Power consumption management device, program, and power consumption management method

Abstract

In the present invention, a power consumption management device (100) comprises: a collection unit (111) for collecting the power consumption and information regarding the performance of servers (210, 220, 230); and a selection unit (112) for selecting an item of performance information to be a candidate for an explanatory variable in a power consumption estimation model (121) in which the explanatory variable is an item of performance information and power consumption is the objective variable. The invention is characterized in that, in selecting the item, the selection unit (112) excludes items of performance information in which the change in value is less than or equal to a prescribed value and further excludes items using multivariate analysis technique.

Description

[Technical Field]

[0001] The present invention relates to a power consumption management device, a program, and a power consumption management method for managing the power consumption of a server installed in a data center or the like. [Background technology]

[0002] As computer use expands and network services become more widespread, the power consumption of the servers that provide these services is increasing. Reducing power consumption is required not only from the perspective of cost, but also from the perspective of climate change risk. Conventional measures and methods for reducing power consumption include system consolidation using virtualization technology and reducing CPU (Central Processing Unit) power consumption using DVFS (Dynamic Voltage and Frequency Scaling). These methods reduce power consumption by monitoring or estimating server power consumption and then controlling the server hardware and software.

[0003] The main methods for monitoring server power consumption are to physically measure it or to estimate it from server performance information. One physical measurement method is to use a power meter, but this is not suitable for monitoring a large number of servers at once due to the installation costs and the need to shut down the server during installation. Another method is to obtain power consumption from a power sensor installed on the server via IPMI (Intelligent Platform Management Interface). However, not all servers are equipped with power sensors, and the measurement accuracy depends on the model.

[0004] When estimating power consumption from server performance information, performance information is acquired from the hardware or OS, and power consumption is estimated from the acquired performance values. The technology described in Non-Patent Document 1 estimates power consumption based on performance information at two levels: software level and hardware level. The technology described in Non-Patent Document 2 estimates power consumption based on 158 pieces of performance information. [Prior art documents] [Non-patent literature]

[0005] [Non-Patent Document 1] Eva Garcia-Martin, et al., "Estimation of energy consumption in machine learning," Journal of Parallel and Distributed Computing, 134 (2019), pp.75-88. [Non-patent document 2] Ziyu Shen, et al., "PCP-2LSTM: Two Stacked LSTM-based Prediction Model for Power Consumption in Data Centers," 2019 Seventh International Conference on Advanced Cloud and Big Data (CBD), pp.13-18. Summary of the Invention [Problem to be solved by the invention]

[0006] By using server performance information, it is possible to estimate power consumption while minimizing dependency on the server model. However, simply collecting performance information increases power consumption. In an experiment conducted by the inventor, collecting server performance information increased power consumption by approximately 6%, which is a power consumption that cannot be ignored in large-scale data centers with many servers. When collecting performance information for server management, not just for power consumption estimation, it is desirable to take power consumption into consideration when collecting it. The present invention has been made in view of the above background, and an object of the present invention is to enable a reduction in the power consumption required for collecting performance information used to estimate the power consumption of a server. [Means for solving the problem]

[0007] In order to solve the above-mentioned problems, a power consumption control device according to the present invention comprises: In operation a collection unit that collects performance information and power consumption of a server; and a selection unit that selects items of the performance information as explanatory variables in a power consumption estimation model in which the items of the performance information are explanatory variables and the power consumption is a target variable. a model generation unit that selects explanatory variables from the candidates and generates the power consumption estimation model; an estimation unit that estimates power consumption of the server using the power consumption estimation model generated by the model generation unit; and a server control unit that controls the server so that the power consumption of the server is reduced based on the power consumption estimated by the estimation unit. the selection unit excludes an item whose change in value is equal to or less than a predetermined value from among the items of the performance information, and further excludes the item by using a multivariate analysis technique to select the items. The collection unit newly collects the performance information for the items selected by the selection unit. It is characterized by: [Effects of the Invention]

[0008] According to the present invention, it is possible to reduce the power consumption required for collecting performance information used to estimate the power consumption of a server. [Brief explanation of the drawings]

