A data center AI energy consumption analysis and optimization method and system

Abstract

The present invention relates to the field of data center technology, specifically a data center AI energy consumption analysis optimization method and system, including the following steps: obtaining data center status information, collecting CPU occupancy, memory usage and disk I / O frequency, monitoring computer room temperature, humidity and cabinet energy consumption, identifying servers with abnormal power fluctuations, abnormal loads and abnormal temperature changes, and establishing an energy consumption monitoring data set. The present invention, by collecting key operating parameters of the data center and implementing data cleaning and abnormal point elimination, enhances the data center's monitoring capabilities for power fluctuations, abnormal loads and temperature changes, making the analysis of energy consumption more refined and more targeted, and directly comparing and analyzing with energy efficiency benchmarks, can quickly identify equipment with low operating efficiency, and quickly take measures to optimize and adjust, and improve the energy efficiency management level of the data center through intelligent data processing and real-time optimization feedback.

Description

Technical Field

[0001] The present invention relates to the field of data center technology, and in particular to a data center AI energy consumption analysis and optimization method and system. Background Art

[0002] A data center is a globally collaborative network of specific devices used to transmit, accelerate, display, calculate and store data information on the Internet network infrastructure. It ensures efficient processing and secure storage of data while guaranteeing the performance of servers and storage devices. It covers multiple aspects from physical infrastructure (such as buildings, power supply, cooling systems) to software-defined networks and virtualization technologies.

[0003] Among them, the AI ​​energy consumption analysis and optimization method of data centers refers to the use of artificial intelligence technology to analyze and optimize the energy consumption of data centers, usually including the use of machine learning algorithms to predict and manage the energy needs of data centers, as well as optimize the operating status of their cooling systems and servers, thereby reducing overall energy usage. In addition, AI energy consumption analysis can help automatically adjust the operation of data centers, such as adjusting cooling systems and server loads to achieve optimal energy efficiency, thereby reducing energy costs and reducing the environmental impact of data centers. It is one of the key technologies to promote the development of data centers.

[0004] Although existing technologies provide multi-dimensional management and operation, from physical foundations to virtualization technology, they are still insufficient in real-time data analysis and instant optimization and adjustment. Current technologies mostly rely on preset management strategies and lack dynamic learning and adaptation of real-time operation data, which makes data centers slow to respond to emergencies or unusual load fluctuations and unable to effectively adapt to environmental changes, resulting in inefficient energy utilization and even waste of resources or equipment loss. These problems not only increase operating costs, but also limit the potential of data centers in energy management and environmental adaptability. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings existing in the prior art and propose a data center AI energy consumption analysis and optimization method and system.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: a data center AI energy consumption analysis and optimization method, comprising the following steps: S1: Obtain data center status information, collect CPU occupancy, memory usage, and disk I / O frequency, monitor computer room temperature, humidity, and cabinet energy consumption, identify servers with abnormal power fluctuations, abnormal loads, and abnormal temperature changes, and establish an energy consumption monitoring data set; S2: calling the energy consumption monitoring data set, calculating task power consumption, task load power ratio and server resource utilization, comparing the energy efficiency benchmark value, and obtaining energy efficiency deviation analysis results; S3: Based on the energy efficiency deviation analysis result, identify server devices with abnormal power fluctuation, load imbalance and temperature control abnormality, associate the abnormality type with the server device number, and obtain abnormal energy consumption classification indicators; S4: Based on the abnormal energy consumption classification index, analyze the impact of task migration cost and load balancing adjustment, evaluate the cooling power optimization point, screen the optimal direction of energy consumption efficiency, and obtain energy efficiency optimization direction information; S5: Based on the energy efficiency optimization direction information, adjust the CPU operating frequency, task load balancing and computer room cooling efficiency, analyze the adjusted energy efficiency utilization level and temperature and humidity change trend, evaluate the adjustment effect of the total power consumption, and obtain the energy consumption management optimization result.

[0007] The improvements of the present invention are that the energy consumption monitoring data set includes power load data, fan speed records, and power module input parameters; the energy efficiency deviation analysis results include a list of high-power consumption servers and a list of inefficient resource utilization servers; the abnormal energy consumption classification indicators include abnormal power fluctuation data, abnormal load fluctuation data, and abnormal temperature change data; the energy efficiency optimization direction information includes benefit impact analysis results, load adjustment direction, and cooling optimization points; the energy consumption management optimization results include energy efficiency improvement information, temperature and humidity change analysis results, and total power adjustment effect.

[0008] The present invention is improved in that the steps of establishing the energy consumption monitoring data set are specifically as follows: S111: Obtain data center status information, collect the CPU occupancy rate, memory usage and disk I / O frequency of the server, synchronously monitor the temperature, humidity and cabinet energy consumption of the computer room, and clean the collected data to remove invalid and abnormal data to obtain a server monitoring data set; S112: Based on the server monitoring data set, analyze the power load, fan speed, and power module input power of the server, determine the fluctuation range of each parameter, and use the formula: ; Calculate the server power fluctuation degree, construct the server load fluctuation curve, and obtain the server power load fluctuation data, where: Represents the server power fluctuation coefficient, Representative The power load of the server, represents the average power load of the server group, Represents the number of servers; S113: Call the server power load fluctuation data, combine the changes in fan speed and power module input power, calculate the power fluctuation rate of the server at multiple time nodes, filter out server devices with excessive fluctuation rates, and combine the room temperature change trend to analyze the impact of temperature anomalies and establish an energy consumption monitoring data set.

