A server resource scheduling method and a server

Abstract

The application relates to a server resource scheduling method and a server, wherein the server scheduling method comprises the following steps: acquiring resource demand and current temperatures of sub-servers, inputting the resource demand and the current temperatures of the sub-servers into a pre-trained prediction model, and outputting predicted temperatures and predicted loadable amounts of the sub-servers by the prediction model; and distributing the resource demand to the sub-servers by using a dynamic load balancing algorithm and the predicted loadable amounts of the sub-servers. The application effectively avoids server crashes or disconnections caused by overloading of the sub-servers.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of server resource scheduling, and particularly relates to a server resource scheduling method and a server. BACKGROUND

[0002] There are many factors affecting the temperature of a server, such as a mechanical heat dissipation system, an ambient temperature, etc. When the temperature of a server is too high, a mechanical heat dissipation system is used first to dissipate heat. When the heat dissipation is insufficient, the load of the server is reduced, or even the server is controlled to stop serving, so as to reduce the temperature of the server as soon as possible.

[0003] However, directly reducing the load of the server will cause the server to be unable to provide sufficient resources for the client in a short time, greatly affecting the utilization of server resources. SUMMARY

[0004] The present application actively adjusts the load of each sub-server, so that each sub-server works in a suitable temperature range.

[0005] In a first aspect, the present application provides a server resource scheduling method, which comprises:

[0006] obtaining resource demand and current temperature of each sub-server, and inputting the resource demand and the current temperature of each sub-server into a pre-trained prediction model, wherein the prediction model outputs predicted temperature and predicted load capacity of each sub-server;

[0007] allocating the resource demand to each sub-server by using a dynamic load balancing algorithm and the predicted load capacity of each sub-server.

[0008] In some embodiments, the obtaining resource demand and current temperature of each sub-server, and inputting the resource demand and the current temperature of each sub-server into a pre-trained prediction model, wherein the prediction model outputs predicted temperature and predicted load capacity of each sub-server, comprises:

[0009] predicting temperature of each sub-server when the resource demand is allocated to each sub-server by using a load balancing algorithm and standard load capacity of each sub-server;

[0010] obtaining a decay weight based on the predicted temperature of each sub-server;

[0011] multiplying the standard load capacity of each sub-server by the decay weight to obtain the predicted load capacity.

[0012] In some embodiments, the obtaining a decay weight based on the predicted temperature of each sub-server comprises:

[0013] configuring the decay weight as 1 when the predicted temperature of the sub-server is less than a first temperature threshold;

[0014] In a case where the predicted temperature of the sub-server is greater than the first temperature threshold, the attenuation weight is calculated according to the size of the predicted temperature, wherein the predicted temperature is negatively correlated with the attenuation weight.

[0015] In some embodiments, the attenuation weight is calculated according to the size of the predicted temperature, comprising:

[0016] confirming a temperature interval in which the predicted temperature is located;

[0017] confirming the attenuation weight based on the temperature interval.

[0018] In some embodiments, the attenuation weight is confirmed based on the temperature interval, comprising:

[0019] assigning a standard weight interval to each temperature interval;

[0020] using the predicted temperature and a larger endpoint value of the temperature interval in which the predicted temperature is located to perform mean interpolation;

[0021] obtaining the attenuation weight according to the result of the interpolation and a first formula:

[0022] the first formula: y min +(1-x i -x min )×(y max -y min );

[0023] x max -x min

[0024] wherein y min is a smaller endpoint value of the standard weight interval, y max is a larger endpoint value of the standard weight interval, x min is a smaller endpoint value of the temperature interval, x max is a larger endpoint value of the temperature interval, and x i is the result of the interpolation.

[0025] In some embodiments, the temperature interval comprises a first temperature interval, a second temperature interval and a third temperature interval; the first temperature interval corresponds to a first attenuation weight, the second temperature interval corresponds to a second attenuation weight, and the third temperature interval corresponds to a third attenuation weight; wherein,

[0026] the temperature of the first temperature interval is less than the temperature of the second temperature interval, the temperature of the second temperature interval is less than the temperature of the third temperature interval; and the first attenuation weight is less than the second attenuation weight, and the second attenuation weight is less than the third attenuation weight.

