Server power consumption modeling method based on CPU working status temperature estimation

Abstract

The present invention relates to a server power consumption modeling method for estimating temperature based on CPU working state. The power consumption modeling process uses the least squares method to fit the actual measured results of running a specific test program on the server to accurately obtain the coefficients of the server power consumption mathematical model taking into account the temperature variable. In order to avoid additional temperature measurements in actual use, a CPU core temperature estimation method based on the CPU working state is designed. The CPU working state is judged according to the change in the discreteness of the average value of the CPU utilization time series, and then the CPU core temperature is estimated. In order to obtain a more accurate CPU working state judgment result, a CPU time series length and CPU working state judgment threshold parameter optimization method is designed to obtain the optimal parameter value that minimizes the model calculation error. The modeling process proposed by this method is universal, and the final modeling result can significantly improve the accuracy of power consumption calculation without introducing additional measurement equipment.

Description

Technical Field

[0001] The invention relates to a server power consumption modeling method, and in particular to a server power consumption modeling method based on CPU working state estimated temperature. Background Art

[0002] In recent years, technologies such as the Internet of Things, cloud computing, artificial intelligence, and 5G have developed rapidly. As the basic supporting facilities, the number and scale of data centers continue to grow, and the proportion of energy consumption has also increased year by year. It is estimated that by 2025, the proportion of power consumption of data centers in my country will reach 4.1% of the national power consumption. The huge energy consumption of data centers has become an important factor restricting their operational economy. How to reduce the energy consumption of data centers has become a problem of extensive concern and research in the domestic and foreign industrial and academic circles.

[0003] Among all kinds of electrical equipment in data centers, IT equipment energy consumption accounts for about 50% of the total energy consumption. Among them, servers, as core basic equipment for performing computing, storage, communication and other functions, account for about 70% of the total energy consumption of IT equipment. In addition, reducing server energy consumption can also reduce the energy consumption of other equipment (such as cooling equipment) necessary to maintain safe and stable operation of servers. Therefore, energy saving and consumption reduction of servers becomes the key. The reasonable design and optimal configuration of server energy saving and consumption reduction methods need to be based on accurate and efficient service power consumption models. Therefore, accurately establishing a server power consumption model is of great significance for energy saving and consumption reduction in data centers.

[0004] Among the energy consumption components of servers, CPU power consumption accounts for the largest proportion, and the activities of other components in the server are highly coupled with CPU activities. Therefore, existing studies mostly model server power consumption based on CPU utilization. The linear model based on CPU utilization is simple and intuitive and is widely used, but it cannot characterize the nonlinear characteristics of server power consumption caused by the application of CPU technologies such as hyperthreading technology and turbo frequency technology in recent years. In response to this problem, relevant studies have proposed solutions such as piecewise linear models, exponential models, and polynomial models, but they still cannot explain the physical mechanism of the nonlinear characteristics of server power consumption. In fact, as the power density of servers continues to increase, the thermoelectric coupling phenomenon is becoming more and more significant. In addition to CPU utilization, the impact of temperature on server power consumption cannot be ignored. Due to the delayed characteristics of temperature changes, power consumption changes with CPU utilization also have a delay. The existing server power consumption model based only on CPU utilization cannot accurately capture the dynamic characteristics of power consumption delay changes, resulting in large errors. Relevant scholars introduced temperature variables in the process of server power consumption modeling to capture the dynamic laws of power consumption. However, introducing temperature variables as model input will increase additional measurement equipment and temperature measurement errors. In summary, the existing server power consumption modeling methods cannot restore the dynamic characteristics of power consumption delay changes affected by temperature under the premise of using only CPU utilization as the model input, resulting in large model calculation errors and limiting its promotion and application. Summary of the invention

[0005] The purpose of the present invention is to overcome the shortcomings of the prior art and propose a server power consumption modeling method based on CPU working state temperature estimation, which can accurately capture the dynamic characteristics of power consumption delay changes affected by temperature and reduce model errors.