[0009] The present invention is improved in that the steps of obtaining the energy efficiency deviation analysis result are specifically as follows: S211: calling the energy consumption monitoring data set, calculating the task power consumption, task load power ratio and server resource utilization of the server, obtaining the energy consumption distribution of multiple servers under different task load conditions, and obtaining the server task energy consumption information; S212: Based on the server task energy consumption information, compare and analyze with the energy efficiency benchmark value, calculate the deviation of task power consumption, task load power ratio and server resource utilization, using the formula: ; Get the energy efficiency deviation of the server ,in, Representative The task power consumption of each server, Representative Resource utilization of each server, Representative Energy efficiency benchmark values ​​for servers, Represents the number of servers; S213: Call the energy efficiency deviation degree of the server, screen the servers whose energy efficiency deviation exceeds the threshold, determine whether their task power consumption and resource utilization are in an abnormal range, and mark the servers with high task power consumption or low resource utilization to obtain energy efficiency deviation analysis results.

[0010] The present invention is improved in that the steps of obtaining the abnormal energy consumption classification index are specifically as follows: S311: Based on the energy efficiency deviation analysis result, extract the historical power consumption, task load and ambient temperature data of the server, calculate the power consumption change rate, load fluctuation amplitude and ambient temperature gradient change rate in the difference time period, and obtain the server energy consumption fluctuation parameter; S312: Based on the server energy consumption fluctuation parameters, determine whether the power consumption change rate, load fluctuation amplitude and temperature gradient change rate of the server exceed the threshold value, identify servers with abnormal power fluctuation, load imbalance and abnormal temperature control, and use the formula: ; Calculate the abnormal energy consumption deviation value of the server , get the abnormal energy consumption classification data of the server, where Representative The power consumption change rate of the server, Representative The load fluctuation of the server, Representative The ambient temperature gradient change rate of each server, Represents the number of servers; S313: Based on the abnormal energy consumption classification data of the server, the server devices with abnormal power fluctuation, load imbalance and temperature control abnormality are classified according to the abnormal category associated with the server number to obtain the abnormal energy consumption classification index.

[0011] The present invention is improved in that the step of acquiring the energy efficiency optimization direction information is specifically as follows: S411: Based on the abnormal energy consumption classification index, extract the server task migration cost, load balancing adjustment impact value and computer room cooling power data, analyze the energy consumption required for task migration, the energy consumption change caused by load balancing adjustment and the optimization potential point of computer room cooling power, and obtain the energy consumption adjustment impact parameter; S412: Based on the energy consumption adjustment influencing parameters, the energy consumption benefit ratio of the adjustment direction is calculated, and the direction with the best energy consumption benefit ratio is selected, using the formula: ; Get the optimal energy efficiency adjustment direction, where represents the energy efficiency ratio, Representative The energy consumption of task migration of each server, Represents the load balancing adjustment impact value, represents the migration cost, Represents the optimization impact of the cooling power in the equipment room, Represents the number of servers to which the task is migrated. Represents the number of servers adjusted for cooling optimization in the computer room; S413: calling the optimal energy efficiency adjustment direction, combining the server task scheduling rules and the computer room cooling adjustment plan, identifying the adjustment path that matches the current energy consumption optimization target, and obtaining energy efficiency optimization direction information.

[0012] The present invention is improved in that the steps of obtaining the energy consumption management optimization result are specifically as follows: S511: Based on the energy efficiency optimization direction information, adjust the CPU operating frequency of the server, collect the current load level, task execution priority and power consumption trend of the server, calculate the energy consumption change after the CPU frequency is adjusted, and obtain the CPU frequency optimization value; S512: Based on the CPU frequency optimization value, calculate the energy consumption change under the task load balance distribution, select the optimal load distribution mode, and use the formula: ; Get the load balancing optimization result, where: Represents the change in energy consumption under load adjustment. Representative The load on the server, Represents the CPU frequency after the server is adjusted. Represents the maximum computing capacity of the server. represents the server power consumption before adjustment, Represents the adjustment cycle duration, Represents the number of servers; S513: calling the load balancing optimization result, analyzing the temperature and humidity change trend after the cooling efficiency of the computer room is adjusted, evaluating the adjustment effect of the total power consumption, and obtaining the energy consumption management optimization result.

