Server power consumption control method and device, equipment, medium and program product

Abstract

The invention provides a server power consumption control method which can be applied to the field of artificial intelligence. The server power consumption control method comprises the steps that operation state information of a plurality of servers in a server cluster is analyzed, belief distribution corresponding to preset power consumption influence factors is determined, and the belief distribution is used for representing the trust level of the preset power consumption influence factors on the power consumption influence degree; the belief distributions of the plurality of servers are fused according to the respective information source confidence of the plurality of servers to obtain comprehensive state information, and the information source confidence represents the credibility of the running state information in the servers; inputting the comprehensive state information into an action selection model so as to select an optimization control operation corresponding to the comprehensive state information from a plurality of power consumption control operations by utilizing the action selection model; and controlling the server cluster to execute optimization control operation so as to adjust the power consumption of the plurality of servers. The invention further provides a server power consumption control device and equipment, a storage medium and a program product.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence, specifically to the field of server power consumption control technology, and more specifically to a server power consumption control method, device, equipment, medium and program product. Background Art

[0002] Server clusters undertake massive data processing tasks, and they generate a lot of power consumption during operation. In order to improve energy efficiency, related technologies monitor the operating status of servers in real time. When the power consumption of the server is detected to be too high, users can use the controller to adjust the power consumption accordingly.

[0003] In the process of realizing the concept of the present disclosure, the inventors found that there are at least the following problems in the related technology: due to the limited accuracy of the controller and the fact that the user can only perform passive adjustments through the controller during the adjustment process, the adjustment of the server power consumption is not timely enough and the efficiency is relatively low. Summary of the invention

[0004] In view of the above problems, the present disclosure provides a method, apparatus, device, medium and program product for controlling server power consumption.

[0005] According to the first aspect of the present disclosure, a server power consumption control method is provided, comprising: analyzing the operating status information of each of a plurality of servers in a server cluster, determining a belief distribution corresponding to a preset power consumption influencing factor, wherein the belief distribution is used to characterize the trust level of the degree of influence of the preset power consumption influencing factor on power consumption; fusing the belief distributions of the plurality of servers according to their respective source confidences to obtain comprehensive status information, wherein the source confidence characterizes the degree of trust in the operating status information in the servers; inputting the comprehensive status information into an action selection model, so as to select an optimization control operation corresponding to the comprehensive status information from a plurality of power consumption control operations using the action selection model; and controlling the server cluster to execute the optimization control operation to adjust the power consumption of the plurality of servers.

[0006] According to an embodiment of the present disclosure, the above-mentioned analysis of the operating status information of each of the multiple servers in the server cluster to determine the belief distribution corresponding to the preset power consumption influencing factors includes: using a belief allocation function to analyze the operating status information of each of the multiple servers in the above-mentioned server cluster to determine the belief distribution of each of the above-mentioned servers, wherein the above-mentioned belief distribution includes a belief value for each assumption condition corresponding to the above-mentioned belief allocation function, and the above-mentioned assumption condition is obtained by combining multiple of the above-mentioned preset power consumption influencing factors.

[0007] According to an embodiment of the present disclosure, the belief distributions of the multiple servers are fused according to the respective source confidences of the multiple servers to obtain comprehensive status information, including: assigning a weight value to the belief distribution of each server according to the respective source confidences of the multiple servers, wherein the weight value is proportional to the source confidence; utilizing information combination rules, based on the weight value of each belief distribution, weighted fusion of the multiple belief distributions is performed to obtain the comprehensive status information, wherein the comprehensive status information includes belief values ​​for each of the power consumption influencing factors.

[0008] According to an embodiment of the present disclosure, the above-mentioned comprehensive state information is input into an action selection model to select an optimized control operation corresponding to the above-mentioned comprehensive state information from multiple power consumption control operations using the above-mentioned action selection model, including: inputting the above-mentioned comprehensive state information into the above-mentioned action selection model as a state representation of the above-mentioned action selection model to perform the following operations using the above-mentioned action selection model: using an action value function to estimate the long-term added value of using each of the above-mentioned power consumption control operations based on the above-mentioned comprehensive state information; selecting an optimized control operation from multiple power consumption control operations according to the above-mentioned long-term added value, wherein the above-mentioned optimized control operation is the power consumption control operation with the highest long-term added value.

[0009] According to an embodiment of the present disclosure, the method further includes: obtaining the adjustment state information obtained when the server cluster is running under the control of the optimization control operation; evaluating the adjustment state information according to the operation state information to obtain an evaluation result; in the case where the evaluation result indicates that there is a difference between the adjustment state information and the operation state information, adjusting the parameters in the belief allocation function and the information combination rule respectively based on the state adjustment information, wherein the belief allocation function is used to analyze the operation state information, and the information combination rule is used to fuse multiple belief distributions; verifying the rationality of the adjusted belief allocation function and information combination rule; feeding back the adjustment state information to the action selection model to optimize the weight and bias of the action value function in the action selection model based on the adjustment state information; in response to the received state analysis request, determining the updated comprehensive state information of the server cluster using the verified adjusted belief allocation function and information combination rule; inputting the updated comprehensive state information into the optimized action selection model, and outputting the optimization control operation corresponding to the updated comprehensive state information.

[0010] According to an embodiment of the present disclosure, the above method also includes: preprocessing the above operating status information, wherein the above preprocessing operation includes at least one of denoising processing, standardization processing and normalization processing; wherein the above normalization processing includes mapping the operating parameters in the above operating status information to a preset range based on a reference interval, wherein, for the temperature parameters of the components in the above operating status information, the above reference interval is determined based on the heat dissipation characteristics of the above components.

[0011] According to an embodiment of the present disclosure, the above-mentioned control of the server cluster to perform the above-mentioned optimization control operation to adjust the power consumption of the above-mentioned multiple servers includes: determining operating parameters associated with the above-mentioned optimization control operation, wherein the above-mentioned operating parameters include at least one of fan speed, processor utilization and memory usage; controlling the above-mentioned server cluster to adjust the above-mentioned operating parameters to reduce the power consumption of the above-mentioned multiple servers.