[0006] The present invention solves the technical problem by the following technical solutions:

[0007] A server power consumption modeling method for estimating temperature based on CPU working status, characterized by comprising the following steps:

[0008] Step 1, establishing a mathematical model of server power consumption taking into account temperature variables, which takes the actual measured value of CPU utilization and the actual measured value of CPU core temperature as input and outputs the calculated value of total server power consumption, and sets the coefficients in the mathematical model of server power consumption taking into account temperature variables;

[0009] Step 2: Estimate the CPU core temperature based on the CPU working state according to the server power consumption mathematical model established in step 1, determine the CPU fluctuation or step different working states through the CPU utilization time series change, and obtain the CPU core temperature estimation value to replace the input of the CPU core temperature actual value of the server power consumption mathematical model considering the temperature variable established in step 1;

[0010] Step 3: Based on the server power consumption mathematical model taking into account the temperature variable established in step 1 and the CPU core temperature estimation value obtained in step 2, with the goal of minimizing the model error, the time series length Δt in step 2 and the CPU working state judgment threshold R are calculated. MVD,th Conduct optimization search;

[0011] Furthermore, the method for establishing a mathematical model of server power consumption taking into account temperature variables comprises the following steps:

[0012] Step 1.1, configuring the equipment required for measuring and analyzing the power consumption of the server, including a test program that can simulate a specific computing load, an AIDA64 software detection engine, a power meter that can measure and record the actual value of the total power consumption of the server, and a computing server that can record and analyze the measurement data. The AIDA64 software detection engine is used to obtain the actual value of CPU utilization, the actual value of CPU power, the actual value of CPU core temperature, and the actual value of cooling fan speed;

[0013] Step 1.2, run the test program on the test server and collect and record relevant data;

[0014] Step 1.3, according to the measured value of CPU power consumption, the measured value of CPU utilization and the measured value of CPU core temperature, the CPU utilization coefficient a and the temperature coefficient b are obtained by fitting using the least square method according to the corresponding relationship between the three 0 、b 1 :

[0015]

[0016] in, t n CPU power consumption at the moment; P idle The idle power consumption of CPU is set to a constant value; t n CPU utilization at the moment; t n CPU core temperature at the moment;

[0017] Step 1.4: Calculate the power consumption of the server cooling fan based on the measured values ​​of the CPU power consumption and the total power consumption of the server:

[0018]

[0019] in, t n The power consumption of the server cooling fans at the moment; t n Total power consumption of the server at the moment; t n CPU power consumption at the moment; P otherThe power consumption of the rest of the system except the CPU and cooling fan;

[0020] Step 1.5, according to the cooling fan power consumption calculation data and the cooling fan speed measured value, the fan power consumption coefficient c is obtained by using the least squares method according to the corresponding relationship between the two. 0 、c 1 、c 2 and c 3 :

[0021]

[0022] in, t n The power consumption of the server cooling fans at the moment; t n The speed of the server cooling fans at the moment;

[0023] Step 1.6: According to the measured value of the cooling fan speed and the measured value of the CPU core temperature, the fan speed f of the server under low load is obtained by using the least squares method according to the corresponding relationship between the two. min , Fan speed f when the server is under high load max The fan speed switches between constant speed and variable speed working states. min , T max , and the rate of change of speed in the speed range with temperature change k fan :

[0024]

[0025] in, t n The speed of the server cooling fans at the moment; t n CPU core temperature at the moment;

[0026] Step 1.7, according to the measured value of CPU core temperature and the measured value of CPU utilization, according to the corresponding relationship between the steady-state value of CPU core temperature and CPU utilization obtained by using the equivalent thermal parameter model calculation method, the steady-state temperature parameter μ is obtained by least squares fitting. 0 , μ 1 and μ 2 :

[0027]

[0028] in, The CPU utilization is The steady-state value of CPU core temperature in the state; t n CPU usage at a certain moment.

[0029] Moreover, the method for setting the coefficients in the mathematical model of server power consumption taking into account temperature variables is: running a specific test program on the server and measuring the CPU utilization, CPU power, CPU core temperature, cooling fan speed and total server power consumption, and using the least squares method for fitting and setting according to the mathematical relationship.