[0013] A data center AI energy consumption analysis and optimization system, the system comprising: The status monitoring module obtains data center status information, collects server CPU occupancy, memory usage, and disk I / O frequency, and simultaneously monitors the room temperature, humidity, and cabinet energy consumption, identifies server devices with severe power fluctuations, abnormal load fluctuations, and abnormal temperature changes, and establishes an energy consumption monitoring data set; The deviation analysis module calls the energy consumption monitoring data set, calculates the task power consumption, task load power ratio and server resource utilization of the server, compares the energy efficiency benchmark value, and obtains the energy efficiency deviation analysis result; The abnormal classification module identifies server devices with abnormal power fluctuation, load imbalance and temperature control based on the energy efficiency deviation analysis result, and associates the abnormal type with the server device number to obtain an abnormal energy consumption classification index; Based on the abnormal energy consumption classification index, the direction optimization module analyzes the impact value of server task migration cost and load balancing adjustment, evaluates the optimization point of the cooling power of the computer room, selects the adjustment direction with the best energy efficiency ratio, and obtains energy efficiency optimization direction information; Based on the energy efficiency optimization direction information, the management adjustment module adjusts the server CPU operating frequency, task load balancing distribution and computer room cooling efficiency, analyzes the adjusted server energy efficiency utilization level and computer room temperature and humidity change trend, and evaluates the adjustment effect of total power consumption to obtain energy consumption management optimization results.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, by collecting key operating parameters of the data center and implementing data cleaning and outlier elimination, the data center's monitoring capabilities for power fluctuations, load anomalies and temperature changes are enhanced, making the analysis of energy consumption more refined and targeted. Direct comparison and analysis with energy efficiency benchmarks can quickly identify equipment with low operating efficiency and quickly take measures to optimize and adjust. Through intelligent data processing and real-time optimization feedback, the energy efficiency management level of the data center is improved, achieving a dual reduction in cost savings and environmental impact, and improving the automated management and energy efficiency ratio of the data center. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 A flowchart of a data center AI energy consumption analysis and optimization method is proposed for the present invention; Figure 2 A flow chart for establishing an energy consumption monitoring data set in the present invention; Figure 3 A flowchart for obtaining the energy efficiency deviation analysis result in the present invention; Figure 4 This is a flow chart for obtaining abnormal energy consumption classification indicators in the present invention; Figure 5 A flowchart for obtaining energy efficiency optimization direction information in the present invention; Figure 6 This is a flow chart for obtaining the energy consumption management optimization results in the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention 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 invention and are not intended to limit the present invention.

[0017] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating positions or positional relationships, are based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined.

[0018] Example See also Figure 1 The present invention provides a technical solution: a data center AI energy consumption analysis and optimization method, comprising the following steps: S1: Obtain data center status information, collect server CPU occupancy, memory usage, and disk I / O frequency, simultaneously monitor room temperature, humidity, and cabinet energy consumption, clean data and remove abnormal points, identify server devices with severe power fluctuations, abnormal load fluctuations, and abnormal temperature changes based on server power load, fan speed, and power module input power, and establish an energy consumption monitoring data set; S2: Call the energy consumption monitoring data set to calculate the task power consumption, task load power ratio and server resource utilization of the server, compare and analyze the calculation results with the energy efficiency benchmark value, identify the server with high task power consumption or low resource utilization, and obtain the energy efficiency deviation analysis result; S3: Based on the results of energy efficiency deviation analysis, the server power consumption change rate, load fluctuation amplitude and ambient temperature gradient change rate are calculated to identify server devices with abnormal power fluctuation, load imbalance and temperature control abnormality, and associate the abnormal type with the server device number to obtain abnormal energy consumption classification indicators; S4: Based on the abnormal energy consumption classification indicators, analyze the impact values ​​that affect the server task migration cost and load balancing adjustment, evaluate the potential points for optimizing the cooling power of the computer room, and select the adjustment direction with the best energy efficiency ratio to obtain energy efficiency optimization direction information; S5: Based on the energy efficiency optimization direction information, adjust the server CPU operating frequency, task load balancing distribution and computer room cooling efficiency, analyze the adjusted server energy efficiency utilization level and computer room temperature and humidity change trend, and evaluate the adjustment effect of total power consumption to obtain energy consumption management optimization results.

[0019] The energy consumption monitoring data set includes power load data, fan speed records, and power module input parameters. The energy efficiency deviation analysis results include a list of high-power consumption servers and a list of inefficient resource utilization servers. Abnormal energy consumption classification indicators include abnormal power fluctuation data, abnormal load fluctuation data, and abnormal temperature change data. Energy efficiency optimization direction information includes benefit impact analysis results, load adjustment direction, and cooling optimization points. Energy consumption management optimization results include energy efficiency improvement information, temperature and humidity change analysis results, and total power adjustment effects.

[0020] See also Figure 2 ,The specific steps for establishing the energy consumption monitoring data set are: S111: Obtain data center status information, collect the CPU occupancy rate, memory usage and disk I / O frequency of the server, synchronously monitor the temperature, humidity and cabinet energy consumption of the computer room, and clean the collected data to remove invalid and abnormal data to obtain a server monitoring data set; Get the CPU usage of the server, call the CPU usage time data recorded by the server management, and calculate the CPU usage as the ratio of CPU busy time to total time per unit time. For example, if the total CPU busy time of a server is 6 minutes within 10 minutes, the CPU usage is calculated as follows: For memory usage, call the server memory management module to obtain the current memory usage and total memory capacity. The calculation formula is: , if the total memory of a server is 128GB and 96GB is currently used, then the memory usage is The disk I / O frequency is calculated by counting the number of disk reads / writes per unit time. For example, if the number of disk I / Os is 250 within 5 minutes, the disk I / O frequency is times / minute. The room temperature, humidity and cabinet energy consumption are collected through environmental sensors. The temperature unit is Celsius (℃), the humidity is relative humidity percentage (RH%), and the cabinet energy consumption is expressed in watt-hours (Wh). Examples of temperature are 22.5℃, humidity are 45%RH, and cabinet energy consumption are 2000Wh. All collected data are screened and cleaned to remove missing values ​​and abnormal values. For example, if the temperature sensor misreads 80℃, the data can be removed. According to the principles of data integrity and stability, values ​​with excessive deviations in the time series are screened out. For example, if the disk I / O data jumps to 10,000 times / minute in a short period of time, it is considered to be abnormal data. This ensures the stability and availability of the data and obtains the data center server monitoring data set.

