Energy efficiency perception and load distribution method and system for distributed network acceleration nodes

Abstract

The invention provides an energy efficiency perception and load distribution method and system for distributed network acceleration nodes, and relates to the technical field of communication. Inputting the server performance portrait characteristics of each business server into an energy efficiency estimation model, determining the Nth estimated energy efficiency of each business server under different loads, and calculating the Nth estimated energy efficiency and the Nth total business access volume of each business server under different loads and the Nth existing load of each business server according to the Nth estimated energy efficiency and the Nth total business access volume of each business server under different loads; and inputting the data into a business page view distribution model, so that the business page view distribution model distributes the business sub-page views to each business server. Determining the Nth average actual energy efficiency of the plurality of service servers according to the Nth actual energy efficiency of the plurality of service servers; and if the Nth average energy efficiency is smaller than the (N-1) th average energy efficiency, and the difference value between the Nth average energy efficiency and the (N-1) th average energy efficiency is greater than a set difference value threshold value, updating network parameters of the business access volume distribution model by adopting a reinforcement learning model.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a method and system for energy efficiency perception and load distribution of distributed network acceleration nodes. Background Art

[0002] A distributed network is a network architecture consisting of interconnected computer systems located in different locations. There is no central node in the network, and nodes collaborate to share resources and process tasks. Distributed network acceleration nodes are key components in distributed networks, optimizing network performance and accelerating data transmission.

[0003] Among them, the energy efficiency of each server in the distributed network (that is, the ratio between energy consumption and actual processing workload) is related to the operating cost of the distributed network and the user experience, and is an indicator that technical personnel are more concerned about.

[0004] Currently, in some scenarios, it's possible to maintain a constant total number of visits per unit time on a distributed network (e.g., batch processing scenarios where scheduled tasks execute at a fixed frequency, or registration systems that control the number of registrations at a fixed rate). In these scenarios, DNS round-robin can be used to balance the load across servers in the distributed network, thereby improving resource utilization, reducing latency and energy consumption, and ultimately reducing the energy efficiency of the distributed network. However, even with these approaches, the energy efficiency of the distributed network remains relatively high. Summary of the Invention

[0005] The present application provides a method and system for energy efficiency perception and load distribution of distributed network acceleration nodes, which is used to solve the problem of high energy efficiency of each server in a distributed network in the existing technology when the total number of visits per unit time remains unchanged.

[0006] In a first aspect, the present application provides a method for energy efficiency perception and load distribution of a distributed network acceleration node, which is applied to a distributed network acceleration node. The distributed network acceleration node is respectively connected to multiple subordinate business servers. The method provided by the present application includes: Step 1: At the Nth time slice, obtain the server type of each business server among the multiple business servers, the environmental data of each business server at the Nth time slice, and the hardware health of each business server, where N is an integer greater than or equal to 2; Step 2: For each business server, build a server performance profile based on the server type, the environment data of the business server at the Nth time slice, and the hardware health of the business server; Step 3: Input the server performance profile features of each business server into a pre-trained energy efficiency estimation model to determine the Nth estimated energy efficiency of each business server under different loads, wherein the energy efficiency estimation model is trained by inputting multiple first training samples into a first neural network, each first training sample including historical server performance profile features and corresponding historical Nth estimated energy efficiency; Step 4: Obtain the Nth total service access volume of the terminal device set and the Nth existing load of each service server; Step 5: Inputting the Nth estimated energy efficiency of each service server under different loads, the Nth total service access volume, and the Nth existing load of each service server into a pre-trained service access volume distribution model, so that the service access volume distribution model distributes service sub-access volume to each service server according to pre-configured network parameters, wherein the service access volume distribution model is trained by inputting multiple second training samples into a second neural network, each second training sample including the historical Nth estimated energy efficiency of each historical service server under different loads, the historical Nth total service access volume of a historical set of terminal devices, and the historical Nth existing load of each historical service server; Step 6: Detect the Nth actual energy efficiency of each business server, and determine the Nth average actual energy efficiency of the multiple business servers based on the respective Nth actual energy efficiencies of the multiple business servers; Step 7: If the Nth average energy efficiency of the multiple service servers is less than the N-1th average energy efficiency of the multiple service servers, and the difference between the Nth average energy efficiency and the N-1th average energy efficiency is greater than the set difference threshold, then based on the Nth average energy efficiency, the reinforcement learning model is used to update the network parameters of the service access distribution model; Step 8: Add 1 to the value of N and return to step 1 until the difference between the Nth average energy efficiency and the N-1th average energy efficiency is less than or equal to the set difference threshold; Step 9: Repeat steps 1 to 5 until the distributed network acceleration node is powered off.

[0007] In some embodiments, before obtaining the Nth total service access volume of the terminal device set and the Nth existing load of each service server, the method provided by the present application further includes: For each load of each business server, obtain the historical energy efficiency of the business server under load for multiple consecutive time slices within a previous preset time period; Based on the deep confidence DBN network, the temporal dependency of the historical energy efficiency of the business server under load for multiple consecutive time slices within a preset time period is extracted; Based on the historical energy efficiency of the business server under load for multiple consecutive time slices within a previous preset period, and the temporal dependency of the historical energy efficiency of the business server under load for multiple consecutive time slices within a previous preset period, the energy efficiency variation characteristics of the business server under load are obtained; For each business server, inputting the energy efficiency change characteristics under each load into a pre-trained energy efficiency prediction model to predict the Nth calibrated estimated energy efficiency of the business server under each load in the Nth time slice, wherein the energy efficiency prediction model is trained by inputting multiple third training samples into a long short-term memory network, each third training sample including the historical energy efficiency change characteristics of the historical business server under each load and the corresponding historical Nth calibrated estimated energy efficiency of the historical business server under each load in the historical Nth time slice; For each business server, the Nth estimated energy efficiency under the corresponding load is calibrated according to the Nth calibrated estimated energy efficiency under each load.

