Data Center PUE Diagnosis Method Based on IoT Big Data

By obtaining the temperature distribution curves of hot spot areas of the data center and sub-server side in real time, calculating the data transmission impact coefficient, and optimizing the cooling system, the problem of insufficient micro-level analysis in the improvement of data center energy efficiency is solved, and the energy efficiency improvement and green development of data centers is achieved.

CN118708462BActive Publication Date: 2024-11-12NANJING DEEPCTRLS TECHNOLOGIES CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411197887.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-29
Publication Date
2024-11-12
Estimated Expiration
2044-08-29

AI Technical Summary

Technical Problem

The existing technology lacks micro-level analysis in improving the energy efficiency of data centers, and it is difficult to identify and solve the problems of temperature distribution and cooling efficiency at the micro-level such as server motherboards, resulting in insufficient precision in cooling optimization and insufficient dynamic regulation capabilities.

Method used

By obtaining the temperature distribution curve of the hot spot area of ​​the data center and sub-server side in real time, calculating the data transmission impact coefficient, and calculating the PUE value based on this, identifying energy consumption abnormalities, optimizing the cooling system, and improving energy efficiency.

Benefits of technology

The micro-level refinement analysis and optimization of the data center has been achieved, the PUE value is reduced, the energy efficiency is improved, and the green and sustainable development of the data center has been promoted.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118708462B_ABST
    Figure CN118708462B_ABST
Patent Text Reader

Abstract

A data center PUE diagnosis method based on IoT big data, the method comprising: obtaining temperature distribution curves of multiple hot spots in a data center end and a sub-server end in real time. Obtaining real-time data transmission results between the data center end and the sub-server end, and calculating the data transmission influence coefficients of the data center end and the sub-server end respectively based on the data transmission results combined with the temperature distribution curve. Based on the data transmission influence coefficient, the PUE values ​​of the data center end and the sub-server end are calculated respectively, and the respective PUE values ​​of the data center end and the sub-server end are compared to generate a PUE diagnosis result. The method can perform PUE diagnosis of a data center based on IoT big data, realize refined analysis and optimization of the data center at the micro level, monitor and simulate the temperature distribution of the server motherboard in real time, identify hot spots, improve energy efficiency, and effectively reduce the PUE value.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of Internet of Things big data, and more specifically, relates to a data center PUE diagnosis method based on Internet of Things big data. Background Art

[0002] With the rapid development of information technology, data centers, as core facilities for storing, processing, and transmitting data, are growing in size and number. High energy consumption in data centers is becoming increasingly prominent, a significant factor in operating costs and environmental impacts. Improving data center energy efficiency and reducing its energy consumption have become a global focus. Power Usage Effectiveness (PUE), a key metric for measuring data center energy efficiency, can effectively assess energy efficiency and promote energy efficiency optimization in data centers.

[0003] Currently, data center energy efficiency optimization primarily relies on the following technologies and methods: Overall energy consumption monitoring: This involves monitoring the total energy consumption of the entire data center and IT equipment, calculating the Power Use Effectiveness (PUE) value, and conducting energy efficiency analysis and optimization based on this overall energy consumption data. Cooling system optimization: This involves employing efficient cooling technologies, such as liquid cooling and free cooling, to optimize cooling equipment and strategies and reduce cooling system energy consumption. Data visualization and reporting: This involves using data visualization tools to generate energy efficiency reports that visually display the data center's energy efficiency status and PUE trends.

[0004] Although existing technologies have achieved certain results in improving the energy efficiency of data centers, the following defects still exist: Lack of micro-level analysis: Traditional methods mainly focus on the overall energy consumption data of data centers, ignoring the analysis of temperature distribution and energy efficiency issues at the micro level of server motherboards, making it difficult to identify and solve specific hot spots and low cooling efficiency problems. Cooling optimization is not precise enough: Due to the lack of in-depth analysis of temperature distribution at the micro level, the optimization of the cooling system can often only be carried out at the macro level, making it difficult to achieve precise cooling and maximize energy efficiency. Insufficient dynamic control capabilities: The control strategies of existing intelligent power management systems are relatively fixed, lacking the ability to dynamically adjust according to real-time temperature and energy consumption data, and are difficult to adapt to the complex and changing operating environment of data centers.

[0005] By simulating temperature distribution on motherboards and conducting data center PUE diagnosis based on IoT big data, we can address the shortcomings of existing technologies and achieve refined analysis and optimization of data centers at a microscopic level. Leveraging high-precision temperature sensors and big data analysis, we can monitor and simulate the temperature distribution of server motherboards in real time, identify hotspots, optimize cooling systems, and improve energy efficiency, effectively reducing PUE and achieving green and sustainable data center development. Summary of the Invention

[0006] In order to address the deficiencies in the prior art, the purpose of the present invention is to solve the above-mentioned defects and further propose a data center PUE diagnosis method based on IoT big data.

[0007] The present invention adopts the following technical solutions.