[0021] S112: Based on the server monitoring data set, analyze the server's power load, fan speed, and power module input power, and determine the fluctuation range of each parameter using the formula: ; Calculate the server power fluctuation degree, construct the server load fluctuation curve, and obtain the server power load fluctuation data, where: Represents the server power fluctuation coefficient, Representative The power load of the server, represents the average power load of the server group, Represents the number of servers; Calculate the server's power load, fan speed, and power module input power. The power load is measured by the actual power consumption of the server. The power is calculated based on the server current and voltage. For example, if the operating voltage of a server is 220V and the current is 2.5A, the power load is calculated as The fan speed is measured by the server's built-in sensor in revolutions per minute (RPM). If a fan's measured value is 2800RPM, the fan speed is 2800 RPM. The power module input power is recorded and obtained through the PDU (power distribution unit) in watts (W). To calculate the power fluctuation, first calculate the average power load of each server. The formula is: , if there are three servers with power of 550W, 600W and 580W respectively, then , calculate the power deviation of each server , respectively , and , and then calculate the power fluctuation coefficient: ; The result shows that the server power fluctuation coefficient is 17.78W. If this value exceeds the preset power fluctuation threshold (such as 20W), it indicates that there is obvious power fluctuation in the server group and further analysis of the cause of the load abnormality is required. If the value is lower than the threshold, it means that the server group is running stably.

[0022] S113: calling the server power load fluctuation data, combining the changes in the fan speed and the input power of the power module, calculating the power fluctuation rate of the server at multiple time nodes, screening the server devices with excessive fluctuation rates, and combining the room temperature change trend to analyze the impact of temperature anomalies, and establishing an energy consumption monitoring data set; Combined with the changes in fan speed and power module input power, calculate the power fluctuation rate of the server at each time node. The power fluctuation rate is defined as the change in power load per unit time. For example, if the power load of a server changes from 550 W to 580 W within 1 hour, the power fluctuation rate is calculated as follows: , filter out server devices whose fluctuation rate exceeds the set threshold. The threshold is set based on the server type and historical power fluctuation data. For example, if the data server sets the threshold of 50W / h and the calculated value is 30W / h, it does not exceed the threshold. If the calculated value is 60W / h, it exceeds the threshold range. Combined with the temperature change trend of the computer room, analyze the impact of temperature anomalies. First, count the average temperature around each server. For example, the temperatures of 4 servers in a certain area are 22.5℃, 23.0℃, 22.8℃ and 23.2℃ respectively. The average temperature is calculated as follows: ℃, if the temperature at the location of a server is measured to be 28℃, which is 5.12℃ higher than the average temperature, it is determined to be a temperature anomaly point and an energy consumption monitoring data set is established.

[0023] See also Figure 3 , the specific steps for obtaining the energy efficiency deviation analysis results are as follows: S211: calling the energy consumption monitoring data set, calculating the task power consumption, task load power ratio and server resource utilization of the server, obtaining the energy consumption distribution of multiple servers under different task load conditions, and obtaining the server task energy consumption information; Extract the task execution records of each server, record the task running time, resource allocation and power consumption data, calculate the power consumption of each server under different task loads, obtain the CPU, memory and storage occupancy ratios when the task is running, calculate the energy consumption value of unit task processing, classify the energy consumption data of different servers, and classify and count them according to task type, task duration and load. For example, in a data center, the power consumption of server A when executing a data analysis task is 120W, the CPU occupancy rate is 75%, and the task execution time is 3 hours. Then the total energy consumption of the task is 120W×3h=360Wh. Calculate the task power consumption, task load power ratio and server resource utilization. For example, when server B executes a video transcoding task, the power consumption is 150W, the CPU occupancy rate is 90%, and the task duration is 2 hours. Then its task power consumption is 150W×2h=300Wh. The method for calculating the task load power ratio is the ratio of power consumption to load proportion. For example, the ratio of server B is 150W / 90%=1.67W / %. At the same time, calculate the resource utilization to obtain the energy consumption distribution of different servers.

[0024] S212: Based on the server task energy consumption information, compare and analyze with the energy efficiency benchmark value, calculate the deviation of task power consumption, task load power ratio and server resource utilization, using the formula: ; Get the energy efficiency deviation of the server ,in, Representative The task power consumption of each server, Representative Resource utilization of each server, Representative Energy efficiency benchmark values ​​for servers, Represents the number of servers; Using the energy efficiency benchmark as a reference, calculate the energy efficiency deviation of each server under different task conditions, extract the historical task data of each server, and calculate its average energy efficiency ratio. For example, the average task power consumption of server C in the past 7 days is 500Wh, and the average resource utilization is 70%. Its historical energy efficiency benchmark value can be calculated as 500Wh / 70%=7.14Wh / %. Get the current task operation data. If there are 5 servers involved in the calculation: Server A: Wh, , Wh / %; Server B: Wh, , Wh / %; Server C: Wh, , Wh / %; Server D: Wh, , Wh / %; Server E: Wh, , Wh / %.

[0025] Calculate the energy efficiency ratio difference of each server: ; ; ; ; ; Calculate the energy efficiency deviation: ; The result shows that the currently calculated energy efficiency deviation is 1.258, which is higher than the preset threshold of 1.0, which means that the overall energy efficiency deviation of the server is large, and there are abnormal task power consumption or insufficient resource utilization. It is necessary to further screen the energy efficiency deviation analysis results, mark abnormal servers, and take optimization measures.