[0008] In some implementations, for each service server, calibrating the Nth estimated energy efficiency under the corresponding load according to the Nth calibrated estimated energy efficiency under each load includes: According to the formula , calibrate the Nth estimated energy efficiency under the corresponding load, where is the Nth estimated energy efficiency of the jth load of the i-th business server in the Nth time slice before calibration, The estimated energy efficiency of the Nth calibration for the jth load of the i-th business server in the Nth time slice is, is the calibrated Nth estimated energy efficiency of the jth load of the i-th business server in the Nth time slice, is the first weighting coefficient, is the second weighting coefficient, and .

[0009] In some implementations, obtaining the hardware health of each service server includes: Obtain the wear of each business server's solid-state drive, the accuracy of its operation logs, and the stability of its power supply voltage; The wear of the solid-state hard drive of each business server, the accuracy of the operation log, and the stability of the power supply voltage are input into the pre-trained hardware health determination model to determine the hardware health of each business server, wherein the hardware health determination model is obtained by inputting multiple fourth training samples into the third neural network for training, and each fourth training sample includes the historical wear of the solid-state hard drive of the historical business server, the accuracy of the historical operation log, the stability of the historical power supply voltage, and the corresponding historical actual hardware health.

[0010] In some embodiments, the network parameters of the service access volume distribution model are the state of the reinforcement learning model, updating the network parameters of the service access volume distribution model is the action of the reinforcement learning model, the Nth average energy efficiency is the reward of the reinforcement learning model, and the lower the Nth average energy efficiency, the higher the reward.

[0011] In a second aspect, the present application also provides an energy efficiency perception and load distribution system for a distributed network acceleration node, which is applied to a distributed network acceleration node. The distributed network acceleration node is respectively connected to multiple subordinate business servers. The system provided by the present application includes: a data acquisition unit, configured to acquire, in an Nth time slice, a server type of each of the plurality of business servers, environmental data of each of the business servers in the Nth time slice, and a hardware health status of each of the business servers, where N is an integer greater than or equal to 2; A feature construction unit is used to construct a server performance profile feature for each business server based on the server type of the business server, the environmental data of the business server at the Nth time slice, and the hardware health of the business server; an energy efficiency determination unit, configured to input the server performance profile features of each business server into a pre-trained energy efficiency estimation model to determine an Nth estimated energy efficiency of each business server under different loads, wherein the energy efficiency estimation model is trained by inputting a plurality of first training samples into a first neural network, each first training sample including a historical server performance profile feature and a corresponding historical Nth estimated energy efficiency; The data acquisition unit is further used to obtain the Nth total service access volume of the terminal device set and the Nth existing load of each service server; an access volume allocation unit, configured to input the Nth estimated energy efficiency of each service server under different loads, the Nth total service access volume, and the Nth existing load of each service server into a pre-trained service access volume allocation model, so that the service access volume allocation model allocates service sub-access volumes to each service server according to pre-configured network parameters, wherein the service access volume allocation model is trained by inputting multiple second training samples into a second neural network, each second training sample including the Nth historical estimated energy efficiency of each historical service server under different loads, the Nth historical total service access volume of a historical set of terminal devices, and the Nth historical existing load of each historical service server; an energy efficiency detection unit, configured to detect an Nth actual energy efficiency of each business server, and determine an Nth average actual energy efficiency of the plurality of business servers based on the respective Nth actual energy efficiencies of the plurality of business servers; a parameter updating unit configured to update the network parameters of the service access volume distribution model using a reinforcement learning model based on the Nth average energy efficiency, if the Nth average energy efficiency of the plurality of service servers is less than the N-1th average energy efficiency of the plurality of service servers, and the difference between the Nth average energy efficiency and the N-1th average energy efficiency is greater than a set difference threshold; add 1 to the value of N, and return to step 1 until the difference between the Nth average energy efficiency and the N-1th average energy efficiency is less than or equal to the set difference threshold; The service execution unit is configured to execute steps 1 to 5 in a loop until the distributed network acceleration node is powered off.

[0012] In some embodiments, the data acquisition unit is further configured to acquire, for each load of each business server, a historical energy efficiency of the business server under the load for a plurality of consecutive time slices within a previous preset time period; The device provided by the present application also includes: a relationship extraction unit, which is used to extract the temporal dependency relationship of the historical energy efficiency of the service server under load for multiple consecutive time slices within a previous preset time period based on the deep confidence DBN network; a characteristic obtaining unit, configured to obtain an energy efficiency variation characteristic of the service server under load based on historical energy efficiencies of the service server for a plurality of consecutive time slices under load within a previous preset duration, and a temporal dependency relationship between the historical energy efficiencies of the service server for a plurality of consecutive time slices under load within a previous preset duration; an energy efficiency prediction unit, configured to input, for each service server, the energy efficiency variation characteristics under each load into a pre-trained energy efficiency prediction model, so as to predict the Nth calibrated estimated energy efficiency of the service server under each load in the Nth time slice, wherein the energy efficiency prediction model is trained by inputting a plurality of third training samples into a long short-term memory network, each third training sample including a historical energy efficiency variation characteristic of the historical service server under each load and a corresponding historical Nth calibrated estimated energy efficiency of the historical service server under each load in the Nth historical time slice; The energy efficiency calibration unit is used to calibrate the Nth estimated energy efficiency under the corresponding load for each business server according to the Nth calibrated estimated energy efficiency under each load.

[0013] In some embodiments, the energy efficiency calibration unit is specifically configured to calculate , calibrate the Nth estimated energy efficiency under the corresponding load, where is the Nth estimated energy efficiency of the jth load of the i-th business server in the Nth time slice before calibration, The estimated energy efficiency of the Nth calibration for the jth load of the i-th business server in the Nth time slice is, is the calibrated Nth estimated energy efficiency of the jth load of the i-th business server in the Nth time slice, is the first weighting coefficient, is the second weighting coefficient, and .

