Method for regulating power consumption of server and server

Abstract

The application discloses a kind of server power consumption adjustment method and server, it is related to server technical field, the method compares the current resource matching ratio of server with target optimal resource ratio, to identify still have surplus resource available component after using optimal resource occupancy ratio, based on the maximum power consumption capacity of margin component, target optimal resource ratio, the current resource matching ratio of server, current total power consumption and power consumption ceiling threshold, not only consider the upper limit of the power consumption of each component, also fully combined with the current running condition and performance demand of server, while maintaining the resource balance between each component as far as possible, avoid the performance of overall performance is affected by the excessive frequency reduction of certain component, so as to solve the problem that the overall resource performance is not considered when power consumption ceiling in the power consumption adjustment process of existing scheme for server.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of servers, and in particular to a server power consumption adjustment method and a server. BACKGROUND

[0002] Limiting the total power consumption is the core requirement for data center energy efficiency optimization and stable operation. If not controlled, instantaneous power consumption peaks can cause power overloads, component frequency reductions, and even system crashes. Moreover, limiting the total power consumption in certain scenarios can avoid energy waste and increase system reliability during long-term operation. Typical application scenarios include: in a multi-node server cluster, the total power consumption is capped to ensure that the total power consumption does not exceed a preset threshold, thereby avoiding system crashes caused by overloads.

[0003] Existing solutions rely on processor platform characteristics, and the main control algorithm is limited to the platform, which cannot be freely adjusted. More importantly, the performance balance of the entire machine when the power consumption is capped is not considered. SUMMARY

[0004] The present application provides a server power consumption adjustment method and a server to at least solve the problem that existing solutions do not consider how to balance the performance of the entire machine when capping the power consumption during the server power consumption adjustment process.

[0005] The present application provides a server power consumption adjustment method, which includes: obtaining the resource allocation ratio of each component of the server, the resource allocation ratio being the resource ratio of each component of the server; determining a target optimal resource ratio according to the system version, the execution software version, and the optimal resource ratio mapping relationship of the server at the current time, the optimal resource ratio mapping relationship being the mapping relationship of the system version, the execution software version, and the optimal resource ratio, and the resource ratio being the optimal resource occupation ratio of each component corresponding to the system version and the execution software version of the server; determining a margin component based on the target optimal resource ratio and the resource allocation ratio, the margin component being a component that still has available resources after power consumption adjustment of the component using the optimal resource occupation ratio; and adjusting the power consumption of each margin component of the server based on the maximum power consumption of the margin component, the target optimal resource ratio, the resource allocation ratio, the current total power consumption of the server, and the power consumption cap threshold of the server.

[0006] The present application also provides a server, which includes: a central processing unit, a baseboard controller, and a programmable logic device, the central processing unit and the baseboard controller being in communication, the central processing unit and the programmable logic device being in communication, and the central processing unit being configured to execute any of the server power consumption adjustment methods.

[0007] By the present application, in order to adapt to different system environments and workloads, the present application introduces "optimal resource ratio mapping relationship", which is a mapping relationship between system version, executed software version and optimal resource ratio, which can intelligently determine the optimal resource occupation ratio of each component according to the system version and the executed software version under different server environments. This means that the power consumption adjustment strategy will no longer be fixed, but can be dynamically adjusted according to the actual running software environment, so as to achieve the best performance and power consumption balance in different scenarios; after determining the system version and the executed software version of the current server, the present application will find and determine the target optimal resource ratio suitable for the current environment from the optimal resource ratio mapping relationship. This step ensures the accuracy of the power consumption adjustment strategy, making it more suitable for the actual running state of the server, thereby improving the adjustment effect; then, the present application compares the current resource matching ratio of the server with the target optimal resource ratio to identify components that still have surplus resources available after using the optimal resource occupation ratio, which are referred to as "margin components". In this way, those components that can be further optimized in the power consumption cap adjustment can be accurately located, without affecting the core business of those components that are carrying out key tasks; finally, based on the maximum power consumption capability of the margin component, the target optimal resource ratio, the current resource matching ratio of the server, the current total power consumption and the power consumption cap threshold, the present application proposes a dynamic power consumption adjustment method. This method not only considers the upper limit of the power consumption of each component, but also fully considers the current running state and performance demand of the server, and limits the power consumption of the margin component one by one by reasonably allocating the power consumption difference, until the predetermined power consumption cap threshold is reached, while maintaining the resource balance between each component as much as possible, avoiding excessive frequency reduction of a certain component to affect the overall performance, thereby solving the problem of how to balance the resource performance of the whole machine when the power consumption cap is not considered in the power consumption adjustment process of the server in the prior art. BRIEF DESCRIPTION OF DRAWINGS

[0008] In order to more clearly illustrate the embodiments of the present application, the drawings needed in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creating laborious work.

[0009] Figure 1 A flowchart of a power consumption adjustment method of a server provided by an embodiment of the present application;

[0010] Figure 2 A flowchart of static adjustment provided by an embodiment of the present application;

[0011] Figure 3 A flowchart of dynamic adjustment provided by an embodiment of the present application;

[0012] Figure 4 A structural block diagram of a power consumption adjustment device of a server is provided in the embodiments of the present application. DETAILED DESCRIPTION

[0013] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work fall within the protection scope of the present application.

[0014] It should be noted that, in the description of the present application, the terms “comprise”, “contain” or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. The terms “first”, “second” and the like in the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence.

[0015] In order for those skilled in the art to better understand the technical solutions of the present application, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0016] In combination with the specific application environment architecture or specific hardware architecture on which the execution of the power consumption adjustment method of the server depends, the specific application environment architecture or specific hardware architecture is described herein.

[0017] The embodiments of the present application provide a power consumption adjustment method of a server, as shown in Figure 1 The method comprises the following steps:

[0018] In step S101, the resource matching ratio of each component of the server is obtained, and the resource matching ratio is the resource ratio of each component of the server.

[0019] The components of the server include a central processing unit, a memory, a network card and a hard disk.

[0020] In step S102, the target optimal resource ratio is determined according to the system version, the execution software version and the optimal resource ratio mapping relationship of the server at the current time, the optimal resource ratio mapping relationship is the mapping relationship of the system version, the execution software version and the optimal resource ratio, and the resource ratio is the optimal resource occupation ratio of each component corresponding to the system version and the execution software version of the server.

