Power reduction control device, power reduction control method, power reduction control system, and program

Abstract

An electric power amount reduction control device (100) comprises: an air conditioning control value generation unit (221) that generates an air conditioning control value; an air conditioning control execution unit (240) that causes control of a plurality of air conditioners to be executed using the air conditioning control value; an operation history creation unit (223) that acquires temperature distribution information (64) and the air conditioning electric power consumption amounts of the plurality of air conditioners as a control result and creates operation history information (201); an arrangement pattern calculation unit (310) that calculates an arrangement pattern for a processing load; an area heat generation amount estimation unit (320) that estimates a predicted heat generation amount for each of arrangement control areas; a server electric power consumption amount prediction unit (330) that calculates a server electric power consumption amount for each of the arrangement control areas (30); and an arrangement pattern determination unit (340) that calculates the total of the server electric power consumption amount and the air conditioning electric power consumption amounts in each arrangement pattern and determines an arrangement pattern in which the total is smallest.

Description

[Technical Field]

[0001] The present invention relates to a power reduction control device, a power reduction control method, a power reduction control system, and a program for reducing the amount of power consumed in a data center (hereinafter sometimes referred to as "DC"). [Background technology]

[0002] Air conditioning in data centers (DCs) accounts for a large proportion of the power consumption, and as the number and size of DCs increase, there is a need to reduce this power consumption. In addition, the amount of data processed in DCs is increasing year by year, making it necessary to improve the power consumption efficiency of the DC as a whole (the amount of power consumed by the entire DC for a certain amount of data processing).

[0003] Non-Patent Document 1 discloses a technology that optimizes the power consumption of the entire DC by taking into consideration the power consumption of air conditioners and servers (IT devices). The air conditioning-linked IT load placement optimization method for data centers described in Non-Patent Document 1 collects operational and monitoring information from IT equipment in data centers to predict future changes in the load on the IT equipment and calculates the power increase for air conditioning equipment in response to the increase in power consumption of the IT equipment. Then, an optimization problem is solved to minimize the objective function, which is the amount of power consumed by the data center, so that the load concentration rate on the IT equipment increases over time, i.e., the number of operating IT devices is reduced. This calculates the placement of IT loads (virtual machines) on IT equipment that minimizes the amount of power consumed by the data center. [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] Jun Okitsu and 4 others, "IT Load Placement Optimization Method Linked with Air Conditioning for Eco-Friendly Data Centers," FIT (Forum on Information Technology) 2010, 9th Information Science and Technology Forum, RC-009 Summary of the Invention [Problem to be solved by the invention]

[0005] However, the technology described in Non-Patent Document 1 uses a general-purpose rule-based standard that is not dependent on the equipment conditions that differ for each data center in the air conditioning power model used to calculate the power consumption of the air conditioning equipment.As a result, it was difficult to perform optimization to reduce the total power consumption of the data center, taking into account individual equipment conditions such as the location of the air conditioning equipment, airflow, server configuration within the data center, and thermal cooling efficiency.

[0006] Furthermore, it is assumed that load processing will be performed in a DC environment where CPU (Central Processing Unit) servers, GPU (Graphics Processing Unit) servers, and accelerators coexist, such as FPGAs (Field Programmable Gate Arrays), ASICs (Application Specific Integrated Circuits), and TPUs (Tensor Processing Units).

[0007] The CPU, GPU, and accelerators all require different amounts of power to cool the server during load processing. Specifically, while the CPU's power consumption barely fluctuates within the normal temperature range, the GPU server and accelerators are expected to experience fluctuations in power consumption even within the normal temperature range.

[0008] The inventors of this application set the air intake temperature of the GPU server to 20°C and 33°C, which are within the normal temperature range, and measured the temperature (GPU temperature) of the GPU card in the GPU server, the power consumption, and the fan rotation rate. As a result, as shown in Figure 1, when comparing the air intake temperature at 20°C and 33°C, the GPU temperature rises by approximately 10°C at 33°C. Furthermore, as shown in Figure 2, the power consumption of the GPU card also rises by approximately 20W at 33°C. Furthermore, as shown in Figure 3, the fan rotation rate of the GPU server also rises by approximately 10% at 33°C. The horizontal axis in Figures 1 to 3 represents the time [minutes:seconds] from the start of measurement.

[0009] In the experiment, when the air inlet temperature was 20°C, the power consumption of the four GPU cards was 0.66kWh, and the power consumption of the GPU server itself excluding the GPU cards was 0.33kWh, for a total of 0.99kWh for the entire GPU server. On the other hand, when the air inlet temperature was 33°C, the power consumption of the four GPU cards was 0.72kWh, and the power consumption of the GPU server itself excluding the GPU cards was 0.37kWh, for a total of 1.09kWh for the entire GPU server. These results confirmed that an increase in the air inlet temperature from 22°C to 33°C resulted in an increase in power consumption of the entire GPU server by approximately 9%.

[0010] In other words, previous research has not been able to model the differences in power consumption characteristics required for cooling functions relative to air inlet temperature in environments where CPU servers, GPU servers, accelerators, etc. are mixed. More specifically, when the air conditioning control level is set to "high," the air inlet temperature becomes "low," and server power consumption decreases to "low," but air conditioning power consumption remains "high." On the other hand, when the air conditioning control level is set to "low," the air inlet temperature becomes "high," and server power consumption increases to "high," but air conditioning power consumption remains "low." In other words, there is a trade-off between the amount of air conditioning power consumed by air conditioning control and the amount of server power consumption. Therefore, it is difficult to accurately estimate the overall power consumption of a DC, which consists of server power consumption and air conditioning power consumption. Therefore, it has not been possible to reduce the overall power consumption of a DC by finding an optimal solution for the load processing layout and air conditioning control within the DC.

[0011] The present invention was made in consideration of these points, and its objective is to reduce the total power consumption, consisting of server power consumption and air conditioning power consumption, in an environment where CPU servers, GPU servers, accelerators, etc. are mixed. [Means for solving the problem]

[0012] The power consumption reduction control device according to the present invention is a power consumption reduction control device that controls a CPU server, a GPU server, an accelerator, and a plurality of air conditioners, and has set therein a plurality of placement control areas in which any of the CPU server, the GPU server, and the accelerator are placed, and an air conditioning control area that is an area in which the effect of air conditioning control by the plurality of air conditioners is measured, and the power consumption reduction control device includes an air conditioning control value generation unit that generates air conditioning control values ​​including at least a target temperature to be set for the plurality of air conditioners, an air conditioning control execution unit that executes control of the plurality of air conditioners using the air conditioning control values, a reward calculation unit that calculates a reward using the target temperature as an index for results of control of the plurality of air conditioners by the air conditioning control values ​​in a plurality of placement patterns in which processing loads are placed on the CPU server, the GPU server, and the accelerator, and determines whether the reward satisfies a predetermined condition, and a reward calculation unit that acquires temperature distribution information and air conditioning power consumption of the plurality of air conditioners as control results using the air conditioning control values ​​that are determined to satisfy the predetermined condition, and determines whether the reward satisfies a ... an operation history creating unit that creates operation history information associated with a predicted heat generation amount for each placement control area; an arrangement pattern calculating unit that uses information on the processing loads on the CPU server, the GPU server, and the accelerator to calculate a plurality of arrangement patterns for allocating the new processing loads; an area heat generation amount estimating unit that estimates a predicted heat generation amount for each placement control area by summing up the heat generation amounts when processing loads are allocated to the CPU server, the GPU server, and the accelerator that belong to each of the placement control areas for each of the calculated arrangement patterns; an operation history information extracting unit that uses information on the predicted heat generation amount for each of the placement control areas to refer to the operation history information and extracts the temperature distribution information and the air conditioning power consumption when controlled by the air conditioning control value in each placement pattern; and a server power consumption predicting unit that uses the extracted temperature distribution information and information on the new processing load to calculate a server power consumption amount that is the sum of the power consumption of each CPU server, the power consumption of each GPU server, and the power consumption of each accelerator for each of the placement control areas in each of the placement patterns;and a placement pattern determination unit that adds up the server power consumption amounts of the respective placement control areas, calculates the sum of the total server power consumption amount and the extracted air conditioning power consumption amount, and determines the placement pattern that minimizes the calculated sum as the placement pattern for placing the processing loads. [Effects of the Invention]

[0013] According to the present invention, in an environment where CPU servers, GPU servers, accelerators, etc. coexist, the total power consumption, which is made up of server power consumption and air conditioning power consumption, can be reduced. [Brief explanation of the drawings]

[0014] [Figure 1] This is a graph comparing GPU temperatures when the intake temperature is 20°C and 33°C. [Figure 2] This is a graph comparing the power consumption of GPU cards when the air inlet temperature is 20°C and 33°C. [Figure 3] This is a graph comparing the fan rotation rates of a GPU server when the air intake temperatures are 20°C and 33°C. [Figure 4] 1 is a diagram showing the overall configuration of a power amount reduction control system including a power amount reduction control device according to an embodiment of the present invention; [Figure 5] 1 is a functional block diagram illustrating an example of the configuration of a power amount reduction control device according to an embodiment of the present invention. [Figure 6] FIG. 10 is a diagram for explaining situation classification according to the present embodiment. [Figure 7] FIG. 10 is a diagram for explaining temperature distribution information according to the embodiment. [Figure 8] 10 is a flowchart showing the flow of an operation history information generation process executed by the power amount reduction control device according to the present embodiment. [Figure 9] 10 is a flowchart showing the flow of an arrangement pattern determination process executed by the power amount reduction control device according to the present embodiment. [Figure 10]FIG. 2 is a hardware configuration diagram illustrating an example of a computer that realizes the functions of the power amount reduction control device according to the present embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0015] Next, an embodiment of the present invention (hereinafter referred to as "the present embodiment") will be described. FIG. 4 is a diagram showing the overall configuration of a power amount reduction control system 1 including a power amount reduction control device 100 according to this embodiment.

