Interface). This may be realized by a service provided via a ramming interface or the like. Next, the processing executed by the management computer 2000 will be explained.

[0049] <Workload control processing> FIG. 10 illustrates a workload control process that controls each workload in the data center 1000. FIG. 1 is a flow diagram for explaining an outline of a workload control process. (e.g., every hour), at a predetermined time interval (e.g., at a predetermined time at the start of each time slot), It is executed before the scheduled time (before the scheduled time), or repeatedly at a predetermined timing (a time specified by the user).

[0050] First, the management computer 2000 calculates the amount of electricity (renewable) used to operate the data center 1000. electricity generated and prices, data center The power consumption of the data center 1000 and the delay limit of each workload in the data center 1000 The data update process S1 is executed to predict the distribution between the data points and accumulate the past data. The data update process S1 will be described later in detail.

[0051] The management computer 2000 updates the data based on the data predicted and accumulated in the data update process S1. Each workload in data center 1000 is scheduled to be executed in the nearest time slot. Deploy (input) the workload that will actually be executed. The workload deployment process S2 is executed. The details of the workload deployment process S2 will be described later. The above process is executed repeatedly. Next, the data update process S1 will be described in detail.

[0052] <Data update process> FIG. 11 is a flow diagram illustrating the details of the data update process S1. The power consumption price prediction program 9000 predicts the playback time in each time slot after the current time. The amount of electricity generated from renewable energy and the price per unit of electricity are predicted (S10). Specifically, For example, the power consumption price prediction program 9000 calculates each record of the DC power prediction table 8100. Code time 8120, renewable energy power generation measurement 8140, renewable energy price Each of the 8160 actual measured values ​​is acquired, and the acquired values ​​are processed by a predetermined algorithm (e.g., time series Based on the results of the analysis (performing column analysis, performing machine learning to create a predictive model), Predict the amount of renewable energy generated and the price per unit of electricity in the time slot The power consumption price prediction program 9000 calculates the predicted power generation amount and each price as a DC power forecast. The renewable energy power generation forecast 81 for each time slot of the record in the measurement table 8100 30 and renewable energy price forecast 8150, respectively.

[0053] The power consumption price prediction program 9000 is capable of predicting the reproducibility of the past time slots. The amount of generated energy and the price per unit of electricity are calculated by a specified device (e.g., an external database). The power generation amount and price are obtained from the DC power forecast table 810. The actual renewable energy power generation amount of the record related to the time slot of 0 is 8140 and The actual renewable energy price is stored as an actual value in the field 8160 (S10).

[0054] Furthermore, the power consumption price prediction program 9000 predicts the power consumption price in each time slot after the current time slot. The price per unit of electricity in the power grid is predicted (S11). For example, the power consumption price prediction program 9000 calculates the power consumption price of each record in the DC power prediction table 8100. The time 8120 of the grid price measurement and the grid price measurement 8180 are acquired, and the acquired values ​​are Algorithms (e.g., performing time series analysis or applying machine learning to create predictive models) Based on this, the power per unit power of the power system in each time slot after the present is calculated. The power consumption price prediction program 9000 predicts each predicted price based on the DC power The system price forecast 8170 for each time slot record in the forecast table 8100 is stored. .

[0055] The power consumption price prediction program 9000 uses the power system price in the past time slots The price per unit of electricity of the grid is calculated by a predetermined device (e.g., an external database or server). ) and the obtained price is input to the corresponding time slot of the DC power forecast table 8100. The actual value is stored in the system price actual measurement 8180 of the relevant record (S11).

[0056] Furthermore, the power consumption price prediction program 9000 predicts the power consumption price in each time slot after the current time slot. The power consumption of the entire data center 1000 and the power consumption of the batch workload are forecasted. Specifically, for example, the power consumption price prediction program 9000 predicts the power consumption price in Thailand. The time 8220 of each record in the slot table 8200, the actual power consumption measurement 8240, and Each value of the batch power consumption measurement 8260 is acquired, and each acquired value is processed by a predetermined algorithm. (e.g., perform time series analysis or use machine learning to create predictive models) The power consumption and backup power consumption of the entire data center 1000 in each time slot after the current time slot are calculated. The power consumption price prediction program 9000 predicts the power consumption of a workload. The power consumption measured is calculated by dividing the total power consumption by the power consumption of the record of each time slot in the time slot table 8200. The power consumption prediction 8230 and the batch power consumption prediction 8250 are stored.

[0057] The power consumption price prediction program 9000 uses data from past time slots. The power consumption of the entire center 1000 and the power consumption of the batch workload are calculated based on the power consumption of a specified device (e.g. For example, the amount of power consumed is obtained from an external database or server, and each amount of power consumed is compared with the amount of DC power consumed. The actual power consumption measurement 8240 of the record related to the time slot in the prediction table 8100 and The actual values ​​are stored in the batch power consumption measurement 8260 (S12).

[0058] Furthermore, the power consumption price prediction program 9000 predicts the power consumption price in each time slot after the current time slot. In this way, the delay limit time of the batch workload of the data center 1000 is predicted (S13). Specifically, for example, the power consumption price prediction program 9000 uses a workload table 8500 execution schedule 8550 (time when past workload was actually executed) or the changed execution schedule 8560 and the delay limit time 8570 are obtained, and Run a predefined algorithm (e.g., time series analysis, machine learning, predictive model) on the values. Based on the time slots after the current time, each batch workload is The power consumption price prediction program 9000 predicts the distribution of the delay limit time of the The distribution data of the delay limit time is input to each time slot of the delay limit time prediction distribution table 8300. The delay limit time 8320 and the number 8330 of the record of the The power consumption price prediction program 9000 predicts the number of predicted items 833 in each time slot. Create a new record in the predicted workload table 8650 for the number of 0s, and The data of the time slot and delay limit time are input to the time slot of the predicted workload table 8650. The delay limit time prediction 8670 is stored in the delay limit time prediction 8665 and the delay limit time prediction 8670, respectively. The theory is repeated.