[0027] In some embodiments, the difference between the first and second decay weights is less than the difference between the second and third decay weights.

[0028] In some embodiments, after the resource demand is allocated to each sub-server using the dynamic load balancing algorithm and the predicted load capacity of each sub-server, and before the next resource demand is obtained, the method further comprises:

[0029] periodically predicting the temperature of each sub-server when the resource demand is allocated to each sub-server using the dynamic load balancing algorithm and the predicted load capacity of each sub-server;

[0030] updating the decay weight using the periodically obtained predicted temperature;

[0031] updating the predicted load capacity using the updated decay weight.

[0032] In some embodiments, the pre-trained prediction model comprises a recurrent neural network, wherein,

[0033] The training data of the recurrent neural network comprises a temperature change curve and a load capacity of a preset duration;

[0034] The label of the recurrent neural network comprises an actually measured temperature value.

[0035] In a second aspect, the present application provides a server, comprising:

[0036] a plurality of sub-servers;

[0037] a controller connected to the temperature control system of each sub-server to obtain the current temperature of each sub-server output by the temperature control system; the controller is configured to execute a server resource scheduler to implement the above method.

[0038] The present application first adjusts the predicted load capacity using the predicted temperature information, and then performs task allocation / resource allocation based on the predicted load capacity dynamic load balancing algorithm, so as to dynamically adjust the load capacity of each sub-server according to the temperature and workload, and effectively avoid server crash / line drop caused by sub-server overload. BRIEF DESCRIPTION OF DRAWINGS

[0039] Figure 1 a flowchart of an embodiment of the server resource scheduling method of the present application;

[0040] Figure 2 a flowchart of another embodiment of the server resource scheduling method of the present application;

[0041] Figure 3A flow chart of another embodiment of the server resource scheduling method of the application. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only some of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, any other embodiments obtained by a person skilled in the art without creative work fall within the protection scope of the application.

[0043] In the description of the application, the terms "first", "second" are used only for the purpose of description, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first", "second" can explicitly or implicitly include one or more of the features. In the description of the application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0044] In the description of the application, the term "for example" is used to indicate "as an example, illustration or description". Any embodiment described as "for example" in the application is not necessarily interpreted as more preferred or more advantageous than other embodiments. The following description is given to enable any person skilled in the art to implement and use the application. In the following description, details are listed for the purpose of explanation. It should be understood that a person skilled in the art can realize the application without using these specific details. In other examples, well-known structures and processes will not be described in detail to avoid unnecessary details making the description of the application obscure. Therefore, the application is not intended to be limited to the shown embodiments, but is consistent with the broadest scope of principles and features disclosed.

[0045] Example 1

[0046] Reference Figure 1 In a first aspect, the application provides a server resource scheduling method, which comprises:

[0047] S100, obtaining resource demand and current temperature of each sub-server, and inputting them into a pre-trained prediction model, wherein the prediction model outputs predicted temperature and predicted loadable amount of each sub-server;

[0048] Exemplarily, the server can be composed of multiple sub-servers, each of which has a standard loadable amount calibrated in advance, which is the maximum loadable amount that the sub-server can provide when working in the optimal temperature range.

[0049] In some embodiments, the prediction model can also be a non-neural network model, but a fitted curve, with the independent variable being the resource demand and the dependent variable being the predicted temperature of each sub-server.

[0050] In the embodiments of the present application, the pre-trained prediction model is preferably a neural network. Compared with other models, the neural network can more accurately regress and predict the predicted temperature and the predicted loadable amount. For example, the neural network can be one or a combination of a convolutional neural network (CNN), a recurrent neural network (RNN), and a deep neural network (DNN), which is selected according to actual needs and is not limited herein. In the embodiments, the recurrent neural network is preferred, which has continuity and is more suitable for processing continuous data and performing the prediction function.