[0012] The second aspect of the present disclosure provides a server power consumption control device, including: a belief determination module, which is used to analyze the operating status information of each of the multiple servers in the server cluster, and determine the belief distribution corresponding to the preset power consumption influencing factor, wherein the above belief distribution is used to characterize the trust level of the influence of the above preset power consumption influencing factor on the power consumption; a belief fusion module, which is used to fuse the belief distributions of the multiple servers according to the respective source confidences of the multiple servers, and obtain comprehensive status information, wherein the above source confidence characterizes the degree of trust in the operating status information in the above servers; an operation selection module, which is used to input the above comprehensive status information into an action selection model, so as to select the optimization control operation corresponding to the above comprehensive status information from multiple power consumption control operations using the above action selection model; an operation execution module, which is used to control the above server cluster to execute the above optimization control operation, so as to adjust the power consumption of the above multiple servers.

[0013] A third aspect of the present disclosure provides an electronic device, comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the above method.

[0014] The fourth aspect of the present disclosure further provides a computer-readable storage medium having a computer program or instructions stored thereon, which implements the steps of the above method when the above computer program or instructions are executed by a processor.

[0015] The fifth aspect of the present disclosure further provides a computer program product, including a computer program or instructions, which implement the steps of the above method when the above computer program or instructions are executed by a processor.

[0016] According to the embodiments of the present disclosure, by analyzing the operating status information of each of the multiple servers and fusing the multiple belief distributions obtained by the analysis, the fusion of multi-source data is achieved, so that the comprehensive status information determined is more accurate and comprehensive. The fused comprehensive status information is input into the action selection model, and the optimization control operation corresponding to the comprehensive status information is determined by using the action selection model, so that when facing changing workloads and environmental conditions, the operation of adjusting the power consumption of the server can be determined quickly and accurately. The server cluster is controlled to perform the optimization control operation, and the power consumption of the server is automatically adjusted based on the operating status information detected in real time, which effectively improves the efficiency of power consumption adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The above contents and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0018] Figure 1 The application scenario diagram of the server power consumption control method, apparatus, device, medium and program product according to the embodiment of the present disclosure is schematically shown;

[0019] Figure 2 A flow chart of a method for controlling power consumption of a server according to an embodiment of the present disclosure is schematically shown;

[0020] Figure 3 A flowchart of determining an optimization control operation in a method for controlling power consumption of a server according to an embodiment of the present disclosure is schematically shown;

[0021] Figure 4 A schematic diagram showing a connection relationship of multiple processing modules in a method for controlling power consumption of a server according to an embodiment of the present disclosure;

[0022] Figure 5 A flow chart of a method for controlling power consumption of a server according to another embodiment of the present disclosure is schematically shown;

[0023] Figure 6 The structure block diagram of the server power consumption control device according to the embodiment of the present disclosure is schematically shown;

[0024] Figure 7 A block diagram of an electronic device suitable for implementing a method for controlling server power consumption according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0025] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0026] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise", "include", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components.

[0027] All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0028] When using expressions such as "at least one of A, B, and C, etc.", they should generally be interpreted according to the meaning of the expression commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0029] In the technical solution of the present disclosure, the user information (including but not limited to user personal information, user image information, user device information, such as location information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved are all information and data authorized by the user or fully authorized by all parties, and the collection, storage, use, processing, transmission, provision, disclosure and application of the relevant data comply with relevant laws, regulations and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0030] Server clusters are the core components of modern data centers, responsible for massive data processing and computing tasks. However, these clusters consume a lot of electricity during operation, so how to effectively control their power consumption to improve energy efficiency, reduce operating costs and reduce carbon emissions is a relatively important issue. Power consumption control methods in related technologies mainly rely on static strategies or experience-based heuristic algorithms, but these methods often find it difficult to achieve the optimal balance between energy efficiency and performance when faced with complex and changing workloads.

[0031] Specifically, power consumption control methods in related technologies mainly rely on hardware-level improvements, such as using more efficient power supplies and cooling systems. However, these methods often require huge upfront investments and cannot dynamically adapt to changes in actual workloads.

[0032] Secondly, methods based on intelligent algorithms and data-driven are gradually becoming popular. Fixed policy control methods usually use preset thresholds and rules to manage server power consumption. For example, when the processor utilization exceeds a certain threshold, the fan speed is increased or more computing resources are enabled. This method is simple to implement and easy to deploy; however, it has poor flexibility, cannot dynamically adapt to uncertain workload changes, and easily leads to excessive or insufficient resource configuration.

[0033] Furthermore, the feedback control method usually uses classic control methods such as controllers to adjust power consumption by monitoring the operating status of the server in real time. This method can be adjusted dynamically and has a fast response speed; however, the control accuracy is limited and it is difficult to handle complex nonlinear relationships and multi-objective optimization problems.

[0034] Furthermore, the control method based on machine learning uses regression models or classification models to predict future power consumption requirements, and then pre-adjusts based on the predicted results. This method can improve the prediction accuracy by using historical data, but the model training and updating are complex and rely on a large amount of high-quality training data. For the traditional existing solutions, it is necessary to make up for the shortcomings in dynamic adaptability and uncertainty.

[0035] An embodiment of the present disclosure provides a server power consumption control method, which analyzes the operating status information of each of multiple servers in a server cluster, determines the belief distribution corresponding to the preset power consumption influencing factor, wherein the belief distribution is used to characterize the trust level of the degree of influence of the preset power consumption influencing factor on power consumption; fuses the belief distributions of the multiple servers according to the respective source confidences of the multiple servers to obtain comprehensive status information, wherein the source confidence characterizes the degree of trust in the operating status information in the server; inputs the comprehensive status information into an action selection model, so as to use the action selection model to select the optimization control operation corresponding to the comprehensive status information from multiple power consumption control operations; and controls the server cluster to perform the optimization control operation to adjust the power consumption of the multiple servers.

[0036] like Figure 1 As shown, the application scenario 100 according to this embodiment may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104 and a server 105. The network 104 is used to provide a medium for a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0037] The user can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 through the network 104 to receive or send messages, etc. Various communication client applications can be installed on the first terminal device 101, the second terminal device 102, and the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only for example).

[0038] The first terminal device 101, the second terminal device 102, and the third terminal device 103 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0039] The server 105 may be a server that provides various services, such as a background management server (only as an example) that provides support for websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103. The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device.