[0059] Next, FIG. 12 is a flow diagram illustrating the workload deployment process S2. The parameter determination program 8700 determines the time slots from the most recent time slot onward. Then, a parameter determination process S20 is executed to determine a pattern of the parameter α in the

[0060] In addition, the risk tolerance calculation program 8800 calculates the risk tolerance for each time slot after the most recent time slot. A risk tolerance calculation process S21 is executed to calculate a risk tolerance for a slot.

[0061] Then, the IT workload control program 8900 determines the parameter in the parameter determination process S20. The target value of the power consumption calculated based on the pattern of the parameter α and the risk tolerance are Based on the risk tolerance calculated in the calculation process S21, The batch workload to be used is determined, and the determined batch workload is applied to the interactive workstation. Execute the IT workload control process S22 to be deployed together with the cloud. The following describes the parameter determination process S20, the risk tolerance calculation process S21, and the IT workload. The details of the mode control process S22 will now be described.

[0062] <Parameter determination process> FIG. 13 is a flow diagram illustrating an example of the parameter determination process S20. In the data determination process, the operation policy specified by the user in the user policy table 8400 is executed. To realize this, we deploy the ideal batch workload for each future time slot. The deployment amount of the batch workload is determined by the amount of data that is allocated to each time slot of the batch workload. The parameter α is determined by specifying the execution ratio in the set. The amount of batch workload deployed is determined by identifying the time slot. The amount of power consumed in a given area can be calculated, and the utilization rate of renewable energy (hours) can be calculated. It is possible to calculate the unit renewable energy rate and the cost of using renewable energy. Therefore, the user can specify the parameter α by calculating it. Then, the optimal parameter α that realizes the above operation policy is determined.

[0063] First, the parameter determination program 8700 executes a parameter pair creation process S1000. Then, one or more lists (patterns) of parameters α for each time slot from the most recent time slot onwards are generated. Execute the parameter creation process S1000 to create the parameters. Details will be given later.

[0064] The parameter determination program 8700 performs the parameter creation process S1000. One of the meter α patterns is obtained (S1010).

[0065] The parameter determination program 8700 calculates the parameter α in all time slots. The utilization rate of renewable energy (hourly renewable energy rate f_re) and the renewable energy The cost f_cost relating to the use of energy is calculated (S1020). , the cost of using renewable energy f_cost is the cost of using renewable energy Only electricity costs are considered, but this includes grid power costs and other costs. Good too.

[0066] That is, the parameter determination program 8700 determines the The value obtained by dividing the batch power consumption forecast 8250 by the power consumption forecast 8230 and the batch power consumption The power consumption target value is calculated by adding the value of 8250 and the value multiplied by the parameter α. Renewable energy power generation forecast 8230 of the power forecast table 8100 (data center 10 By referring to the renewable energy available at 00, the power consumption target value can be calculated. Determine whether renewable energy can cover the costs, and if not, This determines that the power grid will cover the portion of the power consumption. can calculate the renewable energy utilization rate (hourly renewable energy rate f_re) In addition, the parameter determination program 8700 predicts the power consumption of renewable energy. This is multiplied by the renewable energy price forecast 8150 in the DC power forecast table 8100. Then, the cost of using renewable energy, f_cost, can be calculated. .

[0067] Next, the parameter determination program 8700 determines whether the control strategy is "COST" or not. For example, the parameter determination program 8700 checks the user policy (S1030). Refer to table 8400 and check that the value of control policy 8450 in the latest record is "COST". Check whether it is possible.

[0068] If the control strategy is "COST" (S1030: YES), the parameter determination program The RAM 8700 executes the process of S1070, and if the control policy is "RE" (S10 30: NO), the parameter determination program 8700 executes the process of S1040.

[0069] In S1040, the parameter determination program 8700 determines, as the first objective function, Set the unit renewable energy rate f_re. In this case, the user decides to use renewable energy. Therefore, we determine the parameter α that maximizes the hourly renewable energy rate f_re. do.

[0070] The parameter determination program 8700 determines the first objective function for all patterns of the parameter α. It is confirmed whether the first objective function has been set for all parameters of the parameter α (S1050). If the turn is set (S1050: YES), the parameter determination program 87 00 executes the process of S1060 and sets the pattern of the parameter α for which the first objective function is not set. If there is a line (S1050: NO), the parameter determination program 8700 The processes from S1010 onwards are repeated to obtain the pattern of the parameter α.

[0071] In S1060, the parameter determination program 8700 executes the parameter creation process S10 Among the multiple parameter α patterns created in 00, the value of the first objective function is the largest. The pattern of the parameter α is identified, and the identified result is stored in the time slot table 8200. Set the parameter α8280 of the record related to each time slot. The decision process S20 ends.

[0072] On the other hand, in S1070, the parameter determination program 8700 sets the first objective function as In this case, the user decides to use renewable energy. In addition, we also place importance on the costs of electricity usage, so we are looking at the actual renewable energy usage rate. If the hourly renewable energy rate is higher than the target rate, the cost is reduced as long as the hourly renewable energy rate does not fall below the target rate. We determine the parameter α that minimizes f_cost. However, in the case of actual renewable energy, If the utilization rate of renewable energy is below the target rate, the utilization rate of renewable energy above the target rate will be increased. The parameters that prioritize achieving the utilization rate and maximize the hourly renewable energy rate f_re are Determine α.

