Energy efficiency ratio tuning method, device, equipment and storage medium for server cluster

Abstract

The embodiments of the present application provide a method, device, equipment and storage medium for energy efficiency ratio tuning of a server cluster, which belongs to the field of computer processing technology. The method includes: based on the existing energy efficiency ratio information and the corresponding control parameter set, constructing a distribution model of the distribution relationship between the control parameters of multiple processors and the energy efficiency ratio of the server cluster, and constructing an acquisition function for evaluating the quality of the control parameter set based on the distribution model; under the premise of optimizing the acquisition function, sampling other control parameter sets other than the existing energy efficiency ratio information based on the distribution model, and determining the new energy efficiency ratio information of the server cluster when working under the sampled control parameter set; determining the target energy efficiency ratio information that meets the preset energy efficiency ratio requirements from the multiple existing energy efficiency ratio information after the update, and controlling the server cluster to work under the target control parameter set. The present application can improve the energy efficiency ratio of the server cluster and reduce energy consumption.

Description

Technical Field

[0001] The present application relates to the field of computer processing technology, and in particular to a method, device, equipment and storage medium for optimizing the energy efficiency ratio of a server cluster. Background Art

[0002] A server cluster refers to a computing environment that connects multiple servers together to provide high availability, scalability, and load balancing. For example, data centers and intelligent computing clusters are both server clusters. A server cluster is usually composed of multiple processor groups, and each processor group includes multiple processors. There are many types of processors, such as central processing units (CPUs), graphics processing units (GPUs), neural network processors (NPUs), etc. Energy efficiency ratio is an important indicator for measuring the energy consumption of server clusters. The larger the energy efficiency ratio, the more computing work the server cluster can complete under the same energy consumption.

[0003] In the related art, the processor with the highest energy consumption is selected from multiple processors and its parameters are adjusted. However, the energy efficiency ratio of a server cluster depends on the balance between different types of processors. The one-sided parameter adjustment method adopted in the related art cannot improve its energy efficiency ratio well, which leads to excessive energy consumption of the server cluster. Summary of the invention

[0004] The main purpose of the embodiments of the present application is to propose a method, device, equipment and storage medium for optimizing the energy efficiency ratio of a server cluster, aiming to improve the energy efficiency ratio of the server cluster and reduce energy consumption.

[0005] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application proposes a method for optimizing the energy efficiency ratio of a server cluster, which is applied to the server cluster. The method includes:

[0006] Acquire multiple different control parameter sets of a processor group and energy efficiency ratio information of a server cluster when operating under any control parameter set, wherein the processor group includes multiple different types of processors, and each control parameter set includes control parameters of each processor in the processor group;

[0007] Based on the existing energy efficiency ratio information and the corresponding control parameter set, a distribution model of the distribution relationship between the control parameters of multiple processors and the energy efficiency ratio of the server cluster is constructed, and a collection function for evaluating the quality of the control parameter set is constructed based on the distribution model;

[0008] Under the premise of optimizing the acquisition function, other control parameter sets other than the existing energy efficiency ratio information are sampled based on the distribution model to determine the new energy efficiency ratio information of the server cluster when it works under the sampled control parameter set;

[0009] The target energy efficiency ratio information satisfying the preset energy efficiency ratio requirement is determined from the updated multiple existing energy efficiency ratio information, and the control parameter set corresponding to the target energy efficiency ratio information is used as the target control parameter set, and the server cluster is controlled to work under the target control parameter set.

[0010] In some embodiments, the processor includes a central processing unit and a graphics processing unit, and the energy efficiency ratio information includes first energy efficiency ratio information;

[0011] Obtain multiple sets of different control parameter sets of the processor group, and energy efficiency ratio information of the server cluster when working under any control parameter set, including:

[0012] Selecting a first control parameter from a plurality of preset first control parameters of a central processing unit, and selecting a second control parameter from a plurality of preset second control parameters of at least one graphics processing unit, wherein the first control parameter and the second control parameter constitute a set of control parameters;

[0013] Under the condition that the central processing unit enters the running state based on the first control parameter and the graphics processing unit enters the running state based on the second control parameter, obtaining a first computing performance value and a first service power consumption value of the server cluster;

[0014] First energy efficiency ratio information is obtained according to the ratio of the first computing performance value to the first service power consumption value.

[0015] In some embodiments, the processor includes a central processing unit and a neural network processor, and the energy efficiency ratio information includes second energy efficiency ratio information;

[0016] Acquiring multiple sets of different control parameter sets of the processor group and energy efficiency ratio information of the server cluster when working under any control parameter set, further comprising:

[0017] Selecting a third control parameter from a plurality of preset third control parameters of at least one neural network processor, wherein the first control parameter and the third control parameter constitute a set of control parameters;

[0018] Under the condition that the central processor enters the running state based on the first control parameter and the neural network processor enters the running state based on the third control parameter, obtaining a second computing performance value and a second service power consumption value of the server cluster;

[0019] Second energy efficiency ratio information is obtained according to the ratio of the second computing performance value to the second service power consumption value.

[0020] In some embodiments, based on the existing energy efficiency ratio information and the corresponding control parameter set, a distribution model of the distribution relationship between the control parameters of multiple processors and the energy efficiency ratio of the server cluster is constructed, including:

[0021] Based on the existing energy efficiency ratio information and the corresponding control parameter set, construct a mean function that characterizes the correlation between multiple energy efficiency ratio information and a covariance function that characterizes the correlation between the control parameter set and the corresponding energy efficiency ratio information;

[0022] According to the mean function and the covariance function, a distribution model is constructed to obtain the distribution relationship between the control parameters of multiple processors and the energy efficiency ratio of the server cluster.

[0023] In some embodiments, the acquisition function comprises a first acquisition function;

[0024] Based on the distribution model, a collection function is constructed to evaluate the quality of the control parameter set, including:

[0025] Determine the current optimal energy efficiency ratio information from the existing energy efficiency ratio information;

[0026] The mean function is updated based on the current optimal energy efficiency ratio information and the preset cumulative distribution function to obtain an updated mean function;

[0027] Updating the covariance function based on a preset probability density function to obtain a first updated covariance function;

[0028] The updated mean function and the first updated covariance function are superimposed to construct a first acquisition function for evaluating the quality of the control parameter set.

[0029] In some embodiments, the acquisition function includes a second acquisition function;

[0030] Based on the distribution model, a collection function is constructed to evaluate the quality of the control parameter set, which also includes:

[0031] The covariance function is updated based on the preset hyperparameters to obtain a second updated covariance function;

[0032] The mean function and the second updated covariance function are superimposed to construct a second acquisition function for evaluating the quality of the control parameter set.