[0021] In step S103, based on the target optimal resource ratio and the resource collocation ratio, a margin component is determined, which is a component that still has remaining resources available after the component is power consumption adjusted using the optimal resource occupancy ratio;

[0022] In step S104, based on the maximum power consumption of the margin component, the target optimal resource ratio, the resource collocation ratio, the current total power consumption of the server, and the power consumption cap threshold of the server, the power consumption of each margin component of the server is adjusted.

[0023] In the above steps, in order to adapt to different system environments and workloads, the present application introduces an "optimal resource ratio mapping relationship", which is a mapping relationship between system versions, executed software versions, and optimal resource ratios. The optimal resource ratio mapping relationship can intelligently determine the optimal resource occupancy ratio of each component according to the system version and the executed software version under different server environments. This means that the power consumption adjustment strategy will no longer be fixed, but can be dynamically adjusted according to the actual running software environment, so as to achieve the best performance and power consumption balance in different scenarios. After the system version and the executed software version of the current server are determined, the present application will search for and determine the target optimal resource ratio suitable for the current environment from the optimal resource ratio mapping relationship. This step ensures the accuracy of the power consumption adjustment strategy, making it more suitable for the actual running state of the server, thereby improving the adjustment effect. Then, the present application compares the current resource collocation ratio of the server with the target optimal resource ratio to identify components that still have remaining resources available after using the optimal resource occupancy ratio. These components are referred to as "margin components". In this way, the components that can be further optimized in the power consumption cap adjustment can be accurately located, without affecting the core business of the components that are currently carrying out critical tasks. Finally, based on the maximum power consumption capability of the margin component, the target optimal resource ratio, the current resource collocation ratio of the server, the current total power consumption, and the power consumption cap threshold, the present application proposes a dynamic power consumption adjustment method. This method not only considers the upper limit of the power consumption of each component, but also fully considers the current running state and performance requirements of the server. By reasonably allocating the power consumption difference, the power consumption of each margin component is limited one by one until the predetermined power consumption cap threshold is reached, while maintaining the resource balance between each component as much as possible to avoid excessive frequency reduction of a component that affects the overall performance, thereby solving the problem of how to balance the resources of the entire machine when the power consumption cap is not considered in the power consumption adjustment process of the server in the existing scheme.

[0024] In an embodiment of the present application, the static adjustment part, as shown in Figure 2 adjusts the power consumption of each margin component of the server based on the maximum power consumption of the margin component, the target optimal resource ratio, the resource collocation ratio, the current total power consumption of the server, and the power consumption cap threshold of the server, including:

[0025] In the case where the number of margin components is multiple, the total adjustable power consumption is determined according to the maximum power consumption of the margin components, the target optimal resource ratio, and the resource matching ratio;

[0026] The total power consumption difference is determined as the difference between the power consumption cap threshold and the current total power consumption;

[0027] In the case where the absolute value of the total power consumption difference is less than the total adjustable power consumption, the power consumption of each margin component is adjusted using the absolute value of the total power consumption difference and the number of margin components;

[0028] In the case where the absolute value of the total power consumption difference is greater than or equal to the total adjustable power consumption, the power consumption of each margin component of the server is adjusted using the target optimal resource ratio and the total adjustable power consumption.

[0029] The application also provides a specific use scenario for adjusting the power consumption of each margin component of the server: In a data center, there are a large number of general-purpose servers based on different processor platforms, and these servers run various types of workloads, including but not limited to high-performance computing, machine learning training, online transaction processing, cloud storage services, etc. In the operation of the data center, there are two main goals: one is to ensure the stable operation of the server under high load, avoiding overheating and power overload problems; the other is to improve energy efficiency and reduce operating costs, especially in areas where power supply is tight or costs are high.

[0030] Specific implementation process: With the help of the server system power consumption cap adjustment method of the application, the administrator of the data center can set the power consumption cap threshold of the server, for example, 1000W, to adapt to the power supply capacity and the load of the cooling system of the data center. The administrator can also customize the resource matching ratio according to different business needs, such as in the machine learning training scenario, it may be more inclined to the high performance of CPU (Central Processing Unit, Central Processing Unit, one of the main hardware of computer, equivalent to the brain of computer, it is responsible for executing instructions in computer program, processing data and controlling various operations of computer) and hard disk, while in the online transaction processing scenario, it may focus more on the coordination of CPU and memory. When the server is running, the power consumption cap adjustment mechanism will dynamically adjust the power consumption of each component of the server according to the real-time load and resource usage. Specifically, when the current total power consumption of the server exceeds the set power consumption cap threshold, the system will first determine which components have power consumption margin according to the maximum power consumption and resource matching ratio of CPU, memory, network card, hard disk, etc., and then calculate the total adjustable power consumption.

[0031] Margin component power consumption adjustment example: system current total power consumption = 1200W; power consumption cap threshold = 1000W; total power consumption difference = 200W; margin components = CPU (max power consumption 400W, current power consumption 350W), memory (max power consumption 160W, current power consumption 150W); resource allocation ratio = CPU: memory = 2:1; target optimal resource ratio = CPU: memory = 1.5:1.

[0032] Process analysis: the power consumption margin of the CPU is 50W, and the power consumption margin of the memory is 10W. Since the resource allocation ratio of the CPU and the memory is 2:1, it is necessary to allocate power consumption between the CPU and the memory so that the adjusted ratio is close to the target optimal resource ratio 1.5:1. Based on the current resource allocation ratio, the total adjustable power consumption is 60W (i.e. the sum of 50W of the CPU and 10W of the memory). The difference between the system current total power consumption and the power consumption cap threshold is 200W. Since the total power consumption difference (200W) is greater than the total adjustable power consumption (60W), the system will make a greater adjustment to the power consumption of the CPU and the memory according to the target optimal resource ratio (1.5:1) and the total adjustable power consumption (60W). For example, in order to achieve the target optimal resource ratio, it may be necessary to further reduce the CPU power consumption to 300W and the memory power consumption to 120W to narrow the gap between the power consumption of the two and more evenly distribute the resources. After static adjustment, if the power consumption still exceeds the threshold, the adjustment agent will start the dynamic adjustment mechanism to adjust the power consumption allocation based on the real-time load, for example, if an increase in network traffic is detected, the dynamic adjustment mechanism will increase the power consumption allocation to the network card, and correspondingly reduce the power consumption of the CPU or the memory, to maintain the overall performance and power consumption balance of the system.