[0073] In addition, the parameter determination program 8700 performs time unit re-entry in the first objective function. The renewable energy rate f_re is equal to or exceeds the renewable energy target rate (meets the minimum renewable energy rate requirements) The parameter determination program 870 sets a constraint that the 0 is the renewable energy target rate 84 of the latest record in the user policy table 8400. The value of 20 will be used as the renewable energy target rate.

[0074] The parameter determination program 8700 determines the first objective function based on the pattern of all parameters α. It is confirmed whether the first objective function has been executed for all parameters (S1080). If the pattern is α (S1080: YES), the parameter determination program The RAM 8700 executes the process of S1090 and sets the parameters for which the first objective function is not set. If there is a pattern of α (S1080: NO), the parameter determination program 8700 Then, the process from S1010 onwards is repeated to obtain the pattern of the parameter α.

[0075] In S1100, the parameter determination program 8700 determines parameters that satisfy the constraint conditions. Check whether a pattern of parameter α exists. If it exists (S1100: YES), the parameter determination program 8700 goes to S11 10, if there is no pattern of parameter α that satisfies the constraints (S 1100: NO), the parameter determination program 8700 executes the process of S1120.

[0076] In S1110, the parameter determination program 8700 executes the parameter creation process S10 Among the multiple parameter α patterns created in 10, the value of the first objective function is the smallest. The pattern of the parameter α in this case is identified, and the identification result is stored in the time slot table 82. The parameter α8280 of the record related to each time slot of 00 is set. The meter determination process S20 ends.

[0077] In S1120, the parameter determination program 8700 determines, as the second objective function, Set the hourly renewable energy rate f_re.

[0078] The parameter determination program 8700 creates the parameter in the parameter creation process S1010. Among the multiple parameter α patterns, the parameter α that maximizes the value of the second objective function is selected. The pattern of the meter α is identified, and the identification result is stored in each time slot table 8200. The time slot is set in the parameter α8280 of the record related to the time slot (S1130). The parameter determination process S20 ends.

[0079] <Parameter creation process> 14 and 15 are flow charts for explaining the details of the parameter creation process S1000 (paper (For the sake of space, it is divided into two figures.) In the parameter creation process S1000, all possible A set of parameters α is created. The batch workload is determined by the user and the delay limit time is 8570 Since the execution time is set, it cannot be delayed forever. The amount of batch workloads deployed in a slot cannot be determined completely freely. Therefore, the parameter α in each time slot must also be limited. The parameter α is restricted based on the information on the delay limit time prediction distribution in the delay limit time prediction distribution table 8300. A set of possible parameters α within the range is created. As shown in FIG. 14, the parameter determination program 8700 selects the most recent time slot. Specifically, the parameter determination program 8700 selects a time The lot table 8200 is referenced, and the time 8220 indicates the closest future time to the current time. Select the records that are

[0080] The parameter determination program 8700 determines whether the selected time slot is the last time slot. Specifically, the parameter determination program 87 determines whether the parameter is a 00 means that the selected time slot is the last time slot with pre-set timing. The time slot is checked to see if it is the time slot (for example, 12 hours later).

[0081] If the selected time slot is the last time slot (S2010: YES), The parameter determination program 8700 executes the process of S2060 and selects the time slot. If the time slot is not the last time slot (S2010: NO), the parameter determination program The Ram 8700 runs the S2020.

[0082] In S2060, the parameter determination program 8700 determines that this is the last time slot. For the purpose of the test, all batch workloads are deployed and the selected time slot ( The parameter α of the last time slot is set to 1, and the parameter creation process S1000 is Finish.

[0083] In S2020, the parameter determination program 8700 determines whether the selected time slot is It is determined whether it is the first time slot. Specifically, the parameter determination program 8700 is the time of the record selected in S2000, 8220, which is the closest future time to the current time. Check whether the time is displayed.

[0084] If the selected time slot is the first time slot (S2020: YES), The parameter determination program 8700 executes the process of S2030 and selects the time slot. If the time slot is not the first time slot (S2020:NO), the parameter determination program The RAM 8700 executes the processing of the S2070.

[0085] S2030 to S2050 are for when the selected time slot is the first time slot. In the first time slot, if a batch has already been assigned to that time slot, The workloads are set and delays are calculated based on the information in those batch workloads. The limit time is acquired and power consumption is predicted. On the other hand, S2070 to S2110 are selected This is the process when the selected time slot is not the first time slot. If they are not batches, then there is no batch workload set for those time slots. Therefore, those batch workloads are not yet registered in the workload table 8500. Therefore, it is necessary to predict the batch workload, and the delay limit time prediction is required. Predicts profits and power consumption.

[0086] In step S2030, the parameter determination program 8700 The configured batch workload and the timeslot immediately preceding the selected timeslot Get all the batch workloads currently stored in the queue. The parameter determination program 8700 refers to the workload table 8500 and selects The data of the records related to the time slots in the queue and all records with queue flag 8580 set to "Y" Get the record data.

[0087] The parameter determination program 8700 calculates each batch workload obtained by the S2030. Then, the delay limit time is rearranged in ascending order (earliest order) (S2040). The data determination program 8700 refers to the workload table 8500 and The records are sorted in order of the delay limit time 8570 of each record that is closest to the current time. Replace.

[0088] The parameter determination program 8700 performs the batch workload sorting in S2040. The total workload power consumption forecast value PB is calculated as the total of the power consumption forecast values ​​Pb of Specifically, the parameter determination program 8700 Refer to the 8500 bus and check the power consumption of each record related to the workload sorted by the S2040. The values ​​of the force predictions 8520 are summed up. Then, the process of S2110 is performed.