[0033] In some embodiments, after determining new energy efficiency ratio information when the server cluster operates under the sampled control parameter set, the method further includes:

[0034] Step 1: based on the sampled control parameter set and the new energy efficiency ratio information, update the existing energy efficiency ratio information and the corresponding control parameter set;

[0035] Step 2: Under the premise of optimizing the acquisition function, resample the other control parameter sets other than the updated existing energy efficiency ratio information based on the distribution model, and determine the new energy efficiency ratio information of the server cluster when working under the sampled control parameter set;

[0036] Repeat steps one and two until the preset iteration conditions are met, determine the target energy efficiency ratio information that meets the preset energy efficiency ratio requirements from the updated multiple existing energy efficiency ratio information, and use the control parameter set corresponding to the target energy efficiency ratio information as the target control parameter set to control the server cluster to work under the target control parameter set.

[0037] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application proposes an energy efficiency ratio tuning device for a server cluster, the device comprising:

[0038] an acquisition module, used to acquire multiple sets of different control parameter sets of a processor group and energy efficiency ratio information of a server cluster when operating under any control parameter set, wherein the processor group includes multiple different types of processors, and each set of control parameter sets includes control parameters of each processor in the processor group;

[0039] A construction module is used to construct a distribution model of the distribution relationship between the control parameters of multiple processors and the energy efficiency ratio of the server cluster based on the existing energy efficiency ratio information and the corresponding control parameter set, and to construct a collection function for evaluating the quality of the control parameter set based on the distribution model;

[0040] A sampling module is used to sample other control parameter sets other than the existing energy efficiency ratio information based on the distribution model under the premise of optimizing the acquisition function, and determine the new energy efficiency ratio information of the server cluster when working under the sampled control parameter set;

[0041] The target control module is used to determine the target energy efficiency ratio information that meets the preset energy efficiency ratio requirements from the multiple existing energy efficiency ratio information after update, and control the server cluster to work under the target control parameter set based on the control parameter set corresponding to the target energy efficiency ratio information.

[0042] To achieve the above-mentioned purpose, the third aspect of an embodiment of the present application proposes an electronic device, the electronic device comprising a memory and a processor, the memory storing a computer program, and the processor implementing the method of the above-mentioned first aspect when executing the computer program.

[0043] To achieve the above objectives, the fourth aspect of an embodiment of the present application proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the method of the first aspect is implemented.

[0044] The energy efficiency ratio tuning method, device, equipment and storage medium of the server cluster proposed in the present application obtain multiple groups of different control parameter sets of the processor group and the energy efficiency ratio information of the server cluster when working under any control parameter set, wherein the processor group includes multiple different types of processors, and each group of control parameter sets includes the control parameters of each processor in the processor group; based on the existing energy efficiency ratio information and the corresponding control parameter set, a distribution model of the distribution relationship between the control parameters of multiple processors and the energy efficiency ratio of the server cluster is constructed, and based on the distribution model, a collection function for evaluating the quality of the control parameter set is constructed, so that the tuning device can clarify the collection direction from the multiple groups of control parameter sets according to the collection function to avoid blindly collecting the control parameter sets. Causing energy waste; under the premise of optimizing the acquisition function, sampling other control parameter sets besides the existing energy efficiency ratio information is performed based on the distribution model to determine the new energy efficiency ratio information when the server cluster works under the sampled control parameter set; then, the target energy efficiency ratio information that meets the preset energy efficiency ratio requirements is determined from the updated multiple existing energy efficiency ratio information, and the control parameter set corresponding to the target energy efficiency ratio information is used as the target control parameter set, and the server cluster is controlled to work under the target control parameter set. In this way, by jointly tuning multiple different types of processors, the time it takes for the server cluster to reach the optimal performance is reduced, and at the same time, the waste of resources in the parameter adjustment process caused by trying parameters one by one is reduced, thereby improving the energy efficiency ratio of the server cluster and reducing energy consumption. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a schematic diagram of an optional application scenario of the energy efficiency ratio tuning device for a server cluster provided in an embodiment of the present application;

[0046] Figure 2 It is an optional flow chart of the energy efficiency ratio tuning method of the server cluster provided in the embodiment of the present application;

[0047] Figure 3 yes Figure 2 An optional implementation flowchart of step 101 in ;

[0048] Figure 4 yes Figure 2 Another optional implementation flowchart of step 101 in ;

[0049] Figure 5 yes Figure 2 An optional implementation flowchart of step 102 in FIG. 1 ;

[0050] Figure 6 yes Figure 2 Another optional implementation flowchart of step 102 in;

[0051] Figure 7yes Figure 2 Another optional implementation flowchart of step 102 in ;

[0052] Figure 8 yes Figure 2 An optional implementation flowchart after step 103 in;

[0053] Fig. 9 It is an optional flow chart of the energy efficiency ratio tuning device of the server cluster provided in the embodiment of the present application;

[0054] Fig.10 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0056] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0057] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0058] A server cluster refers to a computing environment that connects multiple servers together to provide high availability, scalability, and load balancing. For example, data centers and intelligent computing clusters are both server clusters. A server cluster is usually composed of multiple processor groups, and each processor group includes multiple processors. There are many types of processors, such as central processing units, graphics processing units, neural network processors, etc. Energy efficiency ratio (EER) is an important indicator for measuring the energy consumption of server clusters. The larger the EER, the more computing work the server cluster can complete under the same energy consumption.

[0059] Among them, data centers and computing clusters, as the core infrastructure for the current development of the information technology industry, bear the huge data processing needs of modern society and have become an important support for the global digital economy. Especially today when artificial intelligence technology is advancing by leaps and bounds, intelligent computing clusters (also referred to as "intelligent computing clusters") have gradually become a key support for promoting artificial intelligence research and industrial development. Intelligent computing clusters are built on advanced intelligent computing processors such as graphics processors and neural network processors, which can efficiently handle complex artificial intelligence tasks and large-scale data computing needs.

[0060] In the related art, the processor with the highest energy consumption is selected from multiple processors and its parameters are adjusted. However, the energy efficiency ratio of a server cluster depends on the balance between different types of processors. The one-sided parameter adjustment method adopted in the related art still cannot improve its energy efficiency ratio well, which leads to excessive energy consumption of the server cluster.