[0008] The first aspect of the present invention discloses a data center PUE diagnosis method based on Internet of Things big data, the method comprising:

[0009] Real-time acquisition of temperature distribution curves of multiple hot spots in the data center and sub-server areas;

[0010] Obtaining real-time data transmission results between the data center and the sub-server, and calculating data transmission influence coefficients of the data center and the sub-server based on the data transmission results and the temperature distribution curve;

[0011] Based on the data transmission impact coefficient, respectively calculating the PUE values ​​of the data center and the sub-server, and comparing the PUE values ​​of the data center and the sub-server to generate a PUE diagnosis result;

[0012] Among them, the data transmission result is the real-time pressure value of the data transmission between the i-th sub-server end and the data center end, and the data transmission impact coefficient is used to characterize the impact of the data transmission between the data center end and the sub-server end on the PUE of the data center.

[0013] Furthermore, the formula for the data transmission result is:

[0014] ;

[0015] ;

[0016] Where, is the data transmission result of the i-th (i≠0) sub-server at time t, is the sum of the data transmission results of i (i≠0) sub-servers, i (i≠0) is the sub-server number, t is any time, is the unit time, is a fixed constant, which represents the ratio between the energy consumed by the server to the energy consumed by receiving data. Unit time From the data center end The total amount of data received by each sub-server, Indicates unit time From the data center to the The total amount of data sent by each sub-server.

[0017] Furthermore, the formula for the data transmission influence coefficient is:

[0018] ;

[0019] Where, is the data transmission result of the i-th (i≠0) sub-server at time t, is the data transmission influence coefficient of the i-th sub-server in the k-th hotspot area at time t, i (i≠0) is the number of the sub-server, t is any time, k is the number of the hotspot area, is the first child server of the i-th (i≠0) at time t The temperature distribution curve corresponding to the hot spot area, when i=0, is the first The temperature distribution curve corresponding to the hot spot area is , For the Normal operating temperature of each hot spot area.

[0020] Furthermore, the PUE values ​​of the data center and the sub-server are calculated based on the data transmission impact coefficient, and the PUE values ​​of the data center and the sub-server are compared to generate a PUE diagnosis result, including:

[0021] Calculate the PUE values ​​of the data center and the sub-server respectively;

[0022] Based on the PUE values ​​of the data center and the sub-server, calculate a ratio between the PUE value of the data center and the sum of the PUE values ​​of the multiple sub-servers, as a basis for generating the PUE diagnosis result;

[0023] The calculation formula for the ratio between the PUE value of the data center and the sum of the PUE values ​​of multiple sub-servers is:

[0024] ;

[0025] Where, is the ratio between the PUE value of the data center and the sum of the PUE values ​​of i (i≠0) sub-servers. is the PUE value of the data center, is the PUE value of the i-th (i≠0) sub-server;

[0026] The calculation formula of the PUE value is:

[0027] ;

[0028] Where, is the PUE value of the i-th (i≠0) sub-server, is the data transmission influence coefficient of the i-th sub-server in the k-th hotspot area at time t, where k is the number of the hotspot area. For the The area of ​​the hotspot area, is the cooling constant.

[0029] Furthermore, the calculating of the PUE values ​​of the data center and the sub-server based on the data transmission impact coefficient, and comparing the PUE values ​​of the data center and the sub-server to generate a PUE diagnosis result further includes:

[0030] Determine whether the ratio between the PUE value of the data center and the sum of the PUE values ​​of multiple sub-servers exceeds a set threshold; if so,

[0031] Determine that the energy consumption of the sub-server is abnormal; if not, then

[0032] It is determined that the energy consumption of the sub-server is normal.

[0033] Furthermore, the method further comprises:

[0034] Establishing a coordinate system with the sub-server end with abnormal energy consumption as the origin, obtaining spatial position information of the volume element, and modeling the temperature field of the sub-server end with abnormal energy consumption by using the Poisson equation or the Laplace equation to construct a multilayer perceptron;

[0035] A loss function is constructed based on the data loss and physical loss of the sub-server end with abnormal energy consumption, and the multi-layer perceptron is trained using the temperature distribution data of multiple locations in the temperature field to obtain a physical information neural network for calculating the temperature data of the sub-server end at any location.

[0036] Furthermore, the services running on the data center side include internal services and data transmission services, and the internal services include triggered services;

[0037] When the data transmission service is running, the trigger service has a lower running priority than the data transmission service.

[0038] A second aspect of the present invention discloses a data center PUE diagnostic device based on IoT big data, the device comprising:

[0039] The temperature curve acquisition module is used to obtain the temperature distribution curves of multiple hot spots in the data center and sub-server in real time;

[0040] An influence coefficient calculation module is used to obtain real-time data transmission results between the data center end and the sub-server end, and calculate the data transmission influence coefficients of the data center end and the sub-server end respectively based on the data transmission results and the temperature distribution curve;

[0041] a PUE diagnosis result generating module, configured to calculate the PUE values ​​of the data center and the sub-server respectively based on the data transmission impact coefficient, and compare the PUE values ​​of the data center and the sub-server to generate a PUE diagnosis result;

[0042] Among them, the data transmission result is the real-time pressure value of the data transmission between the i-th sub-server end and the data center end, and the data transmission impact coefficient is used to characterize the impact of the data transmission between the data center end and the sub-server end on the PUE of the data center.