[0030] Moreover, the coefficients in the mathematical model of server power consumption taking into account temperature variables include CPU utilization coefficient a, temperature coefficient b 0 、b 1 , CPU idle power consumption P idle , fan power consumption coefficient c 0 、c 1 、c 2 、c 3 , the fan speed f of the server under low load min , the fan speed f under high load condition of the server max The fan speed switches between constant speed and variable speed working states. min The maximum temperature threshold T of the fan speed switching between constant speed and variable speed working states max , the rate of change of speed in the speed change range with temperature change k fan , steady-state temperature parameter μ 0 , μ 1 , μ 2 .

[0031] Furthermore, the method for estimating the CPU core temperature based on the CPU working state comprises the following steps:

[0032] Step 2.1, for t n Temperature estimation at time t n CPU utilization time series within Δt before the moment

[0033] Step 2.2, for the CPU utilization time series U, calculate its mean dispersion MVD value:

[0034]

[0035] in, t n The average dispersion of the CPU utilization time series at the moment, which is smaller in the CPU fluctuating working state and larger in the CPU step working state; Δt is the length of the time series; is the CPU utilization at time t;

[0036] Step 2.3, according to According to the changes in the CPU core temperature, it is determined that the CPU is in a fluctuating or stepping working state, and then the CPU core temperature estimate is updated.

[0037]

[0038] in, t n CPU core temperature at the moment; is the estimated value of temperature at the previous moment; t n The average dispersion of the CPU utilization time series at the moment; R MVD,th is the CPU working state judgment threshold; δt is the correction sequence length set to eliminate transient errors; is the CPU utilization at the tth moment; the update of the temperature estimation value is based on the corresponding relationship between the CPU core temperature steady-state value and the CPU utilization obtained in step 1.7.

[0039] Moreover, based on the mathematical model of server power consumption taking into account temperature variables established in step 1 and the estimated value of CPU core temperature obtained in step 2, with the goal of minimizing the model error, the time series length Δt in step 2 and the CPU working state judgment threshold R MVD,th The optimization method includes the following steps:

[0040] Step 3.1, input test time t N Measured value of internal server power consumption CPU utilization measured value and coefficients in a mathematical model of server power consumption that accounts for temperature variation;

[0041] Step 3.2, set the time series length Δt and the CPU working status judgment threshold R MVD,th Initial value;

[0042] Step 3.3, update the time series length Δt and increase the unit value;

[0043] Step 3.4, update the CPU working state judgment threshold R MVD,th Increase the unit value;

[0044] Step 3.5, for t N At each moment, the actual measured value of CPU utilization U m Calculate the corresponding CPU core temperature estimate, and then obtain the server power consumption at all times in the server power consumption mathematical model taking into account the temperature variable

[0045] Step 3.6, use the root mean square error R RMSE Measure the total power consumption of the server P srv,mThe total power consumption model of the server calculates the value P srv Error between:

[0046]

[0047] Among them, t N is the total test time; is the measured value of the total power consumption of the server at time t; Calculate and record R for the total power consumption model of the server at time t. RMSE ;

[0048] Step 3.7, if the CPU working state judgment threshold R MVD,th If the CPU working state judgment threshold R MVD,th When the set upper limit is reached, the CPU working state judgment threshold R MVD,th Reset the initial value and execute step 3.8;

[0049] Step 3.8: If the time series length Δt does not reach the set upper limit, return to step 3.3; if the time series length Δt reaches the set upper limit, execute step 3.9;

[0050] Step 3.9, based on the recorded root mean square error R RMSE The minimum root mean square error value is found in the data, that is, the time series length Δt and the CPU working state judgment threshold R of the most accurate model are MVD,th .

[0051] The advantages and beneficial effects of the present invention are:

[0052] This server power consumption modeling method based on CPU working status temperature estimation, the hardware and software testing tools selected in the testing process, and the model parameter fitting optimization method can be used in most heterogeneous server modeling processes and have strong versatility.