[0009] [Figure 1] 1 is a functional block diagram of a power consumption control device according to an embodiment of the present invention; [Figure 2] 10 is a flowchart of a model generation process of a power consumption estimation model according to the present embodiment. [Figure 3] 10 is a flowchart of a parameter selection process according to the present embodiment. [Figure 4] 10 is a flowchart of a power consumption management process according to the present embodiment. [Figure 5] FIG. 2 is a hardware configuration diagram illustrating an example of a computer that realizes the functions of the power consumption management device according to the present embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0010] A power consumption management device according to an embodiment of the present invention will be described below. The power consumption management device estimates the power consumption of a server (physical server) based on performance information of the server using a power consumption estimation model, which is a model of machine learning and function approximation (regression). Explanatory variables (inputs) of the power consumption estimation model include those related to hardware operation, such as the number of branch prediction errors and the number of cache misses, and those related to the OS, such as CPU utilization rate and memory utilization rate.

[0011] The power consumption management device according to this embodiment further reduces these performance information items (parameters) by using a multivariate analysis technique, excluding parameters that always have constant values. Examples of multivariate analysis techniques include a technique for deleting one of a pair of parameters whose partial correlation coefficient is equal to or greater than a predetermined value, and a technique for reducing parameters using a stepwise method. In this embodiment, a performance information item (parameter) having a constant value means that the change in the parameter value over time is equal to or less than a predetermined value. The power consumption management device generates a power consumption estimation model using some or all of the reduced performance information items (parameters) as explanatory variables and power consumption as the objective variable. The power consumption management device collects server performance information, estimates power consumption using the power consumption estimation model, and controls the server to reduce power consumption.

[0012] Since the power consumption required to collect performance information depends on the number of performance information items (number of parameters), it is desirable to have a small number of items. The power consumption management device selects performance information items that have a strong correlation with the target variable as candidates for explanatory variables. For example, the power consumption management device selects items excluding items whose values remain constant (small changes) or items excluding one of a pair of items with a high degree of correlation as candidates for explanatory variables. The power consumption management device also selects items with a higher evaluation index (described below) than other items as candidates for explanatory variables. The power consumption management device then generates a power consumption estimation model that uses some or all of the candidate explanatory variables as explanatory variables. As a result, the power consumption management device can predict power consumption with high accuracy while reducing the number of items to be collected and the power consumption required for collection.The power consumption management device can control the server appropriately based on the predicted power consumption, and manage power consumption reduction.

[0013] <Configuration of power consumption management device> 1 is a functional block diagram of a power consumption management apparatus 100 according to this embodiment. The power consumption management apparatus 100 collects performance information from a server 210 on which an application runs, and from servers 220 and 230 on which virtual machines (referred to as VMs (Virtual Machines) in FIG. 1) and containers run. The performance information includes performance information items related to hardware operation, such as the number of branch prediction errors and the number of cache misses, and performance information items related to the OS, such as CPU utilization rate and memory utilization rate. The power consumption management apparatus 100 also collects power consumption data of the servers 210, 220, and 230 from IPMIs provided in the servers 210 and 220 and a power meter connected to the server 230.

[0014] 1, the power consumption of servers 210, 220, and 230 can be collected using IPMI, a power meter, etc., but after the generation of power consumption estimation model 121 (described later), there is no need to collect power consumption. For example, once power consumption estimation model 121 for a server is generated, the power consumption of other servers of the same model, configuration, and running the same types of applications can be estimated from performance information using the same power consumption estimation model 121, and IPMI or a power meter is not required. For example, in FIG. 1, server 240 is the same model as server 230, and the power consumption can be estimated from performance information using power consumption estimation model 121 of server 230.

[0015] The power consumption management device 100 is a computer and includes a control unit 110, a storage unit 120, and an input / output unit 180. User interface devices such as a display, keyboard, and mouse are connected to the input / output unit 180. The input / output unit 180 includes a communication device and is capable of sending and receiving data to and from the servers 210, 220, 230, and 240. A media drive may also be connected to the input / output unit 180, enabling data exchange using a recording medium.

[0016] ≪Power consumption management device: Storage section≫ The storage unit 120 includes storage devices such as a read-only memory (ROM), a random access memory (RAM), a solid-state drive (SSD), etc. The storage unit 120 stores collected data 130, training data 140, a power consumption estimation model 121, and a program 128. The collected data 130 is performance information (values of performance information items (parameters)) and power consumption collected from the servers 210, 220, and 230. The collected data 130 is time-series data collected at a predetermined timing, for example, periodically.