[0026] S213: calling the energy efficiency deviation degree of the server, screening the servers whose energy efficiency deviation degree exceeds the threshold, judging whether their task power consumption and resource utilization are in an abnormal range, marking the servers whose task power consumption is high or resource utilization is low, and obtaining the energy efficiency deviation analysis result; Set the energy efficiency deviation threshold to 1.0, filter out all servers with deviations exceeding this value, and determine whether their task power consumption and resource utilization are in the abnormal range. For example, the task power consumption of server D is 650Wh, and the resource utilization is 60%. Calculate its energy efficiency ratio as 650Wh / 60%=10.83Wh / %. If its historical benchmark value is 7.5Wh / %, then the deviation is , when the threshold is exceeded, the servers with high task power consumption or low resource utilization are marked, and the energy efficiency deviation analysis results are obtained.

[0027] See also Figure 4 ,The specific steps for obtaining abnormal energy consumption classification indicators are as follows: S311: Based on the energy efficiency deviation analysis result, extract the historical power consumption, task load and ambient temperature data of the server, calculate the power consumption change rate, load fluctuation amplitude and ambient temperature gradient change rate in the difference time period, and obtain the server energy consumption fluctuation parameter; The power consumption values ​​of each server in different time periods are counted, and the change rate of adjacent data points in the time series is calculated to obtain the power consumption change rate. For example, if the power consumption of a server at 8:00 am and 9:00 am is 320 W and 350 W respectively, then the power consumption change rate is When calculating the load fluctuation range, extract the CPU usage at adjacent time points and calculate its change. For example, if the CPU usage is 60% at a certain moment and changes to 75% at the next moment, the load fluctuation range is When calculating the ambient temperature gradient change rate, extract the room temperature data and calculate the temperature change rate. For example, if the room temperature rises from 22°C to 25°C with a time interval of 2 hours, then the temperature gradient change rate is , obtain the server energy consumption fluctuation parameters.

[0028] S312: Based on the server energy consumption fluctuation parameters, determine whether the power consumption change rate, load fluctuation amplitude, and temperature gradient change rate of the server exceed the threshold value, identify servers with abnormal power fluctuation, load imbalance, and abnormal temperature control, and use the formula: ; Calculate the abnormal energy consumption deviation value of the server , get the abnormal energy consumption classification data of the server, where Representative The power consumption change rate of the server, Representative The load fluctuation of the server, Representative The ambient temperature gradient change rate of each server, Represents the number of servers; Assume that the power consumption change rate threshold is 40W / h, the load fluctuation threshold is 20%, and the temperature change rate threshold is 2.0℃ / h. If the power consumption change rate of a server is 50W / h, the load fluctuation amplitude is 25%, and the temperature gradient change rate is 2.5℃ / h, then the server meets the abnormal conditions. There are 5 servers with the following parameters: Server A: W / h, , ℃ / h; Server B: W / h, , ℃ / h; Server C: W / h, , ℃ / h; Server D: W / h, , ℃ / h; Server E: W / h, , ℃ / h; Calculate the intermediate term for each server : ; ; ; ; ; Calculate the squared term for each server: ; ; ; ; ; Calculate the sum: ; Calculate the mean: ; Calculation results: ; The result shows that the currently calculated server abnormal energy consumption deviation value is 0.757. If the preset abnormal judgment threshold is 0.7, the value exceeds the threshold, indicating that the server cluster has abnormal energy consumption and needs further screening and classification to obtain abnormal energy consumption classification indicators.

[0029] S313: Based on the abnormal energy consumption classification data of the server, the server devices with abnormal power fluctuation, load imbalance and temperature control abnormality are classified according to the abnormal category associated with the server number to obtain the abnormal energy consumption classification index; Extract the device number corresponding to each server, match the servers with abnormal power fluctuation, load imbalance and temperature control, and classify the server abnormalities. For example, the abnormality category of server number S1001 is abnormal power fluctuation, the server number S1002 is load imbalance, the server number S1003 is temperature control abnormality, and the server number S1004 has both load imbalance and power fluctuation abnormalities. Summarize the classified data to obtain the abnormal energy consumption classification index.

[0030] See also Figure 5 ,The specific steps for obtaining energy efficiency optimization direction information are: S411: based on the abnormal energy consumption classification index, extract the server task migration cost, load balancing adjustment impact value and computer room cooling power data, analyze the energy consumption required for task migration, the energy consumption change caused by load balancing adjustment and the optimization potential point of computer room cooling power, and obtain the energy consumption adjustment impact parameter; Extract the server task migration cost, load balancing adjustment impact value and room cooling power data, analyze the server nodes involved in the task migration, calculate the migration energy consumption of each server, extract the CPU, memory and network bandwidth occupancy of the source server to calculate the energy consumption required for data replication and status synchronization during the task migration process. For example, the source server CPU occupies 80%, the memory occupies 16GB, and the data transmission volume is 50GB. Combined with the bandwidth rate, calculate the data transmission time. If the network rate is 1Gbps, the transmission time is seconds, the power consumption is calculated based on the CPU occupancy ratio: Full load power, for example, full load power 200W, then the migration energy consumption is When calculating the impact value of load balancing adjustment, extract the load changes of each server and analyze the load changes of the target server after migration. Assuming that the CPU utilization of the target server is 50% before migration and increases to 75% after migration, calculate the load change ratio. , obtain the load adjustment impact value, analyze the potential points for optimizing the cooling power of the computer room, extract the heat dissipation power of the server cabinet, and calculate the change in the heat dissipation load of the target cabinet after migration. If the cooling power of the target cabinet is 5kW and the heat dissipation of the new server is 1.2kW, the total cooling power demand is 6.2kW, and the energy consumption adjustment impact parameters are obtained.