[0053] By taking temperature variables into account during the modeling process, the model calculation results can accurately restore the dynamic characteristics of server power consumption delay changes affected by temperature delay changes, eliminating the model calculation errors caused by the traditional server power consumption model based only on CPU utilization without considering this dynamic characteristic, thereby improving the accuracy of model calculation.

[0054] By using the CPU core temperature estimation method based on the CPU working state, the CPU utilization time series average dispersion is used as a judgment indicator to judge the CPU fluctuation or step working state, and based on this, the estimated value of the CPU core temperature is updated, which can achieve a more accurate transient temperature estimation. The final modeling result only uses the CPU utilization as the model input to complete the accurate calculation of the server power consumption taking into account the temperature variable, avoiding the introduction of additional measurement equipment and temperature measurement errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 The configuration and operation flow chart of the experimental equipment required for modeling involved in the present invention;

[0056] Figure 2 This is a schematic diagram showing the variation of the MVD value of the CPU working state criterion involved in the present invention with the CPU utilization rate;

[0057] Figure 3 A flow chart of the method for optimizing the CPU time series length and CPU working state determination threshold parameter involved in the present invention;

[0058] Figure 4 A schematic diagram of the change of the model root mean square error during the optimization process of the CPU time series length and the CPU working state determination threshold parameter of the test server involved in the present invention;

[0059] Figure 5 The model usage process involved in the present invention is a calculation flow chart that takes CPU utilization as input and outputs the total power consumption of the server;

[0060] Figure 6 A schematic diagram of the dynamic characteristics of server power consumption delay changes in the actual measurement results involved in the present invention;

[0061] Figure 7 A schematic diagram showing the degree of fit and error of the method involved in the present invention to the measured power consumption value;

[0062] Figure 8 This is a schematic diagram of the effect of eliminating calculation errors of traditional modeling methods using the method of the present invention. DETAILED DESCRIPTION

[0063] The present invention is further described in detail below through specific examples. The following examples are only illustrative and not restrictive, and the protection scope of the present invention cannot be limited thereto.

[0064] A server power consumption modeling method for estimating temperature based on CPU working status, the innovation of which is that the method includes the following three parts:

[0065] Step 1: Method for establishing a mathematical model of server power consumption taking into account temperature variables: Its goal is to comprehensively consider the mutual influence between the internal temperature of the server and the total power consumption of the server, and to build a server power consumption model taking into account temperature variables, which can eliminate the model calculation error problem caused by the dynamic characteristics of the delayed change of server power consumption affected by temperature. The specific steps are as follows:

[0066] Step 1.1, configure the equipment required for measuring and analyzing the power consumption of the server, including a test program that can simulate a specific computing load, an AIDA64 software detection engine, a power meter that can measure and record the actual value of the total power consumption of the server, and a computing server that can record and analyze the measurement data. The AIDA64 software detection engine is used to obtain the actual value of CPU utilization, the actual value of CPU power, the actual value of CPU core temperature, and the actual value of cooling fan speed;

[0067] Step 1.2: Run the test program on the server and collect and record relevant data. The process is as follows: Figure 1 shown.

[0068] Step 1.3, according to the measured value of CPU power consumption, the measured value of CPU utilization and the measured value of CPU core temperature, the CPU utilization coefficient a and the temperature coefficient b are obtained by using the least squares fitting method according to the corresponding relationship between the three 0 、b 1 :

[0069]

[0070] in, t n CPU power consumption at the moment; P idle is the CPU idle power consumption, which is considered to be a constant value; t n CPU utilization at the moment; t n CPU core temperature at the moment;

[0071] Step 1.4: Calculate the power consumption of the server cooling fan based on the measured values ​​of the CPU power consumption and the total power consumption of the server:

[0072]

[0073] in, t n The power consumption of the server cooling fans at the moment; t n Total power consumption of the server at the moment; t n CPU power consumption at the moment; P other The power consumption of the rest of the system except the CPU and cooling fan;

[0074] Step 1.5, according to the calculated value of cooling fan power consumption and the measured value of cooling fan speed, the fan power consumption coefficient c is obtained by using the least squares fitting method according to the corresponding relationship between the two. 0 、c 1 、c 2 and c 3 :