[0061] Alternatively, as shown in Table 1 below, Table 1 is used to characterize an optional related technical parameter tuning example of the energy efficiency ratio tuning device of the server cluster provided in the embodiment of the present application. Specifically, when the processor group of the server cluster includes a central processing unit (CPU) and a neural network processor (NPU), the joint adjustment of the parameters of the CPU-NPU is called the optimization strategy of the energy efficiency of the server cluster. Table 1 shows the impact of the CPU-NPU on the energy efficiency ratio under different control parameters measured on the server cluster. It can be found that when the relative control parameter of the CPU is 0.95 and the relative control parameter of the NPU is 0.64, the energy efficiency ratio of the server cluster reaches the maximum value. "Relative" means compared to the maximum adjustable parameter of the CPU or NPU. It can also be seen from Table 1 that only performing a single control of the CPU or NPU will not be able to find the optimal value of the energy efficiency ratio. In the related art, a simple traversal search method is used, that is, each combination of adjustment parameters is experimentally measured to determine the energy efficiency ratio of the server cluster. Although this method is direct, its efficiency is extremely low, especially when there are many control states of the CPU and NPU, the problem is particularly prominent. For example, when the CPU has M control states and the NPU has N control states, and each experiment is repeated R times, then a total of M*N*R experiments are required. This not only brings huge time costs and resource consumption, but also may delay the real-time requirements of computing tasks.

[0062] Table 1

[0063]

[0064] Based on this, the embodiments of the present application provide a method, apparatus, device and storage medium for optimizing the energy efficiency ratio of a server cluster, aiming to improve the energy efficiency ratio of the server cluster and reduce energy consumption.

[0065] The energy efficiency ratio tuning method, device, equipment and storage medium of the server cluster provided in the embodiment of the present application are specifically explained through the following embodiments. First of all, it should be noted that the energy efficiency ratio tuning method of the server cluster in the embodiment of the present application relates to the field of computer processing technology. It can be applied to the terminal or to the server side, as long as it includes a system of multiple processors of different types. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or a server cluster or a distributed system composed of multiple physical servers, and can also be configured as a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, and big data and artificial intelligence platforms; only examples are given here, and are not limited to the above forms. The embodiment of the present application takes a server cluster as an example, because the server cluster involves a large number of multiple processor control parameter adjustments, and the effect is more significant.

[0066] It should also be noted that in the embodiments of the present application, when it comes to information related to user characteristics such as user basic information or user identity, the user's permission or consent will be obtained first, and the collection, use and processing of these data will comply with relevant laws, regulations and standards. In addition, when the embodiments of the present application need to obtain sensitive personal information of the user, the user's separate permission or separate consent will be obtained first. After the user's separate permission or separate consent is clearly obtained, the necessary data for the normal operation of the embodiments of the present application will be obtained. For example, the energy efficiency ratio tuning device of the server cluster of the embodiment of the present application will obtain the consent of the relevant personnel who manage the server cluster before determining the target control parameter set of the server cluster using the energy efficiency ratio tuning method of the server cluster proposed in the embodiment of the present application, otherwise it will not be possible to obtain multiple sets of different control parameter sets. In addition, all relevant data obtained by the embodiments of the present application are authorized data after obtaining consent, which will not be repeated here one by one.

[0067] Next, the application scenario of the energy efficiency ratio tuning device of the server cluster in the embodiment of the present application is described, such as Figure 1 As shown, Figure 1 is a schematic diagram of an optional application scenario of the energy efficiency ratio tuning device for a server cluster provided in an embodiment of the present application, Figure 1In the example, the server cluster includes multiple servers such as server a, server b and server c (for example only and does not limit the number of servers), and multiple different types of processors (processor groups) are arranged in the server cluster, and these different types of processors may be arranged in the same server or in different servers; obtain the corresponding control parameter set of the processor group, wherein the control parameter set includes the control parameters of each processor of different types; input the obtained multiple sets of different control parameter sets into the energy efficiency ratio tuning device of the server cluster proposed in the embodiment of the present application (for the convenience of description, it may also be referred to as "tuning device"), the tuning device determines and outputs the target control parameter set that can make the energy efficiency ratio of the server cluster the highest without trying all control parameter sets by applying the energy efficiency ratio tuning method of the server cluster proposed in the embodiment of the present application (for the convenience of description, it may also be referred to as "tuning method"), and applies the target control parameter set to the server cluster, and the energy consumption of the server cluster that enters the working state based on the target control parameter set is the lowest among all control parameter sets, that is, the energy efficiency ratio is the highest.

[0068] After understanding the example application scenario of the energy efficiency ratio tuning device for a server cluster proposed in the present application, the energy efficiency ratio tuning method for a server cluster proposed in an embodiment of the present application is described next.

[0069] In the embodiments of the present application, the optimization device will be described from the perspective of: Figure 2 As shown, Figure 2 is an optional flowchart of the tuning method provided in the embodiment of the present application. Figure 2 The method in the embodiment may include but is not limited to the following steps 101 to 104. When the tuning device executes the tuning method, the specific process is as follows. It should be noted that the present embodiment Figure 2 The order of step 101 to step 104 is not specifically limited, and the order of steps can be adjusted or some steps can be reduced or added according to actual needs.

[0070] Step 101, obtain multiple groups of different control parameter sets of the processor group, and energy efficiency ratio information of the server cluster when working under any control parameter set, wherein the processor group includes multiple different types of processors, and each group of control parameter sets includes the control parameters of each processor in the processor group.

[0071] The following is a detailed description of step 101.

[0072] Among them, data centers and intelligent computing clusters are common server clusters. A data center is composed of a variety of facilities such as servers, storage devices, switches, routers, firewalls, etc. Usually, most of the facilities in the data center are equipped with processors, so that a variety of facilities work together to achieve efficient data processing capabilities of the data center. A computing cluster is a computer system that connects multiple computers (or nodes) through a high-speed network, and each node is equipped with a processor so that the computing cluster distributes computing tasks to multiple nodes, thereby improving overall computing power. The server cluster to which the tuning method proposed in the embodiment of the present application is applied can be set according to actual conditions. The embodiment of the present application only illustrates two common server clusters, and does not specifically limit this.

[0073] Among them, a processor group refers to a set consisting of multiple processors, and the processor group in the embodiment of the present application includes at least two different types of processors. Different types of processors differ in architecture, performance, power consumption, etc. Exemplarily, different types of processors include but are not limited to central processing units, graphics processing units, neural network processors, digital signal processors, etc., and the embodiments of the present application do not limit the specific types of processors.

[0074] Exemplarily, (CPU+GPU), (CPU+NNP), (CPU+DSP) are all processor groups that may be used in a server cluster. It should be noted that no matter how the multiple processors included in the processor group are combined, the processor group usually includes a core processor, i.e., a CPU; the specific processor type included in the processor group can be set according to actual conditions, and the embodiments of the present application do not limit this.