[0017] The training data 140 is training data used when generating the power consumption estimation model 121, which will be described later. The explanatory variables of the training data 140 are performance information, and the objective variable is power consumption. The training data 140 may be composed of data on performance information items (parameters) that serve as explanatory variables for the power consumption estimation model 121 from the collected data 130, and power consumption. These parameters may be further subjected to optimal preprocessing (selection of parameters that serve as explanatory variables) for each model of the server to be controlled and each application (service) on the server, and the performance information and power consumption corresponding to the selected parameters may be used as the training data 140. Alternatively, the training data 140 may be composed of newly collected data on explanatory variables and power consumption separately from the collected data 130.

[0018] The power consumption estimation model 121 is a model based on machine learning technology or function approximation (regression), and uses performance information as an explanatory variable (input) and power consumption as a response variable (output). The program 128 includes descriptions of a model generation process (see FIG. 2) and a parameter selection process (see FIG. 3), which will be described later.

[0019] <Power consumption control device: control unit> The control unit 110 includes a CPU, and is provided with a collection unit 111 , a selection unit 112 , a model generation unit 113 , an estimation unit 114 , and a server control unit 115 . The collection unit 111 collects performance information (parameter values) and power consumption from the servers 210, 220, and 230 and stores the collected data 130. The collection unit 111 collects power consumption when preparing learning data 140 and generating the power consumption estimation model 121. After the power consumption estimation model 121 is generated, the collection unit 111 does not collect the power consumption of the servers 210, 220, and 230, except, for example, when updating the power consumption estimation model 121 or when generating the power consumption estimation model 121 for the servers 210, 220, and 230 of a new model or with a new configuration. After the power consumption estimation model 121 is generated, the collection unit 111 collects performance information that is an explanatory variable of the power consumption estimation model 121, but does not collect other performance information.

[0020] The selection unit 112 selects candidate parameters (explanatory variables of the power consumption estimation model 121) to be used for predicting power consumption from among the performance information items (parameters). The selection unit 112 deletes parameters for which the change in the collected performance information (parameter values) is less than a predetermined value. The selection unit 112 also calculates a partial correlation coefficient between two parameters, and deletes one of the parameters if the P value (significance probability) of the non-correlation test is less than a predetermined value (e.g., 0.05) and the partial correlation coefficient is greater than or equal to a predetermined value (e.g., 0.5).

[0021] The selection unit 112 also applies a stepwise method to the remaining parameters using performance information as an explanatory variable and power consumption as a response variable, and selects candidate parameters to be used for prediction by leaving only parameters whose P-values are equal to or less than a predetermined value (e.g., 0.05) among the partial regression coefficients obtained and deleting the others. The remaining parameters are parameters excluding parameters whose parameter value changes are equal to or less than a predetermined value, or parameters excluding one of a pair of parameters whose partial correlation coefficient is equal to or greater than a predetermined value.

[0022] For example, the selection unit 112 selects candidate parameters to be used for prediction using a stepwise method such as a variable increasing / decreasing method or a variable decreasing / increasing method for the remaining parameters. Alternatively, for example, the selection unit 112 may select one parameter from the remaining parameters, delete the selected parameter, and then delete the remaining parameters one by one by repeating this process to maximize the evaluation index of the remaining parameter group. The evaluation index of the parameter group is an index based on the information criterion of the model that uses the parameter group as explanatory variables, and a model with a large evaluation index among multiple models is a desirable model. Examples of information criteria include the Akaike's Information Criterion (AIC) and the Bayesian Information Criterion (BIC).

[0023] The model generation unit 113 selects explanatory variables from the parameters selected by the selection unit 112 and generates a power consumption estimation model 121 using the training data 140. The explanatory variables of the power consumption estimation model 121 are a group of parameters included in the parameters (parameter group) selected by the selection unit 112, and the parameter group is selected depending on the server model and / or the application (service) on the server. In addition, a parameter group that increases the estimation accuracy is selected. For example, the model generation unit 113 may calculate power consumption using multiple power consumption estimation models 121 generated in preprocessing of model generation and adopt the power consumption estimation model 121 with the highest accuracy. Note that the model generation unit 113 may generate the power consumption estimation model 121 using all of the parameters selected by the selection unit 112 as explanatory variables.