[0075]

[0076] in, t n The power consumption of the server cooling fans at the moment; t n The speed of the server cooling fans at the moment;

[0077] Step 1.6, based on the cooling fan and CPU core temperature measurement data, the least squares method is used to fit the corresponding relationship between the two to obtain the required parameter, the fan speed f of the server under low load and high load conditions. min 、f max , the temperature threshold T at which the fan speed switches between constant speed and variable speed working states min , T max And the rate of change of speed in the speed range with temperature change k fan :

[0078]

[0079] in, t n The speed of the server cooling fans at the moment; t n CPU core temperature at the moment;

[0080] Step 1.7, according to the measured value of CPU core temperature and the measured value of CPU utilization, according to the corresponding relationship between the steady-state value of CPU core temperature and CPU utilization obtained by using the equivalent thermal parameter model calculation method, the steady-state temperature coefficient μ is obtained by least squares fitting. 0 , μ 1 and μ 2 :

[0081]

[0082] in, The CPU utilization is The steady-state value of CPU core temperature in the state; t n CPU usage at a certain moment.

[0083] Step 2, CPU core temperature estimation method based on CPU working status: Its goal is to judge the CPU working status through the change of CPU utilization time series, and estimate the CPU core temperature according to the temperature change law under different working states of CPU fluctuation or step, so as to avoid the additional investment and measurement error caused by introducing temperature measurement device in the use of model. The specific steps are as follows:

[0084] Step 2.1, for t n Temperature estimation at time t n CPU utilization time series within Δt before the moment

[0085] Step 2.2: Calculate the mean dispersion of the CPU utilization time series U:

[0086]

[0087] in, t n The dispersion of the average value of the CPU utilization time series at the moment is smaller when the CPU is in a fluctuating working state and larger when the CPU is in a step working state, such as Figure 2 As shown; Δt is the length of the time series; is the CPU utilization at time t;

[0088] Step 2.3, according to According to the changes in the CPU core temperature, it is determined that the CPU is in a fluctuating or stepping working state, and then the estimated value of the CPU core temperature is updated.

[0089]

[0090] in, t n CPU core temperature at the moment; It is the estimated value of CPU core temperature at the previous moment; t n The average dispersion of the CPU utilization time series at the moment; R MVD,th is the CPU working state judgment threshold; δt is the correction sequence length set to eliminate transient errors; is the CPU utilization at the tth moment; the update of the temperature estimation value is based on the corresponding relationship between the CPU core temperature steady-state value and the CPU utilization obtained in step 1.7.

[0091] Step 3, CPU time series length and CPU working state judgment threshold parameter optimization method: Its goal is to optimize the important parameters Δt and R in the CPU core temperature estimation method based on the CPU working state. MVD,thOptimize to achieve the maximum accuracy of the model. The specific steps are:

[0092] Step 3.1, input test time t N Measured value of internal server power consumption CPU utilization measured value And the model coefficients a and b obtained according to the method described in step 1 0 、b 1 , P idle 、c 0 、c 1 、c 2 、c 3 、f min 、f max , T min , T max , k fan , μ 0 , μ 1 and μ 2 ;

[0093] Step 3.2, set Δt and R MVD,th Initial value;

[0094] Step 3.3, update Δt to increase the unit value;

[0095] Step 3.4, Update R MVD,th Increase the unit value;

[0096] Step 3.5, for t N At each moment, according to the method described in step 2, use U m Calculate the corresponding CPU core temperature estimate Then, the model obtained by the method described in step 1 is used to calculate the server power consumption at all times

[0097] Step 3.6, use the root mean square error R RMSE Measure the measured value P srv,m The model calculated value P srv Error between:

[0098]

[0099] Among them, t N is the total test time; The actual power consumption data of the server at time t; Calculate the server power consumption data for the model at time t, calculate and record R RMSE ;

[0100] Step 3.7, if R MVD,th If the upper limit is not reached, return to step 3.4.MVD,th When the set upper limit is reached, R MVD,th Reset the initial value and execute step 3.8;

[0101] Step 3.8: If Δt does not reach the set upper limit, return to step 3.3; if Δt reaches the set upper limit, execute step 3.9;

[0102] Step 3.9, according to the record R RMSE The data is used to find the minimum root mean square error value, that is, the most accurate Δt and R MVD,th , complete the CPU time series length Δt and CPU working state judgment threshold R MVD,th Parameter optimization process.

