Resource management method and system for green cloud computing and storage medium

Abstract

The application relates to the technical field of cloud computing resources, and discloses a resource management method and system for green cloud computing and a storage medium. The method comprises the following steps: performing weight optimization on CPU and memory data through a coyote optimization algorithm to obtain a server energy consumption coefficient matrix; using a DDQN decision to allocate the CPU of a virtual machine based on the matrix to obtain a resource allocation parameter table; adjusting the storage location of a data backup copy to obtain a backup node configuration table; monitoring the PUE values of servers to obtain a list of servers with excessive power consumption; and using a particle swarm algorithm to calculate a virtual machine migration path to obtain a load redistribution table. The application solves the problems of inaccurate energy consumption prediction, lack of intelligent scheduling strategy, separation of backup storage and main task scheduling, and low efficiency of abnormal power consumption processing in the existing cloud computing resource management, and improves the energy utilization efficiency of the cloud computing environment and the intelligent level of resource scheduling.

Description

Technical Field

[0001] This application relates to the field of cloud computing resource technology, and in particular to a resource management method, system and storage medium for green cloud computing. Background Technology

[0002] Existing cloud computing resource management technologies mainly adopt static scheduling strategies based on historical load data, and achieve simple allocation of virtual machines by monitoring CPU utilization thresholds and memory usage. At the same time, they use data backup strategies with a fixed number of replicas and load balancing algorithms based on time windows for resource scheduling. These technologies can meet the basic resource allocation needs in traditional cloud computing environments.

[0003] However, existing technologies have shortcomings: First, the accuracy of energy consumption prediction is insufficient. Existing methods make scheduling decisions based on only a single resource indicator and lack accurate modeling of the comprehensive energy consumption characteristics of CPU and memory resources. Second, there is a lack of intelligent dynamic scheduling mechanisms. Traditional rule-based scheduling algorithms cannot adapt to the complex and ever-changing cloud computing load patterns. Third, data backup strategies are independent of the main task scheduling, and the selection of backup copy storage locations is not coordinated with the overall resource utilization efficiency for optimization.

[0004] When power consumption exceeds the standard in a cloud computing environment, existing technologies lack a systematic problem localization and resolution mechanism. They cannot accurately identify the specific server causing the abnormal energy consumption, nor can they formulate the optimal virtual machine migration strategy based on global optimization goals. This limitation leads to low efficiency in handling energy consumption problems, which in turn affects the green energy-saving effect of the entire cloud computing system. Summary of the Invention

[0005] This application provides a resource management method, system, and storage medium for green cloud computing, which addresses the problems of inaccurate energy consumption prediction, lack of intelligent scheduling strategies, separation of backup storage and main task scheduling, and low efficiency in handling power consumption anomalies in existing cloud computing resource management, thereby improving the energy utilization efficiency and the level of intelligent resource scheduling in the cloud computing environment.

[0006] Firstly, this application provides a resource management method for green cloud computing. The method includes: weighting and optimizing CPU utilization data and memory usage data using the coyote optimization algorithm to obtain a server energy consumption coefficient matrix; performing DDQN decision processing on the virtual machine CPU allocation ratio based on the server energy consumption coefficient matrix to obtain a virtual machine resource allocation parameter table; adjusting the data backup copy storage location using the virtual machine resource allocation parameter table to obtain a backup node distribution configuration table; monitoring the real-time power consumption value of each server using the backup node distribution configuration table to obtain a list of servers exceeding power consumption limits; and calculating virtual machine migration paths for the list of servers exceeding power consumption limits using the particle swarm optimization algorithm to obtain a server load redistribution table.

[0007] Secondly, this application provides a resource management system for green cloud computing, the resource management system for green cloud computing comprising:

[0008] The processing module is used to perform weighted optimization processing on CPU utilization data and memory usage data using the coyote optimization algorithm to obtain the server energy consumption coefficient matrix.

[0009] The decision module is used to perform DDQN decision processing on the virtual machine CPU allocation ratio based on the server energy consumption coefficient matrix to obtain the virtual machine resource allocation parameter table.

[0010] The adjustment module is used to adjust the storage location of the data backup copy in the virtual machine resource allocation parameter table to obtain the backup node distribution configuration table.

[0011] The monitoring module is used to perform PUE monitoring on the real-time power consumption value of each server according to the backup node distribution configuration table, and obtain a list of servers with excessive power consumption.

[0012] The calculation module is used to perform virtual machine migration path calculation on the list of servers with excessive power consumption using the particle swarm optimization algorithm, and obtain a server load redistribution table.

[0013] Thirdly, a resource management device for green cloud computing is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the resource management device for green cloud computing to execute the aforementioned resource management method for green cloud computing.

[0014] Fourthly, a computer-readable storage medium is provided, wherein instructions are stored therein, which, when executed on a computer, cause the computer to perform the aforementioned resource management method for green cloud computing.