[0014] In some embodiments, the data acquisition unit is specifically used to obtain the wear of the solid-state hard disk of each business server, the accuracy of the operation log, and the stability of the power supply voltage; the wear of the solid-state hard disk of each business server, the accuracy of the operation log, and the stability of the power supply voltage are input into a pre-trained hardware health determination model to determine the hardware health of each business server, wherein the hardware health determination model is obtained by inputting multiple fourth training samples into the third neural network for training, and each fourth training sample includes the historical wear of the solid-state hard disk of the historical business server, the accuracy of the historical operation log, the stability of the historical power supply voltage, and the corresponding historical actual hardware health.

[0015] In some embodiments, the network parameters of the service access volume distribution model are the state of the reinforcement learning model, updating the network parameters of the service access volume distribution model is the action of the reinforcement learning model, the Nth average energy efficiency is the reward of the reinforcement learning model, and the lower the Nth average energy efficiency, the higher the reward.

[0016] The present application provides a method and system for energy efficiency perception and load distribution of distributed network acceleration nodes. For each business server, a server performance profile feature is constructed based on the server type of the business server, the environmental data of the business server in the Nth time slice, and the hardware health of the business server. It can be understood that the processing capabilities and energy efficiency of business servers of different server types are also different; the processing capabilities of the same business server under different environmental data (such as temperature and / or humidity) are different. For example, excessively high temperature will cause the business server to reduce its frequency, and different temperatures will have different reduction amplitudes on the business server, which in turn will lead to a decrease in the computing power of the business server. Excessively high humidity will corrode the hardware of the business server, and different humidity will corrode the hardware of the business server to different degrees, resulting in a decrease in the processing power of the business server. In addition, when the hardware health of the business server is different, the corresponding processing capabilities are also different. Therefore, the reliability of the server performance profile feature constructed in the above manner is high.

[0017] The server performance profile features of each business server are input into a pre-trained energy efficiency estimation model to determine the Nth estimated energy efficiency of each business server under different loads. Because the server performance profile features are highly reliable, and because the energy efficiency estimation model is trained by inputting multiple first training samples into a first neural network, each first training sample includes historical server performance profile features and the corresponding historical Nth estimated energy efficiency. This allows for a high degree of reliability in determining the Nth estimated energy efficiency of each business server under different loads.

[0018] Then, the Nth total service access volume of the terminal device set and the Nth existing load of each service server are obtained. Furthermore, the energy efficiency of the server varies under different loads (e.g., when the load is between 0% and 60%, each 10% increase in load only results in a 6% increase in power consumption; when the load is between 60% and 80%, each 10% increase in load results in a 12% to 15% increase in power consumption; and when the load is greater than 80%, each 5% increase in load results in a 15% increase in power consumption). Therefore, properly allocating the Nth total service access volume to each service server (i.e., assigning different loads to different service servers) can reduce the energy efficiency of each service server.

[0019] Furthermore, the Nth estimated energy efficiency of each service server under different loads, the Nth total service visits, and the Nth existing load of each service server are input into a pre-trained service visit volume distribution model, so that the service visit volume distribution model allocates service sub-visits to each service server based on pre-configured network parameters. Because the service visit volume distribution model is trained by inputting multiple second training samples into the second neural network, each second training sample includes the historical Nth estimated energy efficiency of each historical service server under different loads, the historical Nth total service visits of a historical set of terminal devices, and the historical Nth existing load of each historical service server. In this way, the Nth total service visits can be reasonably distributed to each service server.

[0020] Step 6: Detect the Nth actual energy efficiency of each business server, and determine the Nth average actual energy efficiency of the multiple business servers based on the respective Nth actual energy efficiencies of the multiple business servers; Step 7: If the Nth average energy efficiency of the multiple service servers is less than the N-1th average energy efficiency of the multiple service servers, and the difference between the Nth average energy efficiency and the N-1th average energy efficiency is greater than the set difference threshold, indicating that the Nth average energy efficiency of the multiple service servers can still be reduced, then the network parameters of the service access distribution model are updated using the reinforcement learning model based on the Nth average energy efficiency. The value of N is incremented by 1, and the process returns to step 1 until the difference between the Nth average energy efficiency and the N-1th average energy efficiency is less than or equal to the set difference threshold. This minimizes the average energy efficiency of the multiple service servers.

[0021] Finally, steps 1 through 5 are repeated until the distributed network acceleration node is powered off. When each service server operates at the lowest average energy efficiency, operating costs for the distributed network are reduced while ensuring efficiency and performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0023] Figure 1 A schematic diagram of the interaction between the distributed network acceleration node provided in the embodiment of the present application and a terminal device set and multiple service servers; Figure 2 A flow chart of the energy efficiency perception and load distribution method for distributed network acceleration nodes provided in an embodiment of the present application; Figure 3 This is a functional module block diagram of the energy efficiency perception and load distribution system for distributed network acceleration nodes provided in an embodiment of the present application. DETAILED DESCRIPTION

[0024] To make the purpose, technical solutions, and advantages of the embodiments of the present application more clear, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments made by ordinary technicians in this field based on the inspiration of these embodiments fall within the scope of protection of this application.

[0025] The terms "first," "second," "third," "fourth," and so forth (if any) in the specification and claims of this application and in the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequential sequence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the application described herein can be implemented in orders other than those illustrated or described herein. In addition, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or elements is not necessarily limited to those steps or elements explicitly listed, but may include other steps or elements not explicitly listed or inherent to such process, method, product, or apparatus.

[0026] Explanation of technical terms involved in this patent application: Energy efficiency of business servers: refers to the proportional relationship between the energy consumed by business servers when providing computing services and the actual effective work completed. It reflects the efficiency of the server in converting electrical energy into useful computing power.