[0043] A third aspect of the present invention discloses a terminal, comprising a processor and a storage medium, characterized in that:

[0044] The storage medium is used to store instructions;

[0045] The processor is configured to operate according to the instructions to execute the steps of the method of the first aspect.

[0046] A fourth aspect of the present invention discloses a computer-readable storage medium having a computer program stored thereon, wherein the program implements the steps of the method described in the first aspect when executed by a processor.

[0047] The beneficial effects of the present invention are that, compared with the prior art, the present invention has the following advantages:

[0048] By obtaining the temperature distribution curves of multiple hot spots in the data center and sub-server in real time, and obtaining the real-time data transmission results between the data center and sub-server, the data transmission impact coefficients of the data center and sub-server are calculated based on the data transmission results and the temperature distribution curves of the two. Finally, based on the data transmission impact coefficients, the PUE values ​​of the data center and sub-server are calculated respectively, and the PUE values ​​of the data center and sub-server are compared to generate the final PUE diagnosis results. This method can perform data center PUE diagnosis based on IoT big data, realize refined analysis and optimization of the data center at the micro level, monitor and simulate the temperature distribution of the server motherboard in real time, identify hot spots, optimize the cooling system, improve energy efficiency, and effectively reduce the PUE value, thus achieving the green and sustainable development of the data center. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1This is a flow chart of the data center PUE diagnosis method based on IoT big data.

[0050] Figure 2 This is a diagram of the distributed architecture between the data center and sub-server sides. DETAILED DESCRIPTION

[0051] The present application will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present invention and are not intended to limit the scope of protection of the present application.

[0052] like Figure 1 and Figure 2 As shown, in one embodiment, the present invention provides a data center PUE diagnosis method based on IoT big data, comprising the following steps:

[0053] Step S110 , obtaining temperature distribution curves of multiple hot spots in the data center and the sub-server in real time.

[0054] Step S120 , obtaining real-time data transmission results between the data center and the sub-server, and calculating the data transmission influence coefficients of the data center and the sub-server based on the data transmission results and the temperature distribution curve.

[0055] The data transmission result is the real-time pressure value of the data transmission between the i-th sub-server and the data center, and the data transmission impact coefficient is used to characterize the impact of the data transmission between the data center and the sub-server on the PUE of the data center.

[0056] In one embodiment, the formula for data transmission result is:

[0057] ,

[0058] .

[0059] Where, is the data transmission result of the i-th (i≠0) sub-server at time t, is the sum of the data transmission results of i (i≠0) sub-servers, i (i≠0) is the sub-server number, t is any time, is the unit time, is a fixed constant, which represents the ratio between the energy consumed by the server to the energy consumed by receiving data. Unit time From the data center end The total amount of data received by each sub-server, Indicates unit time From the data center to the The total amount of data sent by each sub-server.

[0060] The formula for the data transmission impact coefficient is:

[0061] .

[0062] Where, is the data transmission result of the i-th (i≠0) sub-server at time t, is the data transmission influence coefficient of the i-th sub-server in the k-th hotspot area at time t, i (i≠0) is the number of the sub-server, t is any time, k is the number of the hotspot area, is the first child server of the i-th (i≠0) at time t The temperature distribution curve corresponding to the hot spot area, when i=0, is the first The temperature distribution curve corresponding to the hot spot area is , For the Normal operating temperature of each hot spot area.

[0063] In step S130 , based on the data transmission impact coefficient, the PUE values ​​of the data center and the sub-server are calculated respectively, and the PUE values ​​of the data center and the sub-server are compared to generate a PUE diagnosis result.

[0064] In one embodiment, the present invention provides a data center PUE diagnosis method based on IoT big data. The method calculates the PUE values ​​of the data center and the sub-server based on the data transmission impact coefficient, and compares the PUE values ​​of the data center and the sub-server to generate a PUE diagnosis result. The method specifically includes the following steps:

[0065] Step S131 , respectively calculating the PUE values ​​of the data center side and the sub-server side.

[0066] Step S132 : Based on the PUE values ​​of the data center and the sub-servers, a ratio between the PUE value of the data center and the sum of the PUE values ​​of the multiple sub-servers is calculated as a basis for generating a PUE diagnosis result.

[0067] The calculation formula for the ratio of the PUE value of the data center to the sum of the PUE values ​​of multiple sub-servers is:

[0068] .

[0069] Where, is the ratio between the PUE value of the data center and the sum of the PUE values ​​of i (i≠0) sub-servers. is the PUE value of the data center, is the PUE value of the i-th (i≠0) sub-server.

[0070] The calculation formula for PUE value is:

[0071] .

[0072] Where, is the PUE value of the i-th (i≠0) sub-server, is the data transmission influence coefficient of the i-th sub-server in the k-th hotspot area at time t, where k is the number of the hotspot area. For the The area of ​​the hotspot area, is the cooling constant.