[0089] On the other hand, in S2070, the parameter determination program 8700 The number of batch workloads deployed (executed) on the batch Specifically, the parameter determination program 8700 refers to the delay limit time forecast distribution table 8300 and selects the selected time slot The delay limit time 8320 and the number 8330 of the record related to the item are obtained.

[0090] The parameter determination program 8700 determines the batch work in the selected time slot. The estimated power consumption per load, Pb, is calculated as the batch power consumption for the selected time slot. The predicted power consumption is calculated by dividing the predicted power consumption by the number of batch workloads calculated by S2070. , is calculated (S2080).

[0091] Specifically, the parameter determination program 8700 uses the time slot table 8200 and obtain the batch power consumption forecast 8250 for the record related to the selected time slot. Then, the acquired power consumption value is divided by the number 8330 acquired in S2070.

[0092] The parameter determination program 8700 calculates the predicted value of the batch power consumption as a predicted work The data is stored in the load table 8650 (S2085).

[0093] Specifically, the parameter determination program 8700 selects the record having the latest predicted time 8660. In the code, the record with the time slot 8665 that matches the selected time slot The predicted value Pb calculated in S2080 is stored in the power consumption prediction 8670 of the mode.

[0094] The parameter determination program 8700 obtains the predicted power consumption value Pb in S2070. The batch workload and the batch workload of the previous time slot that is currently in the queue The batch workload that has been accumulated as a parameter is then acquired. The 8700 sorts the acquired batch workloads in order of shortest delay limit time ( S2090).

[0095] Specifically, for example, the parameter determination program 8700 determines the workload table 8 500 and obtains the data of the record whose queue flag 8580 is "Y". The parameter determination program 8700 refers to the predicted workload table 8650, Data of the record whose predicted time 8660 is the latest and whose queue flag 8675 is "Y", The predicted time 8660 is the latest and is the same as the selected time slot 8665. The parameter determination program 8700 obtains the data of the record. The delay limit time prediction 8675 of the table 8650 and the workload table 8 The delay limit time of 8570 for the 500 records is to be sorted.

[0096] The parameter determination program 8700 is used to sort the batch workloads in the S2090. The total workload power consumption forecast value PB is calculated as the total of the power consumption forecast values ​​Pb of Specifically, the parameter determination program 8700 performs the queuing in S2090 (S2100). -flag 8580 is "Y", predicted time 8660 is the latest and queue flag 8675 is "Y " or the predicted time 8660 is the latest and is the same as the selected time slot. The power consumption forecast 8520 or power consumption forecast 8670 for each record that was previously set as 8665 The predicted values ​​Pb shown in FIG.

[0097] In S2110, the parameter determination program 8700 executes S2050 or S2100 Among the workloads for which the predicted power consumption value Pb is calculated, Identify all workloads that cannot be scheduled (delayed) in a later timeslot. The sum of the predicted power consumption values ​​Pb of each workload (non-delayable workload power consumption forecast Specifically, the parameter determination program 8700 calculates the total measurement value PB'. The workload table 8500 and the predicted workload table 8650 are referenced to determine the delay limit. The time between the selected time slot 8570 and the delay limit time prediction 8675 is the same as the time between the selected time slot 8570 and the delay limit time prediction 8675. Identify workloads for which the records are the same, and estimate the power consumption of each identified workload. The total value of Pb is the total predicted power consumption of the non-delayable workloads, PB'. The power consumption due to the switch workload is always consumed in that time slot.

[0098] Then, as shown in FIG. 15, the parameter determination program 8700 PB' / PB is set as the lower limit α_min of the slot parameter α (S212 0).

[0099] In this way, the parameter determination program 8700 performs batch processing with a short delay limit time. This will give priority to the cloud load and make it run at an earlier time.

[0100] Then, the parameter determination program 8700 selects the selected parameter that is equal to or greater than the lower limit α_min. The value of α for the time slot is arbitrarily determined to be one or more (S2130).

[0101] The parameter determination program 8700 sets the power consumption judgment value P to 0 (S214 0).

[0102] The parameter determination program 8700 determines the power consumption of each workload for the judgment value P. The predicted values ​​Pb of the workloads are added in the order sorted in S2040 or S2090. (S2150, S2160). The parameter determination program 8700 performs this addition as a judgment The value P is the multiplication of PB calculated in S2050 or S2100 and α set in S2130. Repeat until the value is exceeded (S2170: NO).

[0103] If the power consumption value P is equal to or greater than the product of PB and α (S2170: YES), The parameter determination program 8700 assigns the workload that was not multiplied to the queue. Specifically, the parameter determination program 8700 accumulates the work The load table 8500 and the predicted workload table 8650 are referenced, and the The queue flag 8580 or queue flag 86 for the records related to the workload that did not Set 80 to "Y".

[0104] The parameter determination program 8700 selects the next time slot of the selected time slot. and repeats the process from S2010 onwards (S2190).

[0105] In addition, the parameter determination program 8700 selects the time stamp in S2130. Multiple values ​​can be used as the lot α value (for example, if the lower limit α_min is 0.1, 0.2, 0.3, 0.4, ..., 1) and set the S2140 or lower for each. By carrying out the following process, multiple patterns of the parameter α are created.

[0106] <Risk tolerance calculation process> FIG. 16 is a flow diagram illustrating an example of the risk tolerance calculation process S21. The risk tolerance calculation program 8800 calculates the risk tolerance for each type determined in the parameter determination process S20. Based on the parameter α of each time slot, the power consumption target value of each time slot is calculated. Each power consumption target value is stored in the power consumption target value 8270 of the time slot table 8200. Deliver (S3000).

[0107] For example, the risk tolerance calculation program 8800 may use the time slot table 8200 as Refer to the power consumption forecast 8230 for each time slot and record the power consumption forecast 8230 for each time slot. The value obtained by dividing 250 and multiplying the batch power consumption 8250 and the parameter α is Add.