[0051] In actual applications, in order to train the neural network, training data can be prepared, and the training data can be divided into a training set, a test set, and a validation set. Each training data is labeled with a “predicted temperature”. After iterative training, the prediction model obtains a set of converged parameters. Then, the predicted temperature and the predicted loadable amount of each sub-server can be output according to the input resource demand and the current temperature of each sub-server.

[0052] S200, using a dynamic load balancing algorithm and the predicted loadable amount of each sub-server, the resource demand is distributed to each sub-server.

[0053] The dynamic load balancing algorithm is an algorithm that determines task allocation according to the real-time load state of the server. Unlike the static load balancing algorithm, the dynamic load balancing algorithm can adjust according to the current performance and state of the server, thereby more effectively utilizing resources and improving the overall performance of the system.

[0054] On the basis of the dynamic balancing algorithm, the present application updates the predicted loadable amount using the predicted temperature information, and then distributes tasks based on the predicted loadable amount using the dynamic balancing algorithm.

[0055] The present application first adjusts the predicted loadable amount using the predicted temperature information, and then distributes tasks based on the predicted loadable amount using the dynamic load balancing algorithm, thereby dynamically adjusting the load of each sub-server according to the temperature and the workload, and effectively avoiding server crashes / line drops caused by overloading of the sub-server.

[0056] Reference Figure 2 In some embodiments, S100, the resource demand and the current temperature of each sub-server are input into a pre-trained prediction model, and the prediction model outputs the predicted temperature and the predicted loadable amount of each sub-server, including:

[0057] S101, when allocating the resource demand to each sub-server using a load balancing algorithm and a standard load capacity of each sub-server, a predicted temperature of each sub-server;

[0058] The standard load capacity can be determined by calibration, i.e., the load capacity of the server when the server works in an optimal environment. Step S101 can be implemented by a neural network model.

[0059] S102, obtaining a decay weight based on the predicted temperature of each sub-server;

[0060] In the present application, the higher the predicted temperature, the greater the decay weight. In the present application, the predicted temperature and the decay weight are not in a linear relationship, but should change at a greater rate with the increase of the predicted temperature. For example, an exponential relationship.

[0061] S103, multiplying the standard load capacity of each sub-server by the decay weight to obtain a predicted load capacity. That is, as the predicted temperature increases, the predicted load capacity decreases, and decreases at a faster rate.

[0062] In some embodiments, S102, obtaining a decay weight based on the predicted temperature of each sub-server, comprises:

[0063] In the case where the predicted temperature of the sub-server is less than a first temperature threshold, the decay weight is configured to be 1;

[0064] In the case where the predicted temperature of the sub-server is greater than the first temperature threshold, the decay weight is calculated according to the size of the predicted temperature, wherein the predicted temperature and the decay weight are negatively correlated.

[0065] In the present application, below the first temperature threshold means that the server works in a better temperature range and does not need to be attenuated. Above the first temperature threshold, the decay weight is calculated according to the size of the predicted temperature.

[0066] Exemplarily, the present application gives several methods for determining the decay weight, which are described in the following embodiments:

[0067] Referring to Figure 3 In some embodiments, the decay weight is calculated according to the size of the predicted temperature, comprising:

[0068] S1021, determining a temperature range in which the predicted temperature is located;

[0069] The decay weight is determined based on the temperature range.

[0070] That is, we first divide multiple temperature intervals. Exemplarily, the temperature intervals cannot be divided equally, but the higher the temperature, the smaller the temperature interval range, and the lower the temperature, the larger the temperature interval range. This is more in line with the fact that the server is more affected by temperature at high temperature and less affected by temperature at low temperature.

[0071] Specifically, the step of confirming the attenuation weight based on the temperature interval comprises:

[0072] S10221, assigning a standard weight interval to each temperature interval;

[0073] S10222, using the predicted temperature and the larger endpoint value of the temperature interval where the predicted temperature is located to perform mean interpolation;

[0074] S10223, obtaining the attenuation weight according to the result of the interpolation and a first formula:

[0075] The first formula is:

[0076] Wherein, y min is the smaller endpoint value of the standard weight interval, y max is the larger endpoint value of the standard weight interval, x min is the smaller endpoint value of the temperature interval, x max is the larger endpoint value of the temperature interval, x i is the result of the interpolation.