[0073] In one embodiment, the data center PUE diagnosis method based on IoT big data provided by the present invention calculates the PUE values ​​of the data center and the sub-server based on the data transmission impact coefficient, and compares the PUE values ​​of the data center and the sub-server to generate a PUE diagnosis result. Specifically, the method further includes the following steps:

[0074] Step S133: determine whether the ratio between the PUE value of the data center and the sum of the PUE values ​​of multiple sub-servers exceeds a set threshold.

[0075] Step S134: determine if the energy consumption of the sub-server is abnormal. If not, then

[0076] Step S135: determining whether the energy consumption of the sub-server is normal.

[0077] In one embodiment, the data center PUE diagnostic method based on IoT big data provided by the present invention further includes the following steps:

[0078] In step S210, a coordinate system is established with the sub-server end with abnormal energy consumption as the origin, the spatial position information of the volume element is obtained, and the temperature field of the sub-server end with abnormal energy consumption is modeled by the Poisson equation or the Laplace equation to construct a multilayer perceptron.

[0079] In step S220, a loss function is constructed based on the data loss and physical loss of the sub-server end with abnormal energy consumption, and a multi-layer perceptron is trained using the temperature distribution data of multiple locations in the temperature field to obtain a physical information neural network for calculating the temperature data of the sub-server end at any location.

[0080] The above-mentioned data center PUE diagnosis method based on IoT big data obtains the temperature distribution curves of multiple hot spots in the data center and sub-server in real time, and obtains the real-time data transmission results between the data center and sub-server. Subsequently, based on the data transmission results and the temperature distribution curves of the two, the data transmission impact coefficients of the data center and sub-server are calculated respectively. Finally, based on the data transmission impact coefficient, the PUE values ​​of the data center and sub-server are calculated respectively, and the PUE values ​​of the data center and sub-server are compared to generate the final PUE diagnosis results. This method can perform data center PUE diagnosis based on IoT big data, realize refined analysis and optimization of the data center at the micro level, monitor and simulate the temperature distribution of the server motherboard in real time, identify hot spots, optimize the cooling system, improve energy efficiency, and effectively reduce the PUE value, thereby achieving green and sustainable development of the data center.

[0081] In a specific embodiment, the present invention provides a data center PUE diagnostic method based on IoT big data. In PUE diagnosis, common PUE failures include inefficient cooling systems, hot spot problems, inefficient power systems, and inefficient IT equipment energy efficiency. Among them, hot spot problems are the key. The failure phenomenon is manifested in that the temperature of the server motherboard or a specific area is too high, forming a hot spot, resulting in equipment performance degradation or damage, and then causing other PUE failures.

[0082] Considering that internal services consume the same amount of energy over a certain period, and that other triggered services, such as process 3, have lower priority and typically avoid peak data periods, it is necessary to predict and diagnose PUE failures caused by the frequent data transmission and reception at the data center end in IoT big data.

[0083] In this embodiment, the data center PUE diagnosis method based on IoT big data provided by the present invention is applied to the IoT big data scenario, and the process of performing PUE diagnosis during the peak period of data transmission and reception includes steps 1 to 4.

[0084] Step 1: Get the temperature distribution curves of several hot spots in the data center and sub-server in real time, which are and .

[0085] Step 2: Based on real-time data transmission results , and according to the temperature distribution curves of the data center and the sub-server end, the data transmission impact coefficients of the data center and the sub-server end are calculated respectively. The data transmission impact coefficient is used to describe the impact of the data transmission of the sub-server end on the PUE of the data center.

[0086] In some embodiments, the data transmission result Refers to the The real-time pressure value of data transmission between the sub-server and the data center. The formula can be as follows:

[0087] ,

[0088] .

[0089] in, is the data transmission result of the i-th (i≠0) sub-server at time t, is the sum of the data transmission results of i (i≠0) sub-servers, i (i≠0) is the sub-server number, t is any time, is the unit time, It is a fixed constant, which represents the ratio of the energy consumed by the server to the energy consumed by receiving data, usually between 0.3 and 0.6. Indicates unit time From the data center end The total amount of data received by each sub-server, Indicates unit time From the data center to the The total amount of data sent by each sub-server.

[0090] Step 3: Based on the data transmission impact coefficient, calculate the PUE values ​​of the data center and the sub-server respectively.

[0091] Step 4: Compare the PUE values ​​of the data center and the sub-server to generate a PUE diagnosis result.

[0092] The formula for the data transmission impact coefficient is as follows:

[0093] .

[0094] in, is the data transmission result of the i-th (i≠0) sub-server at time t, is the data transmission influence coefficient of the i-th sub-server in the k-th hotspot area at time t, i (i≠0) is the number of the sub-server, t is any time, k is the number of the hotspot area, is the first child server of the i-th (i≠0) at time t The temperature distribution curve corresponding to the hot spot area, when i=0, is the first The temperature distribution curve corresponding to the hot spot area is , For the Normal operating temperature of each hot spot area.

[0095] In step 3, the PUE value can be calculated as follows:

[0096] .