[0108] The risk tolerance calculation program 8800 determines the parameter α in the parameter determination process S20. One of the calculated time slots is selected (S3010).

[0109] The risk tolerance calculation program 8800 takes the parameter α of the selected time slot. Benefit (S3020).

[0110] The risk tolerance calculation program 8800 checks whether the acquired parameter α is 1 or not. (S3030).

[0111] If the acquired parameter α is 1 (S3030: YES), the risk tolerance calculation process The program 8800 executes the process of S3040, and if the obtained parameter α is not 1, (S3030: NO), the risk tolerance calculation program 8800 performs the process of S3070. Execute.

[0112] In S3070, the risk tolerance calculation program 8800 The risk tolerance for the time slot is set to the minimum value (1 in this embodiment) and The selected time slot table 8200 (specifically, the time slot table 8200) The risk tolerance of the record related to the slot is stored in S3060. Processing is carried out.

[0113] In S3040, the risk tolerance calculation program 8800 calculates the risk tolerance from the renewable energy perspective. and executes a process S3040 for calculating a risk tolerance for each viewpoint to calculate a risk tolerance for a cost viewpoint. The details of the viewpoint-specific risk tolerance calculation process S3040 will be described later.

[0114] The risk tolerance calculation program 8800 uses the renewable energy perspective risk calculated by S3040. Based on the risk tolerance and cost perspective risk tolerance, the risk for the selected time slot is The tolerance is calculated (S3050).

[0115] For example, the risk tolerance calculation program 8800 calculates the risk tolerance and cost from the renewable energy perspective. Calculate the multiplication value of the viewpoint risk tolerance or the exponentiation value (e.g., square root) of the multiplication value. The calculation method described here is an example, and the risk tolerance from a renewable energy perspective and the risk tolerance from a cost perspective If the magnitude of each tolerance value is reflected in the risk tolerance for the selected time slot, Other calculation methods may also be used.

[0116] In addition, the risk tolerance calculation program 8800 selects a control policy for the selected time slot. For example, the risk tolerance calculation program 8800 may be reflected in the risk tolerance. gets the control policy 8450 of the latest record in the user policy table 8400. If the control policy is "RE", the renewable energy perspective risk tolerance value or the renewable energy perspective The risk tolerance is multiplied by a predetermined coefficient to obtain the risk tolerance for the selected time slot. It is also possible to use the following.

[0117] The risk tolerance calculation program 8800 determines the parameters by the parameter determination process S20. Determine whether or not the risk tolerance has been calculated for all time slots for which α has been calculated. (S3060).

[0118] If the risk tolerance is calculated for all time slots (S3060: YES), , the risk tolerance calculation process is completed, and there are time slots for which the risk tolerance has not been calculated. In this case (S3060: NO), the risk tolerance calculation program 8800 calculates the risk tolerance as follows: In order to select a time slot that has not been calculated, the processes from S301 onwards are repeated.

[0119] <Risk tolerance calculation process for each viewpoint> FIG. 17 is a flow diagram illustrating the details of the viewpoint-specific risk tolerance calculation process S3040. The risk tolerance calculation program 8800 calculates the renewable energy The value becomes smaller as the amount of power generated by renewable energy sources exceeds the target value for power consumption. Perspective risk tolerance (the renewable energy rate will not reach the target rate, and the utilization of renewable energy will be insufficient) The tolerance for the risk that the risk of the failure does not occur is calculated (S4000).

[0120] For example, the risk tolerance calculation program 8800 may calculate the risk tolerance by multiplying the power consumption by a predetermined negative coefficient. The renewable energy perspective risk tolerance is calculated based on the target value minus the amount of renewable energy generated. The formula shown here is just an example, and other formulas that represent a monotonically decreasing value may be used.

[0121] The risk tolerance calculation program 8800 calculates the renewable energy Check whether the price per unit of electricity of the customer is greater than the price per unit of electricity of the grid (S4010). Specifically, the risk tolerance calculation program 8800 is a DC power prediction test. Refer to Table 8100 and check the renewable energy of the record for the selected time slot. This is confirmed by identifying the values ​​of the price forecast 8150 and the system price forecast 8170.

[0122] The price per unit of renewable energy is higher than the price per unit of grid power. If not (S4010: YES), the risk tolerance calculation program 8800 executes S4030 The price per unit of renewable energy power is calculated based on the price per unit of grid power. If the price is lower than the price (S4010: NO), the risk tolerance calculation program 8800 Then, execute processing of S4020.

[0123] In S4030, the risk tolerance calculation program 8800 The higher the amount of renewable energy generated exceeds the target value for power consumption, the higher the value. Therefore, the cost-perspective risk tolerance (excessive use of renewable energy) This completes the calculation of risk tolerance for each viewpoint. Processing S3040 ends.

[0124] For example, the risk tolerance calculation program 8800 may calculate the risk tolerance by multiplying the power consumption by a factor of (a predetermined positive value coefficient)×( Calculate the renewable energy perspective risk tolerance based on the target power output - renewable energy power output Note that the formula shown here is just an example, and other formulas that represent a monotonically increasing value may be used.

[0125] In S4020, the risk tolerance calculation program 8800 The value is smaller the more the amount of renewable energy generated exceeds the target value for power consumption. The cost-perspective risk tolerance is calculated so that 3040 ends.

[0126] For example, the risk tolerance calculation program 8800 may calculate the risk tolerance by multiplying the coefficient (a predetermined negative value) by the power consumption Calculate the risk tolerance from the perspective of renewable energy based on (the target value of power - the power generation amount of renewable energy). Note that the formula shown here is just an example, and other formulas representing monotonic decrease may also be used.