[0024] One method in which the model generation unit 113 selects a group of parameters according to the server model and application to generate the power consumption estimation model 121 is to select parameters that are correlated with the server power consumption according to the server model and the running application (for example, a correlation coefficient of 0.5 or more) and use them as explanatory variables to generate the power consumption estimation model 121. Another method is to perform principal component analysis to perform dimensional reduction (parameter reduction) to generate estimation models of multiple machine learning algorithms, and to use the estimation model with the highest estimation accuracy as the power consumption estimation model 121. Furthermore, parameters may be reduced using a stepwise method, or these methods may be combined.

[0025] The applications include web applications, database servers, applications with high CPU loads, and applications with large input / output volumes (number of times). Of these types, servers include those that run specific types of applications and those that run a variety of applications. The model generation unit 113 selects a parameter group according to the type of server and the type of application running, and generates the power consumption estimation model 121.

[0026] The learning data 140 used for learning (generation) of the power consumption estimation model 121 may use performance information corresponding to parameters after preprocessing in which the model generation unit 113 selects (reduces) parameters according to the model of the server to be controlled and the applications on the server from the parameters selected by the selection unit 112. The learning data 140 may be composed of the collected data 130, or may be newly collected and composed.

[0027] The estimation unit 114 calculates (estimates) the power consumption from the performance information using the power consumption estimation model 121. Note that this performance information is performance information that serves as an explanatory variable of the power consumption estimation model 121 and is performance information collected by the servers 210, 220, 230, and 240.

[0028] The server control unit 115 controls the servers based on the estimated power consumption. For example, when there are multiple servers running virtual machines or containers with power consumption below a predetermined level, the server control unit 115 consolidates the virtual machines or containers into one server and shuts down the source server to reduce power consumption. For example, the server control unit 115 may lower the clock speed of the server's CPU when power consumption is above a predetermined value. Alternatively, the server control unit 115 may adjust the number of operating power supply modules according to the power consumption of the server group.

[0029] <Model generation process> 2 is a flowchart of a model generation process of the power consumption estimation model 121 according to this embodiment. The power consumption estimation model 121 may be generated only once for servers of the same model or configuration, or for the types of applications running on the servers, or may be updated at predetermined times (for example, periodically). In step S11, the collection unit 111 collects performance information and power consumption of the servers 210, 220, and 230, and stores the collected information in the collected data .

[0030] In step S12, the selection unit 112 executes a parameter selection process (see FIG. 3 described later) to select performance information items (parameters) that are candidates for explanatory variables of the power consumption estimation model 121. In step S13, the model generation unit 113 generates a power consumption estimation model 121. The explanatory variables of the power consumption estimation model 121 are selected from the parameters selected in the parameter selection process in step S12 according to the model of the server to be controlled and the applications on the server. The learning data 140 used for learning the power consumption estimation model 121 may be configured by extracting performance information and power consumption corresponding to the explanatory variables of the power consumption estimation model 121 from the collected data 130 collected in step S11, or may be configured by newly collecting performance information and power consumption.

[0031] ≪Parameter selection process≫ FIG. 3 is a flowchart of the parameter selection process according to this embodiment (see step S12 in FIG. 2). In step S21, the selection unit 112 deletes performance information items (parameters) whose values have not changed or whose value has changed by a predetermined value or less from the performance information collected in step S11 (see FIG. 2).

[0032] In step S22, the selection unit 112 starts the process of executing step S23 for each multivariate analysis method. In step S23, the selection unit 112 deletes parameters according to the multivariate analysis method. For example, in a multivariate analysis method using a partial correlation coefficient, the selection unit 112 calculates a partial correlation coefficient between two parameters, and deletes one of the parameters if the P value of the non-correlation test is a predetermined value (e.g., 0.05) or less and the partial correlation coefficient is a predetermined value (e.g., 0.5) or more. In a multivariate analysis method using a stepwise method, the selection unit 112 applies the stepwise method using performance information as an explanatory variable and power consumption as a target variable, and retains parameters whose P value is a predetermined value (e.g., 0.05) or less from the obtained partial regression coefficients and deletes the others. The selection unit 112 may further delete parameters using other multivariate analysis methods (e.g., principal component analysis or multiple regression analysis).