[0097] in, is the PUE value of the i-th (i≠0) sub-server, is the data transmission influence coefficient of the i-th sub-server in the k-th hotspot area at time t, where k is the number of the hotspot area. For the The area of ​​the hotspot area, is the cooling constant.

[0098] In step 4, the ratio of the two can be calculated as the basis for PUE diagnosis results, as shown below:

[0099] .

[0100] Where, is the ratio between the PUE value of the data center and the sum of the PUE values ​​of i (i≠0) sub-servers. is the PUE value of the data center, is the PUE value of the i-th (i≠0) sub-server. In the PUE diagnosis results, you can observe a period of time For example, if If the value is higher than the set threshold, the energy consumption is judged to be abnormal; otherwise, it is judged to be normal.

[0101] It should be understood that the services run by the server terminal can cover two types: one is internal services, and the other is data transmission services. More broadly and more accurately, using the Go language as an example, the difference in time between the two services can be shown in the following code:

[0102] net.NetSubInst.RegSubs(comm.FrameType, comm.Sub{C: mg.C, S:"storage", Mark: 9999})

[0103] dbchange.RegDBChangeNotify(mg.dbchange)

[0104] go func() {

[0105] mg.ticker = time.NewTicker(time.Duration(mg.interval) * time.Second)

[0106] defer mg.ticker.Stop()

[0107] for {

[0108] select {

[0109] case d :=<- mg.C: / / Data sending and receiving service based on websocket

[0110] mg.process1(d)

[0111] case<- mg.ticker.C: / / A timing service within the program

[0112] mg.process2()

[0113] case<- mg.dbchange: / / Database changes

[0114] mg.process3 ()

[0115] case<- mg.done: / / Program terminated, can be ignored

[0116] return

[0117] }

[0118] }

[0119] }()

[0120] It's easy to understand that process 1 and process 2 in the above code belong to the data transmission service and internal service, respectively. However, it's important to note that process 3, while formally identical to the data transmission service and a triggered (rather than scheduled) service, can be classified as an internal service. This is because process 3 has a lower probability of occurring and consumes fewer resources. More importantly, if process 3 occupies resources, it may interfere with other important services. For example, when data transmission and reception services are in use, they often have to compromise (i.e., have a lower priority) than the data transmission and reception services.

[0121] In some embodiments, obtaining the temperature of a certain part of a server terminal or a sub-server at a certain moment can be specifically achieved by:

[0122] There are multiple components distributed on a motherboard. Some components have higher energy consumption and can be understood as a heat source, while some components have lower energy consumption and can be understood as heat conducting media. In addition, the motherboard itself is also a heat conducting medium.

[0123] The temperature distribution on the motherboard can be expressed using a function. First, a coordinate system is established with the center of the heat source as the origin. Then, when the position is The temperature of the volume element can be expressed by formula (1):

[0124]

[0125] In formula (1), T p is the temperature of the volume element p, T represents the temperature, x, y represent the position of the volume element, and t represents the time. The temperature field of the motherboard can be modeled using the Poisson equation, as shown in Equation (2):

[0126]

[0127] In formula (2), is the temperature field, is the heat source density (i.e. heating power, unit is W), is the thermal conductivity, represents the Laplace operator. On a motherboard, the power consumption of each component can be considered as the heat source density, while the thermal conductivity depends on the thermal conductivity of the motherboard material.

[0128] If the volume element p is located outside the heat source, the Poisson equation can be simplified to the Laplace equation, that is, the right side of equation (2) is equal to 0, as shown in equation (3):

[0129]

[0130] In order to solve Equation (2) and obtain the temperature distribution on the motherboard, the physical information neural network (PINNs) method can be used. PINNs is a method that combines physical equations and neural networks. By embedding physical constraints in the neural network, partial differential equations can be solved efficiently.

[0131] Step 1: Define mathematical model and physical parameters

[0132] The mathematical model can be described by equations (2) and (3), which will not be repeated here. The physical parameters in the model include: heat source heating power q. When defining this parameter, it is necessary to refer to the data sheet of each component to obtain its average or maximum power consumption value under different workloads. Here, it is set to 5W / cm 2 Thermal conductivity refers to the rate of heat transfer per unit area. Assume that the thermal conductivity of component 1 is 5W / K / cm 2, the thermal conductivity of component 2 is 7W / K / cm 2 , the thermal conductivity of the motherboard is 10W / K / cm 2 .

[0133] Boundary conditions define the temperature or heat flow at the system's boundaries and are crucial for solving heat conduction problems. In the server motherboard model, the boundary conditions can be assumed to be constant edge temperature. With the ambient temperature set at 26°C, the motherboard edge temperature is locked at 26°C.

[0134] Step 2: Design PINNs architecture

[0135] To solve Equations (2) and (3), we need to build PINNs. Traditional numerical integration can solve Equations (2) and (3), but it can be challenging when dealing with complex boundary conditions and nonlinear problems. Physically Informed Neural Networks (PINNs) provide a new solution. The following describes how to design a PINN architecture, which primarily consists of a neural network and a loss function.