[0040] It should be noted that the server power consumption control method provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the server power consumption control device provided in the embodiment of the present disclosure can generally be set in the server 105. The server power consumption control method provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105. Correspondingly, the server power consumption control device provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105.

[0041] It should be understood that Figure 1The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0042] The following will be based on Figure 1 The scene described by Figure 2~Figure 5 The server power consumption control method of the disclosed embodiment is described in detail.

[0043] like Figure 2 As shown, the server power consumption control method of this embodiment includes operations S210 to S240.

[0044] In operation S210, the operation status information of each of the plurality of servers in the server cluster is analyzed to determine a belief distribution corresponding to a preset power consumption influencing factor, wherein the belief distribution is used to characterize a confidence level of the influence degree of the preset power consumption influencing factor on power consumption.

[0045] In operation S220, belief distributions of the multiple servers are fused according to the respective source confidences of the multiple servers to obtain comprehensive status information, wherein the source confidence represents the degree of trustworthiness of the running status information in the server.

[0046] In operation S230, the comprehensive state information is input into an action selection model, so as to select an optimized control operation corresponding to the comprehensive state information from a plurality of power consumption control operations using the action selection model.

[0047] In operation S240 , the server cluster is controlled to perform an optimization control operation to adjust power consumption of the plurality of servers.

[0048] According to an embodiment of the present disclosure, the power consumption generated during the operation of the cluster is controlled by collecting and analyzing the operation status information of each server in the server cluster. The operation status information of the server is provided by the baseboard management controller (BMC) in the server, and different BMCs can be regarded as different information sources. The operation status information is obtained by using multiple sensors in the BMC, and the operation status information includes information such as processor utilization, memory usage, temperature, and fan speed.

[0049] According to an embodiment of the present disclosure, the operating status information of each server is analyzed to convert the operating status information into a belief distribution corresponding to a preset power consumption influencing factor. The preset power consumption factors include temperature factors, load factors, and resource occupancy factors. In the process of analyzing the operating status information, information related to each preset power consumption influencing factor is extracted and analyzed one by one to obtain a corresponding belief distribution. The belief distribution is used to quantify the degree of trust in the information under a certain assumption, that is, how high is the level of trust in the degree to which the preset power consumption influencing factor affects power consumption.

[0050] According to an embodiment of the present disclosure, after determining the belief distribution of each server, multiple belief distributions are fused to obtain comprehensive status information of the server cluster. Since the source confidence of each server is different, the weight value assigned to the belief distribution of each server during the fusion process is also different. Among them, the source confidence refers to the evaluation of the reliability and authenticity of the information source. By fusing the belief distribution according to the source confidence, the final comprehensive status information is more accurate.

[0051] According to an embodiment of the present disclosure, the comprehensive state information is input into the action selection model to determine the optimal control operation corresponding to the comprehensive state information using the action selection model. The action selection model may be an algorithm that combines deep learning and reinforcement learning, such as a DQN (Deep Q-Network) model. The core idea of ​​the action selection model is to use a deep neural network to approximate the action value function, thereby making decisions among multiple power consumption control operations. The power consumption control operation is a plurality of means of operating the server learned during the model training process.

[0052] According to an embodiment of the present disclosure, after determining that an optimization control operation corresponding to the comprehensive status information is obtained, the server cluster is controlled to perform the operation to adjust the power consumption of multiple servers and reduce the overall power consumption of the server cluster. The optimization control operation can maximize the power consumption reduction effect.

[0053] According to an embodiment of the present disclosure, by analyzing the operating status information of each of the multiple servers and fusing the multiple belief distributions obtained by the analysis, the fusion of multi-source data is achieved, so that the comprehensive status information determined is more accurate and comprehensive. The fused comprehensive status information is input into the action selection model, and the optimization control operation corresponding to the comprehensive status information is determined by using the action selection model, so that when facing changing workloads and environmental conditions, the operation of adjusting the power consumption of the server can be determined quickly and accurately. The server cluster is controlled to perform the optimization control operation, and the power consumption of the server is automatically adjusted based on the operating status information detected in real time, which effectively improves the efficiency of power consumption adjustment.

[0054] According to an embodiment of the present disclosure, the operating status information of each of the multiple servers in the server cluster is analyzed to determine the belief distribution corresponding to the preset power consumption influencing factors, including: using a belief allocation function to analyze the operating status information of each of the multiple servers in the server cluster to determine the belief distribution of each server, wherein the belief distribution includes a belief value for each assumption condition corresponding to the belief allocation function, and the assumption condition is obtained by combining multiple preset power consumption influencing factors.

[0055] According to an embodiment of the present disclosure, a belief distribution function is used to analyze the operating status information of each of the multiple servers in the server cluster. The belief distribution function is used to define the probability mapping of an event in the interval [0,1], which represents the credibility of the occurrence of the event under a certain assumption. Each assumption corresponds to a belief distribution function, and the belief distribution function is used to calculate the belief value under each assumption. Specifically, the assumption is obtained by combining multiple preset power consumption influencing factors. In the process of determining the assumption, the identification framework is analyzed, and then all possible situations in the identification framework are listed, thereby obtaining multiple assumptions. The identification framework is defined based on the preset power consumption influencing factors.

[0056] For example, according to the preset recognition framework defined by the power consumption influencing factors as {temperature factor, load factor, resource occupancy factor}, multiple assumptions are determined as {temperature factor}, {temperature factor, load factor}, {temperature factor, resource occupancy factor}, {load factor}, {load factor, resource occupancy factor}, {resource occupancy factor}. The belief allocation function is used to calculate the belief value corresponding to each assumption. If there is strong supporting evidence for the "temperature factor" in the operating status information, the trust allocation function may assign a higher belief value to the {temperature factor}.

[0057] According to the embodiment of the present disclosure, the belief distribution of the server is constructed according to the belief value of each assumption condition, and the belief distribution is used to represent the trust level of different assumption conditions of the health status of the server. Assuming that the identification framework is {temperature factor, load factor, resource occupancy factor}, then using m i ({A}) represents each belief value to obtain the belief distribution, where i represents different servers and A represents different assumptions.