[0077] Exemplarily, the temperature interval is (60-75], the standard weight interval is (0.8-0.9], and the predicted temperature is 68 degrees. At this time, mean interpolation can be performed between 68 and 75 to obtain a result of 71.5; then the temperature interval (60-75] is normalized, that is, the data is mapped to the range of 0-1 for processing, and the normalized result is That is, 0.77, at this time, 0.8+(1-0.77)×(0.9-0.8)=0.823.

[0078] In the above calculation formula, as the predicted temperature is closer to the larger endpoint value, the attenuation weight is larger, thereby better adjusting the temperature and confirming the predicted loadable amount.

[0079] In other embodiments, a standard weight is assigned to each temperature interval, for example, the temperature interval is (60-75], and the standard weight is 0.85.

[0080] Exemplarily, the temperature intervals include a first temperature interval, a second temperature interval, and a third temperature interval; the first temperature interval corresponds to a first attenuation weight, the second temperature interval corresponds to a second attenuation weight, and the third temperature interval corresponds to a third attenuation weight; wherein,

[0081] The temperature of the first temperature interval is less than the temperature of the second temperature interval, and the temperature of the second temperature interval is less than the temperature of the third temperature interval; and the first decay weight is less than the second decay weight, and the second decay weight is less than the third decay weight.

[0082] In some embodiments, the difference between the first decay weight and the second decay weight is less than the difference between the second decay weight and the third decay weight.

[0083] In the present embodiment, the decay weights of the same temperature interval are the same, which can reduce the amount of calculation;

[0084] And between different temperature intervals, with the increase of temperature, the decay weight increases nonlinearly, and the slope becomes larger, that is, with the increase of temperature, the decay weight increases in the form of an exponential.

[0085] In some embodiments, after the resource demand is allocated to each sub-server using the dynamic load balancing algorithm and the predicted load capacity of each sub-server, and before the next resource demand is obtained, the method further comprises:

[0086] Periodically predicting the predicted temperature of each sub-server when the resource demand is allocated to each sub-server using the dynamic load balancing algorithm and the predicted load capacity of each sub-server;

[0087] Updating the decay weight using the periodically obtained predicted temperature;

[0088] Updating the predicted load capacity using the updated decay weight.

[0089] In the present embodiment, the frequency of obtaining resource demand can be relatively low, for example, in model training, the task is allocated once every hour or even half a day. In order to better adjust the temperature of each sub-server, the decay weight updating strategy is added in the present application, that is, the predicted temperature of each sub-server is periodically predicted (for example, every 10 minutes), and then the predicted temperature is used to update the decay weight, and then the load capacity is updated.

[0090] In some embodiments, the pre-trained prediction model comprises a recurrent neural network, wherein,

[0091] The training data of the recurrent neural network comprises a temperature change curve and a load capacity of a preset time length;

[0092] The label of the recurrent neural network comprises a measured temperature value.

[0093] It should be noted that the recurrent neural network can be an RNN, an LSTM, a GRU, or other more complex recurrent neural network, and the embodiment is not limited, and preferably an LSTM, which avoids the problem of excessive memory compared to an RNN, and has a smaller calculation amount compared to a GRU.

[0094] In practical applications, the training data and labels can be matched first, and then divided into a training set, a test set and a validation set, and then trained, tested and validated.

[0095] Example 2

[0096] The application provides a server, comprising:

[0097] a plurality of sub-servers;

[0098] a controller connected to the temperature control system of each sub-server to obtain the current temperature of each sub-server output by the temperature control system; the controller is configured to execute a server resource scheduler to implement the following method:

[0099] S100, obtaining the resource demand and the current temperature of each sub-server, and inputting them into a pre-trained prediction model, wherein the prediction model outputs the predicted temperature and the predicted loadable amount of each sub-server;

[0100] S200, using a dynamic load balancing algorithm and the predicted loadable amount of each sub-server to distribute the resource demand to each sub-server.