[0033] The specific use scenario of adjusting the power consumption of each margin component of the server has the beneficial effects: by dynamically adjusting the power consumption of each component of the server, the application can ensure that the resources of the server are optimally allocated under the constraint of power consumption cap, both meeting the performance requirements of high load scenarios and avoiding waste of energy. The power consumption adjustment strategy of the application is not limited by a specific processor platform and can be applied to servers of different architectures, improving the universality and adaptability of the method; combined with the lightweight time series prediction model and the feature explanation engine, the application can intelligently predict future load trends and adjust power consumption allocation in advance, avoiding system instability or downtime caused by sudden peaks in power consumption; users can customize resource matching ratios according to their own business scenarios and needs, thereby better meeting the performance and power consumption requirements in specific scenarios and improving the flexibility and controllability of server operation; by continuously monitoring and adjusting the power consumption of the server, the application helps to reduce the risk of system overheating and power overload, thereby improving the long-term stability and reliability of the server; under the premise of ensuring system performance and stability, the application helps to effectively manage the total power consumption of the data center, reduces energy consumption, and reduces operating costs, while being environmentally friendly; for enterprises operating data centers, using the method of the application not only improves energy efficiency but also improves customer satisfaction by optimizing resource allocation, enhancing their competitiveness in the market.

[0034] In an embodiment of the present application, the power consumption of each margin component of the server is adjusted using the target optimal resource ratio and the adjustable total power consumption, including: dividing the adjustable total power consumption according to the target optimal resource ratio to obtain a division result; and adjusting the power consumption of each margin component of the server using the division result.

[0035] Specifically, by calculating the target optimal resource ratio, it is ensured that the CPU, memory, network, storage and other key components of the server can work at the optimal performance ratio under the condition of power cap. This method avoids simply over-limiting the power consumption of a certain component, which leads to the problem of overall system performance decline or uneven resource utilization, thereby realizing efficient utilization of resources and optimal balance of system performance; by determining the total adjustable power consumption and dividing it based on the optimal resource ratio, this method allows the server to flexibly adjust power consumption distribution according to the actual load situation, rather than being limited to static power consumption limits. This means that even in scenarios of high load or changing resource demand, the server can dynamically adjust power consumption to adapt to business needs while remaining below the set power cap value; this method helps to reduce the thermal load and power supply pressure of the server by reducing the power consumption peak, thereby improving the reliability and stability of the overall hardware. At the same time, by balancing resources, the risk of overheating or downtime caused by excessive power consumption of a certain component is avoided, ensuring the continuous and stable operation of the server in complex working environments. By precisely controlling power consumption, the energy consumption of the server can be effectively reduced, the total energy consumption of the data center can be reduced, and the operating costs can be reduced, while meeting the trend of green computing and energy saving and emission reduction.

[0036] In an embodiment of the present application, the total adjustable power consumption is determined according to the maximum power consumption of the margin component, the target optimal resource ratio and the resource matching ratio, comprising: determining the target ratio as the ratio of the maximum power consumption of the margin component to the value of the corresponding margin component in the resource matching ratio; determining the target difference as the difference between the value of the corresponding margin component in the resource matching ratio and the value of the corresponding margin component in the target optimal resource ratio; determining the adjustable power consumption of each margin component as the product of the target ratio and the target difference; and determining the total adjustable power consumption as the sum of the adjustable power consumption of each margin component.

[0037] Specifically, by calculating the difference between the maximum power consumption of the margin component and the current resource matching ratio, it can be more detailed to identify which components have the space for power consumption adjustment, thereby avoiding the one-size-fits-all power consumption restriction measures and reducing unnecessary negative impact on system performance; The concept of target optimal resource ratio is introduced, which allows the system to maintain the ideal performance matching between key components while the power consumption is capped. This means that even in a limited power consumption environment, the system can run as efficiently as possible, reducing the impact of power consumption capping on business processing capacity. The process of determining the adjustable power consumption amount is essentially an intelligent allocation process based on resource margin. This method helps to ensure that power consumption adjustment does not excessively affect any particular component, but is evenly distributed to all components that can contribute, thereby improving overall adjustment efficiency when the power consumption is capped. Since this method is not limited to a specific processor platform, it can be widely applied to servers of different brands and models, enhancing its application potential in data centers. This means that server administrators can flexibly adjust the power consumption capping strategy according to different hardware configurations and business needs, without worrying about hardware compatibility issues.

[0038] In an embodiment of the present application, the power consumption of each margin component is adjusted by using the absolute value of the total power consumption difference and the number of margin components, including: determining the target adjustment power consumption as the ratio of the absolute value of the total power consumption difference to the number of margin components; and adjusting the power consumption of each margin component using the target adjustment power consumption.

[0039] Specifically, it can ensure that power consumption adjustment is evenly distributed among all available margin components, avoiding excessive adjustment of the power consumption of a single component, thereby preventing the performance of the component from declining too much or unnecessarily affecting other components, making the power consumption adjustment of the entire system more smooth and balanced; By calculating the power adjustment amount of each margin component, the overall impact on system performance can be minimized while ensuring that the total power consumption does not exceed the capped value. This is because power adjustment is based on the margin of the component rather than simply reducing all components by the same proportion, which helps to maintain the performance level of core components such as CPU and memory, thereby ensuring the execution efficiency of critical tasks; This method simplifies the complexity of power management. There is no need to pre-set the power adjustment level or strategy of each component, but to automatically calculate the adjustment amount based on the real-time power consumption difference and the margin of the component, which enables the system to adaptively adjust the power consumption under different load environments, improving management efficiency. In some cases, if a margin component cannot meet the power adjustment requirement due to its own reasons, the method of equally distributing the target adjustment power consumption to all margin components can automatically compensate by redistributing the excess adjustment amount to other components, thereby ensuring the achievement of the overall power consumption target and enhancing the robustness and reliability of the system.

[0040] In an embodiment of the present application, after adjusting the power consumption of each margin component of the server based on the maximum power consumption of the margin component, the target optimal resource ratio, the resource matching ratio, the current total power consumption of the server, and the power consumption cap threshold of the server, the method further comprises: adjusting the power consumption of each component of the server using a neural network model.

[0041] Specifically, the neural network model can learn and predict future power consumption and load trends based on historical data, which means it can predict the power consumption pressure the server will face to some extent, so as to adjust the power consumption allocation in advance and avoid instantaneous power consumption exceeding the standard. The neural network model can analyze dynamic indicators such as CPU power consumption fluctuation rate, memory utilization, I / O (input / output) latency, network packet loss rate, disk queue depth, and system total power consumption change gradient in real time, dynamically adjust the power consumption weight of different components according to these indicators, and ensure that the resource allocation of each component is more reasonable and the performance efficiency ratio is better under the condition of meeting the power consumption cap. Since the model can continuously learn and adapt to new working condition characteristics, even if the load and use scenario of the server changes, the power consumption strategy can be quickly adjusted to maintain stable operation and high performance of the system.