[0016] As shown in FIG. 4, the power consumption reduction control system 1 is configured with a data center (DC10) having a plurality of CPU servers 3, GPU servers 4, accelerators 5, and a plurality of air conditioners 2 housed in a predetermined control area (the "placement control area" described below), and a power consumption reduction control device 100. The plurality of CPU servers 3, GPU servers 4, and accelerators 5 housed in the predetermined control area (placement control area) may be referred to below as a "server group" or collectively as "servers." In FIG. 4 and FIG. 7 described below, the servers 3 are represented by plain hexagons, the GPU servers 4 are represented by hexagons with multiple diagonal lines, and the accelerators 5 are represented by hexagons filled with dots. This server group includes cases where there is at least one CPU server 3, one GPU server 4, and one accelerator 5, as well as cases where there are no GPU servers 4 and the group is composed of CPU servers 3 and accelerators 5, and cases where there are no accelerators 5 and the group is composed of CPU servers 3 and GPU servers 4. The power amount reduction control device 100 may be provided inside the DC 10, or may be provided in a location separate from the DC 10 and control a plurality of DCs 10.

[0017] This power consumption reduction control device 100 may acquire status information and transmit air conditioning control information for the air conditioners 2 (air conditioners "1", "2", and "3" in Figure 4) installed within the DC 10 via an air conditioning management device not shown, or may be directly connected to communicate with each air conditioner 2 without going through the air conditioning management device. In addition, the power consumption reduction control device 100 may acquire status information and transmit control information of the CPU servers 3, GPU servers 4, and accelerators 5 housed as a group of servers provided within the DC 10 via a server management device (not shown), or may be directly connected to each CPU server 3, GPU server 4, and accelerator 5 for communication.

[0018] In the DC 10 of this embodiment, all the servers (CPU servers 3, GPU servers 4, and accelerators 5) accommodated therein are divided into areas where the CPU servers 3, GPU servers 4, and accelerators 5 are respectively placed, as shown in Fig. 4, and these areas are controlled as "placement control areas." These placement control areas 30 are areas that accommodate a group of servers to which processing loads (virtual resources, GPU and accelerator processing, etc.) are placed. Fig. 4 shows an example in which placement control areas "1" to "6" are provided.

[0019] In the DC10, the virtualization infrastructure is constructed on the CPU server 3 and is explained as being operated using containers and VMs. Known open source virtualization infrastructures include OpenStack (registered trademark), software for building cloud environments, and Kubernetes (registered trademark), software for operating and managing containerized workloads and services. OpenStack is primarily used for managing and operating physical machines and virtual machines (VMs). Kubernetes is primarily used for managing and operating containers. In this specification, an application virtualized on a virtualization platform (consisting of one or more containers, one or more VMs, etc.) is referred to as a virtual resource. In Kubernetes, the smallest execution unit of an application is a pod consisting of one or more containers.

[0020] In this embodiment, an "air conditioning control area" is set up in correspondence with the placement control area 30 of the server group, as shown in Fig. 4. The air conditioning control area 20 is a collective area where the room temperature effect of air conditioning control is measured, and faces either the intake side or the exhaust side of each server (CPU server 3, GPU server 4, accelerator 5). The air blown from the air conditioner 2 is blown out from the air conditioning controlled areas 20 on the intake side (air conditioning controlled areas "3," "4," "7," and "8" in FIG. 4) via piping installed, for example, under the floor of DC 10. Then, air whose temperature has risen due to the heat of each server is taken in from the intake of piping installed in the air conditioning controlled areas 20 on the discharge side (air conditioning controlled areas "1," "2," "5," and "6" in FIG. 4), generating an airflow that returns to the air conditioner 2.

[0021] A plurality of sensors (temperature sensors, etc.) are installed in each of the air-conditioning controlled areas 20. Temperature sensors are also installed at the intake ports of the GPU servers 4 and accelerators 5 in each placement controlled area 30. Furthermore, sensors (temperature sensors, etc.) are also installed outside the DC 10. Information obtained from these sensors (sensor information) can be acquired by the power reduction control device 100 via a communication line, etc.

[0022] The power consumption reduction control device 100 according to this embodiment predicts the amount of heat generated (hereinafter referred to as the "predicted amount of heat generated in a placement control area") for each placement control area 30 in a placement pattern in which the load is allocated to each server resource (in this embodiment, this means the CPU server 3, GPU server 4, and accelerator 5) based on creation / deletion schedule information for the virtual resources that become the processing load for the CPU server 3 and the load for the GPU / accelerator. Note that the " / " above means "and / or." The power consumption reduction control device 100 sets multiple levels of air conditioning control values ​​for the air conditioner 2 for each situation (hereinafter referred to as the "Situation") of the external temperature and floor temperature of the DC 10 and the predicted amount of heat generated in the placement control area 30, and stores temperature distribution information and air conditioning power consumption information when controlled at each level. The power reduction control device 100 then calculates the server power consumption in each placement control area 30 ("total server power consumption," as described below) based on temperature distribution information and the like when controlled at each stage, and determines a placement pattern that minimizes the total of the server power consumption and air conditioning power consumption (as described in detail below). The power reduction control device 100 will be described in detail below.

[0023] <Power reduction control device> FIG. 5 is a functional block diagram showing an example of the configuration of the power amount reduction control device 100 according to this embodiment. The power consumption reduction control device 100 predicts the amount of heat generated (predicted heat generation amount) for each placement control area 30 in each placement pattern of the server resources (CPU servers 3, GPU servers 4, accelerators 5), and acquires temperature distribution information 64 and air conditioning power consumption information 65 for when air conditioning control of the air conditioners 2 is performed in that situation. The power consumption reduction control device 100 then uses a learning model to calculate the server power consumption of each CPU server 3, GPU server 4, and accelerator 5, and sums these values ​​to calculate the total power consumption of the CPU servers 3, GPU servers 4, and accelerators 5 for each placement control area 30. The power consumption reduction control device 100 calculates the total server power consumption (the sum of the total power consumption of the CPU servers 3, GPU servers 4, and accelerators 5 in each placement control area 30) and the air conditioning power consumption, determines the placement pattern that minimizes this total, and executes load placement and air conditioning control based on this. The power consumption reduction control device 100 is configured by a computer having a control unit, an input / output unit, and a storage unit (all of which are not shown).

[0024] The input unit inputs and outputs information between each device (each air conditioner 2 and each server (CPU server 3, GPU server 4, accelerator 5)) in the DC 10. This input / output unit is composed of a communication interface that transmits and receives information via a communication line, and an input / output interface that inputs and outputs information between an input device such as a keyboard and an output device such as a monitor (not shown).

[0025] The storage unit is composed of a hard disk, flash memory, RAM (Random Access Memory), and the like. This storage unit temporarily stores programs for executing the functions of the control unit and information necessary for processing by the control unit. This storage unit also stores operation history information 201, which includes the control values ​​of the air conditioners 2 (air-conditioning control value information 63), temperature distribution information 64 as the control results, and air-conditioning power consumption information 65, for each situation in each placement pattern. The storage unit also stores basic power consumption information 301 for calculating the predicted heat generation amount of each server (CPU server 3, GPU server 4, accelerator 5), a CPU server power consumption learning model 302 for predicting the power consumption of the CPU server 3, a GPU server power consumption learning model 303 for predicting the power consumption of the GPU server 4, and an accelerator power consumption learning model 304 for predicting the power consumption of the accelerator 5 (details will be described later).

[0026] The control unit is responsible for all the processes executed by the power amount reduction control device 100, and is configured to include an air conditioning control unit 200 and a server control unit 300, as shown in FIG.

[0027] <Air conditioning control unit> The air conditioning control unit 200 generates operation history information 201 by acquiring temperature distribution information 64 in control turns and calculating air conditioning power consumption information 65 at each air conditioning control stage of each situation (setting the air conditioning control value of each air conditioner 2 in stages). The air conditioning control unit 200 generates operation history information 201 by acquiring information on the predicted heat generation amount for each placement control zone 30 from the server control unit 300 during the operation phase in which load allocation and air conditioning control are actually performed. The air conditioning control unit 200 extracts operation history information 201 corresponding to the corresponding situation (situation classification 62) and outputs the temperature distribution information 64 and air conditioning power consumption information 65 to the server control unit 300. The air conditioning control unit 200 then controls each air conditioner 2 to perform air conditioning control in the placement pattern determined by the server control unit 300. The air conditioning control unit 200 includes a situation recognition unit 210, an operation history information generation unit 220, an operation history information extraction unit 230, and an air conditioning control execution unit 240.

[0028] The situation recognition unit 210 acquires information on external factors, which are parameter elements that make up a situation. The situation recognition unit 210 then divides each external factor into multiple ranges, defines a combination of each range area as one situation, and determines a situation classification 62 that indicates to which situation the situation belongs based on the acquired information on the external factors. The situation recognition unit 210 includes an external factor acquisition unit 211 and a situation determination unit 212 .

[0029] The external factor acquisition unit 211 acquires information on the measurement results of external factors. Here, the external factors are elements that affect the increase or decrease in air conditioning power consumption, and refer to parameter elements that make up the Situation classification 62. Here, the external factors are (1) the average floor temperature in the DC 10 before control, (2) the outside temperature (outside air temperature), and (3) the predicted heat generation amount for each placement control area 30.

[0030] (1) The external factor acquisition unit 211 calculates the average floor temperature in the DC 10 before control as follows: The external factor acquisition unit 211 calculates the average value of the temperatures acquired from the temperature sensors in the air-conditioning controlled zones 20, and calculates the average temperature for each air-conditioning controlled zone 20. The external factor acquisition unit 211 then averages the calculated average temperatures for each air-conditioning controlled zone 20 across the entire floor, and sets the obtained temperature as the average floor temperature.