[0075] Furthermore, the control parameter set is composed of the control parameters of each processor in the processor group. When each processor sets different control parameters, the processor enters a different operating state accordingly. The control parameters of the processor include but are not limited to the processor operating frequency, voltage, current, power consumption limit, etc. Usually, the control parameters included in the control parameter set are of the same type. For example, if a processor group is CPU+NPU, then the control parameter set of the processor group obtained can be the CPU operating frequency and the NPU operating frequency, or the CPU voltage and the NPU voltage. The parameter type of each control parameter in the control parameter set can be adaptively adjusted according to actual conditions. Exemplarily, (f CPU1 ,f NPU1 ) is a set of control parameter sets, and f represents the operating frequency. The embodiment of the present application needs to obtain multiple sets of control parameter sets: (f CPU1 ,f NPU1 )、(f CPU2 ,fNPU2 )......(f CPUn ,f NPUn ).

[0076] Furthermore, multiple groups are randomly selected from the obtained multiple control parameter sets, and energy efficiency ratio information of the server cluster when working under the selected control parameter sets is obtained. The energy efficiency ratio information is an indicator for evaluating the energy utilization efficiency of the server cluster, and the larger the value of the energy efficiency ratio information, the higher the energy utilization rate of the server cluster, and the less energy is wasted.

[0077] Exemplarily, the processor group includes a CPU and a GPU, and the corresponding multiple sets of control parameter sets obtained are (f CPU1 ,f GPU1 )、(f CPU2 ,f GPU2 ), adjust the CPU operating frequency to f CPU1 , adjust the GPU operating frequency to f GPU1 , and determine the energy efficiency ratio information EER1 of the server cluster when it is running in this state; then, adjust the CPU operating frequency to f CPU2 , adjust the GPU operating frequency to f GPU2 , and determine the energy efficiency ratio information EER2 when the server cluster runs in this state.

[0078] Furthermore, by the following formula <1> Define the calculation method of energy efficiency ratio information:

[0079]

[0080] Among them, Performance represents the computing performance value of the server cluster, which can be determined by measuring the number of floating-point calculations Flops completed by the server cluster per second; Power represents the service power value of the server cluster, which can be determined by measuring the total power of the server cluster in watts.

[0081] Furthermore, Flops and Watts can usually be automatically measured based on a benchmark program running on a server cluster. The specific benchmark program selected can be set according to actual conditions, and the present application embodiment does not limit this.

[0082] Then, for the mainstream processor groups, the energy efficiency ratio information of the corresponding server cluster is calculated by the following two methods:

[0083] In some embodiments, Figure 3 As shown, Figure 3 yes Figure 2An optional implementation process of step 101 in the embodiment of the present invention is to obtain multiple different sets of control parameter sets of the processor group and energy efficiency ratio information of the server cluster when working under any set of control parameter sets, including the following steps 201 to 203:

[0084] Step 201, select a first control parameter from a plurality of preset first control parameters of a central processing unit, select a second control parameter from a plurality of preset second control parameters of at least one graphics processing unit, the first control parameter and the second control parameter constitute a set of control parameters.

[0085] Step 202: Under the condition that the central processing unit enters the running state based on the first control parameter and the graphics processing unit enters the running state based on the second control parameter, a first computing performance value and a first service power consumption value of the server cluster are obtained.

[0086] Step 203: Obtain first energy efficiency ratio information according to the ratio of the first computing performance value to the first service power consumption value.

[0087] Steps 201 to 203 are described in detail below.

[0088] Exemplarily, when the processors in the processor cluster include CPUx+GPUy, the multiple preset first control parameters that CPUx can select include f CPU1 、f CPU2 ......f CPUn , GPUy can select multiple preset second control parameters including f GPU1 、f GPU2 ......f GPUn , where the control parameters with the same subscript match, that is, the first control parameter selected is f CPU1 When the second control parameter is selected, the corresponding second control parameter should be f GPU1 Assume that f is selected from a plurality of preset first control parameters CPU2 , select f from multiple preset second control parameters GPU2 , then (f CPU2 , f GPU2 ) is a set of selected first control parameters.

[0089] Furthermore, the operating frequency of CPUx is adjusted to f CPU2 , and adjust the operating frequency of GPUy to f GPU2In this case, the first computing performance value and the first service power consumption value of the server cluster are determined according to a preset benchmark test program, and based on the formula <1> The first energy efficiency ratio information is calculated. Similarly, the first energy efficiency ratio information corresponding to other selected first control parameter sets is determined to obtain multiple sets of existing first energy efficiency ratio information, and in order to ensure that the multiple sets of first energy efficiency ratio information are comparable, after each processor in the processor group is parameter-adjusted according to each first control parameter set, the first computing performance value and the first service power consumption value are measured after the server cluster runs for the same time.

[0090] Furthermore, there may be multiple graphics processors. In this case, the processor group may be CPUx+GPUa+GPUb. Then, the corresponding multiple control parameter sets of the processor group are (f CPUx1 , f GPUa1 、f GPUb1 )、(f CPUx2 , f GPUa2 、f GPUb2 )......(f CPUxn , f GPUan 、f GPUbn ), the number of graphics processors can be set according to actual conditions, and the embodiments of the present application do not impose any specific restrictions on this.

[0091] It is understandable that in actual applications, the CPU and GPU of the processor group can work together. In modern server cluster scenarios, the GPU, especially multiple GPUs, can assist the CPU to better complete tasks such as machine learning and visual data processing. The tuning method proposed in the embodiment of the present application can jointly tune multiple different types of processors in the processor group, which can reduce the time it takes for the server cluster to reach optimal performance, while reducing energy waste in the parameter adjustment process caused by testing parameters one by one.

[0092] In some embodiments, Figure 4 As shown, Figure 4 yes Figure 2 Another optional implementation flowchart of step 101 in the embodiment of the present invention, obtaining multiple different sets of control parameter sets of the processor group and energy efficiency ratio information of the server cluster when working under any set of control parameter sets, further includes the following steps 301 to 303:

[0093] Step 301: Select a third control parameter from a plurality of preset third control parameters of at least one neural network processor, and the first control parameter and the third control parameter constitute a set of control parameters.

[0094] Step 302, under the condition that the central processing unit enters the running state based on the first control parameter and the neural network processor enters the running state based on the third control parameter, obtain the second computing performance value and the second service power consumption value of the server cluster.

[0095] Step 303: Obtain second energy efficiency ratio information according to the ratio of the second computing performance value to the second service power consumption value.

[0096] Steps 301 to 303 are described in detail below.