[0058] According to an embodiment of the present disclosure, the DS evidence theory (Dempster-Shafer evidence theory) can be used to determine the belief distribution. By converting the running state information into the belief distribution, the running state of the current server can be accurately determined.

[0059] According to an embodiment of the present disclosure, belief distributions of multiple servers are fused according to their respective source confidences to obtain comprehensive status information, including: assigning a weight value to the belief distribution of each server according to their respective source confidences, wherein the weight value is proportional to the source confidence; utilizing information combination rules and based on the weight value of each belief distribution, weighted fusion of multiple belief distributions is performed to obtain comprehensive status information, wherein the comprehensive status information includes belief values ​​for each power consumption influencing factor.

[0060] According to an embodiment of the present disclosure, multiple belief distributions are fused, wherein the fusion process involves two dimensions: one dimension is to comprehensively consider the evidence provided by multiple belief distributions on preset power consumption influencing factors, and then obtain a comprehensive estimate of the server cluster status; the other dimension is to consider the information provided by different servers, obtain the confidence of different sources, and fuse the confidence of information from different servers. Specifically, in the fusion process, a corresponding weight value is assigned to the belief distribution of each server according to the confidence of the source, so that multiple belief distributions are fused based on the weight value.

[0061] According to an embodiment of the present disclosure, multiple belief distributions are weighted and fused using information combination rules to obtain comprehensive state information. The information combination rule may be a Dempster combination rule, also known as the Dempster-Shafer synthesis rule, which is a method for combining different evidences. In this embodiment, each belief distribution corresponds to a belief confidence. Multiple belief distributions are weighted and summed to obtain a comprehensive belief distribution corresponding to each hypothesis condition, and then comprehensive state information is generated based on the comprehensive belief distribution.

[0062] According to an embodiment of the present disclosure, in the process of fusing belief distributions, when there are conflicts between multiple belief distributions, the conflicts need to be eliminated to ensure the smooth progress of the fusion process. When calculating the conflict, the conflict factor is calculated based on the belief distribution. Among them, the conflict factor can be expressed as K. When K=0, it means that the two belief distributions are completely non-conflicting. When K=1, it means that the two belief distributions are completely conflicting. In this case, using the Dempster combination rule will result in the belief distributions being unable to fuse. Therefore, in the process of fusing using the Dempster combination rule, the conflict information K is used as part of the normalization factor and is "eliminated" from the final fusion result, that is, the information of the conflicting part is proportionally distributed to the non-conflicting part.

[0063] According to the embodiments of the present disclosure, if the belief value in the belief distribution obtained by the final fusion is greater than 1, it is necessary to normalize the multiple comprehensive belief distributions again. By fusing the belief distributions of different servers, the final comprehensive state information is more accurate and the comprehensiveness of the comprehensive state information is improved. Conflicts are eliminated during the fusion process, ensuring that the final fusion result is more reliable.

[0064] According to an embodiment of the present disclosure, comprehensive state information is input into an action selection model so as to utilize the action selection model to select an optimized control operation corresponding to the comprehensive state information from multiple power consumption control operations, including: using the comprehensive state information as a state representation of the action selection model, and inputting it into the action selection model so as to utilize the action selection model to perform the following operations: utilizing an action value function to estimate the long-term added value of using each power consumption control operation based on the comprehensive state information; and selecting an optimized control operation from multiple power consumption control operations based on the long-term added value, wherein the optimized control operation is the power consumption control operation with the highest long-term added value.

[0065] According to an embodiment of the present disclosure, the comprehensive state information obtained after fusion is used as the state representation of the action selection model and input into the action selection model. The action selection model will perform action value function approximation based on this new state representation input. Among them, the action value function can be the Q function in the action selection model. The approximation of the Q function is usually implemented by a deep neural network. The basic principle is to approximate the Q function Q(s,a) through a neural network, where s is the state representation and a is the action representation. The neural network estimates the optimal Q value through continuous trial and error learning.

[0066] According to an embodiment of the present disclosure, the action selection model estimates the long-term added value, that is, the long-term reward, of taking a certain action a in a certain state s by learning a Q-value function. In this problem, the reward function is usually based on the following aspects: temperature control effect, energy efficiency, and task performance. Regarding the temperature control effect, if the adjusted temperature is kept within a safe range, its long-term added value is high. Regarding energy efficiency, if the heat dissipation strategy adopted can reduce energy consumption while maintaining a good temperature control effect, its long-term added value is high. Regarding task performance, if the heat dissipation strategy adopted can maintain the efficient working state of the server, its long-term added value is high.

[0067] According to an embodiment of the present disclosure, based on the Q-value estimation of the current state, the optimal action is selected, which generally refers to a power consumption adjustment strategy or a resource allocation scheme. Specifically, by comparing the long-term added values ​​of different power consumption control operations, the power consumption control operation with the highest long-term added value is selected as the optimized control operation. Among them, the optimized control operation can be to adjust the fan speed or limit the processor utilization to optimize the heat dissipation performance of the server. By analyzing the current and predicted states using the action selection model and combining the comprehensive state information fused by the DS evidence, the power consumption of the server cluster can be accurately adjusted.

[0068] like Figure 3 As shown, the process of determining the optimal control operation includes four steps. In step S310, the operating status information of each server in the server cluster is obtained. In operation S320, the operating status information is analyzed using DS evidence theory to construct the belief distribution of each server. In operation S330, the belief distribution is input into the action selection model to approximate the action selection function. In operation S340, the optimal control operation is selected according to the function approximation result.

[0069] According to the embodiments of the present disclosure, the power consumption of the server cluster can be accurately adjusted by analyzing the current and predicted states through the action selection model and combining the multi-source data fused by the DS evidence theory. This dynamic adjustment can maximize the energy utilization efficiency while ensuring the quality of service according to the real-time load conditions and expected demand.

[0070] According to the embodiments of the present disclosure, the action selection model selects the optimal control strategy and adjustment parameters based on the belief distribution, thereby optimizing the power consumption and performance of the system. This intelligent decision-making can quickly respond to changing workloads and environmental conditions, improving the efficiency and stability of the overall system.