[0042] In an embodiment of the present application, the dynamic adjustment part adjusts the power consumption of each component of the server using a neural network model, as shown in Figure 3 The method comprises the following steps:

[0043] Obtain multi-dimensional working condition information in a preset time period, and the multi-dimensional working condition information at least includes CPU power consumption fluctuation rate, memory utilization, I / O latency, and network packet loss rate;

[0044] Process the multi-dimensional working condition information using a neural network model to obtain a dynamic correction coefficient matrix, and the dynamic correction coefficient matrix represents the influence weight of each parameter of the multi-dimensional working condition information on each component;

[0045] Adjust the power consumption of each component based on the dynamic correction coefficient matrix.

[0046] Specifically, the neural network model can dynamically adjust the power consumption limits of each component based on real-time multi-dimensional working condition information such as central processor power consumption fluctuation rate, memory utilization, input / output latency, network packet loss rate, etc. This means that the system can respond to load changes in real time, avoiding the low efficiency problem of static power consumption cap strategy in dynamic load scenarios. Through the trained neural network model, the load trend in the future period of time can be predicted, and the power consumption limit can be adjusted in advance to avoid the system performance decline or instability caused by sudden power consumption exceeding. For example, the long-short term neural network model is particularly good at processing time series data and can learn the long-term dependence of load changes to provide more accurate predictions. The dynamic correction coefficient matrix can reflect the weight of each parameter on the power consumption of different components, which means that when the power consumption is capped, it can be more intelligent to decide which components should reduce power consumption and which should remain unchanged, so as to limit the total power consumption while maintaining the efficient operation of key tasks, achieving the optimal balance between power consumption and performance. Based on specific business scenarios and system configurations, the neural network model can learn and optimize specific power consumption adjustment strategies, allowing users to adjust model parameters according to their own needs to achieve personalized power consumption management and meet the energy efficiency and performance requirements in specific business scenarios. The dynamic correction coefficient matrix can be fused with the basic resource ratio generated by the static topology channel.

[0047] Power fluctuation rate refers to the degree of change in the power consumption of a server or its components over a certain period of time as the workload changes. It is commonly used as an indicator of system power consumption stability, reflecting the dynamic characteristics of system power consumption over time. High power fluctuation rate means that the system power consumption fluctuates greatly in a short period of time, which may be caused by sudden changes in workload or unstable resource allocation within the system.

[0048] In an embodiment of the present application, the method further comprises: in the case where the number of margin components is 1, determining the adjustable power consumption as the adjustable power consumption; determining the total power consumption difference as the difference between the power consumption cap threshold of the server and the current total power consumption of the server; in the case where the absolute value of the total power consumption difference is less than the adjustable power consumption, adjusting the power consumption of the margin component using the total power consumption difference; in the case where the absolute value of the total power consumption difference is greater than or equal to the adjustable power consumption, adjusting the power consumption of the margin component using the total adjustable power consumption.

[0049] Specifically, when the absolute value of the total power difference is less than the adjustable power consumption, using the more accurate total power difference to directly adjust the power consumption of the margin component can ensure that the system power consumption quickly and accurately approaches or reaches the power cap threshold, avoiding performance loss or power consumption rebound caused by excessive adjustment. If the absolute value of the total power difference is greater than or equal to the total adjustable power consumption, the total adjustable power consumption is used for adjustment. This method avoids the "one-size-fits-all" of power consumption adjustment, preventing the power consumption of the margin component from being adjusted to a too low level at once, thereby affecting the stability and user experience of the system. It ensures that the power consumption adjustment is both effective and does not exceed the necessary range. This strategy can dynamically adjust according to the actual power consumption of the server, whether the system power consumption suddenly increases or gradually increases, and can be adjusted continuously through fine-tuning rather than drastic changes, making the implementation of power capping smoother and reducing unnecessary performance impact. In the case where only one margin component needs to be adjusted, this method simplifies the adjustment logic of power capping, avoids complex multi-component resource allocation calculations, reduces implementation difficulty and cost, and also improves the response speed and efficiency of the system.

[0050] In an embodiment of the present application, the method further comprises: during the process of adjusting the power consumption of the margin component using the total power difference, or during the process of adjusting the power consumption of the margin component using the total adjustable power consumption, determining a power consumption comparison result, which is a comparison result of the absolute value of the total power difference and the preset power consumption, or a comparison result of the total adjustable power consumption and the preset power consumption; and determining that the power consumption of the margin component does not need to be adjusted in the case where the comparison result represents that the absolute value of the total power difference is less than the preset power consumption, or that the total adjustable power consumption is less than the preset power consumption.

[0051] Specifically, by accurately calculating the total power difference or the total adjustable power consumption, this method can more finely control the power consumption of the server, avoid performance loss caused by excessive adjustment, and at the same time ensure that the power consumption does not exceed the preset threshold, achieving the best balance between power consumption and performance; when the absolute value of the total power difference is less than the preset power consumption, or the total adjustable power consumption is less than the preset power consumption, it indicates that the system power consumption has approached or is below the cap value, and at this time there is no need to further adjust the power consumption of the margin component, which can avoid unnecessary power consumption adjustment and reduce the impact on system stability and performance. The power consumption comparison result serves as the basis for decision whether to adjust, which can quickly judge the system power consumption state, reduce the invalid cycle in the adjustment period, and improve the response speed and efficiency of power consumption adjustment. Users can customize the preset power consumption value, flexibly adjust the power capping strategy according to their own business needs and the power consumption characteristics of the server. This not only increases the user's control over the power consumption management of the server, but also makes the power capping more adaptable to different business scenarios.

[0052] In an embodiment of the present application, the method further comprises: recording the power consumption adjustment value of each component and the corresponding adjustment time in a log.

[0053] Specifically, the log can provide detailed operation history, which helps to troubleshoot when the system is abnormal. By checking the log, engineers can understand which components have their power consumption adjusted at a certain time point, and the power consumption change before and after the adjustment, so as to quickly locate the problem source. The power consumption adjustment value and time recorded in the log can help analyze the performance change trend of the system. For example, if the CPU is adjusted for power consumption during high load, but the response speed of the system does not decrease significantly, it may indicate that the power consumption adjustment strategy effectively balances performance and power consumption, or it means that other resources have become the bottleneck. Through analysis of the log, it can be identified which components have high power consumption under certain workloads, and in fact, they do not always need such high performance. Based on this information, resource allocation can be optimized, such as reducing the power consumption of the CPU in a light load environment, and appropriately increasing it in a heavy load, to ensure reasonable use of resources.