[0033] <Power consumption management process> 4 is a flowchart of the power consumption management process according to this embodiment. In the power consumption management process, steps S32 to S34 are repeatedly executed while the server is in operation. In step S31, if any of the servers 210, 220, 230, 240 is operating (step S31→YES), the collection unit 111 proceeds to step S32, and if any of the servers is stopped (step S31→NO), the collection unit 111 ends the power consumption management process.

[0034] In step S32, the collection unit 111 collects performance information from the servers 210, 220, 230, and 240. This performance information is performance information of parameters that serve as explanatory variables of the power consumption estimation model 121. In step S33, the estimation unit 114 calculates (estimates) the power consumption using the power consumption estimation model 121 based on the performance information collected in step S32. In step S34, server control unit 115 controls servers 210, 220, 230, and 240 based on the power consumption estimated in step S33.

[0035] <Features of the power consumption management device> When generating a power consumption estimation model 121 that estimates power consumption from server performance information, the power consumption management apparatus 100 selects candidate explanatory variables for the power consumption estimation model 121. During the selection process, the power consumption management apparatus 100 excludes, from collectable performance information, parameters whose value changes are equal to or less than a predetermined value. Next, the power consumption management apparatus 100 excludes one of the parameter pairs whose partial correlation coefficient is greater than a predetermined value, and further excludes the parameter using a stepwise method. Subsequently, the power consumption management apparatus 100 may generate a power consumption estimation model 121 that uses all of the parameters selected by the selection unit 112 as explanatory variables, or the model generation unit 113 may perform preprocessing of parameter reduction (explanatory variable selection) depending on the model of the server to be controlled and the applications (services) on that server, and generate a power consumption estimation model 121 that uses the reduced parameters as explanatory variables.

[0036] Examples of parameters whose values do not change (whose value change is less than a predetermined value) and are deleted in step S21 include alignment-faults (the number of accesses to unaligned data) and swap-used / swap-free (used size / free size of swap area). Examples of parameters whose partial correlation coefficients are greater than a predetermined value and are deleted in step S23 include page-faults (the number of page faults) and file_locks_wri (the number of file locks during writing). Also, examples of parameters deleted in step S23 using the stepwise method include LLC-prefetches (the number of prefetches from main memory to the last level cache) and memory-usage-cach (memory cache size).

[0037] According to an experiment conducted by the inventor, of the total 146 parameters preset as performance information, 34 parameters were deleted by deleting parameters whose value change was less than a predetermined value and then calculating and deleting the partial correlation coefficients. 60 parameters were further deleted using the stepwise method, and 52 parameters were selected. In other words, 94 parameters were deleted from the 146 parameters. Examples of parameters that were finally selected include branch-loads (number of branch prediction unit reads), LLC-loads (number of LLC reads), context-switches, CPU utilization, memory utilization, and number of writes.

[0038] The power consumption management device 100 can estimate power consumption with high accuracy from performance information of a small number of parameters. Since the power consumption management device 100 collects a small number of performance information items, it can reduce the power consumption of the server required for collection. Furthermore, the power consumption management device 100 controls the server so that power consumption is reduced based on the estimated power consumption. As a result, the power consumption management device 100 can efficiently reduce power consumption.

[0039] <<Variation: Multivariate analysis method>> In the above-described embodiment, the multivariate analysis method uses a method for calculating partial correlation coefficients between pairs of parameters and a stepwise method, but either one may be used. For example, after removing parameters whose value changes by a predetermined value or less, the remaining parameters may be further removed using the stepwise method, and the remaining parameters may be used as candidates for explanatory variables of the power consumption estimation model 121. Note that when using the stepwise method, it is desirable to calculate partial correlation coefficients and remove parameters with high correlations to avoid multicollinearity before using the stepwise method.

[0040] <<Variation: Uses other than power consumption estimation>> In the above-described embodiment, the power consumption management device 100 collects performance information on some or all of the parameters selected in the parameter selection process (see FIG. 3 ), estimates power consumption, and controls the server so as to reduce the server's power consumption. The collected performance information may be used for other purposes. For example, it may be used to monitor the performance of a server or an application running on a server. If the selected parameters are insufficient for performance monitoring, additional parameters may be collected. By monitoring performance based on the parameters selected in the parameter selection process, it becomes possible to perform performance monitoring with reduced power consumption.