[0136] First, build a neural network, specifically a multilayer perceptron (MLP). An MLP consists of multiple neural layers (e.g., four layers), each with multiple nodes. Because the problem involves two-dimensional spatial coordinates (x, y) and time t, the input layer should have three nodes, one for each input variable. The output layer should have only one node, because the prediction is for a single scalar value, namely, the temperature T.

[0137] The number of layers and nodes in the middle layer is usually determined by the complexity of the problem and experiments. A common starting point is to set 1 to 3 middle layers. Taking setting 2 middle layers as an example, the ReLU (Rectified Linear Unit) function is selected as the activation function, which can effectively alleviate the gradient vanishing problem. Sigmoid, tanh, or other activation functions can also be used.

[0138] Secondly, define the loss function. After building the MLP, it is necessary to integrate physical laws into the neural network. The specific method is to construct the loss function. This step is the key difference between PINNs and ordinary deep neural networks.

[0139] The loss function consists of two parts, one is data loss and the other is physical loss.

[0140]

[0141] In formula (4), is the total loss function, and are coefficients, which are 0.4 and 0.6 respectively in this model. Indicates data loss, Indicates physical loss.

[0142] The following describes the calculation methods for data loss and physical loss respectively:

[0143] The data loss part is consistent with that of ordinary deep neural networks, which means using existing data to correct the neural network. Its calculation method is to subtract the measured data (i.e. the temperature of each position of the motherboard, historical data or laboratory data) from the neural network output, and then take the root mean square value of all data, that is, , rms is the root mean square function, T o is the temperature value in the training set (explained in step 3), T PINNs is the output of the neural network.

[0144] The physical loss part is calculated by the following formula:

[0145]

[0146] In formula (5), Indicates physical loss, is the temperature field, is the heat source density (i.e. heating power, unit is W), is the thermal conductivity, represents the Laplace operator, x and y are the position of the volume element, and t represents time. Different heat source density values ​​q and thermal conductivity k are used when the coordinates are located on different components (for example, on a heat source or on the motherboard). This residual R is used to construct the loss function. Ideally, R is zero because the solution provided by the neural network must satisfy the original heat conduction equation. Therefore, the root mean square (RMS) value of this residual can be equated with the physical loss, ensuring that the neural network output satisfies the heat conduction equation as closely as possible.

[0147] The code to define the loss function is as follows:

[0148] # Define the loss function, named lossFunc

[0149] def lossFunc(model, inputs, targets):

[0150] # Calculate model predictions

[0151] predictions = model(inputs)

[0152] # Calculate physical loss and data loss

[0153] physics_loss = ... # Calculate according to formula (4)

[0154] data_loss = rms(model.predict(inputs) – targets) # Take the residual root mean square value

[0155] # The total loss is the sum of the two

[0156] total_loss = 0.6*physics_loss + 0.4*data_loss

[0157] return total_loss

[0158] In the input, model is the name of the neural network model, inputs is the location data on the server motherboard, covering the coordinates of every volume element on the motherboard, especially those covered by components, and targets is the measured temperature data. After establishing the MLP and defining the loss function, you can begin the neural network training process.

[0159] Step 3: Train the neural network

[0160] The PINNs training process is basically the same as that of ordinary deep neural networks. The specific training process is as follows:

[0161] First, we need to obtain data for training the neural network. For example, we can divide the motherboard into grids and collect a data point in each grid. A data point is represented by a vector [x, y, t, T, q, k], where x, y, and t are the location and time of the point, T is the temperature, and q is the heat source density at the point. If the point is on the heat source, then q = 5W / cm 2 , otherwise q = 0. K represents the thermal conductivity, which has been explained in step 1 and will not be repeated here. Through extensive experiments, the training data required for training PINNs can be obtained. The data is divided into a training set and a test set. The training set is used for training the neural network, and the test set is used for evaluating the neural network.

[0162] During training, the Adam optimizer is used, and the number of training epochs can be set to 1000. After training is complete, the PINNs model can be used to simulate the temperature distribution of the server motherboard. Assuming the location to be calculated is p(x, y) and the time is t, the PINNs result is: T = model.predict((x, y, t)). In this result, T represents the temperature, model represents the model, predict represents the prediction, x and y represent the location of the volume element, and t represents the time, which is the predicted temperature of the model at the corresponding volume element location.

[0163] The data center PUE diagnostic device based on IoT big data provided by the present invention is described below. The data center PUE diagnostic device based on IoT big data described below and the data center PUE diagnostic method based on IoT big data described above can be referenced to each other.

[0164] In one embodiment, the present invention provides a data center PUE diagnostic device based on IoT big data, including a temperature curve acquisition module, an influence coefficient calculation module, and a PUE diagnostic result generation module.

[0165] The temperature curve acquisition module is used to obtain the temperature distribution curves of multiple hot spots in the data center and sub-server in real time.

[0166] The influence coefficient calculation module is used to obtain the real-time data transmission results between the data center and the sub-server, and calculate the data transmission influence coefficients of the data center and the sub-server based on the data transmission results and the temperature distribution curve.