[0103] The above-mentioned CPU time series length and CPU working state judgment threshold parameter optimization method process is as follows Figure 3 shown.

[0104] Figure 4 The model building process for a specific test server is shown. The model root mean square error R is the optimal parameter for the CPU time series length and the CPU working status judgment threshold. RMSE With Δt and R MVD ,th changes. According to step 3.9, the root mean square error R RMSE Δt and R at minimum MVD ,th is taken as the optimal value.

[0105] Through the above-mentioned whole process including the method of establishing the mathematical model of server power consumption taking into account the temperature variable, the method of estimating the CPU core temperature based on the CPU working status, and the method of optimizing the CPU time series length and the CPU working status judgment threshold parameter, the establishment of a new type of server power consumption based on the CPU working status estimated temperature can be completed, which can achieve more accurate server power consumption calculation.

[0106] Figure 5 The figure shows the calculation process of the total power consumption of the server using the CPU utilization as input during the model usage. n The power consumption at time t is first obtained. n The CPU utilization time series U within the time Δt before the moment, and the CPU core temperature is estimated according to the step 2 of the present invention The parameters Δt and R MVD , th is obtained according to step 3 of the present invention before the model is used; then, t n CPU utilization at any time and Substitute the server power consumption mathematical model taking into account the temperature variable established in step 1 of the present invention, and finally output t nAccurate calculation of the total power consumption of the server at the moment By continuously updating the CPU utilization time series U, the server power consumption at each moment of the entire process can be accurately calculated.

[0107] Figure 6 The dynamic characteristics of server power consumption delay change in actual measurement results are shown, which causes large errors in the traditional server power consumption model based only on CPU utilization. Figure 6 (a) It can be seen that under the drive of the test program, The working state includes two obvious step changes and continuous short-term random fluctuations. Figure 6 (a) The left sub-graph selects the CPU working status around 650s for analysis. The fluctuation changes from 33% point ① to 42% point ② and then back to about 31%. Since the fluctuation change cycle is much smaller than the temperature dynamic change time constant, the CPU core temperature The change in the corresponding server power consumption is negligible. The change is only due to Some contributions, such as Figure 6 (b) and (c) are shown in the left sub-figure. At 750s, Produces a step change, Following the temperature delay change characteristics, it slowly rises from about 56.8℃ to about 59.1℃, such as Figure 6 (b) It also rises from about 128W to about 145.8W. Figure 6 By comparing the operating points ①② and ③④ before and after the step change, it can be observed that although the operating points ①④ The same is 34%, operating points ②③ Same as 42%, but corresponding to server power consumption However, there is a huge difference of 19W (the power consumption of ①② is 126.2W and 125.8W respectively, and the power consumption of ③④ is 145W and 144.8W respectively). Before and after the step change, the corresponding distribution of server power consumption and CPU utilization data shows stratification, such as Figure 6 As shown in (d), the traditional server power consumption model based only on CPU utilization will produce significant errors, affecting the calculation accuracy.

[0108] Figure 7 The model established by the present invention shows the degree of fit and error of the measured server power consumption value. The polynomial modeling method with the highest accuracy among the traditional modeling methods of the present invention (taking cubic polynomial as an example) is selected to compare the fit and error of the same data. It can be seen that compared with the traditional model, the model established by the present invention has higher calculation accuracy and smaller error.

[0109] Figure 8 The present invention shows the effect of eliminating errors in traditional modeling methods by taking temperature variables into account and performing temperature estimation. The polynomial modeling method with the highest accuracy among traditional modeling methods is selected (taking cubic polynomial as an example). The output data before and after the step are plotted and compared with the server power consumption-CPU utilization scatter plot. It can be seen that compared with the direct correspondence between CPU utilization and server power consumption in the traditional model, the model established by the present invention introduces temperature variables and CPU working state criteria, making its calculation results more consistent with the actual situation. Figure 6 (d) shows the server power consumption stratification phenomenon in actual situations, which reduces the model calculation error.