[0071] According to an embodiment of the present disclosure, a server cluster is controlled to perform an optimization control operation to adjust the power consumption of multiple servers, including: determining operating parameters associated with the optimization control operation, wherein the operating parameters include at least one of fan speed, processor utilization, and memory usage; and controlling the server cluster to adjust the operating parameters to reduce the power consumption of multiple servers.

[0072] According to an embodiment of the present disclosure, after determining the optimization control operation, since the optimization control operation determined according to the action selection model is in the form of a vector, the server cannot be directly controlled based on the vector. Determine the operating parameters associated with the optimization control operation. Specifically, the operating parameters are parameters actually involved in the operation of the server cluster and are adjustable. Control the server cluster to adjust the operating parameters to reduce the power consumption of multiple servers. For example, the speed of the cooling fan in the server can be increased to reduce the temperature of the server during operation. By determining the operating parameters, it is convenient to automatically adjust the power consumption of the server.

[0073] According to an embodiment of the present disclosure, after executing the optimization control operation, the execution effect can also be evaluated, the heat dissipation performance indicators such as temperature stability, accuracy of fan speed control, etc. can be calculated, and the room for improvement can be determined to provide further strategies for reducing power consumption.

[0074] According to an embodiment of the present disclosure, the server power consumption control method also includes: obtaining adjustment state information obtained when the server cluster runs under the control of the optimization control operation; evaluating the adjustment state information according to the operation state information to obtain an evaluation result; when the evaluation result indicates that there is a difference between the adjustment state information and the operation state information, adjusting the parameters in the belief allocation function and the information combination rule based on the state adjustment information, respectively, wherein the belief allocation function is used to analyze the operation state information, and the information combination rule is used to fuse multiple belief distributions; verifying the rationality of the adjusted belief allocation function and information combination rule; feeding back the adjustment state information to the action selection model to optimize the weight and bias of the action value function in the action selection model based on the adjustment state information; in response to the received state analysis request, determining the updated comprehensive state information of the server cluster using the verified adjusted belief allocation function and information combination rule; inputting the updated comprehensive state information into the optimized action selection model, and outputting the optimization control operation corresponding to the updated comprehensive state information.

[0075] According to an embodiment of the present disclosure, after the optimization control operation is executed, the adjustment state information obtained by the server cluster under the control of the optimization control operation is obtained. The adjustment state information is evaluated according to the operation state information, and the specific evaluation means may be to determine whether there is a difference between the two. If there is a difference, it means that the optimization control operation has produced an effect. In addition, the evaluation means may also be to re-judge the power consumption generated by the current server according to the adjustment state information to determine whether the current power consumption is within the range of stable operation to generate an evaluation result.

[0076] According to an embodiment of the present disclosure, when the evaluation result indicates that there is no difference between the adjustment state information and the running state information, it means that the optimization control operation has not produced any effect. The optimization control operation is marked and fed back to the action selection model. The weight of the optimization control operation is reduced so that a lower long-term added value is assigned to it when the action value function is approximated. In addition, if in a new server operating environment, it is determined for multiple consecutive times that there is no difference between the adjustment state information and the running state information, it means that the parameters of the current function and model are relatively inaccurate, then the step length of the preset length is increased during the parameter adjustment process, thereby increasing the span of the parameter adjustment, so that it can learn quickly to adapt to the new operating environment.

[0077] According to an embodiment of the present disclosure, when the evaluation result indicates that there is a difference between the adjustment state information and the operation state information, it is determined that the optimization control operation has produced an effect. Feedback learning is performed based on the state adjustment information to promote continuous learning and optimization of the system. Among them, feedback learning is divided into two parts, one part is for DS evidence theory, and the other part is for action selection model.

[0078] According to the embodiments of the present disclosure, for the DS evidence theory, the parameters in the belief allocation function and the information combination rule are adjusted based on the state adjustment information, so that the belief distribution generated by the belief allocation function is more accurate, and the information combination rule is more comprehensive and accurate when fusing multiple belief distributions. After adjusting the belief allocation function and the information combination rule, it is also necessary to verify their rationality to reduce the occurrence of conflicts. Specifically, simulation data can be used to test the adjusted belief allocation function and combination rules, generate different input belief distributions, combine them, and observe whether the results are reasonable.

[0079] According to the embodiments of the present disclosure, for the action selection model, the weight and bias of the action value function in the action selection model are optimized based on the adjustment state information. Specifically, the model's loss function can be used to complete the parameter adjustment of the model. By adjusting the parameters of the action selection model, the prediction of the action value function is made closer to the actual action value, thereby improving the performance of the action value function.

[0080] According to the embodiment of the present disclosure, after both parts of feedback learning are processed, if a state analysis request is received, the updated comprehensive state information of the server cluster can be determined by using the adjusted belief allocation function and information combination rule that have been verified, and the updated comprehensive state information is input into the optimized action selection model, and the optimized control operation corresponding to the updated comprehensive state information is output. After the optimized control operation is obtained, the server cluster is controlled to perform the optimized control operation to reduce the power consumption of multiple servers.

[0081] According to the embodiments of the present disclosure, feedback learning is performed based on the adjustment status information, so that each processing module can continuously learn and optimize, continuously improve the decision-making ability and prediction accuracy, and cope with the increasingly complex server cluster management challenges.

[0082] like Figure 4 As shown, the server power consumption control method mainly includes a state monitoring module 410 , a belief distribution fusion module 420 , an action selection module 430 , an execution control module 440 , and a feedback learning module 450 .

[0083] Among them, the state monitoring module 410 is responsible for real-time collection of the operating status information of the server cluster, such as processor utilization, memory usage, temperature, fan speed, etc. The belief distribution fusion module 420 uses the DS evidence theory to fuse multi-source data from different sensors to improve the accuracy of state estimation. The action selection module 430 learns and makes decisions on the input comprehensive state information and determines the optimal control operation. The execution control module 440 adjusts the operating parameters of the server cluster according to the optimal control operation selected by the action selection module 430 to achieve power consumption control. The feedback learning module 450 evaluates the execution results and uses the feedback for continuous learning and optimization of the model.