[0101] Referring to Figure 2 In some embodiments, S100, obtaining the resource demand and the current temperature of each sub-server, and inputting them into a pre-trained prediction model, wherein the prediction model outputs the predicted temperature and the predicted loadable amount of each sub-server, comprises:

[0102] S101, predicting the predicted temperature of each sub-server when the resource demand is distributed to each sub-server using a load balancing algorithm and the standard loadable amount of each sub-server;

[0103] S102, obtaining a decay weight based on the predicted temperature of each sub-server;

[0104] S103, multiplying the standard loadable amount of each sub-server by the decay weight to obtain the predicted loadable amount.

[0105] Referring to Figure 3 In some embodiments, S102, obtaining a decay weight based on the predicted temperature of each sub-server, comprises:

[0106] In the case where the predicted temperature of the sub-server is less than a first temperature threshold, the decay weight is configured to be 1.

[0107] In a case where the predicted temperature of the sub-server is greater than the first temperature threshold, the attenuation weight is calculated according to the size of the predicted temperature, wherein the predicted temperature is negatively correlated with the attenuation weight.

[0108] In some embodiments, the attenuation weight is calculated according to the size of the predicted temperature, comprising:

[0109] S1021, confirming a temperature interval in which the predicted temperature is located;

[0110] S1022, confirming the attenuation weight based on the temperature interval.

[0111] In some embodiments, S1022, the attenuation weight is confirmed based on the temperature interval, comprising:

[0112] S10221, assigning a standard weight interval to each temperature interval;

[0113] S10222, using the predicted temperature and the larger endpoint value of the temperature interval in which the predicted temperature is located to perform mean interpolation;

[0114] S10223, obtaining the attenuation weight according to the result of the interpolation and a first formula:

[0115] The first formula is:

[0116] wherein y min is a smaller endpoint value of the standard weight interval, y max is a larger endpoint value of the standard weight interval, x min is a smaller endpoint value of the temperature interval, x max is a larger endpoint value of the temperature interval, and x i is the result of the interpolation.

[0117] In some embodiments, the temperature interval comprises a first temperature interval, a second temperature interval, and a third temperature interval; the first temperature interval corresponds to a first attenuation weight, the second temperature interval corresponds to a second attenuation weight, and the third temperature interval corresponds to a third attenuation weight; wherein,

[0118] the temperature of the first temperature interval is less than the temperature of the second temperature interval, the temperature of the second temperature interval is less than the temperature of the third temperature interval; and the first attenuation weight is less than the second attenuation weight, and the second attenuation weight is less than the third attenuation weight.

[0119] In some embodiments, the difference between the first attenuation weight and the second attenuation weight is less than the difference between the second attenuation weight and the third attenuation weight.

[0120] In some embodiments, after the resource demand is allocated to each sub-server using the dynamic load balancing algorithm and the predicted load capacity of each sub-server, and before the next resource demand is obtained, the method further comprises:

[0121] periodically predicting a predicted temperature of each sub-server when the resource demand is allocated to each sub-server using the dynamic load balancing algorithm and the predicted load capacity of each sub-server;

[0122] updating the decay weight using the predicted temperature periodically obtained;

[0123] updating the predicted load capacity using the updated decay weight.

[0124] In some embodiments, the pre-trained prediction model comprises a recurrent neural network, wherein,

[0125] the training data of the recurrent neural network comprises a temperature change curve and a load capacity in a preset time period;

[0126] the label of the recurrent neural network comprises a measured temperature value.

[0127] It should be noted that in the above embodiments, the description of each embodiment has its own focus, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0128] Those skilled in the art will understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0129] The present application is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, as well as combinations of flows and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded computer, or other programmable data processing devices to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices produce a device that implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The functions specified in one or more flows and / or blocks Figure 1 The functions specified in one or more flows and / or blocks

[0130] These computer program instructions can also be stored in a computer- readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions Figure 1 one or more functions specified in the flow or flows and / or blocks Figure 1 one or more functions specified in the flow or flows and / or blocks

[0131] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions Figure 1 one or more functions specified in the flow or flows and / or blocks Figure 1 one or more functions specified in the flow or flows and / or blocks

[0132] While the preferred embodiments of the application have been described, additional variations and modifications can be made to the preferred embodiments by those skilled in the art once they learn of the basic inventive concepts. Therefore, the appended claims are intended to cover all such additional variations and modifications as fall within the scope of the application.