[0031] S412: Based on the energy consumption adjustment influencing parameters, calculate the energy consumption benefit ratio of the adjustment direction, select the direction with the best energy consumption benefit ratio, and use the formula: ; Get the optimal energy efficiency adjustment direction, where represents the energy efficiency ratio, Representative The energy consumption of task migration of each server, Represents the load balancing adjustment impact value, represents the migration cost, Represents the optimization impact of the cooling power in the equipment room, Represents the number of servers to which the task is migrated. Represents the number of servers adjusted for cooling optimization in the computer room; There are three servers with task migration. The energy consumption of task migration is: Server A: Wh, Server B: Wh, Server C: Wh; Load balancing adjustment impact value: Server A: , Server B: , Server C: ; Migration costs: Wh; Server cooling power optimization impact: Server X: kW, server Y: kW Calculation results: ; ; ; The result shows that the calculated energy efficiency ratio is 37.6. If the preset efficiency ratio judgment threshold is 30, the current adjustment direction meets the optimization requirements and can be used to screen the optimal energy efficiency adjustment direction.

[0032] S413: calling the optimal energy efficiency adjustment direction, combining the server task scheduling rules and the computer room cooling adjustment plan, identifying the adjustment path that matches the current energy consumption optimization target, and obtaining energy efficiency optimization direction information; Combined with the server task scheduling rules and the computer room cooling adjustment plan, the adjustment path is analyzed, the migration plan with the highest energy efficiency ratio is screened, and the server task allocation is adjusted. For example, servers with low task migration energy consumption and small load balancing impact are selected for scheduling. When selecting the target server, priority is given to cabinets with a load increase of no more than 10% after migration and a large cooling power optimization space. If the load of the target server A increases by 6% and the cooling power optimization amount is 1.5kW, and the load of server B increases by 12% and the cooling power optimization amount is 1.2kW, server A is given priority to obtain energy efficiency optimization direction information.

[0033] See also Figure 6 , the specific steps for obtaining the energy consumption management optimization results are: S511: Based on the energy efficiency optimization direction information, adjust the CPU operating frequency of the server, collect the current load level, task execution priority and power consumption trend of the server, calculate the energy consumption change after the CPU frequency is adjusted, and obtain the CPU frequency optimization value; Adjust the server CPU operating frequency, extract the current load level, task execution priority and power consumption parameters, extract the server's current CPU utilization, temperature, power consumption and number of executed tasks, and determine the CPU frequency adjustment direction based on the task priority. If a server's CPU utilization exceeds 80% and the power consumption is higher than 120W, reduce the frequency to reduce power consumption. If the utilization is lower than 30%, increase the frequency appropriately to maintain task execution efficiency. Calculate the energy consumption change after CPU frequency adjustment. Assume that the CPU frequency is 2.5GHz before adjustment and is reduced to 2.0GHz after adjustment. The CPU power consumption is reduced from 120W to 95W. The adjustment period is 1 hour, then the energy consumption is reduced. kWh, get the optimized CPU frequency value.

[0034] S512: Based on the CPU frequency optimization value, calculate the energy consumption change under the task load balancing distribution, select the optimal load distribution mode, and use the formula: ; Get the load balancing optimization result, where: Represents the change in energy consumption under load adjustment. Representative The load on the server, Represents the CPU frequency after the server is adjusted. Represents the maximum computing capacity of the server. represents the server power consumption before adjustment, Represents the adjustment cycle duration, Represents the number of servers; If there are 3 groups of servers, their load : Server A: Task / time, Server B: Task / time, Server C: Task / hour; CPU frequency : After adjustment of server A: GHz, after adjustment of server B: GHz, server C after adjustment: GHz; Maximum computing power Unified to 100 tasks / hour; Power consumption before adjustment : Server A: 120W, Server B: 110W, Server C: 105W; Total power consumption: ; Adjustment cycle Set to 1 hour; Calculate the load balancing factor for each server after adjustment: ; Compute the squared term: ; Substitute into the calculation: ; calculate : ; A negative value means that the energy consumption of the server after load balancing optimization is reduced by 326.1W, that is, the energy consumption of the server is significantly reduced after adjustment, indicating that the balanced allocation strategy is effective. The calculation result can intuitively reflect the degree of energy consumption optimization. A negative value indicates a decrease in energy consumption, and a positive value indicates that the optimization has failed, indicating that after the current task load balancing strategy is optimized, the energy consumption is reduced by 326.1W, which meets the energy consumption management optimization goal.

[0035] S513: calling the load balancing optimization result, analyzing the temperature and humidity change trend after the cooling efficiency of the computer room is adjusted, evaluating the adjustment effect of the total power consumption, and obtaining the energy consumption management optimization result; The temperature and humidity change trend after the computer room cooling efficiency is adjusted, the heat dissipation load change of the server cabinet is analyzed, and the operation data of the cooling system before and after the adjustment is extracted. If the heat dissipation requirements of servers A, B and C before adjustment are 4.5kW, 4.2kW ​​and 4.0kW respectively, and after adjustment they are reduced to 3.8kW, 3.5kW and 3.2kW, the total heat dissipation power is reduced kW, combined with the room temperature and humidity monitoring data, evaluate the total power consumption adjustment effect and obtain energy consumption management optimization results.