[0084] like Figure 4 As shown, the state monitoring module 410 is connected to the belief distribution fusion module 420, and the various operating state information collected in real time by the state monitoring module 410 is passed to the belief distribution fusion module 420. The belief distribution fusion module 420 is connected to the action selection module 430, and the belief distribution fusion module 420 performs fusion processing on the collected multi-source data and passes the fused high-credibility comprehensive state information to the action selection module 430. The action selection module 430 is connected to the execution control module 440, and the action selection module 430 determines the optimization control operation through the deep Q network according to the fused state information, and passes the optimization control operation to the execution control module 440. The execution control module 440 is connected to the feedback learning module 450, and the execution control module 440 adjusts the operating parameters of the server cluster according to the received optimization control operation.

[0085] According to an embodiment of the present disclosure, after executing the optimization control operation, the state monitoring module 410 collects the execution results and new state information, and transmits the information to the feedback learning module 450. The feedback learning module 450 connects the belief distribution fusion module 420 and the action selection module 430, and the feedback learning module 450 transmits the state information after feedback to the belief distribution fusion module 420 to adjust the evidence fusion rule. The execution results are used for continuous learning and optimization of the action selection module 430, and the action selection module 430 is updated to improve the accuracy of the next round of decision-making.

[0086] According to the embodiments of the present disclosure, by designing a feedback learning module, and using the feedback learning module to interact with the belief distribution fusion module and the action selection module, a better strategy and adjustment parameters can be learned to improve the system efficiency and power consumption control effect. The system can be learned and optimized according to the actual execution results to adapt to dynamically changing workloads and environmental requirements.

[0087] According to an embodiment of the present disclosure, the server power consumption control method also includes: preprocessing the operating status information, wherein the preprocessing operation includes at least one of denoising processing, standardization processing and normalization processing; wherein the normalization processing includes mapping the operating parameters in the operating status information to a preset range based on a reference interval, wherein, for the temperature parameters of the components in the operating status information, the reference interval is determined based on the heat dissipation characteristics of the components.

[0088] According to an embodiment of the present disclosure, the collected operating status information is preliminarily cleaned and preprocessed, such as denoising, standardization, and normalization. For temperature data, normalization processing can be performed based on the heat dissipation characteristics of the server. Specifically, the temperature range of each component is determined based on the heat dissipation characteristics of the server. For example, the temperature range of the processor may be between 40°C and 80°C, and the temperature range of the hard disk may be between 30°C and 50°C. These temperature ranges are used as reference intervals for normalization of temperature parameters so that subsequent modules can better understand and use these data.

[0089] like Figure 5 As shown, the server power consumption control method executed in this embodiment includes an operation status information collection unit 501, a data preprocessing unit 502, a belief distribution generation unit 503, a belief distribution fusion unit 504, a fusion conflict resolution unit 505, a fusion result input unit 506, an action value function approximation unit 507, an optimization control operation selection unit 508, an optimization control operation execution unit 509, an operation feedback acquisition unit 510, an operation feedback evaluation unit 511 and a feedback learning unit 512.

[0090] like Figure 5As shown, after the feedback learning unit 512 performs feedback learning according to the evaluation results, the feedback learning results are transmitted to the data preprocessing unit 502, the belief distribution generating unit 503, the belief distribution fusion unit 504 and the action value function approximation unit 507 respectively. The feedback learning results are transmitted to the data preprocessing unit 502 to update the reference interval in the data preprocessing, so that the normalized result is more reliable. The feedback learning results are transmitted to the belief distribution generating unit 503 to update the parameters in the belief allocation function used when generating the belief distribution, so that the generated belief distribution is more accurate. The feedback learning results are transmitted to the belief distribution generating unit 503 to update the parameters in the information combination rule used when fusing the belief distribution, so that the integrated state information obtained by fusion is more comprehensive and reliable. The feedback learning results are transmitted to the action value function approximation unit 507 to optimize the weight and bias of the action value function, so that when the action value function is approximated, the selected optimization control strategy is more accurate.

[0091] According to the embodiments of the present disclosure, the action selection model in deep reinforcement learning is used to approximate the action value function through a neural network to achieve optimal decision-making on the power consumption control strategy of the server cluster. The DS evidence theory is used to fuse data from multiple monitoring points to improve the accuracy and comprehensiveness of the system status. The power consumption of the server is dynamically adjusted to achieve dynamic power consumption adjustment based on real-time monitoring data and prediction models, so that the server cluster can achieve the best energy efficiency ratio and performance under different load and environmental conditions. Using feedback and learning mechanisms, the system can be learned and optimized according to the actual execution results to adapt to dynamically changing workloads and environmental requirements.

[0092] Based on the above server power consumption control method, the present disclosure also provides a server power consumption control device. Figure 6 The device is described in detail.

[0093] like Figure 6 As shown, the server power consumption control device 600 of this embodiment includes a belief determination module 610 , a belief fusion module 620 , an operation selection module 630 and an operation execution module 640 .

[0094] The belief determination module 610 is used to analyze the operation status information of each of the multiple servers in the server cluster, and determine the belief distribution corresponding to the preset power consumption influencing factor, wherein the belief distribution is used to characterize the trust level of the preset power consumption influencing factor on the degree of power consumption. In one embodiment, the belief determination module 610 can be used to perform the operation S210 described above, which will not be repeated here.

[0095] The belief fusion module 620 is used to fuse the belief distributions of multiple servers according to the respective source confidences of the multiple servers to obtain comprehensive status information, wherein the source confidence represents the degree of trust in the running status information in the server. In one embodiment, the belief fusion module 620 can be used to perform the operation S220 described above, which will not be repeated here.

[0096] The operation selection module 630 is used to input the comprehensive state information into the action selection model, so as to select the optimal control operation corresponding to the comprehensive state information from multiple power consumption control operations using the action selection model. In one embodiment, the operation selection module 630 can be used to perform the operation S230 described above, which will not be repeated here.

[0097] The operation execution module 640 is used to control the server cluster to execute the optimization control operation to adjust the power consumption of multiple servers. In one embodiment, the operation selection module 630 can be used to execute the operation S240 described above, which will not be described in detail here.

[0098] According to an embodiment of the present disclosure, the belief determination module 610 includes a belief determination submodule.