[0133] Obviously, numerous modifications and variations of the present application are possible in light of the above teachings. It is therefore to be understood that within the scope of the appended claims and their equivalents, the application can be practiced otherwise than as specifically described.

Claims

1. A server resource scheduling method, characterized in that, include: The resource requirements and the current temperature of each sub-server are obtained and input into a pre-trained prediction model. The prediction model outputs the predicted temperature and predicted load capacity of each sub-server. The resource requirements are allocated to each sub-server using a dynamic load balancing algorithm and the predicted load capacity of each sub-server. The resource demand and the current temperature of each sub-server are obtained and input into a pre-trained prediction model. The prediction model outputs the predicted temperature and predicted load capacity of each sub-server, including: The predicted temperature of each sub-server is calculated when the resource demand is allocated to each sub-server using a load balancing algorithm and the standard load capacity of each sub-server. The attenuation weight is obtained based on the predicted temperature of each sub-server; The predicted load capacity is obtained by multiplying the attenuation weight by the standard load capacity of each sub-server; the attenuation weight is obtained based on the predicted temperature of each sub-server, including: If the predicted temperature of the sub-server is less than the first temperature threshold, configure the attenuation weight to be 1; If the predicted temperature of the sub-server is greater than a first temperature threshold, the attenuation weight is calculated based on the predicted temperature, wherein the predicted temperature is negatively correlated with the attenuation weight; the calculation of the attenuation weight based on the predicted temperature includes: Confirm the temperature range in which the predicted temperature falls; The attenuation weight is determined based on the temperature range; The determination of the attenuation weight based on the temperature range includes: Assign a standard weighted interval to each temperature range; Mean interpolation is performed using the larger endpoint value of the predicted temperature and the temperature range in which the predicted temperature falls. The attenuation weight is obtained based on the interpolation result and the first formula: First formula: ; Among them, y min y represents the smaller endpoint of the standard weighted interval. max x represents the larger endpoint of the standard weighted interval. min x represents the smaller endpoint of the temperature range. max x represents the larger endpoint of the temperature range. i This is the result of interpolation.

2. The server resource scheduling method as described in claim 1, characterized in that, The temperature range includes a first temperature range, a second temperature range, and a third temperature range; the first temperature range corresponds to a first attenuation weight, the second temperature range corresponds to a second attenuation weight, and the third temperature range corresponds to a third attenuation weight; wherein... The temperature in the first temperature range is lower than the temperature in the second temperature range, and the temperature in the second temperature range is lower than the temperature in the third temperature range; and the first attenuation weight is lower than the second attenuation weight, and the second attenuation weight is lower than the third attenuation weight.

3. The server resource scheduling method as described in claim 2, characterized in that, The difference between the first attenuation weight and the second attenuation weight is less than the difference between the second attenuation weight and the third attenuation weight.

4. The server resource scheduling method as described in any one of claims 1-3, characterized in that, After allocating the resource demand to each sub-server using a dynamic load balancing algorithm and the predicted load capacity of each sub-server, and before obtaining the next resource demand, the method further includes: The predicted temperature of each sub-server is predicted when the resource demand is allocated to each sub-server using a dynamic load balancing algorithm and the predicted load capacity of each sub-server periodically. The decay weights are updated using the predicted temperatures obtained periodically. The predicted load capacity is updated using the updated decay weights.

5. The server resource scheduling method as described in claim 1, characterized in that, Pre-trained prediction models include: recurrent neural networks, where, The training data for the recurrent neural network includes: a temperature change curve of a preset duration and a load amount; The labels of the recurrent neural network include: measured temperature values.

6. A server, characterized in that, include: Multiple sub-servers; The controller connects to the temperature control system of each sub-server to obtain the current temperature of each sub-server output by the temperature control system. The controller is configured to execute a server resource scheduler to implement the method as described in any one of claims 1-5.

Images