[0054] In an embodiment of the present application, the method further comprises: acquiring the temperature of each component in real time; and reducing the power consumption of the corresponding component when the temperature of the component exceeds a temperature threshold.

[0055] Specifically, through real-time temperature monitoring, the system can dynamically adjust the power consumption to cope with temperature changes, avoiding system instability or hardware damage caused by overheating; the temperature threshold can be flexibly set according to hardware characteristics and environmental conditions, ensuring effective temperature control under various working conditions; reducing the damage of high temperature to hardware helps to prolong the service life of key components such as CPU, memory, hard disk, etc. in the server; when the temperature approaches the threshold, the system automatically reduces the power consumption, reducing the need for cooling, thereby reducing the energy consumption of the cooling system and improving the overall energy efficiency; in a non-high load scenario, power consumption can be adjusted through temperature feedback to avoid unnecessary energy waste.

[0056] After the preliminary static adjustment of weight allocation and power consumption, a closed-loop dynamic weight optimization mechanism is constructed through a lightweight time series prediction model (i.e. long short-term neural network model) and a feature explanation engine embedded in the adjustment agent program: input layer: real-time acquisition of multi-dimensional working condition feature vector (sampling period 1s) includes dynamic indicators such as CPU power fluctuation rate, memory utilization, I / O waiting time, network packet loss rate, disk queue depth, and system total power consumption change gradient.

[0057] Processing layer: dual-channel feature fusion architecture is adopted; static topology channel: analyze hardware configuration to generate basic resource proportion; dynamic working condition channel: predict future 5s load trend through long short-term neural network model, output dynamic correction coefficient matrix; decision layer: execute dynamic rebalancing of resource weight:

[0058] Wfinal=Wstatic⊙(1+α⋅Mlstm);

[0059] Wherein, Wfinal is the final weight, Wstatic is the initial weight, Mlstm is the neural network model weight, and a is the learning rate decay factor (initial value 0.7), realizing the weighted fusion of static proportion and dynamic prediction.

[0060] Feature explanation engine: integrated SHAP value (SHapley Additive exPlanations, a tool for explaining the output of machine learning models) analysis module, visualizing the weight decision basis; dynamically generating adjustment strategy report (example): "the current network packet loss rate rises by 22%, and the dynamic channel suggests that the network card weight is increased by 30%; the disk queue depth is lower than the threshold, and it is suggested to release 5% power allowance".

[0061] The BMC (Baseboard Management Controller) and CPLD (Complex Programmable Logic Device) of the server are used to realize system power cap related data collection, processing (threshold judgment, information alarm and record), calibration, etc. The BMC communicates with the power adjustment agent program under the operating system through the KCS channel (Keyboard Controller Style, commonly used for direct access of software to the BMC during operating system running to perform system management tasks), and transmits power cap related information. The power adjustment agent program executes the power adjustment strategy for balancing performance. The strategy first performs statistics on the CPU, memory, network card, hard disk, etc. of the current system, calculates the optimal proportion of resource collocation of each component, and then performs power adjustment on each component according to the weight distribution of power adjustment based on the above, so as to achieve the goal of power cap.

[0062] Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiments can be realized by means of software and the necessary general hardware platform, of course, it can also be realized by hardware, but in many cases the former is a better embodiment.

[0063] The embodiments of the present application also provide a server power adjustment device, as shown in Figure 4 The device comprises:

[0064] The acquisition unit 41 is configured to acquire the resource collocation proportion of each component of the server, and the resource collocation proportion is the resource proportion of each component of the server.

[0065] The first processing unit 42 is configured to determine a target optimal resource ratio according to a system version, an execution software version and an optimal resource ratio mapping relationship of the server at a current time, the optimal resource ratio mapping relationship being a mapping relationship among the system version, the execution software version and the optimal resource ratio, and the resource ratio being an optimal resource occupation ratio of each component corresponding to the system version and the execution software version of the server;

[0066] The second processing unit 43 is configured to determine a margin component based on the target optimal resource ratio and the resource collocation ratio, the margin component being a component that still has a remaining resource available after power consumption adjustment of the component using the optimal resource occupation ratio;

[0067] The third processing unit 44 is configured to adjust power consumption of each margin component of the server based on a maximum power consumption of the margin component, the target optimal resource ratio, the resource collocation ratio, a current total power consumption of the server and a power consumption cap threshold of the server.

[0068] In an embodiment of the present application, the third processing unit comprises: a first determining module configured to determine an adjustable power consumption total amount according to the maximum power consumption of the margin component, the target optimal resource ratio and the resource collocation ratio when the number of the margin components is multiple; a second determining module configured to determine a total power consumption difference value as a difference between the power consumption cap threshold and the current total power consumption; a first processing module configured to adjust the power consumption of each margin component by using an absolute value of the total power consumption difference value and the number of the margin components when the absolute value of the total power consumption difference value is less than the adjustable power consumption total amount; and a second processing module configured to adjust the power consumption of each margin component of the server by using the target optimal resource ratio and the adjustable power consumption total amount when the absolute value of the total power consumption difference value is greater than or equal to the adjustable power consumption total amount.

[0069] By dynamically adjusting the power consumption of each component of the server, the application can ensure that the resources of the server are optimally allocated under the constraint of the power consumption cap, meeting the performance requirements of high-load scenarios and avoiding waste of energy. The power consumption adjustment strategy of the application is not limited to a specific processor platform and can be applied to servers of different architectures, improving the universality and adaptability of the method. By combining a lightweight time series prediction model and a feature explanation engine, the application can intelligently predict future load trends and adjust power consumption allocation in advance, avoiding system instability or downtime caused by sudden power consumption peaks. Users can customize resource matching ratios according to their business scenarios and needs, better meeting performance and power consumption requirements in specific scenarios and improving the flexibility and controllability of server operation. By continuously monitoring and adjusting server power consumption, the application helps reduce the risk of system overheating and power overload, improving the long-term stability and reliability of the server. Under the premise of ensuring system performance and stability, the application helps data centers effectively manage total power consumption, reducing energy consumption and operating costs while being environmentally friendly. For enterprises operating data centers, using the method of the application not only improves energy efficiency but also enhances customer satisfaction and market competitiveness by optimizing resource allocation.

[0070] In an embodiment of the application, the second processing module includes: a first processing submodule for dividing the total adjustable power consumption according to the target optimal resource ratio to obtain a division result; and a second processing submodule for adjusting the power consumption of each component of the server using the division result.