[0027] The following describes in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0028] The embodiment of the present application provides a method for energy efficiency perception and load distribution of a distributed network acceleration node, which is applied to a distributed network acceleration node. Figure 1 As shown in the figure, the distributed network acceleration nodes are connected to multiple subordinate business servers and terminal device collections. In scenarios such as batch processing where scheduled tasks are executed at a fixed frequency and registration systems that control the number of registrations at a fixed rate, the total number of visits per unit time by the terminal device collection to multiple business servers remains unchanged, as shown in the figure. Figure 2 As shown, the method provided in the embodiment of the present application includes: Step 1: At the Nth time slice, obtain the server type of each business server in the plurality of business servers, the environmental data of each business server at the Nth time slice, and the hardware health of each business server, where N is an integer greater than or equal to 2.

[0029] It should be noted that the length of the Nth time slice can be, but is not limited to, 1 minute, 2 minutes, 5 minutes, 10 minutes, etc., and is not limited here. The business server is used to process tasks corresponding to batch jobs, or the business server is used to process registration services, etc., and is not limited here.

[0030] In addition, the method for obtaining the hardware health of each business server can be specifically implemented as follows: obtaining the wear degree of the solid-state hard disk of each business server, the accuracy of the operation log, and the stability of the power supply voltage; inputting the wear degree of the solid-state hard disk of each business server, the accuracy of the operation log, and the stability of the power supply voltage into a pre-trained hardware health determination model to determine the hardware health of each business server. The hardware health determination model is obtained by inputting multiple fourth training samples into the third neural network for training, and each fourth training sample includes the historical wear degree of the solid-state hard disk of the historical business server, the accuracy of the historical operation log, the stability of the historical power supply voltage, and the corresponding historical actual hardware health. In this way, the obtained hardware health is highly reliable.

[0031] Step 2: For each business server, a server performance profile feature is constructed based on the server type of the business server, the environmental data of the business server in the Nth time slice, and the hardware health of the business server.

[0032] Understandably, the processing power and energy efficiency of business servers of different server types vary. The processing power of the same business server varies under different environmental conditions (such as temperature and / or humidity). For example, excessively high temperatures can cause the business server to throttle, and different temperatures can cause different throttling rates, leading to a decrease in the server's processing power. Excessive humidity can corrode the server's hardware (for example, high humidity can oxidize the copper foil, solder joints, and connectors on circuit boards, increasing contact resistance; high humidity can also increase dust accumulation on circuit boards; and high humidity can also cause trace amounts of moisture to adsorb on the circuit board surface, reducing insulation resistance, causing a slight increase in leakage current, and increasing static power consumption). Different humidity levels can erode the server's hardware to varying degrees, resulting in a decrease in the server's processing power. Furthermore, different levels of hardware health in a business server correspond to different processing capabilities. Therefore, the server performance profile constructed using the above method is highly reliable.

[0033] Step 3: Input the server performance profile features of each business server into the pre-trained energy efficiency estimation model to determine the Nth estimated energy efficiency of each business server under different loads, wherein the energy efficiency estimation model is obtained by inputting multiple first training samples into the first neural network for training, and each first training sample includes historical server performance profile features and the corresponding historical Nth estimated energy efficiency.

[0034] Because the server performance profile features are highly reliable, and because the energy efficiency estimation model is trained by inputting multiple first training samples into a first neural network, each first training sample includes historical server performance profile features and the corresponding Nth historical estimated energy efficiency. This allows for highly reliable determination of the Nth estimated energy efficiency of each business server under different loads.

[0035] It should be noted that the method provided in the embodiment of the present application may further include: Step A: For each load of each service server, obtain the historical energy efficiency of the service server under the load for a plurality of consecutive time slices within a previous preset time period.

[0036] For example, if the Nth time slice is 9:00-9:05, then for the load j1 under the business server i1, the historical energy efficiency of the previous consecutive time slices of 8:55-9:00, 8:50-8:55, 8:45-8:50, etc. within the previous day (i.e., the previous preset duration) is calculated; for the load j2 under the business server i1, the historical energy efficiency of the previous consecutive time slices of 8:55-9:00, 8:50-8:55, 8:45-8:50, etc. within the previous day (i.e., the previous preset duration) is calculated, and so on. As well as the historical energy efficiency of the load j1 under the business server i2 in the previous consecutive time slices of 8:55-9:00, 8:50-8:55, 8:45-8:50, etc. within the previous day (i.e., the previous preset duration); the historical energy efficiency of the load j2 under the business server i2 in the previous consecutive time slices of 8:55-9:00, 8:50-8:55, 8:45-8:50, etc. within the previous day (i.e., the previous preset duration), and so on.

[0037] Step B: Based on the deep confidence DBN network, extract the temporal dependency of the historical energy efficiency of the business server in multiple consecutive time slices under load within the previous preset time period.

[0038] For example, extract the time series dependency of the historical energy efficiency of the business server i1 in the previous preset time period and under the load j1 in the previous multiple time slices; for example, extract the time series dependency of the historical energy efficiency of the business server i1 in the previous preset time period and under the load j2 in the previous multiple time slices; for example, extract the time series dependency of the historical energy efficiency of the business server i2 in the previous preset time period and under the load j1 in the previous multiple time slices; for example, extract the time series dependency of the historical energy efficiency of the business server i2 in the previous preset time period and under the load j2 in the previous multiple time slices, and so on.

[0039] Step C: Based on the historical energy efficiency of the business server in multiple consecutive time slices under load within the previous preset time period, and the temporal dependence of the historical energy efficiency of the business server in multiple consecutive time slices under load within the previous preset time period, the energy efficiency change characteristics of the business server under load are obtained.

[0040] The principle of step C is the same as that of the above-mentioned steps A and B, and will not be described in detail here.

[0041] Step D: For each service server, the energy efficiency variation characteristics under each load are input into the pre-trained energy efficiency prediction model to predict the Nth calibrated estimated energy efficiency of the service server under each load in the Nth time slice. It can be understood that the Nth calibrated estimated energy efficiency obtained based on steps A through D above is highly reliable.

[0042] Among them, the energy efficiency prediction model is obtained by inputting multiple third training samples into the long short-term memory network for training. Each third training sample includes the historical energy efficiency change characteristics of the historical business server under each load and the corresponding historical N-th calibrated estimated energy efficiency of the historical business server under each load in the historical N-th time slice.