[0036] A data center AI energy consumption analysis and optimization system, the system comprising: The status monitoring module obtains data center status information, collects server CPU occupancy, memory usage, and disk I / O frequency, and simultaneously monitors the room temperature, humidity, and cabinet energy consumption, identifies server devices with severe power fluctuations, abnormal load fluctuations, and abnormal temperature changes, and establishes an energy consumption monitoring data set; The deviation analysis module calls the energy consumption monitoring data set, calculates the task power consumption, task load power ratio and server resource utilization of the server, compares the energy efficiency benchmark value, and obtains the energy efficiency deviation analysis result; The abnormal classification module identifies server devices with abnormal power fluctuation, load imbalance and temperature control based on the energy efficiency deviation analysis results, and associates the abnormal type with the server device number to obtain the abnormal energy consumption classification index; Based on the abnormal energy consumption classification indicators, the direction optimization module analyzes the impact values ​​that affect the server task migration cost and load balancing adjustment, evaluates the optimization points of the cooling power in the computer room, selects the adjustment direction with the best energy efficiency ratio, and obtains the energy efficiency optimization direction information; Based on the energy efficiency optimization direction information, the management adjustment module adjusts the server CPU operating frequency, task load balancing distribution and computer room cooling efficiency, analyzes the adjusted server energy efficiency utilization level and computer room temperature and humidity change trend, and evaluates the adjustment effect of total power consumption to obtain energy consumption management optimization results.

[0037] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A data center AI energy consumption analysis and optimization method, characterized in that: The following steps are involved: S1: Obtain data center status information, collect CPU occupancy, memory usage, and disk I / O frequency, monitor computer room temperature, humidity, and cabinet energy consumption, identify servers with abnormal power fluctuations, abnormal loads, and abnormal temperature changes, and establish an energy consumption monitoring data set; S2: calling the energy consumption monitoring data set, calculating task power consumption, task load power ratio and server resource utilization, comparing the energy efficiency benchmark value, and obtaining energy efficiency deviation analysis results; S3: Based on the energy efficiency deviation analysis result, identify the server devices with abnormal power fluctuation, load imbalance and temperature control abnormality, associate the abnormality type with the server device number, and obtain the abnormal energy consumption classification index; S4: Based on the abnormal energy consumption classification index, analyze the impact of task migration cost and load balancing adjustment, evaluate the cooling power optimization point, screen the optimal direction of energy consumption efficiency, and obtain energy efficiency optimization direction information; S5: Based on the energy efficiency optimization direction information, adjust the CPU operating frequency, task load balancing and computer room cooling efficiency, analyze the adjusted energy efficiency utilization level and temperature and humidity change trends, evaluate the adjustment effect of the total power consumption, and obtain energy consumption management optimization results.

2. The data center AI energy consumption analysis and optimization method according to claim 1 is characterized in that: The energy consumption monitoring data set includes power load data, fan speed records, and power module input parameters. The energy efficiency deviation analysis results include a list of high-power consumption servers and a list of inefficient resource utilization servers. The abnormal energy consumption classification indicators include abnormal power fluctuation data, abnormal load fluctuation data, and abnormal temperature change data. The energy efficiency optimization direction information includes benefit impact analysis results, load adjustment direction, and cooling optimization points. The energy consumption management optimization results include energy efficiency improvement information, temperature and humidity change analysis results, and total power adjustment effect.

3. The data center AI energy consumption analysis and optimization method according to claim 1 is characterized in that: The steps for establishing the energy consumption monitoring data set are specifically as follows: S111: Obtain data center status information, collect the CPU occupancy rate, memory usage and disk I / O frequency of the server, synchronously monitor the temperature, humidity and cabinet energy consumption of the computer room, and clean the collected data to remove invalid and abnormal data to obtain a server monitoring data set; S112: Based on the server monitoring data set, analyze the power load, fan speed, and power module input power of the server, determine the fluctuation range of each parameter, and use the formula: ； Calculate the server power fluctuation degree, construct the server load fluctuation curve, and obtain the server power load fluctuation data, where: Represents the server power fluctuation coefficient, Representative The power load of the server, represents the average power load of the server group, Represents the number of servers; S113: Call the server power load fluctuation data, combine the changes in fan speed and power module input power, calculate the power fluctuation rate of the server at multiple time nodes, filter out server devices with excessive fluctuation rates, and combine the room temperature change trend to analyze the impact of temperature anomalies and establish an energy consumption monitoring data set.

4. The data center AI energy consumption analysis and optimization method according to claim 1 is characterized in that: The steps for obtaining the energy efficiency deviation analysis result are specifically as follows: S211: calling the energy consumption monitoring data set, calculating the task power consumption, task load power ratio and server resource utilization of the server, obtaining the energy consumption distribution of multiple servers under different task load conditions, and obtaining the server task energy consumption information; S212: Based on the server task energy consumption information, compare and analyze with the energy efficiency benchmark value, calculate the deviation of task power consumption, task load power ratio and server resource utilization, using the formula: ； Get the energy efficiency deviation of the server ,in, Representative The task power consumption of each server, Representative Resource utilization of each server, Representative Energy efficiency benchmark values ​​for servers, Represents the number of servers; S213: Call the energy efficiency deviation degree of the server, screen the servers whose energy efficiency deviation exceeds the threshold, determine whether their task power consumption and resource utilization are in an abnormal range, and mark the servers with high task power consumption or low resource utilization to obtain energy efficiency deviation analysis results.