[0097] Exemplarily, when the processors in the processor cluster include CPUx+NPUy, the plurality of preset second control parameters selectable by CPUx include f CPU1 、f CPU2 ......f CPUn , NPUy can select multiple preset third control parameters including f NPU1 、f 1PU2 ......f NPUn , where the control parameters with the same subscript match, that is, the selected second control parameter is f CPU1 When the third control parameter is selected, the corresponding NPU1 Assume that f is selected from a plurality of preset second control parameters CPU2 , select f from multiple preset third control parameters NPU2 , then (f CPU2 , f NPU2 ) is a set of selected second control parameters.

[0098] Furthermore, the operating frequency of CPUx is adjusted to f CPU2 , and adjust the operating frequency of NPUy to f NPU2 In this case, the second computing performance value and the second service power consumption value of the server cluster are determined according to a preset benchmark test program, and based on the formula <1> The second energy efficiency ratio information is calculated. Similarly, the second energy efficiency ratio information corresponding to other selected second control parameter sets is determined to obtain multiple sets of existing second energy efficiency ratio information, and in order to ensure that the multiple sets of second energy efficiency ratio information are comparable, after each processor in the processor group is parameter-adjusted according to each second control parameter set, the second computing performance value and the second service power consumption value are measured after the server cluster runs for the same time.

[0099] Furthermore, there may be multiple graphics processors. In this case, the processor group may be CPUx+NPUa+NPUb. Then, the corresponding multiple control parameter sets of the processor group are (f CPUx1 , f NPUa1 、f NPUb1 )、(f CPUx2 , fNPUa2 、f NPUb2 )......(f CPUxn , f NPUan 、f NPUbn ), the number of graphics processors can be set according to actual conditions, and the embodiments of the present application do not impose any specific restrictions on this.

[0100] It is understandable that in actual applications, the CPU and NPU of the processor group can work together. In modern server cluster scenarios, the NPU, especially multiple NPUs, can assist the CPU to better complete tasks such as edge computing and artificial intelligence reasoning. The tuning method proposed in the embodiment of the present application can jointly tune multiple different types of processors in the processor group, which can reduce the time it takes for the server cluster to reach optimal performance, while reducing energy waste in the parameter adjustment process caused by testing parameters one by one.

[0101] Step 102, based on the existing energy efficiency ratio information and the corresponding control parameter set, construct a distribution model of the distribution relationship between the control parameters of multiple processors and the energy efficiency ratio of the server cluster, and construct a collection function for evaluating the quality of the control parameter set based on the distribution model.

[0102] The following is a detailed description of step 102.

[0103] In some embodiments, a distribution model of the distribution relationship between the control parameters of multiple processors and the energy efficiency ratio of the server cluster is constructed based on the Bayesian optimization principle. Since the relationship between the energy efficiency ratio information of the server cluster and the control parameters of multiple different types of processors cannot be described by a simple analytical formula, the embodiment of the present application constructs a distribution model based on Bayesian optimization, and determines a collection function for evaluating the quality of the control parameter set based on the distribution model, so that the tuning device can then clarify the collection direction from multiple groups of control parameter sets according to the collection function, avoiding blind collection of control parameter sets and causing energy waste.

[0104] In some embodiments, as shown in Figure 5, Figure 5 yes Figure 2 An optional implementation flowchart of step 102 in the embodiment of the present invention is to construct a distribution model of the distribution relationship between the control parameters of multiple processors and the energy efficiency ratio of the server cluster based on the existing energy efficiency ratio information and the corresponding control parameter set, including the following steps 401 to 402:

[0105] Step 401 : Based on the existing energy efficiency ratio information and the corresponding control parameter set, a mean function characterizing the correlation between multiple energy efficiency ratio information and a covariance function characterizing the correlation between the control parameter set and the corresponding energy efficiency ratio information are constructed.

[0106] Step 402: construct a distribution model for the distribution relationship between the control parameters of multiple processors and the energy efficiency ratio of the server cluster based on the mean function and the covariance function.

[0107] Steps 401 to 402 are described in detail below.

[0108] In some embodiments, a distribution model is constructed to approximate the true objective function. Gaussian Process (GP) is a commonly used proxy model in Bayesian optimization, which describes the probability distribution of the objective function through the mean function μ(x) and k(x, x′), as shown in the following formula: <2> As shown:

[0109] EER(x)~GP(μ(x),k(x,x′)) <2>

[0110] Where x = (f CPU ,f NPU ) is the input control parameter set, x′ is another control parameter set different from x; GP stands for Gaussian process; μ(x)=E[EER(x)] is the mean function that characterizes the correlation between multiple energy efficiency ratio information; k(x,x′)=E[(EER(x)-μ(x))(EER(x′)-μ(x′))] is the covariance function that characterizes the correlation between the control parameter set and the corresponding energy efficiency ratio information; E represents the expected value.

[0111] It should be noted that random forests, neural networks, etc. can also be selected as proxy models in the Bayesian optimization process. The specific selection can be made according to the actual situation, and the embodiments of the present application do not limit this. The distribution model constructed based on the existing energy efficiency ratio information and the corresponding control parameter set can capture the complex interactions between multiple different types of processors, and then consider the impact of the interaction on the energy consumption of the server cluster, which helps to reduce the exploration of unknown control parameter sets in the subsequent sampling process, thereby reducing computing costs and unnecessary energy waste.

[0112] In some embodiments, Figure 6 As shown, Figure 6 yes Figure 2 In another optional implementation flowchart of step 102 in the embodiment, the acquisition function includes a first acquisition function, and the acquisition function for evaluating the quality of the control parameter set is constructed based on the distribution model, including the following steps 501 to 504:

[0113] Step 501: determine the current optimal energy efficiency ratio information from the existing energy efficiency ratio information.

[0114] Step 502: Update the mean function based on the current optimal energy efficiency ratio information and the preset cumulative distribution function to obtain an updated mean function.

[0115] Step 503: Update the covariance function based on a preset probability density function to obtain a first updated covariance function.

[0116] Step 504 : superimpose the updated mean function and the first updated covariance function to construct a first acquisition function for evaluating the quality of the control parameter set.

[0117] Steps 501 to 504 are described in detail below.

[0118] In some embodiments, the current optimal energy efficiency ratio information is determined from the currently available multiple energy efficiency ratio information. Usually, the energy efficiency ratio information with the largest energy efficiency ratio is selected as the current optimal energy efficiency ratio information. Next, the following formula is constructed: <3> The first acquisition function shown is the expected improvement (EI):

[0119] EI(x)=E[max(0,EER(x)-EER best )]

[0120] =(u(x)-EER best )φ(Z)+σ(x)φ(Z) <3>

[0121] Among them, EER best is the current optimal energy efficiency ratio information; E represents the expected value; Z = (μ(x)-EER best ) / σ(x); μ(x) and σ(x) are the predicted mean and standard deviation of the Gaussian process at x, respectively. The predicted mean is calculated by the mean function, and the standard deviation is calculated by the square root of the covariance function; Φ(Z) is the preset cumulative distribution function (CDF), (μ(x)-EER best )Φ(Z) is the updated mean function; φ(Z) is the preset probability density function (PDF), and σ(x)φ(Z) is the first updated covariance function.