[0043] Step E: For each service server, calibrate the Nth estimated energy efficiency under the corresponding load based on the Nth calibrated estimated energy efficiency under each load. This can further improve the reliability of the Nth estimated energy efficiency under the corresponding load.

[0044] For example, according to the formula , calibrate the Nth estimated energy efficiency under the corresponding load, where is the Nth estimated energy efficiency of the jth load of the i-th business server in the Nth time slice before calibration, The estimated energy efficiency of the Nth calibration for the jth load of the i-th business server in the Nth time slice is, is the calibrated Nth estimated energy efficiency of the jth load of the i-th business server in the Nth time slice, is the first weighting coefficient, is the second weighting coefficient, and .For example, ; .

[0045] Step 4: Obtain the Nth total service access volume of the terminal device set and the Nth existing load of each service server.

[0046] It's important to note that server energy efficiency varies under different loads (for example, when the load is between 0% and 60%, every 10% increase in load only results in a 6% increase in power consumption; when the load is between 60% and 80%, every 10% increase in load results in a 12%-15% increase in power consumption; and when the load is greater than 80%, every 5% increase in load results in a 15% increase in power consumption). Therefore, properly distributing the total Nth business traffic to each business server (i.e., assigning different loads to different business servers) can reduce the energy efficiency of each business server.

[0047] Step 5: Input the Nth estimated energy efficiency of each service server under different loads, the Nth total service access volume, and the Nth existing load of each service server into the pre-trained service access volume distribution model, so that the service access volume distribution model distributes service sub-access volume to each service server according to the pre-configured network parameters.

[0048] Among them, the service access volume distribution model is obtained by inputting multiple second training samples into the second neural network for training, and each second training sample includes the historical Nth estimated energy efficiency of each historical service server under different loads, the historical Nth total service access volume of the historical terminal device set, and the historical Nth existing load of each historical service server.

[0049] Since the service access volume distribution model is trained by inputting multiple second training samples into the second neural network, each second training sample includes the historical N-th estimated energy efficiency of each historical service server under different loads, the historical N-th total service access volume of a historical set of terminal devices, and the historical N-th existing load of each historical service server. In this way, the N-th total service access volume can be reasonably distributed to each service server.

[0050] Step 6: Detect the Nth actual energy efficiency of each business server, and determine the Nth average actual energy efficiency of the multiple business servers based on the respective Nth actual energy efficiencies of the multiple business servers.

[0051] Step 7: If the Nth average energy efficiency of multiple business servers is less than the N-1th average energy efficiency of multiple business servers, and the difference between the Nth average energy efficiency and the N-1th average energy efficiency is greater than the set difference threshold, then according to the Nth average energy efficiency, the reinforcement learning model is used to update the network parameters of the business access distribution model.

[0052] In this way, the average energy efficiency of multiple business servers can be minimized.

[0053] Step 8: Add 1 to the value of N and return to step 1 until the difference between the Nth average energy efficiency and the N-1th average energy efficiency is less than or equal to the set difference threshold.

[0054] Step 9: Repeat steps 1 to 5 until the distributed network acceleration node is powered off.

[0055] Exemplarily, the network parameters of the service access volume distribution model are the state of the reinforcement learning model, updating the network parameters of the service access volume distribution model is the action of the reinforcement learning model, the Nth average energy efficiency is the reward of the reinforcement learning model, and the lower the Nth average energy efficiency, the higher the reward.

[0056] It can be understood that when each business server operates at the lowest average energy efficiency, the operating cost of the distributed network is saved and the efficiency and performance of the distributed network are guaranteed.

[0057] In summary, the embodiment of the present application provides a method for energy efficiency perception and load distribution of a distributed network acceleration node. For each business server, a server performance profile feature is constructed based on the server type of the business server, the environmental data of the business server in the Nth time slice, and the hardware health of the business server. It can be understood that the processing capabilities and energy efficiency of business servers of different server types are also different; the processing capabilities of the same business server under different environmental data (such as temperature and / or humidity) are different. For example, excessively high temperature will cause the business server to reduce its frequency, and different temperatures will have different reduction amplitudes on the business server, thereby causing the computing power of the business server to decrease. Excessively high humidity will corrode the hardware of the business server, and different humidity will corrode the hardware of the business server to different degrees, resulting in a decrease in the processing power of the business server. In addition, when the hardware health of the business server is different, the corresponding processing capabilities are also different. Therefore, the reliability of the server performance profile feature constructed in the above manner is high.

[0058] The server performance profile features of each business server are input into a pre-trained energy efficiency estimation model to determine the Nth estimated energy efficiency of each business server under different loads. Because the server performance profile features are highly reliable, and because the energy efficiency estimation model is trained by inputting multiple first training samples into a first neural network, each first training sample includes historical server performance profile features and the corresponding historical Nth estimated energy efficiency. This allows for a high degree of reliability in determining the Nth estimated energy efficiency of each business server under different loads.

[0059] Then, the Nth total service access volume of the terminal device set and the Nth existing load of each service server are obtained. Furthermore, the energy efficiency of the server varies under different loads (e.g., when the load is between 0% and 60%, each 10% increase in load only results in a 6% increase in power consumption; when the load is between 60% and 80%, each 10% increase in load results in a 12% to 15% increase in power consumption; and when the load is greater than 80%, each 5% increase in load results in a 15% increase in power consumption). Therefore, properly allocating the Nth total service access volume to each service server (i.e., assigning different loads to different service servers) can reduce the energy efficiency of each service server.

[0060] Furthermore, the Nth estimated energy efficiency of each service server under different loads, the Nth total service visits, and the Nth existing load of each service server are input into a pre-trained service visit volume distribution model, so that the service visit volume distribution model allocates service sub-visits to each service server based on pre-configured network parameters. Because the service visit volume distribution model is trained by inputting multiple second training samples into the second neural network, each second training sample includes the historical Nth estimated energy efficiency of each historical service server under different loads, the historical Nth total service visits of a historical set of terminal devices, and the historical Nth existing load of each historical service server. In this way, the Nth total service visits can be reasonably distributed to each service server.