[0110] Although the embodiments and drawings of the present invention are disclosed for illustrative purposes, those skilled in the art will appreciate that various substitutions, changes and modifications are possible without departing from the spirit and scope of the present invention and the appended claims. Therefore, the scope of the present invention is not limited to the contents disclosed in the embodiments and drawings.

Claims

1. A server power consumption modeling method based on CPU working status estimation temperature, Features: The following steps are involved: Step 1, establishing a mathematical model of server power consumption taking into account temperature variables, which takes the actual measured value of CPU utilization and the actual measured value of CPU core temperature as input and outputs the calculated value of total server power consumption, and sets the coefficients in the mathematical model of server power consumption taking into account temperature variables; Step 2: Estimate the CPU core temperature based on the CPU working state according to the server power consumption mathematical model established in step 1, determine the CPU fluctuation or step different working states through the CPU utilization time series change, and obtain the CPU core temperature estimation value to replace the input of the CPU core temperature actual value of the server power consumption mathematical model considering the temperature variable established in step 1; Step 3: Based on the server power consumption mathematical model taking into account the temperature variable established in step 1 and the CPU core temperature estimation value obtained in step 2, with the goal of minimizing the model error, the time series length Δt in step 2 and the CPU working state judgment threshold R are calculated. MVD,th Conduct optimization search; The method for estimating the CPU core temperature based on the CPU working state comprises the following steps: Step 2.1, for t n Temperature estimation at time t n CPU utilization time series within Δt before the moment Step 2.2, for the CPU utilization time series U, calculate its mean dispersion MVD value: in, t n The average dispersion of the CPU utilization time series at the moment, which is smaller in the CPU fluctuating working state and larger in the CPU step working state; Δt is the length of the time series; is the CPU utilization at time t; Step 2.3, according to According to the changes in the CPU core temperature, it is determined that the CPU is in a fluctuating or stepping working state, and then the CPU core temperature estimate is updated. in, t n CPU core temperature at the moment; is the estimated temperature at the previous moment; t n The average dispersion of the CPU utilization time series at the moment; R MVD,th is the CPU working state judgment threshold; δt is the correction sequence length set to eliminate transient errors; is the CPU utilization at the tth moment; the update of the temperature estimation value is based on the corresponding relationship between the CPU core temperature steady-state value and the CPU utilization obtained in step 1.

7.

2. A server power consumption modeling method for estimating temperature based on CPU working status according to claim 1, Features: The method for establishing a mathematical model of server power consumption taking into account temperature variables comprises the following steps: Step 1.1, configuring the equipment required for measuring and analyzing the power consumption of the server, including a test program that can simulate computing load, an AIDA64 software detection engine, a power meter that can measure and record the actual value of the total power consumption of the server, and a computing server that can record and analyze the measurement data. The AIDA64 software detection engine is used to obtain the actual value of CPU utilization, the actual value of CPU power, the actual value of CPU core temperature, and the actual value of cooling fan speed; Step 1.2, run the test program on the test server and collect and record relevant data; Step 1.3, according to the measured value of CPU power consumption, the measured value of CPU utilization and the measured value of CPU core temperature, the CPU utilization coefficient a and the temperature coefficient b are obtained by fitting using the least square method according to the corresponding relationship between the three 0 , b 1 : in, t n CPU power consumption at the moment; P idle The idle power consumption of CPU is set to a constant value; t n CPU utilization at the moment; t n CPU core temperature at the moment; Step 1.4: Calculate the power consumption of the server cooling fan based on the measured values ​​of the CPU power consumption and the total power consumption of the server: in, t n The power consumption of the server cooling fans at the moment; t n Total power consumption of the server at the moment; t n CPU power consumption at the moment; P other The power consumption of the rest of the system except the CPU and cooling fan; Step 1.5, according to the cooling fan power consumption calculation data and the cooling fan speed measured value, the fan power consumption coefficient c is obtained by using the least squares method according to the corresponding relationship between the two. 0 、c 1 、c 2 and c 3 : in, t n The power consumption of the server cooling fans at the moment; t n The speed of the server cooling fans at the moment; Step 1.6: According to the measured value of the cooling fan speed and the measured value of the CPU core temperature, the fan speed f of the server under low load is obtained by using the least squares method according to the corresponding relationship between the two. min , Fan speed f when the server is under high load max The fan speed switches between constant speed and variable speed working states. min , T max , and the rate of change of speed in the speed range with temperature change k fan : in, t n The speed of the server cooling fans at the moment; t n CPU core temperature at the moment; Step 1.7, according to the measured value of CPU core temperature and the measured value of CPU utilization, according to the corresponding relationship between the steady-state value of CPU core temperature and CPU utilization obtained by using the equivalent thermal parameter model calculation method, the steady-state temperature parameter μ is obtained by least squares fitting. 0 , μ 1 and μ 2 : in, The CPU utilization is The steady-state value of CPU core temperature in the state; t n CPU usage at a certain moment.