[0122] In some embodiments, Figure 7 As shown, Figure 7 yes Figure 2 In another optional implementation flowchart of step 102 in the embodiment, the acquisition function includes a second acquisition function, and the acquisition function for evaluating the quality of the control parameter set is constructed based on the distribution model, and further includes the following steps 601 to 602:

[0123] Step 601, updating the covariance function based on preset hyperparameters to obtain a second updated covariance function.

[0124] Step 602: superimpose the mean function and the second updated covariance function to construct a second acquisition function for evaluating the quality of the control parameter set.

[0125] Steps 601 to 602 are described in detail below.

[0126] In some embodiments, the following formula is constructed <4> The second acquisition function shown is the upper confidence limit (Upper Confidence Bound, UCB):

[0127] UCB(x)=μ(x)+k·σ(x) <4>

[0128] Wherein, μ(x) and σ(x) are the predicted mean and standard deviation of the Gaussian process at x, respectively; k·σ(x) is the second updated covariance function; k is a hyperparameter used to control the balance between exploration and utilization. The common value range of k is between 1 and 10. In practical applications, a smaller k value can be used for a smoother objective function, while for a highly uncertain or complex objective function, a larger k value is usually selected to enhance exploration. That is, the specific k value can be set according to the actual situation, and the embodiments of the present application do not limit this.

[0129] Step 103 , under the premise of optimizing the acquisition function, sampling other control parameter sets other than the existing energy efficiency ratio information based on the distribution model, and determining new energy efficiency ratio information when the server cluster works under the sampled control parameter set.

[0130] The following is a detailed description of step 103.

[0131] Furthermore, compared with the related art in which multiple control parameters in Table 1 are debugged one by one, the embodiment of the present application can determine the sampling direction from multiple control parameter sets based on the distribution model after completing the determination of the acquisition function and on the premise of optimizing the acquisition function, and then sample other control parameter sets outside the existing energy efficiency ratio information, so as to achieve the purpose of selecting the most valuable sampling from other control parameter sets whose energy efficiency ratio information has not been calculated, thereby reducing repeated debugging of parameters and energy consumption.

[0132] When the sampling function is the first sampling function, the next set of control parameters for sampling is selected by maximizing EI, that is, the next set of control parameters is x * =argmaxEI(x). According to the EI calculation formula, when μ(x)>EER bestWhen the EI value is large, it indicates a high expectation for improvement at that point; while when σ(x) is large, even if μ(x) is not high, EI may still be large. This can encourage the distribution model to explore regions of high uncertainty. Thus, by balancing exploration (understanding unknown regions) and exploitation (fine-tuning in known good regions), the sampling accuracy can be improved.

[0133] When the sampling function is the second sampling function, the next sampling point is selected by maximizing UCB(x), that is, x * = argmax UCB(x).

[0134] Furthermore, after determining the set of regulated parameters for the new sampling, the energy efficiency ratio information corresponding to the newly sampled set of regulated parameters is obtained by a method similar to that in step 101, so that the distribution model can output better prediction results based on the newly sampled data later.

[0135] Step 104: Determine the target energy efficiency ratio information that meets the preset energy efficiency ratio requirements from the updated multiple existing energy efficiency ratio information, and use the set of regulated parameters corresponding to the target energy efficiency ratio information as the target set of regulated parameters to control the server cluster to operate under the target set of regulated parameters.

[0136] The following provides a detailed description of step 104.

[0137] In some embodiments, the optimal one is selected from the original energy efficiency ratio information and the energy efficiency ratio information corresponding to the newly sampled set of regulated parameters as the target energy efficiency ratio information; then, the set of regulated parameters corresponding to the target energy efficiency ratio information is used as the target set of regulated parameters, and the regulated parameters of each processor are adjusted according to the target set of regulated parameters, thereby completing the energy efficiency ratio optimization of the server cluster.

[0138] Exemplarily, the existing set of regulated parameters and the corresponding energy efficiency ratio information are (f CPU1 , f NPU1 , EER1), (f CPU2 , f NPU2 , EER2), and the newly sampled set of regulated parameters and the corresponding energy efficiency ratio information are (f CPU3 , f NPU3 , EER3), and EER1 < EER2 < EER3. Then, EER3 is determined as the target energy efficiency ratio information, and the CPU regulated parameters in the processor group are adjusted to f CPU3 , and the NPU regulated parameters in the processor group are adjusted to f NPU3 . In this state, the energy efficiency ratio of the server cluster is optimal.

[0139] The preset energy efficiency ratio requirement may be to select the one with the largest value from the currently existing control parameter set; or to select a preset number of control parameter sets from the currently existing control parameter set. The energy efficiency ratio requirement may be adaptively adjusted according to actual conditions, and the embodiments of the present application do not limit this.

[0140] It can be understood that the tuning method proposed in the embodiment of the present application can not only effectively avoid performance bottlenecks caused by improper regulation of a single computing unit, but also more accurately balance the relationship between computing speed and energy consumption, thereby achieving more efficient server cluster energy efficiency ratio tuning.

[0141] In some embodiments, Figure 8 As shown, Figure 8 yes Figure 2 An optional implementation flowchart after step 103 in the embodiment further includes the following steps 701 to 703 after determining the new energy efficiency ratio information of the server cluster when operating under the sampled control parameter set:

[0142] Step 701: based on the sampled control parameter set and new energy efficiency ratio information, update the existing energy efficiency ratio information and the corresponding control parameter set.

[0143] Step 702, under the premise of optimizing the acquisition function, resample the other control parameter sets other than the updated existing energy efficiency ratio information based on the distribution model, and determine the new energy efficiency ratio information when the server cluster works under the sampled control parameter set.

[0144] Step 703, repeat steps 701 and 702 until the preset iteration condition is reached, determine the target energy efficiency ratio information that meets the preset energy efficiency ratio requirements from the updated multiple existing energy efficiency ratio information, and use the control parameter set corresponding to the target energy efficiency ratio information as the target control parameter set to control the server cluster to work under the target control parameter set.

[0145] Steps 701 to 703 are described in detail below.