[0031] (2) The external temperature is information obtained from a temperature sensor installed outside the DC10. (3) The predicted amount of heat generation for each placement control area 30 is information calculated by the server control unit 300 (details will be described later). When the external factor acquisition unit 211 acquires the information on the external factors, it outputs the information to the situation determination unit 212 .

[0032] The situation determination unit 212 determines to which situation category 62 the information acquired by the external factor acquisition unit 211 belongs. Each external factor is divided into multiple ranges between the minimum and maximum values ​​according to the characteristics of the external factor. A combination of the ranges into which each external factor is divided is defined as one Situation. This will be explained below with reference to FIG. 6.

[0033] As shown in FIG. 6, each external factor is defined as a "factor," and a range to be divided is defined (hereinafter referred to as a "division definition"). For example, the external factor for "factor 1" shown in Situation classification 62 in Figure 6 is "average floor temperature," and the division definition is "0-48 degrees divided into 6." The external factor for "factor 2" is "outside temperature," and the division definition is "0-48 degrees divided into 6." The external factor for "factor 3" is "predicted heat generation amount in location control area 1," and the division definition is "0-200W divided into 20." Similarly, the external factor for "factor 8" is "predicted heat generation amount in location control area 6," and the division definition is "0-200W divided into 20."

[0034] Here, it is assumed that the information on the external factors acquired by the Situation determination unit 212 is the external factor information 61 shown in FIG. 6. In this case, since the value of "factor1" (floor average temperature) is "25," the Situation determination unit 212 determines that the "range" is included in the "24-32 range" (24 degrees or more and less than 32 degrees), and sets the "factor range identifier" to "factor1-4." This "factor range identifier" is information that identifies the range to which the temperature belongs, for example, by dividing 0-48 degrees into six parts, such as "factor1-1" for 0 degrees or more and less than 8 degrees, "factor1-2" for 8 degrees or more and less than 16 degrees, and "factor1-3" for 16 degrees or more and less than 24 degrees. The same applies to the other "factors."

[0035] The situation determination unit 212 combines information on the "factor range identifiers" of the external factors to form a "Situation classification," and determines that the situation classification is "factor1-4_factor2-4_factor3-4_factor4-4_factor5-5_factor6-5_factor7-4_factor8-4." In this way, the situation determination unit 212 determines the "situation classification" based on the acquired information on the external factors.

[0036] Returning to Figure 5, the operation history information generation unit 220 generates operation history information 201 using temperature distribution information 64 and air conditioning power consumption information 65 as the results of controlling the air conditioner 2 based on air conditioning control values ​​that divide the control of the air conditioner 2 into multiple stages in each situation. The operation history information generation unit 220 includes an air conditioning control value generation unit 221 , a remuneration calculation unit 222 , and an operation history creation unit 223 .

[0037] The air conditioning control value generator 221 generates air conditioning control values ​​for each of a plurality of stages of control of the air conditioner 2 for each situation. Specifically, the air conditioning control value generation unit 221 divides each control parameter (for example, set temperature (target temperature), air volume, etc.) that can be set or changed for each air conditioner 2 into M stages between upper and lower limit values. Then, it combines the parameters for each stage to generate air conditioning control value information 63 to be controlled for (one) air conditioner 2. Based on this generated air conditioning control value information 63, the air conditioning control execution unit 240 executes air conditioning control in a number of patterns, such as controlling the air conditioners in the order "1" → "2" → "3", controlling a combination of air conditioners "1" and "2", "air conditioners "2" and "3", or "1" and "3", or simultaneously controlling air conditioners "1", "2", and "3".

[0038] The reward calculation unit 222 calculates a reward (temperature reward) as an index for evaluating the results of control performed using the air conditioning control value generated by the air conditioning control value generation unit 221. The reward calculation unit 222 then determines whether the control result satisfies a predetermined reward, in other words, whether the air conditioning control value satisfies a predetermined condition.

[0039] This reward calculation unit 222 defines two types of rewards, a high temperature alert reward and a low temperature alert reward, for each air conditioning controlled area 20, and calculates the reward for the control result for each turn. The high temperature alert reward is applied when the temperature before control is higher than the target temperature, i.e., when the room temperature is high and the temperature is controlled to decrease. The low temperature alert reward is applied when the temperature before control is lower than the target temperature, i.e., when the room temperature is too low and the temperature is controlled to increase.

[0040] When calculating the reward, the reward calculation unit 222 calculates the reward using the difference between the "target temperature for the turn" and the "temperature after the turn control", that is, the deviation between the target temperature and the current temperature, as an index. For example, in the case of a high temperature alert reward, if the "temperature after turn control" is below the "target temperature for the turn," the reward is "100%." ​​Also, for every 1 degree increase in the "temperature after turn control" from the "target temperature for the turn," the reward is "-10%." Note that this reward is not limited to the above values ​​and can be set arbitrarily. On the other hand, for low temperature vigilance rewards, if the "temperature after turn control" is equal to or higher than the "target temperature for the turn," the reward is "100%." ​​In addition, the reward is "-10%" for every -1 degree drop in the "temperature after turn control" from the "target temperature for the turn." Note that this reward is not limited to the above values ​​and can be set arbitrarily. When the remuneration calculated by the remuneration calculation unit 222 is equal to or greater than a predetermined threshold (%), it satisfies the predetermined condition and is deemed to have passed, and when it is less than the predetermined threshold (%) it is deemed to have failed.

[0041] The reward calculation unit 222 may be configured to determine a final pass if both the high temperature alert reward and the low temperature alert reward are passed. For example, if the initial temperature before control of the air conditioner 2 is higher than the target temperature and control is performed using an air conditioning control value based on the high temperature alert reward, the temperature may exceed a predetermined pass threshold, but the control may result in the target temperature being exceeded and the temperature becoming too low. In this case, excessive air conditioning power consumption may be consumed. Therefore, if the temperature before control is lower than the target temperature, control is performed based on the low temperature alert reward until a pass is determined. In this way, by the reward calculation unit 222 determining a pass based on both the high temperature alert reward and the low temperature alert reward, it is possible to select an air conditioning control value that can be controlled within an appropriate range.

[0042] Furthermore, if the average temperature (average floor temperature) of the sensors in each air-conditioning control zone 20 after the turn control is outside a specified range (for example, 5°C to 35°C), the reward calculation unit 222 rejects the air-conditioning control value. Furthermore, for a GPU server 4, if the GPU temperature is outside a specified range that is set in advance based on the relationship between the GPU temperature (GPU card temperature) and the power consumption of the GPU server, the reward calculation unit 222 rejects the air-conditioning control value. For an accelerator 5, if the accelerator temperature is outside a specified range that is set in advance based on the relationship between the accelerator temperature (temperature inside the accelerator) and the power consumption, the reward calculation unit 222 rejects the air-conditioning control value. This is a process to avoid unnecessary storage of operational history that results in temperatures that are inappropriate for managing the equipment in the DC 10 and are unlikely to be used when actually processing loads on the DC 10.

[0043] The operation history creation unit 223 acquires temperature distribution information 64 and air conditioning power consumption information 65 as a result of the air conditioning control execution unit 240 controlling each air conditioner 2 based on the air conditioning control value information 63 generated by the air conditioning control value generation unit 221 in each situation. That is, the operation history creation unit 233 causes the air conditioning control execution unit 240 to execute control for each air conditioning control value of each combination pattern in which each parameter is divided into M stages, generated by the air conditioning control value generation unit 221 in each Situation, and obtains temperature distribution information 64 and air conditioning power consumption information 65 when the air conditioning control value is executed. The operation history creation unit 223 excludes from the creation of the operation history information 201 any air conditioning control value and its control results that the remuneration calculation unit 222 has determined not to satisfy a predetermined condition and therefore has failed.

[0044] FIG. 7 is a diagram illustrating temperature distribution information 64 according to this embodiment. The temperature distribution information 64 is temperature information measured over each time transition of the control turn by the temperature sensor 44 provided on the air inlet side of the GPU server 4 and the temperature sensor 55 provided on the air inlet side of the accelerator 5, as shown in FIG. 7. The temperature sensors 44 and 55 measure temperatures after their identification information is associated in advance with the identification information of the GPU server 4 and the accelerator 5. From the start to the end of the turn, all temperature sensors 44 and 45 on the floor measure temperatures over each time transition (at predetermined time intervals), and the resulting information is generated as temperature distribution information 64. In the following, the temperature measured by the temperature sensor 44 installed on the inlet side of the GPU server 4 will be referred to as the "GPU inlet temperature," and the temperature measured by the temperature sensor 55 installed on the inlet side of the accelerator 5 will be referred to as the "accelerator inlet temperature."

[0045] The air conditioning power consumption information 65 is the total power consumption of each air conditioner 2 measured by a power consumption measurement means (not shown) that monitors the air conditioners 2 during the turn in which the air conditioning control execution unit 240 controls each air conditioner 2 based on the air conditioning control value information 63. For example, the power consumption of each air conditioner 2 is measured over time (at predetermined time intervals), and the total of the power consumption measured during that turn is calculated as the air conditioning power consumption information 65.

[0046] The operation history creation unit 233 creates operation history information 201 that associates the Situation (Situation classification 62) and air conditioning control value information 63 when the air conditioning control execution unit 240 executes air conditioning control with the temperature distribution information 64 and air conditioning power consumption information 65 obtained as the control results, and stores the information in a memory unit (not shown).

[0047] In the operation phase, the operation history information extraction unit 230 acquires information on the predicted heat generation amount of each placement control area 30 from the server control unit 300 (area heat generation amount estimation unit 320). Then, the operation history information extraction unit 230 determines the situation classification 62 at the start of the control turn via the situation recognition unit 210. The operation history information extraction unit 230 then extracts temperature distribution information 64 and air conditioning power consumption information 65, which are the results of control using each piece of air conditioning control value information 63 in the confirmed situation classification 62, from the operation history information 201. The operation history information extraction unit 230 outputs the extracted temperature distribution information 64 and air conditioning power consumption information 65 to the server control unit 300 (server power consumption prediction unit 330, arrangement pattern determination unit 340).