[0061] Step 6: Detect the Nth actual energy efficiency of each business server, and determine the Nth average actual energy efficiency of the multiple business servers based on the respective Nth actual energy efficiencies of the multiple business servers; Step 7: If the Nth average energy efficiency of the multiple service servers is less than the N-1th average energy efficiency of the multiple service servers, and the difference between the Nth average energy efficiency and the N-1th average energy efficiency is greater than the set difference threshold, indicating that the Nth average energy efficiency of the multiple service servers can still be reduced, then the network parameters of the service access distribution model are updated using the reinforcement learning model based on the Nth average energy efficiency. The value of N is incremented by 1, and the process returns to step 1 until the difference between the Nth average energy efficiency and the N-1th average energy efficiency is less than or equal to the set difference threshold. This minimizes the average energy efficiency of the multiple service servers.

[0062] Finally, steps 1 through 5 are repeated until the distributed network acceleration node is powered off. When each service server operates at the lowest average energy efficiency, operating costs for the distributed network are reduced while ensuring efficiency and performance.

[0063] See also Figure 3, the embodiment of the present application also provides an energy efficiency perception and load distribution system for distributed network acceleration nodes, which is applied to distributed network acceleration nodes, and the distributed network acceleration nodes are respectively connected to multiple subordinate business servers. It should be noted that the energy efficiency perception and load distribution system for distributed network acceleration nodes provided by the embodiment of the present application has the same basic principles and technical effects as the above embodiments. For the sake of brief description, for parts not mentioned in the embodiment of the present application, please refer to the corresponding content in the above embodiments. Figure 3 As shown, the system provided by the embodiment of the present application includes a data acquisition unit, a feature construction unit, an energy efficiency determination unit, an access volume allocation unit, an energy efficiency detection unit, a parameter updating unit and a service execution unit, wherein: a data acquisition unit, configured to acquire, in an Nth time slice, a server type of each of the plurality of business servers, environmental data of each of the business servers in the Nth time slice, and a hardware health status of each of the business servers, where N is an integer greater than or equal to 2; A feature construction unit is used to construct a server performance profile feature for each business server based on the server type of the business server, the environmental data of the business server at the Nth time slice, and the hardware health of the business server; an energy efficiency determination unit, configured to input the server performance profile features of each business server into a pre-trained energy efficiency estimation model to determine an Nth estimated energy efficiency of each business server under different loads, wherein the energy efficiency estimation model is trained by inputting a plurality of first training samples into a first neural network, each first training sample including a historical server performance profile feature and a corresponding historical Nth estimated energy efficiency; The data acquisition unit is further used to obtain the Nth total service access volume of the terminal device set and the Nth existing load of each service server; an access volume allocation unit, configured to input the Nth estimated energy efficiency of each service server under different loads, the Nth total service access volume, and the Nth existing load of each service server into a pre-trained service access volume allocation model, so that the service access volume allocation model allocates service sub-access volumes to each service server according to pre-configured network parameters, wherein the service access volume allocation model is trained by inputting multiple second training samples into a second neural network, each second training sample including the Nth historical estimated energy efficiency of each historical service server under different loads, the Nth historical total service access volume of a historical set of terminal devices, and the Nth historical existing load of each historical service server; an energy efficiency detection unit, configured to detect an Nth actual energy efficiency of each business server, and determine an Nth average actual energy efficiency of the plurality of business servers based on the respective Nth actual energy efficiencies of the plurality of business servers; a parameter updating unit configured to update the network parameters of the service access volume distribution model using a reinforcement learning model based on the Nth average energy efficiency, if the Nth average energy efficiency of the plurality of service servers is less than the N-1th average energy efficiency of the plurality of service servers, and the difference between the Nth average energy efficiency and the N-1th average energy efficiency is greater than a set difference threshold; add 1 to the value of N, and return to step 1 until the difference between the Nth average energy efficiency and the N-1th average energy efficiency is less than or equal to the set difference threshold; The service execution unit is configured to execute steps 1 to 5 in a loop until the distributed network acceleration node is powered off.

[0064] In some embodiments, the data acquisition unit is further configured to acquire, for each load of each business server, a historical energy efficiency of the business server under the load for a plurality of consecutive time slices within a previous preset time period; The device provided by the present application also includes: a relationship extraction unit, which is used to extract the temporal dependency relationship of the historical energy efficiency of the service server under load for multiple consecutive time slices within a previous preset time period based on the deep confidence DBN network; a characteristic obtaining unit, configured to obtain an energy efficiency variation characteristic of the service server under load based on historical energy efficiencies of the service server for a plurality of consecutive time slices under load within a previous preset duration, and a temporal dependency relationship between the historical energy efficiencies of the service server for a plurality of consecutive time slices under load within a previous preset duration; an energy efficiency prediction unit, configured to input, for each service server, the energy efficiency variation characteristics under each load into a pre-trained energy efficiency prediction model, so as to predict the Nth calibrated estimated energy efficiency of the service server under each load in the Nth time slice, wherein the energy efficiency prediction model is trained by inputting a plurality of third training samples into a long short-term memory network, each third training sample including a historical energy efficiency variation characteristic of the historical service server under each load and a corresponding historical Nth calibrated estimated energy efficiency of the historical service server under each load in the Nth historical time slice; The energy efficiency calibration unit is used to calibrate the Nth estimated energy efficiency under the corresponding load for each business server according to the Nth calibrated estimated energy efficiency under each load.

[0065] In some embodiments, the energy efficiency calibration unit is specifically configured to calculate , calibrate the Nth estimated energy efficiency under the corresponding load, where is the Nth estimated energy efficiency of the jth load of the i-th business server in the Nth time slice before calibration, The estimated energy efficiency of the Nth calibration for the jth load of the i-th business server in the Nth time slice is, is the calibrated Nth estimated energy efficiency of the jth load of the i-th business server in the Nth time slice, is the first weighting coefficient, is the second weighting coefficient, and .