[0041] Other variations In the above-described embodiment, the power consumption management device 100 generates the power consumption estimation model 121, estimates power consumption using this model, and controls the server. The generation of the power consumption estimation model 121, the estimation of power consumption, and the control of the server may be performed by different devices. For example, the power consumption estimation model 121 generated by the power consumption estimation model generation device may be used to estimate power consumption based on performance information collected by the server management device, and the server may be controlled. Note that a collection unit provided in the server management device collects performance information of performance information items that are explanatory variables of the power consumption estimation model 121 from the server.

[0042] Although several embodiments of the present invention have been described above, these embodiments are merely illustrative and do not limit the technical scope of the present invention. The present invention can take on various other embodiments, and various modifications such as omissions and substitutions can be made without departing from the spirit of the present invention. These embodiments and their modifications are included within the scope and spirit of the invention described in this specification, etc., and are also included in the invention described in the claims and their equivalents.

[0043] <Hardware configuration> The power consumption management device 100 according to this embodiment is realized by a computer 900 having a configuration as shown in Fig. 5, for example. Fig. 5 is a hardware configuration diagram showing an example of the computer 900 that realizes the functions of the power consumption management device 100 according to this embodiment. The computer 900 includes a CPU 901, a ROM 902, a RAM 903, an SSD 904 (referred to as an SSD (Solid State Drive) in Fig. 5), an input / output interface 905 (referred to as an input / output I / F (Interface) in Fig. 5), a communication interface 906 (referred to as a communication I / F in Fig. 5), and a media interface 907 (referred to as a media I / F in Fig. 5). The computer 900 may include a hard disk instead of the SSD 904, or may include a hard disk in addition to the SSD 904.

[0044] The CPU 901 operates based on a program stored in the ROM 902 or the SSD 904, and performs control by the control unit 110 in Fig. 1. 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. 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 interface 905. The CPU 901 acquires data from the input device 910 via the input / output interface 905, and outputs generated data to the output device 911.

[0045] The SSD 904 stores programs executed by the CPU 901 and data used by the programs. The communication interface 906 receives data from devices (not shown) (e.g., servers 210, 220, 230) via a communication network and outputs the data to the CPU 901, and also transmits data generated by the CPU 901 to the servers 210, 220, 230 via the communication network.

[0046] The media interface 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 the program from the recording medium 912 onto the RAM 903 via the media interface 907, and executes the loaded program. The recording medium 912 is an optical recording medium such as a Blu-ray Disc (registered trademark), a magneto-optical recording medium such as an MO (Magneto Optical disk), a magnetic recording medium, a conductive memory tape medium, a semiconductor memory, or the like.

[0047] For example, when the computer 900 functions as the power consumption management device 100 according to this embodiment, the CPU 901 of the computer 900 executes the program 128 (see FIG. 1) loaded onto the RAM 903, thereby realizing the functions of the power consumption management device 100. The CPU 901 reads the program from the recording medium 912 and executes it. Alternatively, the CPU 901 may read the program from another device via a communication network, or may install the program 128 from the recording medium 912 onto the SSD 904 and execute it.

[0048] Effect The effects of the power consumption management apparatus 100 will be described below. The power consumption management apparatus 100 according to this embodiment includes a collection unit 111 that collects performance information and power consumption of a server, and a selection unit 112 that selects performance information items that are candidates for explanatory variables in a power consumption estimation model 121 that uses performance information items as explanatory variables and power consumption as a response variable. The selection unit 112 excludes items whose value changes are equal to or less than a predetermined value from among the performance information items, and then uses a multivariate analysis technique to further exclude items and select items.

[0049] Such a power consumption management device 100 makes it possible to select explanatory variables with a small number of items that have a strong relationship (high evaluation index) with the power consumption, which is the objective variable, as explanatory variables for the power consumption estimation model 121. Performance information, which is input when estimating power consumption, can be collected with little power consumption, and explanatory variables (performance information items, parameters) that can be estimated with high accuracy can be identified.

[0050] The multivariate analysis method according to this embodiment is a method for calculating partial correlation coefficients between pairs of items, and the selection unit 112 excludes one of the items when the partial correlation coefficient of the pair of items is equal to or greater than a predetermined value.