[0099] The belief determination submodule is used to use the belief distribution function to analyze the operating status information of multiple servers in the server cluster and determine the belief distribution of each server, wherein the belief distribution includes the belief value of each assumption condition corresponding to the belief distribution function, and the assumption condition is obtained by combining multiple preset power consumption influencing factors.

[0100] According to an embodiment of the present disclosure, the belief fusion module 620 includes a weight allocation submodule and a belief fusion submodule.

[0101] The weight allocation submodule is used to allocate a weight value to the belief distribution of each server according to the respective source confidence of multiple servers, wherein the weight value is proportional to the source confidence.

[0102] The belief fusion submodule is used to perform weighted fusion on multiple belief distributions based on the weight value of each belief distribution by utilizing information combination rules to obtain comprehensive state information, wherein the comprehensive state information includes belief values ​​for various power consumption influencing factors.

[0103] According to an embodiment of the present disclosure, the operation selection module 630 includes an information input submodule, an additional determination submodule and an action selection submodule.

[0104] The information input submodule is used to input the comprehensive state information into the action selection model as the state representation of the action selection model.

[0105] The additional determination submodule is used to estimate the long-term additional value of using each power consumption control operation based on the comprehensive state information by using the action value function.

[0106] The action selection submodule is used to select an optimized control operation from a plurality of power consumption control operations according to the long-term added value, wherein the optimized control operation is the power consumption control operation with the highest long-term added value.

[0107] According to an embodiment of the present disclosure, the server power consumption control device 600 further includes an information acquisition module, an information evaluation module, a parameter adjustment module, a rationality verification module, a function optimization module, an update determination module and an operation determination module.

[0108] The information acquisition module is used to obtain the adjustment status information obtained when the server cluster runs under the control of the optimization control operation.

[0109] The information evaluation module is used to evaluate the adjustment status information according to the operation status information to obtain an evaluation result.

[0110] A parameter adjustment module, used for adjusting the parameters in the belief allocation function and the information combination rule respectively based on the state adjustment information when the evaluation result indicates that there is a difference between the adjustment state information and the operation state information, wherein the belief allocation function is used to analyze the operation state information, and the information combination rule is used to fuse multiple belief distributions;

[0111] The rationality verification module is used to verify the rationality of the adjusted belief allocation function and information combination rules.

[0112] The function optimization module is used to feed back the adjustment state information to the action selection model to optimize the weight and bias of the action value function in the action selection model based on the adjustment state information.

[0113] The update determination module is used to determine the updated comprehensive status information of the server cluster in response to the received status analysis request by using the verified adjusted belief allocation function and information combination rule.

[0114] The operation determination module is used to input the updated comprehensive state information into the optimized action selection model and output the optimized control operation corresponding to the updated comprehensive state information.

[0115] According to an embodiment of the present disclosure, the server power consumption control device 600 further includes a preprocessing module.

[0116] A preprocessing module is used to preprocess the operating status information, wherein the preprocessing operation includes at least one of denoising, standardization and normalization; wherein the normalization includes mapping the operating parameters in the operating status information to a preset range based on a reference interval, wherein, for the temperature parameters of the components in the operating status information, the reference interval is determined based on the heat dissipation characteristics of the components.

[0117] According to an embodiment of the present disclosure, the operation execution module 640 includes a parameter determination submodule and a parameter control submodule.

[0118] The parameter determination submodule is used to determine an operating parameter associated with the optimization control operation, wherein the operating parameter includes at least one of a fan speed, a processor utilization rate, and a memory usage rate.

[0119] The parameter control submodule is used to control the server cluster to adjust the operating parameters to reduce the power consumption of multiple servers.

[0120] According to an embodiment of the present disclosure, any multiple modules of the belief determination module 610, the belief fusion module 620, the operation selection module 630, and the operation execution module 640 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of these modules can be combined with at least part of the functions of other modules and implemented in one module. According to an embodiment of the present disclosure, at least one of the belief determination module 610, the belief fusion module 620, the operation selection module 630, and the operation execution module 640 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or can be implemented by hardware or firmware such as any other reasonable way of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware, and firmware, or in an appropriate combination of any of them. Alternatively, at least one of the belief determination module 610, the belief fusion module 620, the operation selection module 630 and the operation execution module 640 may be at least partially implemented as a computer program module, and when the computer program module is executed, the corresponding function may be executed.

[0121] like Figure 7As shown, the electronic device 700 according to an embodiment of the present disclosure includes a processor 701, which can perform various appropriate operations and processes according to a program stored in a read-only memory (ROM) 702 or a program loaded from a storage part 708 to a random access memory (RAM) 703. The processor 701 may include, for example, a general-purpose microprocessor (such as a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (for example, an application-specific integrated circuit (ASIC)), etc. The processor 701 may also include an onboard memory for caching purposes. The processor 701 may include a single processing unit or multiple processing units for performing different operations of the method flow according to an embodiment of the present disclosure.

[0122] In RAM 703, various programs and data required for the operation of electronic device 700 are stored. Processor 701, ROM 702 and RAM 703 are connected to each other via bus 704. Processor 701 performs various operations of the method flow according to the embodiment of the present disclosure by executing the program in ROM 702 and / or RAM 703. It should be noted that the program can also be stored in one or more memories other than ROM 702 and RAM 703. Processor 701 can also perform various operations of the method flow according to the embodiment of the present disclosure by executing the program stored in one or more memories.

[0123] According to an embodiment of the present disclosure, the electronic device 700 may further include an input / output (I / O) interface 705, which is also connected to the bus 704. The electronic device 700 may further include one or more of the following components connected to the input / output (I / O) interface 705: an input portion 706 including a keyboard, a mouse, etc.; an output portion 707 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 708 including a hard disk, etc.; and a communication portion 709 including a network interface card such as a LAN card, a modem, etc. The communication portion 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the input / output (I / O) interface 705 as needed. A removable medium 711, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 710 as needed, so that a computer program read therefrom is installed into the storage portion 708 as needed.

[0124] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0125] According to an embodiment of the present disclosure, the computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, it may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus or a device. For example, according to an embodiment of the present disclosure, the computer-readable storage medium may include the ROM 702 and / or RAM 703 described above and / or one or more memories other than ROM 702 and RAM 703.