[0167] The PUE diagnosis result generation module is used to calculate the PUE values ​​of the data center and the sub-server respectively based on the data transmission impact coefficient, and compare the PUE values ​​of the data center and the sub-server to generate a PUE diagnosis result.

[0168] The data transmission result is the real-time pressure value of the data transmission between the i-th sub-server and the data center, and the data transmission impact coefficient is used to characterize the impact of the data transmission between the data center and the sub-server on the PUE of the data center.

[0169] In this embodiment, the data center PUE diagnostic device based on IoT big data provided by the present invention has a data transmission result formula as follows:

[0170] ,

[0171] .

[0172] Where, is the data transmission result of the i-th (i≠0) sub-server at time t, is the sum of the data transmission results of i (i≠0) sub-servers, i (i≠0) is the sub-server number, t is any time, is the unit time, is a fixed constant, which represents the ratio between the energy consumed by the server to the energy consumed by receiving data. Unit time From the data center end The total amount of data received by each sub-server, Indicates unit time From the data center to the The total amount of data sent by each sub-server.

[0173] In this embodiment, the data center PUE diagnostic device based on IoT big data provided by the present invention has a data transmission impact coefficient formula as follows:

[0174] .

[0175] Where, is the data transmission result of the i-th (i≠0) sub-server at time t, is the data transmission influence coefficient of the i-th sub-server in the k-th hotspot area at time t, i (i≠0) is the number of the sub-server, t is any time, k is the number of the hotspot area, is the first child server of the i-th (i≠0) at time t The temperature distribution curve corresponding to the hot spot area, when i=0, is the first The temperature distribution curve corresponding to the hot spot area is , For the Normal operating temperature of each hot spot area.

[0176] In this embodiment, the data center PUE diagnostic device based on IoT big data provided by the present invention, the PUE diagnostic result generation module is specifically used to:

[0177] Calculate the PUE values ​​of the data center and sub-server respectively.

[0178] Based on the PUE values ​​of the data center and the sub-servers, the ratio between the PUE value of the data center and the sum of the PUE values ​​of multiple sub-servers is calculated as the basis for generating the PUE diagnosis result.

[0179] The calculation formula for the ratio of the PUE value of the data center to the sum of the PUE values ​​of multiple sub-servers is:

[0180] .

[0181] Where, is the ratio between the PUE value of the data center and the sum of the PUE values ​​of i (i≠0) sub-servers. is the PUE value of the data center, is the PUE value of the i-th (i≠0) sub-server.

[0182] The calculation formula for PUE value is:

[0183] ;

[0184] Where, is the PUE value of the i-th (i≠0) sub-server, is the data transmission influence coefficient of the i-th sub-server in the k-th hotspot area at time t, where k is the number of the hotspot area. For the The area of ​​the hotspot area, is the cooling constant.

[0185] In this embodiment, the data center PUE diagnostic device based on IoT big data provided by the present invention, the PUE diagnostic result generation module is further configured to:

[0186] Determine whether the ratio between the PUE value of the data center and the sum of the PUE values ​​of multiple sub-servers exceeds the set threshold. If so,

[0187] Determine if the energy consumption of the sub-server is abnormal. If not, then

[0188] Determine that the energy consumption of the sub-server is normal.

[0189] In this embodiment, the data center PUE diagnostic device based on IoT big data provided by the present invention further includes a physical information network model construction module for:

[0190] A coordinate system is established with the sub-server end with abnormal energy consumption as the origin to obtain the spatial position information of the volume element, and the temperature field of the sub-server end with abnormal energy consumption is modeled through the Poisson equation or the Laplace equation to construct a multi-layer perceptron.

[0191] A loss function is constructed based on the data loss and physical loss of the sub-server end with abnormal energy consumption, and the multi-layer perceptron is trained with the temperature distribution data of multiple locations in the temperature field to obtain a physical information neural network for calculating the temperature data of the sub-server end at any location.

[0192] In this embodiment, the present invention provides a data center PUE diagnostic device based on IoT big data, and the services running on the data center end include internal services and data transmission services, and the internal services include triggered services.

[0193] When the data transmission service is running, the trigger service runs at a lower priority than the data transmission service.

[0194] The applicant of the present invention has made a detailed explanation and description of the implementation examples of the present invention in conjunction with the drawings in the specification. However, those skilled in the art should understand that the above implementation examples are only preferred implementation plans of the present invention, and the detailed description is only to help readers better understand the spirit of the present invention, and is not a limitation on the scope of protection of the present invention. On the contrary, any improvements or modifications based on the inventive spirit of the present invention should fall within the scope of protection of the present invention.