[0048] In the learning phase, the air conditioning control execution unit 240 controls the air conditioner 2 in the above-mentioned multiple patterns for each situation using the air conditioning control values ​​generated by the air conditioning control value generation unit 221. Furthermore, the air conditioning control execution unit 240 executes air conditioning control of each air conditioner 2 in the optimal arrangement pattern determined by the server control unit 300 in the operation phase.

[0049] <Server control unit> The server control unit 300 calculates the amount of heat generated for each placement control area 30 (the predicted amount of heat generated for the placement control area 30) in a placement pattern in which the load is allocated to each server resource (CPU server 3, GPU server 4, accelerator 5) based on schedule information for generating / deleting virtual resources that become processing loads on the CPU and loads on the GPU / accelerator.The server control unit 300 then calculates the total server power consumption (total server power consumption) by adding up the server power consumption for each placement control area 30 based on temperature distribution information 64 obtained from the air conditioning control unit 200 and other information on air conditioning control at each stage for each situation.The server control unit 300 calculates the sum of the total server power consumption and the air conditioning power consumption, and determines the placement pattern that minimizes this sum. The server control unit 300 includes an arrangement pattern calculation unit 310 , a zone heat generation amount estimation unit 320 , a server power consumption amount prediction unit 330 , and an arrangement pattern determination unit 340 .

[0050] At the start of each control turn, the allocation pattern calculation unit 310 acquires a creation / deletion schedule (hereinafter referred to as "load processing schedule information") for the virtual resources that become the processing load on the CPU server 3 and the load on the GPU / accelerator. Then, the allocation pattern calculation unit 310 calculates an allocation pattern in which the new load is allocated to each server resource (CPU server 3, GPU server 4, accelerator 5) based on the most recent resource usage status (e.g., usage rate of CPU, GPU, accelerator, etc.). After allocating the load to each server resource (CPU server 3, GPU server 4, accelerator 5), the allocation pattern calculation unit 310 ensures that the resource occupation amount of each server resource is equal to or less than the load capacity (upper limit) x a predetermined threshold.

[0051] The area heat generation amount estimation unit 320 predicts the power consumption of each server resource (CPU server 3, GPU server 4, accelerator 5) for each placement pattern calculated by the placement pattern calculation unit 310, by referring to the basic power consumption information 301. Then, the area heat generation amount estimation unit 320 calculates the predicted total heat generation amount for each placement control area 30 in each placement pattern, based on the server placement configuration for each placement control area 30. This basic power consumption information 301 is, for example, the reference power consumption in a normal state at a predetermined temperature (18° C.), without taking into account changes in server power consumption due to temperature changes at the intake ports of each server resource.

[0052] Specifically, for example, assume that the load processing schedule information is for 12 Pods of CPU processing "a," 5 Pods of GPU processing "b," and 10 Pods of FPGA processing "c" to be executed in placement control area "1." Here, assume that the basic power consumption information 301 indicates that one Pod per server for CPUs consumes 100 watts, one GPU per server consumes 5 kw, and one FPGA per server consumes 20 watts. In this case, the power consumption of CPU server 3 in placement control area "1" is 1.2 kw, the power consumption of GPU server 4 is 5 kw, and the power consumption of FPGAs is 0.2 kw.

[0053] Then, the zone heat generation amount estimation unit 320 calculates the heat generation amount W of the corresponding placement control zone 30 by the following formula (1). Heat generation amount in the placement control area W = CPU server power consumption in the placement control area × kc + GPU server power consumption in the placement control area × kg + accelerator power consumption in the placement control area × ka Equation (1) Note that kc, kg, and ka are coefficients that are determined by measuring in advance the load on the air conditioning cooling in each room within DC 10 when the CPU server 3, GPU server 4, and accelerator 5 are each in operation. The area heat generation estimation unit 320 calculates the heat generation amount of each placement control area 30 by adding up the heat generation amounts of the CPU server 3, GPU server 4, and accelerator 5 using the above formula (1), and uses this as the predicted heat generation amount of the placement control area 30. The zone heat generation amount estimation unit 320 then outputs the calculated predicted heat generation amount for each allocation control zone 30 to the air conditioning control unit 200 (operation history information extraction unit 230).

[0054] At the start of each control turn, the server power consumption prediction unit 330 calculates the total power consumption (total CPU server power consumption, total GPU server power consumption, total accelerator power consumption) for each placement control area 30 for each of the CPU server 3, GPU server 4, and accelerator 5 using load processing schedule information (information on new processing load) for the CPU server 3, GPU server 4, and accelerator 5 and temperature distribution information 64 of the placement pattern obtained from the air conditioning control unit 200 (operation history information extraction unit 230). For each placement pattern, the server power consumption prediction unit 330 adds up the total CPU server power consumption, the total GPU server power consumption, and the total accelerator power consumption in that placement control area 30, and calculates the server power consumption for each placement control area 30. The server power consumption prediction unit 330 includes a CPU power consumption prediction unit 331 , a GPU power consumption prediction unit 332 , and an accelerator power consumption prediction unit 333 .

[0055] At the start of each control turn, the CPU power amount prediction unit 331 acquires information on the amount of virtual resources to be newly allocated based on the load processing schedule information (e.g., the number of CPU cores) and the resource usage status (e.g., CPU usage rate) of the CPU server 3 at that time. Then, the CPU power amount prediction unit 331 predicts the power consumption of each CPU server 3 using the CPU server power amount learning model 302. More specifically, when switching control turns, the CPU power amount prediction unit 331 deletes virtual resources (e.g., Pods) whose processing has finished in the previous control turn. Then, at the start of the current control turn, the CPU power amount prediction unit 311 acquires the resource usage rates (e.g., CPU usage rates, memory usage rates, etc.) of each CPU server 3 excluding the deleted Pods. Based on the load processing schedule information, the CPU power amount prediction unit 311 calculates the predicted values ​​of resource usage rates when new Pods are placed on each CPU server 3. Then, the CPU power amount prediction unit 311 inputs these predicted values ​​of resource usage rates into the CPU server power amount learning model 302, thereby calculating the power consumption of each CPU server 3. Furthermore, the CPU power prediction unit 331 calculates the total CPU server power consumption by adding up the power consumption calculated for each CPU server 3 in each placement control area 30 based on the placement configuration of the CPU servers 3 in each placement control area 30.

[0056] Here, the CPU server power consumption learning model 302 is a learning model that uses the resource usage status of the CPU server 3 (for example, CPU usage rate, memory usage rate, etc.) as input information and the power consumption amount of the CPU server 3 as output information. This CPU server power consumption learning model 302 is created in advance using the resource usage status of the CPU server 3 and the server power consumption amount, which is the result information at that time, as learning data.

[0057] The GPU power amount prediction unit 332 predicts the GPU server power consumption of each GPU server 4 using the GPU server power amount learning model 303 based on the type of processing load (hereinafter referred to as "load type") that is newly scheduled to be processed, obtained from the load processing schedule information, the GPU inlet temperature, the number of GPU cards, etc. Furthermore, the GPU power amount prediction unit 332 calculates the total GPU server power consumption by adding up the power consumption of each GPU server 4 in each placement control area 30 based on the placement configuration of the GPU servers 4 in each placement control area 30. The load type is, for example, image processing, machine learning processing, network processing, virtual space processing, etc., depending on the purpose of execution of the GPU server 4, and each load type can be identified using the load processing schedule information. It is also assumed that each GPU server 4 executes a single type of application based on the load processing schedule information.

[0058] This GPU server power consumption learning model 303 has a method for directly predicting GPU server power consumption (one-stage method) and a method for predicting GPU server power consumption in two stages via GPU temperature (GPU card temperature) (two-stage method).

[0059] In the (single-stage) method, one learning model is used as the GPU server power consumption learning model 303 . This GPU server power consumption learning model 303 is a learning model that uses the GPU inlet temperature, load type, and number of GPU cards as input information and the power consumption of the GPU server 4 as output information. This GPU server power consumption learning model 303 is created in advance using the GPU inlet temperature, load type, number of GPU cards, and information on the power consumption of the GPU server 4 at that time as learning data.

[0060] When the one-stage method is adopted, the GPU power consumption prediction unit 332 predicts the GPU server power consumption of each GPU server 4 that has not been assigned processing at the beginning of the turn, using the GPU server power consumption learning model 303 based on the GPU inlet temperature, load type, and number of GPU cards. At the beginning of the turn, the GPU inlet temperature is determined using the current inlet temperature (GPU inlet temperature) of each GPU server 4, and thereafter, the GPU inlet temperature information of each GPU server 4 indicated in the temperature distribution information 64 obtained from the air conditioning control unit 200 (temperature distribution information 64 starting from the same temperature as the current GPU inlet temperature) is used (the same applies to the two-stage system).

[0061] In the (two-stage method), two learning models (a first GPU learning model 303a and a second GPU learning model 303b) are used as the GPU server power consumption learning model 303. The first GPU learning model 303a is a learning model that uses the GPU inlet temperature, load type, and number of GPU cards as input information and the GPU temperature as output information. This first GPU learning model 303a is created in advance using the GPU inlet temperature, load type, number of GPU cards, and the GPU temperature at that time as learning data.

[0062] The second GPU learning model 303b is a learning model that uses the GPU temperature as input information and the power consumption of the GPU server 4 as output information. This second GPU learning model 303b is created in advance using the GPU temperature and the power consumption of the GPU server 4 at that time as learning data.

[0063] When the two-stage method is adopted, the GPU power consumption prediction unit 332 predicts the GPU temperature for each GPU server 4 that has not been assigned processing at the beginning of a turn, using the first GPU learning model 303a, based on the GPU inlet temperature, load type, and number of GPU cards.The GPU power consumption prediction unit 332 then predicts the GPU server power consumption of each GPU server 4, based on the predicted GPU temperature, using the second GPU learning model 303b.