5. The data center AI energy consumption analysis and optimization method according to claim 1 is characterized in that: The steps for obtaining the abnormal energy consumption classification index are specifically as follows: S311: Based on the energy efficiency deviation analysis result, extract the historical power consumption, task load and ambient temperature data of the server, calculate the power consumption change rate, load fluctuation amplitude and ambient temperature gradient change rate in the difference time period, and obtain the server energy consumption fluctuation parameter; S312: Based on the server energy consumption fluctuation parameters, determine whether the power consumption change rate, load fluctuation amplitude and temperature gradient change rate of the server exceed the threshold value, identify servers with abnormal power fluctuation, load imbalance and abnormal temperature control, and use the formula: ； Calculate the abnormal energy consumption deviation value of the server , get the abnormal energy consumption classification data of the server, where Representative The power consumption change rate of the server, Representative The load fluctuation of the server, Representative The ambient temperature gradient change rate of each server, Represents the number of servers; S313: Based on the abnormal energy consumption classification data of the server, the server devices with abnormal power fluctuation, load imbalance and temperature control abnormality are classified according to the abnormal category associated with the server number to obtain the abnormal energy consumption classification index.

6. The data center AI energy consumption analysis and optimization method according to claim 1 is characterized in that: The steps for obtaining the energy efficiency optimization direction information are specifically as follows: S411: Based on the abnormal energy consumption classification index, extract the server task migration cost, load balancing adjustment impact value and computer room cooling power data, analyze the energy consumption required for task migration, the energy consumption change caused by load balancing adjustment and the optimization potential point of computer room cooling power, and obtain the energy consumption adjustment impact parameter; S412: Based on the energy consumption adjustment influencing parameters, the energy consumption benefit ratio of the adjustment direction is calculated, and the direction with the best energy consumption benefit ratio is selected, using the formula: ； Get the optimal energy efficiency adjustment direction, where represents the energy efficiency ratio, Representative The energy consumption of task migration of each server, Represents the load balancing adjustment impact value, represents the migration cost, Represents the optimization impact of the cooling power in the equipment room, Represents the number of servers to which the task is migrated. Represents the number of servers adjusted for cooling optimization in the computer room; S413: calling the optimal energy efficiency adjustment direction, combining the server task scheduling rules and the computer room cooling adjustment plan, identifying the adjustment path that matches the current energy consumption optimization target, and obtaining energy efficiency optimization direction information.

7. The data center AI energy consumption analysis and optimization method according to claim 1 is characterized in that: The steps for obtaining the energy consumption management optimization result are specifically as follows: S511: Based on the energy efficiency optimization direction information, adjust the CPU operating frequency of the server, collect the current load level, task execution priority and power consumption trend of the server, calculate the energy consumption change after the CPU frequency is adjusted, and obtain the CPU frequency optimization value; S512: Based on the CPU frequency optimization value, calculate the energy consumption change under the task load balance distribution, select the optimal load distribution mode, and use the formula: ； Get the load balancing optimization result, where: Represents the change in energy consumption under load adjustment. Representative The load on the server, Represents the CPU frequency after the server is adjusted. Represents the maximum computing capacity of the server. represents the server power consumption before adjustment, Represents the adjustment cycle duration, Represents the number of servers; S513: calling the load balancing optimization result, analyzing the temperature and humidity change trend after the cooling efficiency of the computer room is adjusted, evaluating the adjustment effect of the total power consumption, and obtaining the energy consumption management optimization result.

8. A data center AI energy consumption analysis and optimization system, characterized in that: According to any one of claims 1 to 7, the data center AI energy consumption analysis and optimization method is implemented, and the system includes: The status monitoring module obtains data center status information, collects server CPU occupancy, memory usage, and disk I / O frequency, and simultaneously monitors the room temperature, humidity, and cabinet energy consumption, identifies server devices with severe power fluctuations, abnormal load fluctuations, and abnormal temperature changes, and establishes an energy consumption monitoring data set; The deviation analysis module calls the energy consumption monitoring data set, calculates the task power consumption, task load power ratio and server resource utilization of the server, compares the energy efficiency benchmark value, and obtains the energy efficiency deviation analysis result; The abnormal classification module identifies server devices with abnormal power fluctuation, load imbalance and temperature control based on the energy efficiency deviation analysis result, and associates the abnormal type with the server device number to obtain an abnormal energy consumption classification index; Based on the abnormal energy consumption classification index, the direction optimization module analyzes the impact value of server task migration cost and load balancing adjustment, evaluates the optimization point of the cooling power of the computer room, selects the adjustment direction with the best energy consumption benefit ratio, and obtains energy efficiency optimization direction information; Based on the energy efficiency optimization direction information, the management adjustment module adjusts the server CPU operating frequency, task load balancing distribution and computer room cooling efficiency, analyzes the adjusted server energy efficiency utilization level and computer room temperature and humidity change trend, and evaluates the adjustment effect of total power consumption to obtain energy consumption management optimization results.

Images