[0066] In some embodiments, the data acquisition unit is specifically used to obtain the wear of the solid-state hard disk of each business server, the accuracy of the operation log, and the stability of the power supply voltage; the wear of the solid-state hard disk of each business server, the accuracy of the operation log, and the stability of the power supply voltage are input into a pre-trained hardware health determination model to determine the hardware health of each business server, wherein the hardware health determination model is obtained by inputting multiple fourth training samples into the third neural network for training, and each fourth training sample includes the historical wear of the solid-state hard disk of the historical business server, the accuracy of the historical operation log, the stability of the historical power supply voltage, and the corresponding historical actual hardware health.

[0067] In some embodiments, the network parameters of the service access volume distribution model are the state of the reinforcement learning model, updating the network parameters of the service access volume distribution model is the action of the reinforcement learning model, the Nth average energy efficiency is the reward of the reinforcement learning model, and the lower the Nth average energy efficiency, the higher the reward.

[0068] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for energy efficiency perception and load distribution of distributed network acceleration nodes, characterized in that: Applied to a distributed network acceleration node, each of which is in communication with a plurality of subordinate service servers, the method includes: Step 1: At the Nth time slice, obtain the server type of each of the multiple business servers, the environmental data of each of the business servers at the Nth time slice, and the hardware health of each of the business servers, where N is an integer greater than or equal to 2; Step 2: For each of the business servers, construct a server performance profile feature based on the server type of the business server, the environmental data of the business server in the Nth time slice, and the hardware health of the business server; Step 3: Inputting the server performance profile features of each of the business servers into a pre-trained energy efficiency estimation model to determine the Nth estimated energy efficiency of each of the business servers under different loads, wherein the energy efficiency estimation model is trained by inputting multiple first training samples into a first neural network, each of the first training samples including historical server performance profile features and corresponding historical Nth estimated energy efficiency; Step 4: Obtain the Nth total service access volume of the terminal device set and the Nth existing load of each of the service servers; Step 5: Inputting the Nth estimated energy efficiency of each of the service servers under different loads, the Nth total service access volume, and the Nth existing load of each of the service servers into a pre-trained service access volume distribution model, so that the service access volume distribution model distributes service sub-access volume to each of the service servers according to pre-configured network parameters, wherein the service access volume distribution model is trained by inputting multiple second training samples into a second neural network, each of the second training samples including the historical Nth estimated energy efficiency of each historical service server under different loads, the historical Nth total service access volume of a historical set of terminal devices, and the historical Nth existing load of each of the historical service servers; Step 6: Detecting the Nth actual energy efficiency of each of the business servers, and determining the Nth average actual energy efficiency of the plurality of business servers based on the respective Nth actual energy efficiencies of the plurality of business servers; Step 7: If the Nth average energy efficiency of the plurality of service servers is less than the N-1th average energy efficiency of the plurality of service servers, and the difference between the Nth average energy efficiency and the N-1th average energy efficiency is greater than a set difference threshold, then, based on the Nth average energy efficiency, a reinforcement learning model is used to update the network parameters of the service access volume distribution model; Step 8: Add 1 to the value of N, and return to step 1 until the difference between the Nth average energy efficiency and the N-1th average energy efficiency is less than or equal to the set difference threshold; Step 9: Execute steps 1 to 5 repeatedly until the distributed network acceleration node is powered off.

2. The method according to claim 1, characterized in that Before obtaining the Nth total service access volume of the terminal device set and the Nth existing load of each of the service servers, the method further includes: For each load of each of the business servers, obtaining the historical energy efficiency of the business server under the load for a plurality of consecutive time slices within a previous preset time period; Extracting, based on the deep confidence DBN network, the temporal dependency of the historical energy efficiency of the service server in a plurality of consecutive time slices under the load within a previous preset time period; Obtaining an energy efficiency variation characteristic of the service server under the load based on historical energy efficiencies of the service server in a plurality of consecutive time slices within a previous preset duration under the load, and a temporal dependency relationship between the historical energy efficiencies of the service server in a plurality of consecutive time slices within a previous preset duration under the load; For each of the business servers, inputting the energy efficiency change characteristics under each load into a pre-trained energy efficiency prediction model to predict the Nth calibrated estimated energy efficiency of the business server under each load in the Nth time slice, wherein the energy efficiency prediction model is trained by inputting multiple third training samples into a long short-term memory network, each of the third training samples including the historical energy efficiency change characteristics of the historical business server under each load and the corresponding historical Nth calibrated estimated energy efficiency of the historical business server under each load in the Nth time slice; For each of the service servers, the Nth estimated energy efficiency under the corresponding load is calibrated according to the Nth calibrated estimated energy efficiency under each load.

3. The method according to claim 2, characterized in that For each of the service servers, calibrating the Nth estimated energy efficiency under the corresponding load according to the Nth calibrated estimated energy efficiency under each load includes: According to the formula , calibrate the Nth estimated energy efficiency under the corresponding load, where is the Nth estimated energy efficiency of the jth load of the i-th service server in the N-th time slice before calibration, The Nth calibrated estimated energy efficiency of the jth load of the i-th service server in the Nth time slice, is the calibrated Nth estimated energy efficiency of the jth load of the i-th service server in the Nth time slice, is the first weighting coefficient, is the second weighting coefficient, and .

4. The method according to claim 1, wherein Obtaining the hardware health of each business server includes: Obtaining the wear degree of the solid-state hard disk of each of the business servers, the accuracy of the operation log, and the stability of the power supply voltage; The wear degree of the solid-state hard disk of each business server, the accuracy of the operation log, and the stability of the power supply voltage are input into the pre-trained hardware health determination model to determine the hardware health of each business server, wherein the hardware health determination model is obtained by inputting multiple fourth training samples into the third neural network for training, and each of the fourth training samples includes the historical wear degree of the solid-state hard disk of the historical business server, the accuracy of the historical operation log, the stability of the historical power supply voltage, and the corresponding historical actual hardware health.