Claims

1. A data center PUE diagnosis method based on IoT big data, characterized in that: The method comprises: Obtain temperature distribution curves of multiple hot spots in the data center and sub-server in real time; Acquire the real-time data transmission results between the data center and the sub-server, and calculate the data transmission influence coefficients of the data center and the sub-server respectively based on the data transmission results and the temperature distribution curve; Based on the data transmission impact coefficient, respectively calculating the PUE values ​​of the data center end and the sub-server end, and comparing the respective PUE values ​​of the data center end and the sub-server end to generate a PUE diagnosis result; The data transmission result is the real-time pressure value of the data transmission between the ith sub-server and the data center, and the data transmission impact coefficient is used to characterize the impact of the data transmission between the data center and the sub-server on the PUE of the data center; The formula for the data transmission influence coefficient is: ; In the formula, is the data transmission result of the i-th (i≠0) sub-server at time t, is the data transmission influence coefficient of the i-th sub-server in the k-th hotspot area at time t, i (i≠0) is the number of the sub-server, t is any time, k is the number of the hotspot area, is the i-th (i≠0) sub-server at time t The temperature distribution curve corresponding to the hot spot area, when i=0, is the first The temperature distribution curve corresponding to the hot spot area is: , For the Normal operating temperature of a hot spot area.

2. The data center PUE diagnosis method based on IoT big data according to claim 1 is characterized in that: The formula for the data transmission result is: ; ; In the formula, is the data transmission result of the i-th (i≠0) sub-server at time t, is the sum of the data transmission results of i (i≠0) sub-servers, i (i≠0) is the sub-server number, t is any time, is the unit time, is a fixed constant, which represents the ratio between the energy consumption of sending data and receiving data by the server. Unit time From the data center end The total amount of data received by each sub-server. Indicates unit time From the data center to the The total amount of data sent by each sub-server.

3. The data center PUE diagnosis method based on IoT big data according to claim 1 is characterized in that: The calculating the PUE values ​​of the data center end and the sub-server end respectively based on the data transmission impact coefficient, and comparing the PUE values ​​of the data center end and the sub-server end to generate a PUE diagnosis result, includes: Calculate the PUE values ​​of the data center and the sub-server respectively; Based on the PUE values ​​of the data center and the sub-server, a ratio between the PUE value of the data center and the sum of the PUE values ​​of the multiple sub-servers is calculated as a basis for generating the PUE diagnosis result; The calculation formula for the ratio between the PUE value of the data center and the sum of the PUE values ​​of multiple sub-servers is: ; In the formula, is the ratio between the PUE value of the data center and the sum of the PUE values ​​of the i (i≠0) sub-servers. is the PUE value of the data center, is the PUE value of the i-th (i≠0) sub-server; The calculation formula of the PUE value is: ; In the formula, is the PUE value of the i-th (i≠0) sub-server, is the data transmission influence coefficient of the ith sub-server in the kth hotspot area at time t, where k is the number of the hotspot area. For the The area of ​​the hotspot, is the cooling constant.

4. The data center PUE diagnosis method based on IoT big data according to claim 3 is characterized in that: The method further comprises: calculating the PUE values ​​of the data center and the sub-server based on the data transmission impact coefficient, and comparing the PUE values ​​of the data center and the sub-server to generate a PUE diagnosis result. Determine whether the ratio between the PUE value of the data center and the sum of the PUE values ​​of multiple sub-servers exceeds a set threshold; if so, Determine that the energy consumption of the sub-server is abnormal; if not, then It is determined that the energy consumption of the sub-server is normal.

5. The data center PUE diagnosis method based on IoT big data according to claim 4 is characterized in that: The method further comprises: A coordinate system is established with the sub-server end with abnormal energy consumption as the origin, the spatial position information of the volume element is obtained, and the temperature field of the sub-server end with abnormal energy consumption is modeled by Poisson's equation or Laplace's equation to construct a multi-layer perceptron; A loss function is constructed based on the data loss and physical loss of the sub-server end with abnormal energy consumption, and the multi-layer perceptron is trained through the temperature distribution data of multiple positions in the temperature field to obtain a physical information neural network for calculating the temperature data of the sub-server end at any position.

6. The data center PUE diagnosis method based on IoT big data according to any one of claims 1 to 5, characterized in that: The services running on the data center end include internal services and data transmission services, and the internal services include triggered services; When the data transmission service is running, the priority of the trigger service is lower than that of the data transmission service.

7. A data center PUE diagnostic device based on IoT big data, used to execute any of the methods described in claims 1-6, characterized in that: The device comprises: The temperature curve acquisition module is used to obtain the temperature distribution curves of multiple hot spots in the data center and the sub-server in real time; An influence coefficient calculation module, used to obtain the real-time data transmission results between the data center end and the sub-server end, and calculate the data transmission influence coefficients of the data center end and the sub-server end respectively based on the data transmission results combined with the temperature distribution curve; A PUE diagnosis result generating module, configured to calculate the PUE values ​​of the data center end and the sub-server end respectively based on the data transmission influence coefficient, and compare the respective PUE values ​​of the data center end and the sub-server end to generate a PUE diagnosis result; Among them, the data transmission result is the real-time pressure value of the data transmission between the i-th sub-server end and the data center end, and the data transmission impact coefficient is used to characterize the impact of the data transmission between the data center end and the sub-server end on the PUE of the data center.

8. A terminal comprising a processor and a storage medium; characterized in that: The storage medium is used to store instructions; The processor is configured to operate according to the instructions to execute the steps of the method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.