[0064] The GPU power prediction unit 332 calculates the total power consumption of the GPU servers in each placement control area 30 by adding up the predicted power consumption of each GPU server 4 in each placement control area 30 based on the placement configuration of the GPU servers 4 in that placement control area 30.

[0065] The accelerator power prediction unit 333 predicts the accelerator power consumption of each accelerator 5 using the accelerator power learning model 304 based on the type of accelerator processing load (load type) scheduled for new processing, the accelerator inlet temperature, the number of accelerator processing circuits, etc., which are obtained from the load processing schedule information. Furthermore, the accelerator power prediction unit 333 calculates the total accelerator power consumption by adding up the power consumption of each accelerator 5 in each placement control area 30 based on the placement configuration of the accelerators 5 in each placement control area 30. The load type is a type of load depending on the purpose of execution of the accelerator 5, such as image processing, machine learning processing, internet processing, encryption processing, etc., and the load type can be identified using the load processing schedule information. It is also assumed that each accelerator 5 executes a single type of application based on the load processing schedule.

[0066] Like the GPU server power consumption learning model 303, this accelerator power consumption learning model 304 has a method for directly predicting accelerator power consumption (one-stage method) and a method for predicting accelerator power consumption in two stages via accelerator temperature (temperature inside the accelerator) (two-stage method).

[0067] In the (single-stage) method, one learning model is used as the accelerator power amount learning model 304. This accelerator power amount learning model 304 is a learning model that uses the accelerator inlet temperature, load type, and number of accelerator processing circuits as input information and the power consumption of the accelerator 5 as output information. This accelerator power amount learning model 304 is created in advance using the accelerator inlet temperature, load type, number of accelerator processing circuits, and information on the power consumption of the accelerator 5 at that time as learning data.

[0068] When the one-stage method is adopted, the accelerator power prediction unit 333 predicts the accelerator power consumption of each accelerator 5 that has not been assigned a processing assignment at the beginning of the turn, using the accelerator power learning model 304 based on the accelerator inlet temperature, the load type, and the number of accelerator processing circuits. At the beginning of the turn, the accelerator inlet temperature is determined using the current inlet temperature of each accelerator 5 (accelerator inlet temperature), and thereafter, the accelerator inlet temperature information of each accelerator 5 indicated in the temperature distribution information 64 obtained from the air conditioning control unit 200 (temperature distribution information 64 starting from the same temperature as the current accelerator inlet temperature) is used (the same applies to the two-stage system).

[0069] In the (two-stage) method, two learning models (a first accelerator learning model 304a and a second accelerator learning model 304b) are used as the accelerator power amount learning model 304. The first accelerator learning model 304a is a learning model that uses the accelerator inlet temperature, load type, and number of accelerator processing circuits as input information and the accelerator temperature as output information. This first accelerator learning model 304a is created in advance using the accelerator inlet temperature, load type, number of accelerator processing circuits, and the accelerator temperature at that time as learning data.

[0070] The second accelerator learning model 304b is a learning model that uses the accelerator temperature as input information and the power consumption of the accelerator 5 as output information. This second accelerator learning model 304b is created in advance using the accelerator temperature and information on the power consumption of the accelerator 5 at that time as learning data.

[0071] When the two-stage method is adopted, the accelerator power prediction unit 333 predicts the accelerator temperature for each accelerator 5 that has not been assigned a processing assignment at the beginning of a turn, using the first accelerator learning model 304a, based on the accelerator inlet temperature, load type, and number of accelerator processing circuits.The accelerator power prediction unit 333 then predicts the accelerator power consumption of each accelerator 5, using the second accelerator learning model 304b, based on the predicted accelerator temperature.

[0072] The accelerator power prediction unit 333 calculates the total accelerator power consumption in each placement control area 30 by adding up the predicted power consumption of each accelerator 5 in each placement control area 30 based on the placement configuration of the accelerator 5 in that placement control area 30.

[0073] Then, the server power consumption prediction unit 330 adds up the total CPU server power consumption, the total GPU server power consumption, and the total accelerator power consumption in each placement control area 30 for each placement pattern, and calculates the server power consumption for each placement control area 30.

[0074] The placement pattern determination unit 340 adds up the server power consumption amounts of each placement control area 30 in each placement pattern, and calculates the total server power consumption amount (total server power consumption amount). The placement pattern determination unit 340 calculates the sum of the calculated total server power consumption amount and the air conditioning power consumption amount for that placement pattern acquired from the air conditioning control unit 200, and determines the placement pattern that minimizes this sum.

[0075] <Processing flow> Next, the flow of processing executed by the power amount reduction control device 100 according to this embodiment will be described. Here, the operation history information generation process performed in the learning phase will be described with reference to Fig. 8. Also, the allocation pattern determination process performed in the operation phase will be described with reference to Fig. 9.

[0076] <<Operation history information generation process>> FIG. 8 is a flowchart showing the flow of the operation history information generation process executed by the power amount reduction control device 100 according to this embodiment.

[0077] First, the air conditioning control value generation unit 221 of the air conditioning control unit 200 (operation history information generation unit 220) of the power consumption reduction control device 100 generates air conditioning control values ​​divided into multiple stages for control parameters (e.g., set temperature (target temperature), air volume, etc.) that can be set or changed for each air conditioner 2 (step S1). Specifically, the air conditioning control value generator 221 divides each parameter into M stages between an upper limit value and a lower limit value, and generates air conditioning control value information 63 for each air conditioner 2 by combining parameters for each stage.

[0078] Then, based on the generated air conditioning control value information 63, the air conditioning control execution unit 240 executes air conditioning control in multiple patterns (step S2). For example, for each air conditioning control value, the air conditioning control execution unit 240 executes air conditioning control in multiple patterns, such as controlling the air conditioners in the order of "1" → "2" → "3", controlling air conditioners in combinations of "1" and "2", "2" and "3", or "1" and "3", or simultaneously controlling air conditioners "1", "2", and "3".

[0079] Next, in step S3, the operation history information generator 220 (reward calculator 222) calculates a reward (temperature reward) as an index for evaluating the results of air conditioning control performed using the air conditioning control value generated by the air conditioning control value generator 221. The reward calculator 222 then determines whether the control result satisfies a predetermined reward, in other words, whether the air conditioning control value satisfies a predetermined condition. The reward calculation unit 222 judges the application as successful if the calculated reward is equal to or greater than a predetermined threshold and meets predetermined conditions, such as the average floor temperature, GPU temperature, accelerator temperature, etc. after the control turn being within a specified range.

[0080] Next, the operation history information generation unit 220 (operation history creation unit 223) acquires temperature distribution information 64 and air conditioning power consumption information 65 as a result of the air conditioning control execution unit 240 controlling each air conditioner 2 using the air conditioning control value information 63 generated by the air conditioning control value generation unit 221 in each Situation and judged to be acceptable by the remuneration calculation unit 222 (step S4). The temperature distribution information 64 is information obtained by measuring, at predetermined time intervals, the GPU inlet temperature measured by the temperature sensor 44 provided on the inlet side of the GPU server 4 and the accelerator inlet temperature measured by the temperature sensor 55 provided on the inlet side of the accelerator 5. The air conditioning power consumption information 65 is the total power consumption of each air conditioner 2 measured in a predetermined control cycle.

[0081] Then, the operation history creation unit 223 creates operation history information 201 by associating the Situation (Situation classification 62) and air conditioning control value information 63 when the air conditioning control execution unit 240 executes air conditioning control with the temperature distribution information 64 and air conditioning power consumption information 65 obtained as the control results (step S5), and stores the information in the memory unit.

[0082] The power amount reduction control device 100 performs the process of generating this operation history information 201 in advance during the learning phase prior to the operation phase.

[0083] <<Layout Pattern Determination Process>> FIG. 9 is a flowchart showing the flow of the allocation pattern determination process executed by the power amount reduction control device 100 according to this embodiment.

[0084] First, the server control unit 300 (disposition pattern calculation unit 310) of the power consumption reduction control device 100 acquires load processing schedule information, and at the start of each control turn, calculates a disposition pattern for distributing new loads to each server resource (CPU server 3, GPU server 4, accelerator 5) based on the most recent resource usage status (e.g., usage rates of CPU, GPU, accelerator, etc.) (step S10).

[0085] Next, the area heat generation amount estimation unit 320 predicts the power consumption of each server resource (CPU server 3, GPU server 4, accelerator 5) for each placement pattern calculated by the placement pattern calculation unit 310, by referring to the basic power consumption information 301. Then, the area heat generation amount estimation unit 320 calculates the total predicted heat generation amount for each placement control area 30 in each placement pattern (predicted heat generation amount for the placement control area) based on the server placement configuration for each placement control area 30 (step S11). The zone heat generation amount estimation unit 320 then outputs the calculated predicted heat generation amount for each allocation control zone 30 to the air conditioning control unit 200 (operation history information extraction unit 230).

[0086] Next, the operation history information extraction unit 230 of the air conditioning control unit 200 acquires information on the predicted heat generation amount of each placement control area 30 from the server control unit 300 (area heat generation amount estimation unit 320). Then, the operation history information extraction unit 230 determines the situation classification 62 at the start of the control turn via the situation recognition unit 210 (step S12). The operation history information extraction unit 230 extracts temperature distribution information 64 and air conditioning power consumption information 65, which are the results of control using each air conditioning control value information 63 in the determined situation classification 62, from the operation history information 201 (step S13). The operation history information extraction unit 230 outputs the extracted temperature distribution information 64 and air conditioning power consumption information 65 to the server control unit 300.

[0087] Next, the server power consumption prediction unit 330 of the server control unit 300 calculates the total power consumption (total CPU server power consumption, total GPU server power consumption, total accelerator power consumption) for each placement control area 30 for each of the CPU server 3, GPU server 4, and accelerator 5, based on the load processing schedule information for the CPU server 3, GPU server 4, and accelerator 5, using the temperature distribution information 64 of the placement pattern acquired from the air conditioning control unit 200 (operation history information extraction unit 230) (step S14).