3. A server power consumption modeling method for estimating temperature based on CPU working status according to claim 1, Features: The method for setting the coefficients in the mathematical model of server power consumption taking into account temperature variables is: running a test program on the server and measuring CPU utilization, CPU power, CPU core temperature, cooling fan speed and total server power consumption, and fitting and setting them using the least squares method based on the mathematical relationship.

4. A server power consumption modeling method for estimating temperature based on CPU working status according to claim 1, Features: The coefficients in the mathematical model of server power consumption taking into account temperature variables include CPU utilization coefficient a, temperature coefficient b 0 , b 1 , CPU idle power consumption P idle , fan power consumption coefficient c 0 、c 1 、c 2 、c 3 , the fan speed f of the server under low load min , the fan speed f under high load condition of the server max The fan speed switches between constant speed and variable speed working states. min The maximum temperature threshold T of the fan speed switching between constant speed and variable speed working states max , the rate of change of speed in the speed change range with temperature change k fan , steady-state temperature parameter μ 0 , μ 1 , μ 2 .

5. A server power consumption modeling method for estimating temperature based on CPU working status according to claim 1, Features: Based on the server power consumption mathematical model taking into account temperature variables established in step 1 and the CPU core temperature estimation value obtained in step 2, with the goal of minimizing the model error, the time series length Δt in step 2 and the CPU working state judgment threshold R are calculated. MVD,th The optimization method includes the following steps: Step 3.1, input test time t N Measured value of internal server power consumption CPU utilization measured value and coefficients in a mathematical model of server power consumption that accounts for temperature variation; Step 3.2, set the time series length Δt and the CPU working status judgment threshold R MVD,th Initial value; Step 3.3, update the time series length Δt and increase the unit value; Step 3.4, update the CPU working state judgment threshold R MVD,th Increase the unit value; Step 3.5, for t N At each moment, the actual measured value of CPU utilization U m Calculate the corresponding CPU core temperature estimate, and then obtain the server power consumption at all times in the server power consumption mathematical model taking into account the temperature variable Step 3.6, use the root mean square error R RMSE Measure the total power consumption of the server P srv,m The total power consumption model of the server calculates the value P srv Error between: Among them, t N is the total test time; is the measured value of the total power consumption of the server at time t; Calculate and record R for the total power consumption model of the server at time t. RMSE ; Step 3.7, if the CPU working state judgment threshold R MVD,th If the CPU working state judgment threshold R MVD,th When the set upper limit is reached, the CPU working state judgment threshold R MVD,th Reset the initial value and execute step 3.8; Step 3.8: If the time series length Δt does not reach the set upper limit, return to step 3.3; if the time series length Δt reaches the set upper limit, execute step 3.9; Step 3.9, based on the recorded root mean square error R RMSE The minimum root mean square error value is found in the data, that is, the time series length Δt and the CPU working state judgment threshold R of the most accurate model are MVD,th .

Images