[0071] Specifically, by calculating the target optimal resource ratio, it is ensured that the CPU, memory, network, storage and other key components of the server can work at the best performance ratio under the condition of power consumption cap. This method avoids simply over-limiting the power consumption of a certain component, which leads to a decline in overall system performance or uneven resource utilization, thereby achieving efficient resource utilization and optimal balance of system performance; by determining the total adjustable power consumption and dividing it based on the optimal resource ratio, this method allows the server to flexibly adjust power consumption distribution according to actual load conditions, rather than being limited to static power consumption limits. This means that even in scenarios of high load or changing resource demand, the server can dynamically adjust power consumption to adapt to business needs while remaining below the set power consumption cap value; this method helps reduce the thermal load and power stress of the server by reducing power consumption peaks, thereby improving the reliability and stability of the overall hardware. At the same time, by balancing resources, the risk of overheating or downtime caused by excessive power consumption of a certain component is avoided, ensuring the continuous and stable operation of the server in complex working environments. By precisely controlling power consumption, the energy consumption of the server can be effectively reduced, the total energy consumption of the data center can be reduced, and the operating costs can be reduced, while meeting the trend of green computing and energy saving and emission reduction.

[0072] In an embodiment of the present application, the first determining module comprises: a first determining submodule for determining a target ratio value as a ratio of the maximum power consumption of the margin component to the value of the corresponding margin component in the resource matching ratio; a second determining submodule for determining a target difference value as a difference between the value of the corresponding margin component in the resource matching ratio and the value of the corresponding margin component in the target optimal resource ratio; a third determining submodule for determining an adjustable power consumption of each margin component as a product of the target ratio value and the target difference value; and a fourth determining submodule for determining a total adjustable power consumption as a sum of the adjustable power consumptions of the margin components.

[0073] Specifically, by calculating the difference between the maximum power consumption of the margin component and the current resource matching ratio, it can be more detailed to identify which components have the space for power consumption adjustment, thereby avoiding the one-size-fits-all power consumption limiting measures and reducing the unnecessary negative impact on system performance; the concept of target optimal resource ratio is introduced, which allows the system to maintain the ideal performance matching between key components while the power consumption is capped. This means that even in a limited power consumption environment, the system can run as efficiently as possible, reducing the impact of power consumption capping on business processing capacity. The process of determining the total adjustable power consumption is essentially an intelligent allocation process based on resource margin. This method helps to ensure that power consumption adjustment does not excessively affect any particular component, but is evenly distributed to all possible contributing components, thereby improving the overall adjustment efficiency when the power consumption is capped. Since this method is not limited to a specific processor platform, it can be widely applicable to servers of different brands and models, enhancing its application potential in data centers. This means that server administrators can flexibly adjust the power consumption capping strategy according to different hardware configurations and business needs, without worrying about hardware compatibility issues.

[0074] In an embodiment of the present application, the first processing module comprises: a third processing submodule for determining a target adjustment power consumption as a ratio of the absolute value of the total power consumption difference to the number of margin components; and a fourth processing submodule for adjusting the power consumption of each margin component using the target adjustment power consumption.

[0075] Specifically, it can ensure that the power consumption adjustment is evenly distributed among all available margin components, avoiding excessive adjustment of the power consumption of a single component, thereby preventing the performance of the component from declining too much or having an unnecessary impact on other components, making the power consumption adjustment of the entire system more smooth and balanced; by calculating the power consumption adjustment amount of each margin component, the overall impact on system performance can be minimized while ensuring that the overall power consumption does not exceed the cap value. This is because the power consumption adjustment is based on the margin of the component rather than simply reducing all components by the same proportion, which helps to maintain the performance level of core components such as CPU and memory, thereby ensuring the execution efficiency of critical tasks; this method simplifies the complexity of power management. There is no need to pre-set the power consumption adjustment level or strategy of each component, but to automatically calculate the adjustment amount based on the real-time power consumption difference and the margin of the component, which enables the system to adaptively adjust the power consumption in different load environments, improving management efficiency. In some cases, if a margin component cannot meet the power consumption adjustment requirement due to its own reasons, the method of equally distributing the target adjustment power consumption to all margin components can automatically compensate by redistributing the excess adjustment amount to other components, thereby ensuring the achievement of the overall power consumption target, enhancing the robustness and reliability of the system.

[0076] In an embodiment of the present application, the device further comprises a fourth processing unit for adjusting the power consumption of each component of the server using a neural network model after adjusting the power consumption of each margin component of the server based on the maximum power consumption of the margin component, the target optimal resource ratio, the resource matching ratio, the current total power consumption of the server, and the power consumption cap threshold of the server.

[0077] Specifically, the neural network model can learn and predict future power consumption and load trends based on historical data, which means it can predict to some extent the power consumption pressure the server will face, thereby adjusting the power consumption allocation in advance to avoid instantaneous power consumption exceeding the standard. The neural network model can analyze dynamic indicators such as CPU power consumption fluctuation rate, memory utilization, I / O (input / output) latency, network packet loss rate, disk queue depth, and system total power consumption change gradient in real time, dynamically adjust the power consumption weight of different components according to these indicators, and ensure that the resource allocation of each component is more reasonable and achieves a better performance efficiency ratio under the condition of meeting the power consumption cap. Since the model can continuously learn and adapt to new working condition characteristics, even if the load and usage scenario of the server changes, it can quickly adjust the power consumption strategy to maintain stable operation and high performance of the system.

[0078] In an embodiment of the present application, the fourth processing unit comprises: a third processing module configured to acquire multi-dimensional working condition information in a preset time period, the multi-dimensional working condition information comprising at least a central processor power fluctuation rate, a memory utilization rate, an input / output waiting time, and a network packet loss rate; a fourth processing module configured to process the multi-dimensional working condition information by using a neural network model to obtain a dynamic correction coefficient matrix, the dynamic correction coefficient matrix representing the influence weight of each parameter of the multi-dimensional working condition information on each component; and a fifth processing module configured to adjust the power consumption of each component based on the dynamic correction coefficient matrix.