[0051] According to the power consumption control device 100, it becomes possible to reduce the explanatory variables using partial correlation coefficients.

[0052] The multivariate analysis method according to this embodiment is a stepwise method, and the selection unit 112 further selects items using the stepwise method with the remaining items after selection as explanatory variables and power consumption as a response variable.

[0053] According to the power consumption control device 100, it becomes possible to use a stepwise method to reduce the explanatory variables by leaving explanatory variables that have a strong relationship with the objective variable (high evaluation index).

[0054] The power consumption management device 100 according to this embodiment further includes a model generation unit 113 that generates a power consumption estimation model 121 using the items selected by the selection unit 112 from the performance information and power consumption collected by the collection unit 111, or items further reduced from the selected items depending on the model of the server and / or the services on the server, as explanatory variables, and power consumption as a response variable.

[0055] According to the power consumption management device 100, a power consumption estimation model 121 is generated in which the items selected by the selection unit 112 are used as explanatory variables and power consumption is used as a response variable. The number of performance information items that serve as input to the power consumption estimation model 121 is small, and can be collected with low power consumption, allowing for highly accurate estimation.

[0056] The collection unit 111 according to this embodiment further includes an estimation unit 114 that, after generating the power consumption estimation model 121, collects only performance information that is an explanatory variable of the power consumption estimation model 121, and estimates the power consumption of the server using the power consumption estimation model 121 based on the collected performance information.

[0057] According to the power consumption management apparatus 100, performance information can be collected with low power consumption, and power consumption can be estimated with high accuracy. [Explanation of symbols]

[0058] 100 Power consumption management device 111 Collection Department 112 Selection section 113 Model Generation Unit 114 Estimation Department 115 Server control unit 121 Power Consumption Estimation Model 128 programs 130 Collected Data 140 training data 210,220,230,240 servers

Claims

1. A collection unit that collects performance information and power consumption of a running server; a selection unit that selects the performance information items as candidates for explanatory variables in a power consumption estimation model in which the performance information items are explanatory variables and the power consumption is a response variable; a model generation unit that selects explanatory variables from the candidates and generates the power consumption estimation model; an estimation unit that estimates the power consumption of the server using the power consumption estimation model generated by the model generation unit; a server control unit that controls the server so that the power consumption of the server is reduced based on the power consumption estimated by the estimation unit, The selection unit Among the performance information items, excluding the items whose value change is equal to or less than a predetermined value, using a multivariate analysis technique to select the items while further excluding the items; The collection unit newly collects the performance information for the items selected by the selection unit. A power consumption control device characterized by:

2. the multivariate analysis method is a method for calculating a partial correlation coefficient between pairs of the items, The selection unit excludes one of the items when a partial correlation coefficient of the pair of items is equal to or greater than a predetermined value.

2. The power consumption control device according to claim 1.

3. the multivariate analysis method is a stepwise method, The selection unit further selects the items using the stepwise method with the remaining selected items as explanatory variables and power consumption as a response variable.

3. The power consumption control device according to claim 1 or 2.

4. The model generation unit uses, as explanatory variables, items further reduced from the items selected by the selection unit according to the model of the server and / or the services on the server, The collection unit newly collects the performance information on the items that the model generation unit uses as explanatory variables.

2. The power consumption control device according to claim 1.

5. the collection unit, after generating the power consumption estimation model, collects only performance information that is an explanatory variable of the power consumption estimation model; The estimation unit estimates the power consumption of the server using the power consumption estimation model based on the collected performance information.

5. The power consumption control device according to claim 4.

6. A program for causing a computer to function as the power consumption management device according to any one of claims 1 to 5.

7. The power consumption control device collecting performance information and power consumption of a running server; selecting the performance information items as candidates for explanatory variables in a power consumption estimation model in which the performance information items are used as explanatory variables and the power consumption is used as a response variable; selecting explanatory variables from the candidates and generating the power consumption estimation model; estimating the power consumption of the server using the generated power consumption estimation model; and controlling the server so that the power consumption of the server is reduced based on the estimated power consumption; The step of selecting an item of performance information includes: excluding, from among the performance information items, items whose value changes are equal to or less than a predetermined value; and selecting the items by further excluding the items using a multivariate analysis technique; The new collection of the performance information is performed for the item selected in the step of selecting the item of the performance information. A power consumption management method comprising:

Images