[0146] In some embodiments, after completing the sampling of a set of control parameters, the distribution model can be updated according to the original set of control parameters and the corresponding energy efficiency information, as well as the newly sampled set of control parameters and the corresponding energy efficiency information, and resample other sets of control parameters whose energy efficiency information is unknown; by repeating the steps of iterative model updating and resampling, the optimal output solution is gradually approached, and the accuracy of the target control parameter set finally determined as the output is ensured.

[0147] Among them, after multiple iterations, when the new energy efficiency ratio information is much lower than the existing energy efficiency ratio information value and the improvement of the target energy efficiency ratio information is no longer significant, the optimization process converges and the target control parameter set is returned. Alternatively, the condition for the end of repeated iterations may be that the change between the newly sampled energy efficiency ratio information and the last sampled energy efficiency ratio information is less than a threshold, the maximum number of iterations is reached, the maximum iteration time is reached, etc. The condition for the end of repeated iterations can be adaptively adjusted according to actual conditions, and the embodiments of the present application do not limit this. Compared with related technologies, the distribution model used in the embodiments of the present application usually only needs to be performed dozens of times or even less times to determine the optimal control parameter set of the server cluster. By effectively exploring the search space, the computing resources and time costs required to find the global optimal point are greatly reduced, providing strong support for the green and efficient operation of the server cluster. Exemplarily, assuming that the energy efficiency ratio of the server cluster in Table 1 is 0.85 when running at the default frequency, after the tuning device proposed in the embodiment of the present application is used to optimize the parameters of the processor group of the server cluster in Table 1, an energy efficiency ratio of (1-0.85) / 0.85=17% can be achieved compared to when running at the default frequency. For a server cluster with a typical power consumption of 15MW, this means reducing its average power consumption by 15*17%=2.5MW, thereby helping to save unnecessary energy waste.

[0148] In addition, the iterative process of the distribution model is adaptive. It can dynamically adjust the next search direction according to the existing sample data, which makes the optimization process more intelligent and can gradually approach the global optimal solution in a complex multivariable environment. At the same time, the distribution model supports online optimization, which enables it to maintain high-efficiency energy-efficiency tuning under the ever-changing computing task load, and thus is suitable for real-time application scenarios.

[0149] like Fig. 9 As shown, Fig. 9 This is an optional flow chart of the energy efficiency ratio tuning device of the server cluster provided in the embodiment of the present application. The energy efficiency ratio tuning device of the server cluster includes the following modules 801 to 804:

[0150] Acquisition module 801 is used to obtain multiple groups of different control parameter sets of the processor group and energy efficiency ratio information of the server cluster when working under any control parameter set, wherein the processor group includes multiple different types of processors, and each group of control parameter sets includes the control parameters of each processor in the processor group.

[0151] Construction module 802 is used to construct a distribution model of the distribution relationship between the control parameters of multiple processors and the energy efficiency ratio of the server cluster based on the existing energy efficiency ratio information and the corresponding control parameter set, and to construct a collection function for evaluating the quality of the control parameter set based on the distribution model.

[0152] The sampling module 803 is used to sample other control parameter sets other than the existing energy efficiency ratio information based on the distribution model under the premise of optimizing the acquisition function, and determine new energy efficiency ratio information when the server cluster works under the sampled control parameter set.

[0153] The target control module 804 is used to determine the target energy efficiency ratio information that meets the preset energy efficiency ratio requirements from the multiple existing energy efficiency ratio information after update, and control the server cluster to work under the target control parameter set based on the control parameter set corresponding to the target energy efficiency ratio information.

[0154] The energy efficiency ratio tuning method, device, equipment and storage medium of the server cluster proposed in the present application obtain multiple groups of different control parameter sets of the processor group and the energy efficiency ratio information of the server cluster when working under any control parameter set, wherein the processor group includes multiple different types of processors, and each group of control parameter sets includes the control parameters of each processor in the processor group; based on the existing energy efficiency ratio information and the corresponding control parameter set, a distribution model of the distribution relationship between the control parameters of multiple processors and the energy efficiency ratio of the server cluster is constructed, and based on the distribution model, a collection function for evaluating the quality of the control parameter set is constructed, so that the tuning device can clarify the collection direction from the multiple groups of control parameter sets according to the collection function to avoid blindly collecting the control parameter sets. Causing energy waste; under the premise of optimizing the acquisition function, sampling other control parameter sets besides the existing energy efficiency ratio information is performed based on the distribution model to determine the new energy efficiency ratio information when the server cluster works under the sampled control parameter set; then, the target energy efficiency ratio information that meets the preset energy efficiency ratio requirements is determined from the updated multiple existing energy efficiency ratio information, and the control parameter set corresponding to the target energy efficiency ratio information is used as the target control parameter set, and the server cluster is controlled to work under the target control parameter set. In this way, by jointly tuning multiple different types of processors, the time it takes for the server cluster to reach the optimal performance is reduced, and at the same time, the waste of resources in the parameter adjustment process caused by trying parameters one by one is reduced, thereby improving the energy efficiency ratio of the server cluster and reducing energy consumption.

[0155] The specific implementation of the tuning device is basically the same as the specific implementation of the above-mentioned tuning method, and will not be repeated here.

[0156] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the energy efficiency ratio tuning method of the server cluster when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0157] like Fig.10 As shown, Fig.10: is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application, the electronic device includes:

[0158] The processor 901 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0159] The memory 902 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 902, and the processor 901 calls and executes the energy efficiency ratio tuning method of the server cluster in the embodiment of this application;

[0160] Input / output interface 903, used to implement information input and output;

[0161] Communication interface 904, used to realize communication interaction between the device and other devices, which can be realized by wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0162] A bus 905 that transmits information between the various components of the device (e.g., the processor 901, the memory 902, the input / output interface 903, and the communication interface 904);

[0163] The processor 901 , the memory 902 , the input / output interface 903 and the communication interface 904 are connected to each other in communication within the device via a bus 905 .

[0164] An embodiment of the present application further provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the energy efficiency ratio tuning method of the server cluster is implemented.