[0127] <IT workload control process> Figures 18 and 19 are flowcharts for explaining the IT workload control process S22 (divided into two figures for convenience). In this IT workload control process, the actual batch workload to be deployed is determined so as to approach the target power consumption value in each time slot determined by the calculated parameter α value. At this time, in addition to approaching the target power consumption value, considering the deviation from the predicted power consumption of each batch workload, the batch workload to be deployed is determined so that the sum of the deviations from the predicted power consumption of each batch workload is minimized in the time slots with a small risk tolerance.

[0128] As shown in Figure 18, the IT workload control program 8900 calculates a weight value (the first weight value) related to the risk tolerance in the most recent time slot (S5000).

[0129] For example, the IT workload control program 8900 sets the reciprocal of the risk tolerance in the most recent time slot as the first weight value. Note that the calculation method of the first weight value shown here is just an example, and the first weight value can be a monotonic decreasing function with respect to the risk tolerance.

[0130] The IT workload control program 8900 calculates a weight value (the second weight value) related to the risk tolerance in the time slots after the most recent time slot (S5010). .

[0131] ​​​​​​​​ For example, the IT workload control program 8900 may The inverse of the average value of the risk tolerance in the time slot is set as the second weight value. The calculation method of the second weighting value shown in is an example, and the second weighting value is a simple calculation for the risk tolerance. It can be a decreasing tone function.

[0132] The IT workload control program 8900 calculates the batch The predicted value of the power consumption other than the workload is set to the variable Po (S5020). The IT workload control program 8900 refers to the time slot table 8200. Then, the batch power consumption is calculated from the power consumption prediction 8230 value of the record related to the most recent time slot. The value obtained by subtracting the value of the force prediction 8250 is set as Po.

[0133] The IT Workload Control Program 8900 is set to the nearest timeslot. Batch workload and the currently queued time slot immediately preceding the most recent time slot. The batch workloads stored in the database are acquired (S5030). The IT Workload Control Program 8900 performs the following operations on each record in the Workload Table 8500: The execution schedule 8550 of the queue is referenced, and the record related to the nearest time slot and the queue are The record with the record flag 8580 set to "Y" is obtained.

[0134] The IT workload control program 8900 performs the batch workload control for each batch workload acquired by S5030. The cards are sorted in order of shortest delay limit time (earliest) (S5040). The workload control program 8900 processes each record acquired in S5030 into The delay limit time 8570 of the records related to the time slot is sorted in order of proximity to the current time. Remember.

[0135] Then, the parameter determination program 8700 performs the rearrangement of each batch word for Po. The predicted power consumption of the loads is added up in the rearranged order (S5040). In the parameter determination program 8700, the consumption of each record sorted in S5030 is The power prediction 8520 value is added to Po in order (for batch workloads with the same delay limit time). is added to the predicted power consumption at once).

[0136] The parameter determination program 8700 determines the power consumption of Po in the addition process of S5040. When the target value is exceeded, the predicted power consumption is added to the batch workload time. The time slot ts is identified (S5050). As can be seen from the above, this time slot A batch workload may belong to a single batch workload.

[0137] The parameter determination program 8700 determines the batch workflow that belongs to the time slot ts. Create one or more batch workload sets that divide the tasks into two groups (e.g., If the number of batch workloads belonging to time slot ts is 3, then "0 and 1" and "1 and 2" are ", "2 and 1", and "1 and 0" (S5060).

[0138] The parameter determination program 8700 resets Po to the value in S5020 (S50 70).

[0139] Next, as shown in FIG. 19, the parameter determination program 8700 executes the process in S5060. One of the multiple pairs thus formed is selected (S5080).

[0140] The parameter determination program 8700 selects one group in the set selected in S5080. The batch workload of the group and the batch workload with the delay limit shorter (earlier) than the time slot ts The workload is stored as the first group, and the set selected by S5080 is The batch workload of the other group is the same as that of the other group, and the delay limit time is longer than the time slot ts. The "fast" and "slow" are stored together as a second group (S5090).

[0141] The parameter determination program 8700 determines the power consumption Po of the non-batch workload and the first Is the sum of the power consumption of the batch workload of the group close to the power consumption target value? It is determined whether or not this is the case (S5100).

[0142] For example, the parameter determination program 8700 uses Po set in S5070 and the first Power consumption of batch workload related to the group (power consumption of Workload Table 8500) The total value of the power consumption prediction (obtained from the 8520) is equal to or greater than the power consumption target value, and It is judged whether the error is less than the target value + a specified error tolerance (n%).

[0143] If the sum is close to the power consumption target value (S5100: YES), the parameter determination The determination program 8700 executes the process of S5110, and determines whether the sum is close to the power consumption target value. If not (S5100: NO), the parameter determination program 8700 selects another set. The process from S5080 onwards is repeated to select the desired one.

[0144] In S5110, the parameter determination program 8700 determines the first group and the second group. For the batch workload power consumption of a group, The value of evaluation function g, which indicates the severity of deviation from the measured value, is calculated.

[0145] For example, the parameter determination program 8700 uses the batch workflow acquired in S5080. The value of the evaluation function g for the group of nodes is calculated using the following formula:

[0146] Evaluation function g = (first weight value) × (sum of deviations of the first group) + (second weight value) × (sum of deviations in the second group)

[0147] As a method for calculating the sum of deviations in the first group, the parameter determination program 8700 For each batch workload in the first group, the predicted power consumption (w The power consumption forecast of the load table 8500 (obtained from 8520) and its historical power consumption The average or median predicted value of the workload power consumption forecast distribution table 8600 Calculate the absolute value of the difference between the power 8620 and the probability 8630, and The sum of the deviations in the second group is calculated in the same way.