[0088] Specifically, at the start of each control turn, the server power consumption prediction unit 330 (CPU power consumption prediction unit 331) acquires information on the amount of virtual resources to be newly allocated based on the load processing schedule (e.g., the number of CPU cores) and the resource usage status of the CPU server 3 at that time (e.g., CPU usage rate), and predicts the power consumption of each CPU server 3 using the CPU server power consumption learning model 302. Furthermore, based on the allocation configuration of the CPU servers 3 in each allocation control area 30, the CPU power consumption prediction unit 331 calculates the total CPU server power consumption by adding up the power consumption of each CPU server 3 in each allocation control area 30.

[0089] Furthermore, at the start of each control turn, the server power consumption prediction unit 330 (GPU power consumption prediction unit 332) predicts the GPU server power consumption of each GPU server 4 using the GPU server power consumption learning model 303 based on the load type of new load obtained from the load processing schedule information, the GPU inlet temperature obtained from the temperature distribution information 64, and the number of GPU cards. Furthermore, the GPU power consumption prediction unit 332 calculates the total GPU server power consumption by adding up the power consumption of each GPU server 4 in each placement control area 30 based on the placement configuration of the GPU servers 4 in each placement control area 30.

[0090] Furthermore, at the start of each control turn, the server power consumption prediction unit 330 (accelerator power consumption prediction unit 333) predicts the accelerator power consumption of each accelerator 5 using the accelerator power consumption learning model 304 based on the load type of a new load obtained from the load processing schedule information, the accelerator inlet temperature obtained from the temperature distribution information 64, and the number of accelerator processing circuits. Furthermore, the accelerator power consumption prediction unit 333 calculates the total accelerator power consumption by adding up the power consumption of each accelerator 5 in each placement control area 30 based on the placement configuration of the accelerators 5 in each placement control area 30.

[0091] Then, the server power consumption prediction unit 330 adds up the total CPU server power consumption, the total GPU server power consumption, and the total accelerator power consumption in each placement control area 30 for each placement pattern, and calculates the server power consumption for each placement control area 30 (step S15).

[0092] The placement pattern determination unit 340 sums up the server power consumption of each placement control area 30 in each placement pattern, and calculates the total server power consumption (total server power consumption). The placement pattern determination unit 340 calculates the sum of the calculated total server power consumption and the air conditioning power consumption for that placement pattern acquired from the air conditioning control unit 200, and determines the placement pattern that minimizes this sum (step S16).

[0093] In this way, the power consumption reduction control device 100 can determine a processing load allocation pattern and air conditioning control values ​​that reduce the total power consumption of a data center, which consists of server power consumption and air conditioning power consumption, in a data center environment where CPU servers, GPU servers, accelerators, etc. are mixed.

[0094] <Hardware configuration> The power amount reduction control device 100 according to this embodiment is realized by a computer 900 having a configuration as shown in FIG. 10, for example. 10 is a hardware configuration diagram showing an example of a computer 900 that realizes the functions of the power amount reduction control device 100 according to this embodiment. The computer 900 has a CPU (Central Processing Unit) 901, a ROM (Read Only Memory) 902, a RAM 903, an HDD (Hard Disk Drive) 904, an input / output I / F (Interface) 905, a communication I / F 906, and a media I / F 907.

[0095] The CPU 901 operates based on a program stored in the ROM 902 or the HDD 904, and performs control by the control unit. The ROM 902 stores a boot program executed by the CPU 901 when the computer 900 starts up, programs related to the hardware of the computer 900, and the like.

[0096] The CPU 901 controls an input device 910 such as a mouse or keyboard, and an output device 911 such as a display or printer, via an input / output I / F 905. The CPU 901 acquires data from the input device 910 via the input / output I / F 905, and outputs generated data to the output device 911. Note that a GPU (Graphics Processing Unit) or the like may be used as a processor together with the CPU 901.

[0097] The HDD 904 stores programs executed by the CPU 901 and data used by the programs. The communication I / F 906 receives data from other devices via a communication network (e.g., NW (Network) 920) and outputs the data to the CPU 901, and also transmits data generated by the CPU 901 to other devices via the communication network.

[0098] The media I / F 907 reads a program or data stored in the recording medium 912 and outputs it to the CPU 901 via the RAM 903. The CPU 901 loads a program related to a target process from the recording medium 912 onto the RAM 903 via the media I / F 907, and executes the loaded program. The recording medium 912 is an optical recording medium such as a DVD (Digital Versatile Disc) or a PD (Phase Change Rewritable Disc), a magneto-optical recording medium such as an MO (Magneto Optical Disk), a magnetic recording medium, a semiconductor memory, or the like.

[0099] For example, when a computer 900 functions as the power consumption reduction control device 100 of the present invention, a CPU 901 of the computer 900 executes a program loaded onto a RAM 903 to realize the functions of the power consumption reduction control device 100. Furthermore, data stored in the RAM 903 is stored in an HDD 904. The CPU 901 reads and executes a program related to a target process from a recording medium 912. Alternatively, the CPU 901 may read a program related to a target process from another device via a communication network (NW 920).

[0100] <Effects> The effects of the power amount reduction control device 100 and the like according to the present invention will be described below. The power consumption reduction control device according to the present invention is a power consumption reduction control device 100 that controls a CPU server 3, a GPU server 4, an accelerator 5, and a plurality of air conditioners 2, and has a plurality of placement control areas 30 in which any of the CPU server 3, the GPU server 4, and the accelerator 5 are placed, and an air conditioning control area 20 in which the effect of air conditioning control by the plurality of air conditioners 2 is measured. The power consumption reduction control device 100 has an air conditioning control value generation unit 221 that generates air conditioning control values ​​including at least a target temperature to be set to the plurality of air conditioners 2, and a power consumption reduction control unit 102 that controls the plurality of air conditioners 2 using the air conditioning control values. a reward calculation unit that calculates a reward using a target temperature as an index for the results of the air conditioning control by the air conditioning control execution unit 240 controlling the plurality of air conditioners 2 using air conditioning control values ​​in a plurality of placement patterns in which processing loads are allocated to the CPU server 3, the GPU server 4, and the accelerator 5, and determines whether the reward satisfies a predetermined condition; and a reward calculation unit that acquires temperature distribution information 64 and the air conditioning power consumption of the plurality of air conditioners as the control results using the air conditioning control values ​​that are determined to satisfy the predetermined condition, and calculates a reward for each placement control area in each of the plurality of placement patterns. an operation history creating unit 223 that creates operation history information 201 associated with the predicted heat generation amount of each of the CPU servers 3, the GPU servers 4, and the accelerators 5; an arrangement pattern calculating unit 310 that calculates a plurality of arrangement patterns for allocating new processing loads using information on the processing loads on the CPU servers 3, the GPU servers 4, and the accelerators 5; an area heat generation amount estimating unit 320 that estimates the predicted heat generation amount of each of the arrangement control areas 30 by summing up the heat generation amounts when processing loads are allocated to the CPU servers 3, the GPU servers 4, and the accelerators 5 that belong to each of the arrangement control areas 30 for each of the calculated arrangement patterns; an operation history information extraction unit (230) that refers to operation history information (201) using information on the predicted heat generation amount of 0 and extracts temperature distribution information (64) and air conditioning power consumption when controlled by the air conditioning control value in each placement pattern; a server power consumption prediction unit (330) that uses the extracted temperature distribution information (64) and information on the new processing load to calculate the server power consumption that is the sum of the power consumption of each CPU server (3), the power consumption of each GPU server (4), and the power consumption of each accelerator (5) for each placement control area (30) in each placement pattern;The system is characterized by including an arrangement pattern determination unit 340 that adds up the server power consumption amounts of each of the arrangement control areas 30, calculates the sum of the total server power consumption amount and the extracted air conditioning power consumption amount, and determines the arrangement pattern that minimizes the calculated sum as the arrangement pattern for allocating the processing loads.

[0101] In this way, the power consumption reduction control device 100 can reduce the total power consumption made up of server power consumption and air conditioning power consumption in an environment where a CPU server 3, a GPU server 4, and an accelerator 5 coexist.