[0079] Specifically, the neural network model can dynamically adjust the power consumption limit of each component based on real-time multi-dimensional working condition information (such as central processor power fluctuation rate, memory utilization rate, input / output waiting time, network packet loss rate, etc.), which means that the system can respond to load changes in real time and avoid the low efficiency problem of static power consumption capping strategy in dynamic load scenarios. Through the trained neural network model, the load trend in the future period of time can be predicted, and the power consumption limit can be adjusted in advance to avoid the system performance decline or instability caused by sudden power consumption exceeding. For example, the LSTM (Long Short-Term Memory) model is particularly good at processing time series data and can learn the long-term dependence of load changes to provide more accurate predictions. The dynamic correction coefficient matrix can reflect the weight of the influence of each parameter on the power consumption of different components, which means that when the power consumption is capped, it can be more intelligent to decide which components should reduce power consumption and which should remain unchanged, so as to limit the total power consumption while maintaining the efficient operation of key tasks, achieving the optimal balance between power consumption and performance. Based on specific business scenarios and system configurations, the neural network model can learn and optimize specific power consumption adjustment strategies, allowing users to adjust model parameters according to their own needs to achieve personalized power consumption management and meet the energy efficiency and performance requirements in specific business scenarios, wherein the dynamic correction coefficient matrix can be fused with the basic resource ratio generated by the static topology channel.

[0080] In an embodiment of the present application, the device further comprises: a first determination unit configured to determine an adjustable power consumption of the adjustable power consumption amount when the number of margin components is one; a second determination unit configured to determine a total power consumption difference value as the difference between the power consumption capping threshold of the server and the current total power consumption of the server; a fifth processing unit configured to adjust the power consumption of the margin component by using the total power consumption difference value when the absolute value of the total power consumption difference value is less than the adjustable power consumption; and a sixth processing unit configured to adjust the power consumption of the margin component by using the total adjustable power consumption amount when the absolute value of the total power consumption difference value is greater than or equal to the adjustable power consumption.

[0081] Specifically, when the absolute value of the total power difference is less than the adjustable power consumption, using the more accurate total power difference to directly adjust the power consumption of the margin component can ensure that the system power consumption quickly and accurately approaches or reaches the power cap threshold, avoiding performance loss or power consumption rebound caused by excessive adjustment. If the absolute value of the total power difference is greater than or equal to the total adjustable power consumption, the total adjustable power consumption is used for adjustment. This method avoids the "one-size-fits-all" of power consumption adjustment, preventing the power consumption of the margin component from being adjusted to a too low level at once, thereby affecting the stability and user experience of the system. It ensures that the power consumption adjustment is both effective and does not exceed the necessary range. This strategy can dynamically adjust according to the actual power consumption of the server, whether the system power consumption suddenly increases or gradually increases, and can be adjusted continuously through fine-tuning rather than drastic changes, making the implementation of power capping smoother and reducing unnecessary performance impact. In the case where only one margin component needs to be adjusted, this method simplifies the adjustment logic of power capping, avoids complex multi-component resource allocation calculations, reduces implementation difficulty and cost, and also improves the response speed and efficiency of the system.

[0082] In an embodiment of the present application, the device further comprises: a seventh processing unit for determining a power consumption comparison result in the process of adjusting the power consumption of the margin component using the total power difference, or in the process of adjusting the power consumption of the margin component using the total adjustable power consumption, the power consumption comparison result being a comparison result of the absolute value of the total power difference and the preset power consumption, or a comparison result of the total adjustable power consumption and the preset power consumption; and an eighth processing unit for determining that the power consumption of the margin component does not need to be adjusted in the case where the comparison result represents that the absolute value of the total power difference is less than the preset power consumption, or that the total adjustable power consumption is less than the preset power consumption.

[0083] Specifically, by accurately calculating the total power difference or the total adjustable power consumption, this method can more finely control the power consumption of the server, avoid performance loss caused by excessive adjustment, and at the same time ensure that the power consumption does not exceed the preset threshold, achieving the best balance between power consumption and performance; when the absolute value of the total power difference is less than the preset power consumption, or the total adjustable power consumption is less than the preset power consumption, it indicates that the system power consumption has approached or is lower than the cap value, at which point there is no need to further adjust the power consumption of the margin component, which can avoid unnecessary power consumption adjustment and reduce the impact on system stability and performance. The power consumption comparison result serves as the basis for deciding whether to adjust, which can quickly judge the system power consumption state, reduce invalid cycles in the adjustment period, and improve the response speed and efficiency of power consumption adjustment. Users can customize the preset power consumption value, flexibly adjust the power capping strategy according to their own business needs and the power consumption characteristics of the server. This not only increases the user's control over the server power consumption management, but also makes the power capping more adaptable to different business scenarios.

[0084] In an embodiment of the present application, the device further comprises a ninth processing unit configured to record the power consumption adjustment value of each component and the corresponding adjustment time in a log.

[0085] Specifically, the log can provide detailed operation history, which helps to troubleshoot when the system is abnormal. By checking the log, engineers can understand which components have their power consumption adjusted at a certain time point, as well as the power consumption change before and after the adjustment, so as to quickly locate the problem source. The recorded power consumption adjustment value and time in the log can help analyze the performance change trend of the system. For example, if the CPU is adjusted for power consumption during high load, but the response speed of the system does not decrease significantly, it may indicate that the power consumption adjustment strategy effectively balances performance and power consumption, or it means that other resources have become the bottleneck. Through analysis of the log, it can be identified which components have high power consumption under a certain workload, and in fact, such high performance is not always needed. Based on this information, resource allocation can be optimized, such as reducing the power consumption of the CPU in a light load environment, and appropriately increasing it in a heavy load, to ensure reasonable use of resources.

[0086] In an embodiment of the present application, the device further comprises a tenth processing unit configured to obtain the temperature of each component in real time; and reduce the power consumption of the corresponding component when the temperature of the component exceeds a temperature threshold.

[0087] Specifically, through real-time temperature monitoring, the system can dynamically adjust the power consumption to cope with temperature changes, avoiding system instability or hardware damage caused by overheating; the temperature threshold can be flexibly set according to hardware characteristics and environmental conditions, ensuring effective temperature control under various working conditions; reducing the damage of high temperature to hardware helps to prolong the service life of key components such as CPU, memory, hard disk, etc. in the server; when the temperature approaches the threshold, the system automatically reduces the power consumption, reducing the need for cooling, thereby reducing the energy consumption of the cooling system and improving the overall energy efficiency; in a non-high load scenario, the power consumption can be adjusted through temperature feedback to avoid unnecessary energy waste.

[0088] The description of the features in the embodiments of the power consumption adjustment device of the server can be referred to the related description of the embodiments of the power consumption adjustment method of the server, which will not be repeated here.

[0089] The present application also provides a server comprising: a central processing unit, a baseboard controller and a programmable logic device, the central processing unit and the baseboard controller communicate with each other, the central processing unit and the programmable logic device communicate with each other, the central processing unit is configured to execute any one of the power consumption adjustment methods of the server.