[0148] Note that the evaluation function shown here is an example, and the degree of tolerance for the deviation between the predicted value and the actual value is Any function that takes this into consideration may be used.

[0149] The parameter determination program 8700 determines all of the workload sets created in S5060. It is checked whether the value of the evaluation function g has been calculated for all the work processes (S5120). If the value of the evaluation function g is calculated for the set of parameters (S5120: YES), The determination program 8700 executes the process of S5130 and does not calculate the value of the evaluation function g. If there is a workload pair (S5120: NO), the parameter determination program 870 0 repeats the processes from S5080 onwards to obtain that workload set.

[0150] In S5130, the parameter determination program 8700 The values ​​of the evaluation function g of each pair are compared, and the pair with the minimum value of the evaluation function g is searched for. Identify a first group and a second group to be assigned to the

[0151] Then, the parameter determination program 8700 selects the first group identified in S5130. Deploy the batch workload of S51 to run in the nearest time slot. 40). For example, the parameter determination program 8700 uses the workload table 8500 and executes the execution schedule after the change of the records for each batch workload in the first group. The time of the nearest time slot is set in the queue flag 8560. Set each to "N".

[0152] In addition, the parameter determination program 8700 determines the second group of parameters specified in S5130. Set the batch workload to queue (not run in the nearest timeslot) For example, the parameter determination program 8700 uses the workload table 8 500, and the queue flag of the record related to each batch workload of the second group is Set "Y" to each of 8580. This completes the IT workload control process S22. do.

[0153] After that, the server device 3000 and the storage device 4000 update the workload table 8 Each workload (job) is executed according to the contents of 500 (S5160).

[0154] <Workload Migration Information Screen> Fig. 20 is a diagram showing an example of the workload migration information screen 13000. Workload The migration information screen 13000 indicates that the IT workload control process S22 was not executed. In this case (when the batch job execution timing (time slot) is not changed), A pre-migration information display field 13 in which the power consumption value 13101 for each time slot is displayed 100 and the IT workload control process S22 in each time slot after the process is executed. The post-migration information display field 13200 displays the actual power consumption value 13102 of the IT work An effect display field that displays information indicating the effect of executing the cloud control process S22 13300 and the execution timing (time slot) by IT workload control process S22 Schedule change history display area that displays information about the batch workload for which changes were made 400 and an acknowledgement field 13500 for closing the workload migration information screen 13000. do.

[0155] In each of the pre-migration information display field 13100 and the post-migration information display field 13200, comparison and The actual amount of renewable energy generated in each time slot is displayed. will be done.

[0156] The effect display column 13300 displays the past performance calculated based on the IT workload control process S22. The rate of increase in the utilization rate and cost of renewable energy per unit time for a given period in the past Information such as the rate of decrease is displayed.

[0157] The schedule change history display field 13400 displays the schedule change history by the IT workload control process S22. Information on the batch workload whose execution time slot has been changed 13401 and the most recent Batches that have been shifted from one timeslot to a later timeslot Workload information 13402 is displayed.

[0158] It should be noted that past data is displayed on this workload migration information screen 13000. However, information on the batch workload (power consumption, etc.) for the upcoming time slot is may be displayed.

[0159] As described above, the workload control support system of this embodiment is a system that can calculate the delay limit of each workload. Based on the time and power consumption forecasts, the power consumption forecast for the future timeslots is calculated. The target value is to meet the target value of the renewable energy ratio and the cost conditions related to the use of renewable energy. Based on the calculated target value of the power consumption, the actual power consumption is calculated in the future time slot. Decide when to run each workload and run each workload at the decided timing. Make them carry out the procedure.

[0160] That is, the workload control assistance device of this embodiment sets the target value of power consumption in terms of cost and The target value of power consumption is determined based on the renewable energy rate. This allows the system to determine the timing of each workload to be executed, reducing costs and Considering the renewable energy rate, use of renewable energy and workflow according to user needs It becomes possible to execute the command.

[0161] In this way, the workload control support device of this embodiment is capable of running on renewable energy. Each workload can be cost-effectively controlled.

[0162] In addition, the workload control support device of this embodiment prioritizes either the renewable energy rate or the cost. The control policy that prioritizes the renewable energy ratio is specified by the user. In this case, the pattern of parameter α that optimizes the utilization rate of renewable energy is Based on the identified patterns, the target value of the power consumption is calculated, while emphasizing the cost. If the user specifies a policy to minimize the costs of using renewable energy, A parameter α that optimizes the power consumption is identified, and the target value of the power consumption is set based on the identified pattern. Calculate.

[0163] This allows appropriate consumption decisions to be made based on the choice between prioritizing costs and the renewable energy ratio. A power target can be set.

[0164] In addition, the workload control support device of this embodiment is capable of controlling a workload when a control policy that places importance on cost is specified. If the contract is concluded, the user's renewable energy ratio target will be met and the use of renewable energy will be The target value of the power consumption is calculated based on the pattern of parameter α that optimizes the cost of Put out.

[0165] This will allow the cost-focused electricity target to be set only if the renewable energy ratio target is met. This will ensure the stable use of renewable energy. It is possible.

[0166] In addition, the workload control support device of this embodiment uses a parameter α Based on the parameter α pattern and the predicted value of the power consumption, The target value for power consumption is calculated so as to satisfy the conditions for renewable energy rate and cost.

[0167] This allows the power consumption to be reduced by appropriately allocating the execution timing of each workload that can be delayed. A prediction of the amount can be calculated.