[0126] The embodiment of the present disclosure also includes a computer program product, which includes a computer program, and the computer program includes a program code for executing the method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the server power consumption control method provided by the embodiment of the present disclosure.

[0127] The above functions defined in the system / device of the embodiment of the present disclosure are performed when the computer program is executed by the processor 701. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0128] In one embodiment, the computer program may rely on tangible storage media such as optical storage devices, magnetic storage devices, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of signals on a network medium, and downloaded and installed through the communication part 709, and / or installed from the removable medium 711. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0129] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 709, and / or installed from the removable medium 711. When the computer program is executed by the processor 701, the above functions defined in the system of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the system, device, means, module, unit, etc. described above can be implemented by a computer program module.

[0130] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect through the Internet).

[0131] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0132] It will be appreciated by those skilled in the art that the features described in the various embodiments of the present disclosure may be combined and / or combined in a variety of ways, even if such combinations or combinations are not explicitly described in the present disclosure. In particular, without departing from the spirit and teachings of the present disclosure, the features described in the various embodiments of the present disclosure may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present disclosure.

[0133] The embodiments of the present disclosure are described above. However, these embodiments are only for illustrative purposes and are not intended to limit the scope of the present disclosure. Although the embodiments are described above, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A method for controlling server power consumption, characterized in that: The method comprises: Analyze the operation status information of each of the multiple servers in the server cluster to determine the belief distribution corresponding to the preset power consumption influencing factor, wherein the belief distribution is used to characterize the trust level of the preset power consumption influencing factor on the degree of influence of power consumption; According to the respective source confidences of the multiple servers, the belief distributions of the multiple servers are merged to obtain comprehensive status information, wherein the source confidence represents the degree of trustworthiness of the running status information in the servers; Inputting the comprehensive state information into an action selection model, so as to select an optimized control operation corresponding to the comprehensive state information from a plurality of power consumption control operations using the action selection model; The server cluster is controlled to perform the optimization control operation to adjust the power consumption of the plurality of servers.

2. The method according to claim 1, characterized in that The analyzing the operation status information of each of the multiple servers in the server cluster to determine the belief distribution corresponding to the preset power consumption influencing factors includes: Using the belief allocation function, the operating status information of each of the multiple servers in the server cluster is analyzed to determine the belief distribution of each server, wherein the belief distribution includes the belief value of each assumption condition corresponding to the belief allocation function, and the assumption condition is obtained by combining multiple preset power consumption influencing factors.

3. The method according to claim 1, characterized in that The step of fusing the belief distributions of the multiple servers according to the respective source confidences of the multiple servers to obtain comprehensive status information includes: According to the respective information source confidences of the multiple servers, assigning a weight value to the belief distribution of each server, wherein the weight value is proportional to the information source confidence; The information combination rule is used to perform weighted fusion on a plurality of the belief distributions based on the weight value of each of the belief distributions to obtain the comprehensive state information, wherein the comprehensive state information includes the belief value for each of the power consumption influencing factors.

4. The method according to claim 1, characterized in that: The step of inputting the comprehensive state information into an action selection model, so as to select an optimization control operation corresponding to the comprehensive state information from a plurality of power consumption control operations by using the action selection model, comprises: The comprehensive state information is used as a state representation of the action selection model and input into the action selection model to perform the following operations using the action selection model: estimating, by means of an action-value function, a long-term added value of using each of the power consumption control operations based on the integrated state information; An optimal control operation is selected from the plurality of power consumption control operations according to the long-term added value, wherein the optimal control operation is the power consumption control operation with the highest long-term added value.

5. The method according to any one of claims 1 to 4, characterized in that The method further comprises: Acquire adjustment status information of the server cluster under the control of the optimization control operation; Evaluate the adjustment status information according to the operation status information to obtain an evaluation result; In the case where the evaluation result indicates that there is a difference between the adjustment state information and the operation state information, adjusting the parameters in the belief allocation function and the information combination rule respectively based on the state adjustment information, wherein the belief allocation function is used to analyze the operation state information, and the information combination rule is used to fuse multiple belief distributions; Verify the rationality of the adjusted belief allocation function and information combination rules; Feeding the adjustment state information back to the action selection model to optimize the weight and bias of the action value function in the action selection model based on the adjustment state information; In response to the received status analysis request, determining updated comprehensive status information of the server cluster using the verified adjusted belief allocation function and information combination rule; The updated comprehensive state information is input into the optimized action selection model, and the optimized control operation corresponding to the updated comprehensive state information is output.

6. The method according to claim 1, characterized in that The method further comprises: Preprocessing the running status information, wherein the preprocessing operation includes at least one of denoising, standardization and normalization; The normalization processing includes mapping the operating parameters in the operating status information to a preset range based on a reference interval, wherein, for the temperature parameters of the components in the operating status information, the reference interval is determined based on the heat dissipation characteristics of the components.

7. The method according to claim 1, characterized in that The controlling the server cluster to perform the optimization control operation to adjust the power consumption of the plurality of servers includes: determining an operating parameter associated with the optimization control operation, wherein the operating parameter comprises at least one of a fan speed, a processor utilization, and a memory usage; The server cluster is controlled to adjust the operating parameters to reduce power consumption of the plurality of servers.

8. A server power consumption control device, characterized in that: The device comprises: A belief determination module, used to analyze the operation status information of each of the multiple servers in the server cluster, and determine the belief distribution corresponding to the preset power consumption influencing factor, wherein the belief distribution is used to characterize the trust level of the preset power consumption influencing factor on the degree of influence of power consumption; A belief fusion module, used to fuse the belief distributions of the multiple servers according to the respective source confidences of the multiple servers to obtain comprehensive status information, wherein the source confidence represents the degree of trustworthiness of the running status information in the servers; an operation selection module, configured to input the comprehensive state information into an action selection model, so as to select an optimized control operation corresponding to the comprehensive state information from a plurality of power consumption control operations using the action selection model; The operation execution module is used to control the server cluster to execute the optimization control operation to adjust the power consumption of the multiple servers.

9. An electronic device, comprising: one or more processors; a memory for storing one or more computer programs, It is characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program or instruction stored thereon, characterized in that: When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images