5. The method according to any one of claims 1 to 4, characterized in that: The network parameters of the business access volume distribution model are the state of the reinforcement learning model, updating the network parameters of the business access volume distribution model is the action of the reinforcement learning model, the Nth average energy efficiency is the reward of the reinforcement learning model, and the lower the Nth average energy efficiency, the higher the reward.

6. A distributed network acceleration node energy efficiency perception and load distribution system, characterized in that: Applied to a distributed network acceleration node, each of which is in communication with a plurality of subordinate service servers, the system includes: a data acquisition unit, configured to acquire, in an Nth time slice, a server type of each of the plurality of business servers, environmental data of each of the business servers in the Nth time slice, and a hardware health status of each of the business servers, where N is an integer greater than or equal to 2; a feature construction unit configured to construct, for each of the business servers, a server performance profile feature based on the server type of the business server, the environmental data of the business server in the Nth time slice, and the hardware health of the business server; an energy efficiency determination unit, configured to input the server performance profile characteristics of each of the business servers into a pre-trained energy efficiency estimation model to determine an Nth estimated energy efficiency of each of the business servers under different loads, wherein the energy efficiency estimation model is trained by inputting a plurality of first training samples into a first neural network, each of the first training samples including a historical server performance profile characteristic and a corresponding historical Nth estimated energy efficiency; The data acquisition unit is further configured to acquire the Nth total service access volume of the terminal device set and the Nth existing load of each of the service servers; an access volume allocation unit, configured to input the Nth estimated energy efficiency of each of the service servers under different loads, the Nth total service access volume, and the Nth existing load of each of the service servers into a pre-trained service access volume allocation model, so that the service access volume allocation model allocates service sub-access volumes to each of the service servers according to pre-configured network parameters, wherein the service access volume allocation model is trained by inputting multiple second training samples into a second neural network, each of the second training samples including the historical Nth estimated energy efficiency of each historical service server under different loads, the historical Nth total service access volume of a historical set of terminal devices, and the historical Nth existing load of each of the historical service servers; an energy efficiency detection unit, configured to detect an Nth actual energy efficiency of each of the business servers, and determine an Nth average actual energy efficiency of the plurality of business servers based on the respective Nth actual energy efficiencies of the plurality of business servers; a parameter updating unit, configured to update the network parameters of the service access volume distribution model using a reinforcement learning model based on the Nth average energy efficiency, if the Nth average energy efficiency of the plurality of service servers is less than the N-1th average energy efficiency of the plurality of service servers, and the difference between the Nth average energy efficiency and the N-1th average energy efficiency is greater than a set difference threshold; add 1 to the value of N, and return to step 1 until the difference between the Nth average energy efficiency and the N-1th average energy efficiency is less than or equal to the set difference threshold; The service execution unit is configured to execute step 1 to step 5 in a loop until the distributed network acceleration node is powered off.

7. The system according to claim 6, characterized in that The data acquisition unit is further configured to acquire, for each load of each of the business servers, a historical energy efficiency of the business server under the load for a plurality of consecutive time slices within a previous preset time period; The system further includes: a relationship extraction unit for extracting, based on a deep confidence DBN network, a temporal dependency relationship between the historical energy efficiency of the service server for a plurality of consecutive time slices under the load within a previous preset time period; a characteristic obtaining unit, configured to obtain an energy efficiency variation characteristic of the service server under the load based on historical energy efficiencies of the service server in a plurality of consecutive time slices within a previous preset duration under the load, and a temporal dependency relationship between the historical energy efficiencies of the service server in a plurality of consecutive time slices within a previous preset duration under the load; an energy efficiency prediction unit, configured to input, for each of the service servers, the energy efficiency variation characteristics under each load into a pre-trained energy efficiency prediction model, so as to predict the Nth calibrated estimated energy efficiency of the service server under each load in the Nth time slice, wherein the energy efficiency prediction model is trained by inputting a plurality of third training samples into a long short-term memory network, each of the third training samples including a historical energy efficiency variation characteristic of a historical service server under each load and a corresponding historical Nth calibrated estimated energy efficiency of the historical service server under each load in the Nth time slice; The energy efficiency calibration unit is configured to calibrate the Nth estimated energy efficiency under the corresponding load for each of the service servers according to the Nth calibrated estimated energy efficiency under each load.

8. The system according to claim 7, characterized in that The energy efficiency calibration unit is specifically used to calculate , calibrate the Nth estimated energy efficiency under the corresponding load, where is the Nth estimated energy efficiency of the jth load of the i-th service server in the N-th time slice before calibration, The Nth calibrated estimated energy efficiency of the jth load of the i-th service server in the Nth time slice, is the calibrated Nth estimated energy efficiency of the jth load of the i-th service server in the Nth time slice, is the first weighting coefficient, is the second weighting coefficient, and .

9. The system according to claim 6, wherein: The data acquisition unit is specifically used to obtain the wear degree of the solid-state hard disk of each of the business servers, the accuracy of the operation log, and the stability of the power supply voltage; the wear degree of the solid-state hard disk of each of the business servers, the accuracy of the operation log, and the stability of the power supply voltage are input into a pre-trained hardware health determination model to determine the hardware health of each of the business servers, wherein the hardware health determination model is obtained by inputting multiple fourth training samples into the third neural network for training, and each of the fourth training samples includes the historical wear degree of the solid-state hard disk of the historical business server, the accuracy of the historical operation log, the stability of the historical power supply voltage, and the corresponding historical actual hardware health.

10. The system according to any one of claims 6 to 9, characterized in that: The network parameters of the business access volume distribution model are the state of the reinforcement learning model, updating the network parameters of the business access volume distribution model is the action of the reinforcement learning model, the Nth average energy efficiency is the reward of the reinforcement learning model, and the lower the Nth average energy efficiency, the higher the reward.

Images