[0168] In addition, the workload control support device of this embodiment is capable of reducing the amount of power consumption due to the uncertainty of the predicted value. The risk tolerance that indicates the risk of each workload being consumed is calculated, and the risk tolerance and the power consumption of each workload are compared. Based on the difference between the predicted value and the past power consumption value, and the target value of the power consumption, Determine the timing of the call.

[0169] This allows you to set the appropriate execution timing for your workload by taking into account the uncertainty of the forecast values. In addition, it is possible to determine the amount of power consumption based on the risk of discrepancy between the predicted value and the actual value. With this, you can decide when to run your workload.

[0170] In addition, the workload control support device of this embodiment is configured to calculate the risk tolerance based on the renewable energy It is calculated based on the difference between the predicted power generation amount for each facility and the target power consumption amount.

[0171] This allows us to take into account the risks arising from the uncertainty of renewable energy generation. can be done.

[0172] In addition, the workload control support device of this embodiment is capable of detecting the timing of each workload to be executed. information about each workload executed and its power consumption, or Display.

[0173] This allows users to verify that appropriate decisions have been made regarding the execution of their workloads. It is possible.

[0174] The present invention is not limited to the above-described embodiment, and any other modifications may be made without departing from the spirit and scope of the present invention. The above-described embodiment and modifications are merely examples. The present invention is not limited to these contents as long as the characteristics of the invention are not impaired. Although various embodiments and modifications have been described above, the present invention is not limited to these. Other embodiments that are conceivable within the scope of the technical idea of ​​the present invention are also within the scope of the present invention. is included within the enclosure.

[0175] For example, some of the functions of each device in the present embodiment may be provided in another device, or may be provided separately. The functions of the above-mentioned devices may be provided in the same device.

[0176] The program configuration described in this embodiment is an example. For example, Incorporating a department into another program or integrating multiple programs into one program This is also fine.

[0177] In this embodiment, the delay limit time is used as the executable time. It may be specifically specified.

[0178] In this embodiment, the workload is a workload in a data center. However, the present invention is also applicable to information processing performed in other facilities or networks. It is possible to use. [Explanation of symbols]

[0179] 1 Workload Control System 1000 Data Centers 2000 management computer

Claims

1. A processor and a memory are included. The feasibility of each of multiple power-consuming workloads scheduled to run in the future. Obtain a forecast value for the period and the amount of power consumption, Based on the obtained predicted values ​​of the executable period and the amount of power consumption, The target value of the power consumption in the future time period is set as the ratio of renewable energy to the power consumption in the future time period. Calculate so as to meet the conditions of utilization rate and the cost related to the use of renewable energy. A parameter determination unit for determining Based on the calculated target value of the power consumption, each of the programs to be executed in the future time period is a workload control unit that determines the timing of the workload; A workload control support device comprising:

2. The parameter determination unit A parameter that indicates the policy of which of the above conditions, the utilization rate condition or the cost condition, is to be emphasized. Accept the data specification, If a policy that emphasizes the above utilization rate conditions is specified, the utilization rate of renewable energy Identify patterns in the execution timing of each workload, and optimize the identified performance. calculating a target value of the amount of power consumption for the future time period based on the turn; If a policy that emphasizes the cost condition is specified, the use of renewable energy will be Identifying patterns of when each workload should be executed that optimizes those costs; Calculating a target value of the power consumption amount for the future time period based on the identified pattern. R, 2. The workload control assistance device according to claim 1.

3. The parameter determination unit If a policy that emphasizes the cost condition is specified, the benefits of renewable energy will be To meet minimum utilization requirements and optimize the costs of using renewable energy. Identify patterns in when each workload runs, and then use those patterns to calculating a target value of the power consumption amount for the future time period; 3. The workload control assistance device according to claim 2.

4. The parameter determination unit The execution timing of each workload that satisfies the execution period of each workload obtained above. Create a condition for the tag, Based on the created execution timing condition and the acquired predicted value of the power consumption, The target value of the power consumption amount in the future time period is set based on the condition of the utilization rate and the cost Calculate so as to satisfy the condition 2. The workload control assistance device according to claim 1.

5. A risk that represents a risk due to uncertainty in the predicted value of the power consumption amount in the future time period A risk tolerance calculation unit that calculates a risk tolerance by a predetermined algorithm, The workload control unit calculates the calculated risk tolerance and the future time period. The difference between the predicted power consumption of each workload and the historical power consumption of each workload. and a target value of the power consumption to be calculated based on the calculated power consumption target value. Determine the timing of each workload to be 2. The workload control assistance device according to claim 1.

6. The risk tolerance calculation unit calculates the risk tolerance based on a reproducible risk in the future time period. A predicted value of the amount of power generation related to energy and the calculated amount of power consumption in the future time period 6. The workload control assistance device according to claim 5, wherein the calculation is based on a difference between the target value and the actual value.

7. The workload control unit determines each of the workloads to be executed in the future time period. timing and power consumption information for each of the executed workloads, or 2. The workload control support according to claim 1, further comprising displaying information on the power consumption of each of the workloads. Device.

8. The workload control unit controls each of the workloads to a predetermined device at the determined timing.

2. The workload control support device according to claim 1, wherein the workload control support device causes a program to be executed by a program executing program.

9. An information processing device, The feasibility of each of multiple power-consuming workloads scheduled to run in the future. Obtain a forecast value for the period and the amount of power consumption, Based on the obtained predicted values ​​of the executable period and the amount of power consumption, The target value of the power consumption in the future time period is set as the ratio of renewable energy to the power consumption in the future time period. Calculate so as to meet the conditions of utilization rate and the cost related to the use of renewable energy. A parameter determination process; Based on the calculated target value of the power consumption, each of the programs to be executed in the future time period is Determine the timing of the workloads and execute each of the workloads at the determined timing. Workload control processing and The workload control support method includes:

Images