[0165] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0166] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0167] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0168] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0169] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0170] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0171] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0172] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0173] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0174] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0175] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0176] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A method for optimizing the energy efficiency ratio of a server cluster, characterized in that: Applied to a server cluster, the method comprises: Acquire multiple sets of different control parameter sets of a processor group, and energy efficiency ratio information of the server cluster when operating under any of the control parameter sets, wherein the processor group includes multiple different types of processors, and each set of the control parameter sets includes the control parameters of each processor in the processor group; Based on the existing energy efficiency ratio information and the corresponding control parameter set, a distribution model of the distribution relationship between the control parameters of the plurality of processors and the energy efficiency ratio of the server cluster is constructed, and based on the distribution model, a collection function for evaluating the quality of the control parameter set is constructed; Under the premise of optimizing the acquisition function, sampling other control parameter sets other than the existing energy efficiency ratio information based on the distribution model to determine new energy efficiency ratio information when the server cluster works under the sampled control parameter set; Determine target energy efficiency ratio information that meets the preset energy efficiency ratio requirement from the multiple existing energy efficiency ratio information after update, and control the server cluster to work under the target control parameter set based on the control parameter set corresponding to the target energy efficiency ratio information; The step of constructing a distribution model of the distribution relationship between the control parameters of the processors and the energy efficiency ratio of the server cluster based on the existing energy efficiency ratio information and the corresponding control parameter set includes: Based on the existing energy efficiency ratio information and the corresponding control parameter set, construct a mean function characterizing the correlation between the plurality of energy efficiency ratio information and a covariance function characterizing the correlation between the control parameter set and the corresponding energy efficiency ratio information; Constructing a distribution model for obtaining a distribution relationship between the control parameters of the plurality of processors and the energy efficiency ratio of the server cluster according to the mean function and the covariance function; The acquisition function comprises a first acquisition function; The acquisition function for evaluating the quality of the control parameter set is constructed based on the distribution model, comprising: Determine the current optimal energy efficiency ratio information from the existing energy efficiency ratio information; Updating the mean function based on the current optimal energy efficiency ratio information and a preset cumulative distribution function to obtain an updated mean function; Updating the covariance function based on a preset probability density function to obtain a first updated covariance function; The updated mean function and the first updated covariance function are superimposed to construct the first acquisition function for evaluating the quality of the control parameter set.

2. The method according to claim 1, characterized in that The processor includes a central processing unit and a graphics processing unit, and the energy efficiency ratio information includes first energy efficiency ratio information; The obtaining of a plurality of different control parameter sets of the processor group and energy efficiency ratio information of the server cluster when operating under any of the control parameter sets includes: Selecting a first control parameter from a plurality of preset first control parameters of the central processing unit, and selecting a second control parameter from a plurality of preset second control parameters of at least one of the graphics processing units, wherein the first control parameter and the second control parameter constitute a set of the control parameters; Under the condition that the central processing unit enters a running state based on the first control parameter and the graphics processing unit enters a running state based on the second control parameter, obtaining a first computing performance value and a first service power consumption value of the server cluster; The first energy efficiency ratio information is obtained according to the ratio of the first computing performance value to the first service power consumption value.

3. The method according to claim 2, characterized in that The processor includes a central processing unit and a neural network processor, and the energy efficiency ratio information includes second energy efficiency ratio information; The step of obtaining a plurality of different control parameter sets of the processor group and energy efficiency ratio information of the server cluster when operating under any of the control parameter sets further includes: Selecting a third control parameter from a plurality of preset third control parameters of at least one of the neural network processors, wherein the first control parameter and the third control parameter constitute a set of the control parameters; Under the condition that the central processor enters the running state based on the first control parameter and the neural network processor enters the running state based on the third control parameter, obtaining a second computing performance value and a second service power consumption value of the server cluster; The second energy efficiency ratio information is obtained according to the ratio of the second computing performance value to the second service power consumption value.

4. The method according to claim 1, characterized in that: The acquisition function includes a second acquisition function; The acquisition function for evaluating the quality of the control parameter set is constructed based on the distribution model, and further includes: Updating the covariance function based on preset hyperparameters to obtain a second updated covariance function; The mean function and the second updated covariance function are superimposed to construct the second acquisition function for evaluating the quality of the control parameter set.

5. The method according to claim 1, characterized in that After determining new energy efficiency ratio information of the server cluster when operating under the sampled control parameter set, the method further includes: Step 1: based on the sampled control parameter set and the new energy efficiency ratio information, update the existing energy efficiency ratio information and the corresponding control parameter set; Step 2: Under the premise of optimizing the acquisition function, resample the other control parameter sets other than the updated existing energy efficiency ratio information based on the distribution model, and determine the new energy efficiency ratio information of the server cluster when working under the sampled control parameter set; Repeat step one and step two until a preset iteration condition is reached, determine the target energy efficiency ratio information that meets the preset energy efficiency ratio requirements from the multiple existing energy efficiency ratio information after updating, and use the control parameter set corresponding to the target energy efficiency ratio information as the target control parameter set to control the server cluster to operate under the target control parameter set.

6. A device for optimizing energy efficiency ratio of a server cluster, characterized in that: The device comprises: an acquisition module, configured to acquire a plurality of different control parameter sets of a processor group and energy efficiency ratio information of the server cluster when operating under any of the control parameter sets, wherein the processor group includes a plurality of different types of processors, and each set of the control parameter sets includes the control parameters of each of the processors in the processor group; A construction module, used to construct a distribution model of the distribution relationship between the control parameters of the plurality of processors and the energy efficiency ratio of the server cluster based on the existing energy efficiency ratio information and the corresponding control parameter set, and to construct a collection function for evaluating the quality of the control parameter set based on the distribution model; A sampling module, for sampling the other control parameter sets other than the existing energy efficiency ratio information based on the distribution model under the premise of optimizing the acquisition function, and determining new energy efficiency ratio information when the server cluster works under the sampled control parameter set; A target control module, used to determine the target energy efficiency ratio information that meets the preset energy efficiency ratio requirements from the multiple existing energy efficiency ratio information after the update, and control the server cluster to work under the target control parameter set based on the control parameter set corresponding to the target energy efficiency ratio information; The step of constructing a distribution model of the distribution relationship between the control parameters of the processors and the energy efficiency ratio of the server cluster based on the existing energy efficiency ratio information and the corresponding control parameter set includes: Based on the existing energy efficiency ratio information and the corresponding control parameter set, construct a mean function characterizing the correlation between the plurality of energy efficiency ratio information and a covariance function characterizing the correlation between the control parameter set and the corresponding energy efficiency ratio information; Constructing a distribution model for obtaining a distribution relationship between the control parameters of the plurality of processors and the energy efficiency ratio of the server cluster according to the mean function and the covariance function; The acquisition function comprises a first acquisition function; The acquisition function for evaluating the quality of the control parameter set is constructed based on the distribution model, comprising: Determine the current optimal energy efficiency ratio information from the existing energy efficiency ratio information; Updating the mean function based on the current optimal energy efficiency ratio information and a preset cumulative distribution function to obtain an updated mean function; Updating the covariance function based on a preset probability density function to obtain a first updated covariance function; The updated mean function and the first updated covariance function are superimposed to construct the first acquisition function for evaluating the quality of the control parameter set.

7. An electronic device, characterized in that: The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor implements the method according to any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images