[0090] The embodiment of the present application further provides a computer readable storage medium, which stores a computer program, and the computer program is configured to execute the steps in the power consumption adjustment method of the server.

[0091] In an example embodiment, the computer readable storage medium can include, but is not limited to, a U disk, a read-only memory (ROM), a random access memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and various media that can store computer programs.

[0092] The embodiment of the present application further provides a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the steps in the power consumption adjustment method of the server.

[0093] The embodiment of the present application further provides another computer program product, which comprises a non-volatile computer readable storage medium, and the non-volatile computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps in the power consumption adjustment method of the server.

[0094] The skilled in the art can further realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized in electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been described in general terms in the above description. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0095] The above describes in detail the power consumption adjustment method of the server and the server provided by the present application. The principles and implementation manners of the present application are described by applying specific examples in this paper, and the above description of the examples is only applicable to help understand the method and its core idea of the present application. It should be pointed out that for the ordinary skilled in the art, without departing from the principles of the present application, some improvements and modifications can be made to the present application, and these improvements and modifications also fall within the protection scope of the claims of the present application.

Claims

1. A method for regulating the power consumption of a server, characterized in that, include: Obtain the resource allocation ratio of each component of the server, wherein the resource allocation ratio is the resource ratio of each component of the server; Based on the system version, execution software version, and optimal resource ratio mapping relationship of the server at the current moment, the target optimal resource ratio is determined. The optimal resource ratio mapping relationship is the mapping relationship between the system version, execution software version, and optimal resource ratio. The resource ratio is the optimal resource occupancy ratio of each component corresponding to the system version and execution software version of the server. Based on the target optimal resource ratio and the resource matching ratio, a margin component is determined. The margin component is the component that still has remaining resources available after power consumption adjustment of the component using the optimal resource occupancy ratio. Based on the maximum power consumption of the margin components, the target optimal resource ratio, the resource allocation ratio, the current total power consumption of the server, and the power consumption capping threshold of the server, the power consumption of each of the margin components of the server is adjusted. Based on the maximum power consumption of the margin components, the target optimal resource ratio, the resource allocation ratio, the current total power consumption of the server, and the server's power consumption capping threshold, the power consumption of each of the margin components of the server is adjusted, including: when there are multiple margin components, determining the total adjustable power consumption based on the maximum power consumption of the margin components, the target optimal resource ratio, and the resource allocation ratio; determining the total power consumption difference as the difference between the power consumption capping threshold and the current total power consumption; when the absolute value of the total power consumption difference is less than the total adjustable power consumption, adjusting the power consumption of each margin component using the absolute value of the total power consumption difference and the number of margin components; when the absolute value of the total power consumption difference is greater than or equal to the total adjustable power consumption, adjusting the power consumption of each of the margin components of the server using the target optimal resource ratio and the total adjustable power consumption. Adjusting the power consumption of each of the margin components using the absolute value of the total power consumption difference and the number of margin components includes: determining the target adjustment power consumption as the ratio of the absolute value of the total power consumption difference to the number of margin components; and adjusting the power consumption of each of the margin components using the target adjustment power consumption.

2. The power consumption adjustment method for a server according to claim 1, characterized in that, Using the target optimal resource ratio and the total adjustable power consumption, the power consumption of each of the margin components of the server is adjusted, including: The total adjustable power consumption is divided according to the target optimal resource ratio to obtain the division result; The power consumption of each of the margin components of the server is adjusted using the division results.

3. The power consumption adjustment method for a server according to claim 1, characterized in that, The total adjustable power consumption is determined based on the maximum power consumption of the margin component, the target optimal resource ratio, and the resource allocation ratio, including: The target ratio is determined to be the ratio of the maximum power consumption of the margin component to the value of the margin component corresponding to the resource allocation ratio; The target difference is determined to be the difference between the value of the margin component corresponding to the resource allocation ratio and the value of the margin component corresponding to the target optimal resource ratio; The adjustable power consumption of each of the margin components is determined to be the product of the target ratio and the target difference; The total adjustable power consumption is determined to be the sum of the adjustable power consumption of each of the margin components.

4. The power consumption adjustment method for a server according to claim 1, characterized in that, After adjusting the power consumption of each of the margin components of the server based on the maximum power consumption of the margin components, the target optimal resource ratio, the resource allocation ratio, the current total power consumption of the server, and the power consumption capping threshold of the server, the method further includes: A neural network model is used to adjust the power consumption of each component of the server.

5. The power consumption adjustment method for a server according to claim 4, characterized in that, The power consumption of each component of the server is adjusted using a neural network model, including: Acquire multi-dimensional operating condition information within a preset time period, wherein the multi-dimensional operating condition information includes at least the central processing unit power consumption fluctuation rate, memory utilization rate, input / output latency, and network packet loss rate. The neural network model is used to process the multidimensional operating condition information to obtain a dynamic correction coefficient matrix, which represents the influence weight of each parameter of the multidimensional operating condition information on each component. The power consumption of each component is adjusted based on the dynamic correction coefficient matrix.

6. The power consumption adjustment method for a server according to claim 1, characterized in that, The method further includes: When the number of the margin component is 1, the adjustable power consumption is determined. The total power consumption difference is determined to be the difference between the power consumption capping threshold of the server and the current total power consumption of the server; If the absolute value of the total power consumption difference is less than the adjustable power consumption, the power consumption of the margin component is adjusted using the total power consumption difference. If the absolute value of the total power consumption difference is greater than or equal to the adjustable power consumption, the power consumption of the margin component is adjusted using the total adjustable power consumption.

7. The power consumption adjustment method for a server according to claim 6, characterized in that, The method further includes: During the process of adjusting the power consumption of the margin component using the total power consumption difference, or during the process of adjusting the power consumption of the margin component using the total adjustable power consumption, a power consumption comparison result is determined. The power consumption comparison result is either a comparison between the absolute value of the total power consumption difference and a preset power consumption, or a comparison between the total adjustable power consumption and a preset power consumption. If the comparison result indicates that the absolute value of the total power consumption difference is less than the preset power consumption, or indicates that the total adjustable power consumption is less than the preset power consumption, it is determined that there is no need to adjust the power consumption of the margin component.

8. A server, characterized in that, include: The system includes a central processing unit (CPU), a baseboard controller, and a programmable logic device (PLD), wherein the CPU communicates with the baseboard controller and with the PLD, and the CPU is configured to execute the power consumption regulation method of the server according to any one of claims 1 to 7.

Images