[0102] Furthermore, the power consumption reduction control device according to the present invention is a power consumption reduction control device 100 that controls a CPU server 3, a GPU server 4, an accelerator 5, and a plurality of air conditioners 2 that are possessed by a data center 10, and on a floor of the data center 10, a plurality of placement control areas 30 in which any of the CPU servers 3, the GPU servers 4, and the accelerators 5 are placed as a group of servers to place processing loads, and an air conditioning control area 20 that is an area in which the effect of air conditioning control by the plurality of air conditioners 2 is measured, and the power consumption reduction control device measures the power consumption in the plurality of air conditioning control areas 20. an external factor acquisition unit 211 that acquires information on external factors related to air conditioning control, including an average floor temperature calculated from the average of the temperatures measured, an outside temperature that is the temperature outside the data center 10, and a predicted amount of heat generation in each placement control area 30 that is a predicted amount when a processing load is placed in a server group; a situation determination unit 212 that divides the value of each external factor into a predetermined range width, defines a combination of the ranges divided for each external factor as a situation classification 62, and determines to which situation classification 62 the acquired information on the external factor belongs; In each of the n classes 62, there is an air conditioning control value generation unit 221 that generates air conditioning control values ​​including at least a target temperature to be set for the plurality of air conditioners 2, an air conditioning control execution unit 240 that executes control of the plurality of air conditioners 2 using the air conditioning control values, a reward calculation unit 222 that calculates a reward for evaluating the results of the control of the plurality of air conditioners 2 by the air conditioning control values ​​using the target temperature as an index and determines whether the reward satisfies a predetermined condition, and a GPU inlet temperature indicating the temperature at the inlet of the GPU server 4, and The operation history creation unit 223 acquires temperature distribution information 64 indicating the accelerator inlet temperature, which indicates the temperature at the accelerator inlet, and the air conditioning power consumption of the multiple air conditioners 2 when control is performed using the air conditioning control value, and creates operation history information 201 that associates the temperature distribution information 64 and air conditioning power consumption acquired as the control results with the situation classification 62 and air conditioning control value when air conditioning control is performed. When information on the predicted heat generation amount for each placement control area 30 is acquired, the operation history creation unit 223 determines the current situation classification 62 via the situation determination unit 212,an operation history information extraction unit (230) that refers to operation history information (201) and extracts temperature distribution information (64) and air conditioning power consumption amounts when controlled by air conditioning control values ​​in each placement pattern; an placement pattern calculation unit (310) that acquires load processing schedule information indicating schedules for generating and deleting processing loads for the CPU servers (3), GPU servers (4) and accelerators (5), and calculates a placement pattern for allocating new processing loads to the CPU servers (3), GPU servers (4) and accelerators (5); an area heat generation amount estimation unit (320) that estimates a predicted amount of heat generation in each placement control area (30) by summing up the heat generation amounts when processing loads are allocated to the CPU servers (3), GPU servers (4) and accelerators (5) that belong to each placement control area (30) for each calculated placement pattern; and The system is characterized by comprising a server power consumption prediction unit 330 that calculates a total CPU server power consumption amount obtained by summing up the power consumption amounts of the CPU servers 3 in the placement control areas 30, a total GPU server power consumption amount obtained by summing up the power consumption amounts of the GPU servers in the placement control areas 30, and a total accelerator power consumption amount obtained by summing up the power consumption amounts of the accelerators in the placement control areas 30, and then sums up the total CPU server power consumption amount, the total GPU server power consumption amount, and the total accelerator power consumption amount in the placement control areas 30 to calculate the server power consumption amount for each placement control area 30; and a placement pattern determination unit 340 that sums up the server power consumption amounts for each placement control area 30 in each placement pattern, calculates the sum of the total server power consumption amount and the extracted air conditioning power consumption amount, and determines the placement pattern that minimizes the calculated sum as the placement pattern for allocating the processing loads.

[0103] By doing this, the power consumption reduction control device 100 can reduce the total power consumption of the data center 10, which consists of server power consumption and air conditioning power consumption, in an environment where a CPU server 3, a GPU server 4, and an accelerator 5 are mixed.

[0104] The power consumption reduction control device 100 also includes a CPU server power consumption learning model 302 that takes the resource usage status of the CPU server 3 as input information and outputs the power consumption of the CPU server 3; a GPU server power consumption learning model 303 that takes the GPU inlet temperature, type of processing load, and number of GPU cards of the GPU server 4 as input information and outputs the power consumption of the GPU server 4; and an accelerator power consumption learning model 304 that takes the accelerator inlet temperature, type of processing load, and number of accelerator processing circuits of the accelerator 5 as input information and outputs the power consumption of the accelerator 5, and is characterized in that a server power consumption prediction unit 330 calculates the power consumption of the CPU server 3 using the CPU server power consumption learning model 302, calculates the power consumption of the GPU server 4 using the GPU server power consumption learning model 303, and calculates the power consumption of the accelerator 5 using the accelerator power consumption learning model 304.

[0105] In this way, the power consumption reduction control device 100 can appropriately calculate the power consumption of each of the CPU server 3, GPU server 4, and accelerator 5 using the CPU server power consumption learning model 302, the GPU server power consumption learning model 303, and the accelerator power consumption learning model 304.

[0106] Furthermore, in the power consumption reduction control device 100, instead of the GPU server power consumption learning model 303, a first GPU learning model is provided which takes the GPU inlet temperature, type of processing load, and number of GPU cards of the GPU server 4 as input information and outputs the GPU temperature indicating the temperature of the GPU card, and a second GPU learning model which takes the GPU temperature as input information and outputs the power consumption of the GPU server, and the server power consumption prediction unit 330 calculates the GPU temperature using the first GPU learning model, and then calculates the power consumption of the GPU server 4 using the second GPU learning model.

[0107] By doing this, the power consumption reduction control device 100 can calculate the GPU temperature using the first GPU learning model, and then use the second GPU learning model to preferably calculate the power consumption of the GPU server 4.

[0108] Furthermore, in the power reduction control device 100, instead of the accelerator power learning model 304, there is provided a first accelerator learning model that takes the accelerator inlet temperature, type of processing load, and number of accelerator processing circuits of the accelerator 5 as input information and outputs the accelerator temperature indicating the temperature inside the accelerator, and a second accelerator learning model that takes the accelerator temperature as input information and outputs the power consumption of the accelerator, and the server power consumption prediction unit 330 calculates the accelerator temperature using the first accelerator learning model, and then calculates the power consumption of the accelerator using the second accelerator learning model.

[0109] By doing this, the power consumption reduction control device 100 can calculate the accelerator temperature using the first accelerator learning model, and then use the second accelerator learning model to preferably calculate the power consumption of the accelerator 5.

[0110] The power consumption reduction control device 100 is also provided with basic power consumption information 301 that indicates the reference power consumption at a predetermined temperature for each of the CPU server 3, the GPU server 4, and the accelerator 5, and the area heat generation amount estimation unit 320 uses the basic power consumption information 301 to calculate the power consumption of each of the CPU server 3, the GPU server 4, and the accelerator 5, thereby calculating the heat generation amount of each of the CPU server 3, the GPU server 4, and the accelerator 5.

[0111] In this way, the power consumption reduction control device 100 is equipped with basic power consumption information that indicates the reference power consumption at a specified temperature, making it possible to estimate the heat generation amount of each of the CPU server 3, GPU server 4, and accelerator 5.

[0112] The present invention is not limited to the above-described embodiments, and many modifications can be made by a person skilled in the art within the technical concept of the present invention. [Explanation of symbols]

[0113] 1. Power reduction control system 2 Air conditioner 3 CPU server 4 GPU Server 5. Accelerators 10 Data Center (DC) 20 Air-Conditioned Area 30. Placement Control Area 62 Situation classification 63 Air conditioning control value information 64 Temperature distribution information 65 Air conditioning power consumption information 100 Electricity reduction control device 200 Air conditioning control unit 201 Operational History Information 210 Situational Awareness Department 211 External factor acquisition part 212 Situation judgment section 220 Operation history information generation unit 221 Air conditioning control value generation unit 222 Remuneration Calculation Department 223 Operation History Creation Department 230 Operation History Information Extraction Unit 240 Air conditioning control execution unit 300 Server control unit 301 Basic power consumption information 302 CPU Server Power Consumption Learning Model 303 GPU Server Power Consumption Learning Model 304 Accelerator Power Learning Model 310 Placement pattern calculation unit 320 Regional Heat Generation Estimation Unit 330 Server Power Consumption Prediction Unit 331 CPU Power Prediction Unit 332 GPU Power Prediction Unit 333 Accelerator Power Prediction Unit 340 Placement pattern determination unit

Claims

1. A power consumption reduction control device that controls a CPU server, a GPU server, an accelerator, and a plurality of air conditioners, a plurality of placement control areas in which any of the CPU server, the GPU server, and the accelerator is placed, and an air conditioning control area in which an effect of air conditioning control by the plurality of air conditioners is measured; The power reduction control device includes: an air conditioning control value generation unit that generates air conditioning control values ​​including at least a target temperature to be set to the plurality of air conditioners; an air conditioning control execution unit that executes control of the plurality of air conditioners using the air conditioning control values; a reward calculation unit that calculates a reward for evaluating a result of the air conditioning control execution unit controlling the plurality of air conditioners using the air conditioning control value in a plurality of allocation patterns in which processing loads are allocated to the CPU server, the GPU server, and the accelerator, using the target temperature as an index, and determines whether the reward satisfies a predetermined condition; an operation history creation unit that acquires temperature distribution information and air conditioning power consumption amounts of the plurality of air conditioners as control results based on the air conditioning control values ​​that are determined to satisfy the predetermined conditions, and creates operation history information that corresponds to the predicted heat generation amount of each placement control zone in each of the plurality of placement patterns; an allocation pattern calculation unit that calculates a plurality of allocation patterns for allocating new processing loads using information on processing loads on the CPU server, the GPU server, and the accelerator; a zone heat generation amount estimation unit that estimates a predicted heat generation amount for each of the placement control zones by summing up the heat generation amounts when processing loads are placed on the CPU servers, GPU servers, and accelerators that belong to each of the placement control zones for each of the calculated placement patterns; an operation history information extraction unit that uses information on the predicted heat generation amount of each of the placement control areas, refers to the operation history information, and extracts the temperature distribution information and the air conditioning power consumption amount when controlled by the air conditioning control value in each placement pattern; a server power consumption prediction unit that calculates, for each of the placement control areas in each of the placement patterns, a server power consumption amount that is a total of the power consumption amounts of the CPU servers, the GPU servers, and the accelerators, using the extracted temperature distribution information and information related to the new processing load; an arrangement pattern determination unit that sums up the server power consumption amounts for each of the arrangement control areas in each of the arrangement patterns, calculates the sum of the total server power consumption amount and the extracted air conditioning power consumption amount, and determines the arrangement pattern that minimizes the calculated sum as the arrangement pattern for allocating the processing loads; A power reduction control device comprising:

2. A power consumption reduction control device that controls a CPU server, a GPU server, an accelerator, and a plurality of air conditioners in a data center, On the floor of the data center, a plurality of placement control areas are set in which CPU servers, GPU servers, or accelerators are placed as a group of servers on which processing loads are placed, and an air conditioning control area is set as an area in which the effects of air conditioning control by the plurality of air conditioners are measured; The power reduction control device includes: an external factor acquisition unit that acquires information about external factors related to air conditioning control, including an average floor temperature calculated from an average value of temperatures measured in a plurality of the air conditioning control zones, an external temperature that is the temperature outside the data center, and a predicted amount of heat generation in each of the placement control zones that is a predicted amount when the processing load is placed on the server group; a situation determination unit that divides the value of each of the external factors into a predetermined range width, defines a combination of the ranges divided for each of the external factors as a situation classification, and determines to which situation classification the acquired information of the external factor belongs; an air conditioning control value generator that generates air conditioning control values, including at least a target temperature, to be set for the plurality of air conditioners for each of the situation classifications; an air conditioning control execution unit that executes control of the plurality of air conditioners using the air conditioning control values; a reward calculation unit that calculates a reward for evaluating the result of the air conditioning control execution unit controlling the plurality of air conditioners using the air conditioning control value, using the target temperature as an index, and determines whether the reward satisfies a predetermined condition; and an operation history creation unit that acquires, as control results based on the air-conditioning control value determined to satisfy the predetermined condition, temperature distribution information indicating a GPU inlet temperature indicating the temperature at the inlet of the GPU server and an accelerator inlet temperature indicating the temperature at the inlet of the accelerator, and air-conditioning power consumption of the plurality of air conditioners when control is performed based on the air-conditioning control value, and creates operation history information that associates the temperature distribution information and the air-conditioning power consumption acquired as control results with the situation classification and the air-conditioning control value when air-conditioning control is performed; an operation history information extraction unit that, upon acquiring information on the predicted heat generation amount for each of the placement control areas, determines the current situation classification via the situation determination unit, and refers to the operation history information to extract the temperature distribution information and the air conditioning power consumption amount when controlled by the air conditioning control value in each placement pattern; an allocation pattern calculation unit that acquires load processing schedule information indicating schedules for generating and deleting processing loads for the CPU server, the GPU server, and the accelerator, and calculates an allocation pattern for allocating new processing loads to the CPU server, the GPU server, and the accelerator, respectively; a zone heat generation amount estimation unit that estimates a predicted heat generation amount for each of the placement control zones by summing up the heat generation amounts when processing loads are placed on the CPU servers, GPU servers, and accelerators that belong to each of the placement control zones for each of the calculated placement patterns; a server power consumption prediction unit that calculates, for each of the placement patterns, a total CPU server power consumption amount obtained by summing up the power consumption amounts of the CPU servers in the placement control areas, a total GPU server power consumption amount obtained by summing up the power consumption amounts of the GPU servers in the placement control areas, and a total accelerator power consumption amount obtained by summing up the power consumption amounts of the accelerators in the placement control areas using the load processing schedule information and the extracted temperature distribution information, and then sums up the total CPU server power consumption amount, the total GPU server power consumption amount, and the total accelerator power consumption amount in the placement control areas to calculate the server power consumption amount for each of the placement control areas; an arrangement pattern determination unit that sums up the server power consumption of each of the arrangement control areas for each arrangement pattern, calculates the sum of the total server power consumption and the extracted air conditioning power consumption, and determines the arrangement pattern that minimizes the calculated sum as the arrangement pattern for allocating the processing load; A power reduction control device comprising:

3. The power reduction control device includes: the system includes a CPU server power consumption learning model that takes the resource usage status of the CPU server as input information and the power consumption of the CPU server as output information; a GPU server power consumption learning model that takes the GPU inlet temperature, the type of processing load, and the number of GPU cards of the GPU server as input information and the power consumption of the GPU server as output information; and an accelerator power consumption learning model that takes the accelerator inlet temperature, the type of processing load, and the number of accelerator processing circuits of the accelerator as input information and the power consumption of the accelerator as output information, the server power consumption prediction unit calculates the power consumption of the CPU server using the CPU server power consumption learning model, calculates the power consumption of the GPU server using the GPU server power consumption learning model, and calculates the power consumption of the accelerator using the accelerator power consumption learning model.

3. The power reduction control device according to claim 1 or 2, wherein:

4. The power reduction control device includes: Instead of the GPU server power consumption learning model, a first GPU learning model is provided which takes the GPU inlet temperature, the type of processing load, and the number of GPU cards of the GPU server as input information and outputs a GPU temperature indicating the temperature of the GPU card, and a second GPU learning model is provided which takes the GPU temperature as input information and outputs the power consumption amount of the GPU server, The server power consumption prediction unit calculates the GPU temperature using the first GPU learning model, and then calculates the power consumption of the GPU server using the second GPU learning model.

4. The power reduction control device according to claim 3, wherein:

5. The power reduction control device includes: Instead of the accelerator power amount learning model, a first accelerator learning model is provided which receives as input information the accelerator inlet temperature, the type of processing load, and the number of accelerator processing circuits of the accelerator, and outputs as output information an accelerator temperature indicating a temperature inside the accelerator, and a second accelerator learning model is provided which receives as input information the accelerator temperature, and outputs as output information the amount of power consumed by the accelerator, The server power consumption prediction unit calculates the accelerator temperature using the first accelerator learning model, and then calculates the power consumption of the accelerator using the second accelerator learning model.

4. The power reduction control device according to claim 3, wherein:

6. The power reduction control device includes: basic power consumption information indicating a reference power consumption amount at a predetermined temperature for each of the CPU server, the GPU server, and the accelerator; the area heat generation amount estimating unit calculates the amount of power consumption of each of the CPU server, the GPU server, and the accelerator by using the basic power consumption amount information; and 3. The power reduction control device according to claim 1 or 2, wherein:

7. A power reduction control method for a power reduction control device that controls a CPU server, a GPU server, an accelerator, and a plurality of air conditioners, comprising: a plurality of placement control areas in which any of the CPU server, the GPU server, and the accelerator is placed, and an air conditioning control area in which an effect of air conditioning control by the plurality of air conditioners is measured; The power reduction control device includes: generating air conditioning control values ​​including at least target temperatures to be set to the plurality of air conditioners; a step of controlling the plurality of air conditioners using the air conditioning control values; calculating a reward for evaluating the results of controlling the plurality of air conditioners using the air conditioning control values ​​in a plurality of placement patterns in which processing loads are placed on the CPU server, the GPU server, and the accelerator, using the target temperature as an index, and determining whether the reward satisfies a predetermined condition; obtaining temperature distribution information and the air conditioning power consumption of the plurality of air conditioners as control results based on the air conditioning control values ​​determined to satisfy the predetermined conditions, and creating operation history information associated with the predicted heat generation amount of each placement control area in each of the plurality of placement patterns; calculating a plurality of allocation patterns for allocating the new processing loads using information on the processing loads on the CPU server, the GPU server, and the accelerator; a step of estimating a predicted amount of heat generated in each of the placement control areas by summing up the amounts of heat generated when a processing load is placed on a CPU server, a GPU server, and an accelerator that belong to each of the placement control areas for each of the calculated placement patterns; extracting the temperature distribution information and the air conditioning power consumption amount when controlled by the air conditioning control value in each arrangement pattern by referring to the operation history information using information on the predicted heat generation amount in each of the arrangement control areas; calculating, for each of the placement patterns, a server power consumption amount that is a total of the power consumption amounts of each CPU server, each GPU server, and each accelerator for each placement control area using the extracted temperature distribution information and information related to the new processing load; a step of summing up the server power consumption amounts for each of the placement control areas for each of the placement patterns, calculating the sum of the total server power consumption amount and the extracted air conditioning power consumption amount, and determining the placement pattern that minimizes the calculated sum as the placement pattern for placing the processing load; A power reduction control method comprising:

8. A power consumption reduction control system including a CPU server, a GPU server, an accelerator, and a power consumption reduction control device that controls a plurality of air conditioners, a plurality of placement control areas in which any of the CPU server, the GPU server, and the accelerator is placed, and an air conditioning control area in which an effect of air conditioning control by the plurality of air conditioners is measured; The power reduction control device includes: an air conditioning control value generation unit that generates air conditioning control values ​​including at least a target temperature to be set to the plurality of air conditioners; an air conditioning control execution unit that executes control of the plurality of air conditioners using the air conditioning control values; a reward calculation unit that calculates a reward for evaluating a result of the air conditioning control execution unit controlling the plurality of air conditioners using the air conditioning control value in a plurality of allocation patterns in which processing loads are allocated to the CPU server, the GPU server, and the accelerator, using the target temperature as an index, and determines whether the reward satisfies a predetermined condition; an operation history creation unit that acquires temperature distribution information and air conditioning power consumption amounts of the plurality of air conditioners as control results based on the air conditioning control values ​​that are determined to satisfy the predetermined conditions, and creates operation history information that corresponds to the predicted heat generation amount of each placement control zone in each of the plurality of placement patterns; an allocation pattern calculation unit that calculates a plurality of allocation patterns for allocating new processing loads using information on processing loads on the CPU server, the GPU server, and the accelerator; a zone heat generation amount estimation unit that estimates a predicted heat generation amount for each of the placement control zones by summing up the heat generation amounts when processing loads are placed on the CPU servers, GPU servers, and accelerators that belong to each of the placement control zones for each of the calculated placement patterns; an operation history information extraction unit that uses information on the predicted heat generation amount of each of the placement control areas, refers to the operation history information, and extracts the temperature distribution information and the air conditioning power consumption amount when controlled by the air conditioning control value in each placement pattern; a server power consumption prediction unit that calculates, for each of the placement control areas in each of the placement patterns, a server power consumption amount that is a total of the power consumption amounts of the CPU servers, the GPU servers, and the accelerators, using the extracted temperature distribution information and information related to the new processing load; an arrangement pattern determination unit that sums up the server power consumption amounts for each of the arrangement control areas in each of the arrangement patterns, calculates the sum of the total server power consumption amount and the extracted air conditioning power consumption amount, and determines the arrangement pattern that minimizes the calculated sum as the arrangement pattern for allocating the processing loads; A power reduction control system comprising:

9. A program for causing a computer to function as the power reduction control device according to claim 1 or 2.

Images