[0015] The technical solution provided in this application overcomes the limitations of single resource indicator evaluation in existing technologies by using the coyote optimization algorithm to perform weighted optimization of CPU utilization and memory usage data. This algorithm simulates the social hierarchy of a coyote pack, achieving collaborative optimization of multi-dimensional resource data through Alpha, Beta, and Omega role division, thereby establishing a more accurate server energy consumption coefficient matrix. Based on this energy consumption coefficient matrix, the application employs DDQN decision processing, effectively solving the overestimation problem in traditional reinforcement learning through a dual-network architecture of a deep dual-Q network. Through the alternating update mechanism of the target network and the evaluation network, intelligent decision-making on virtual machine CPU allocation ratios is achieved, significantly improving the accuracy and adaptability of resource allocation. The technical feature of adjusting the storage location of data backup copies using the virtual machine resource allocation parameter table breaks down the technical barriers between traditional backup strategies and main task scheduling. Through data importance rating and cross-regional distribution optimization, deep integration of backup storage and resource scheduling is achieved, ensuring data security while optimizing storage resource utilization efficiency. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of one embodiment of the resource management method for green cloud computing in this application.

[0018] Figure 2 This is a schematic diagram of one embodiment of a resource management system for green cloud computing in this application.

[0019] Figure 3 This is a schematic block diagram of the structure of a resource management device for green cloud computing in an embodiment of the present invention. Detailed Implementation

[0020] This application provides a resource management method, system, and storage medium for green cloud computing. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the resource management method for green cloud computing in this application includes:

[0022] Step S101: Use the coyote optimization algorithm to perform weight optimization on CPU utilization data and memory usage data to obtain the server energy consumption coefficient matrix;

[0023] Step S102: Perform DDQN decision processing on the virtual machine CPU allocation ratio based on the server energy consumption coefficient matrix to obtain the virtual machine resource allocation parameter table;

[0024] Step S103: Adjust the storage location of the data backup copy in the virtual machine resource allocation parameter table to obtain the backup node distribution configuration table;

[0025] Step S104: Perform PUE monitoring on the real-time power consumption value of each server according to the backup node distribution configuration table to obtain a list of servers with excessive power consumption.

[0026] Step S105: The list of servers with excessive power consumption is processed by the particle swarm optimization algorithm to calculate the virtual machine migration path and obtain the server load redistribution table.

[0027] It is understood that the executing entity of this application can be a resource management system for green cloud computing, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiment uses a server as an example for illustration.

[0028] Specifically, the Coyote Optimization Algorithm collects raw CPU utilization data from each server node via SNMP protocol or system monitoring agent. This data is typically expressed as a percentage. Maximum and minimum value scaling standardizes the raw data using a formula: subtracting the minimum value from the raw value and dividing by the difference between the maximum and minimum values ​​yields a standardized CPU utilization vector between 0 and 1. Memory usage data is also obtained through system calls, including metrics such as used memory, available memory, and cached memory. This memory data is concatenated column-wise with the standardized CPU utilization vector to form a multi-dimensional state matrix containing both CPU and memory resource metrics. The Coyote Optimization Algorithm simulates the social hierarchy of a coyote pack in nature. Alpha coyotes, as the pack leader, are responsible for exploring the global optimum, evaluating the quality of each solution by calculating a fitness function value. Beta coyotes, as deputy leaders, assist Alpha in local searches, refining the search direction determined by Alpha. Omega coyotes, as ordinary members, are responsible for random searches to prevent the algorithm from getting trapped in local optima. The fitness function aims to minimize energy consumption, comprehensively considering multiple energy consumption components such as CPU power consumption, memory power consumption, and heat dissipation power consumption. The algorithm initializes the multidimensional state matrix by generating a population, randomly producing several weighted solutions as the initial population. Each weighted solution represents the energy consumption weight coefficient of CPU and memory resources. Through an iterative optimization process, three classes of coyotes (Alpha, Beta, and Omega) collaboratively search, continuously updating the individual solutions in the population until the algorithm converges to obtain the optimal weighted solution. Finally, the optimal weighted solution is linearly combined with the CPU base power consumption and memory power consumption coefficients from the server hardware specifications to generate a server energy consumption coefficient matrix reflecting the energy consumption characteristics of each server. The rows of this matrix represent different servers, and the columns represent the energy consumption coefficients of different resource types.

[0029] DDQN, short for Deep Dual Q-Network, effectively addresses the overestimation bias problem in traditional Q-learning algorithms through a dual-network architecture. The algorithm first transforms the server energy consumption coefficient matrix, flattening the two-dimensional matrix into a one-dimensional state space vector. This is then converted into a multi-dimensional state input format acceptable to neural networks through linear or non-linear mapping. The dual-network architecture comprises two independent deep neural networks: a target network and an evaluation network. The target network calculates the target Q-value as the learning objective, while the evaluation network selects the optimal action based on the current state. Both networks have the same structure but their parameters are updated independently. The evaluation network parameters are updated in real-time, while the target network parameters are periodically copied from the evaluation network. This mechanism reduces instability during the learning process. The ε-greedy policy balances exploring unknown states with utilizing known optimal strategies. When the generated random number is less than ε, a random action is selected for environment exploration; when the random number is greater than or equal to ε, the action with the highest Q-value is selected. An experience replay mechanism establishes a fixed-capacity buffer to store state transition samples. Each sample contains four elements: current state, executed action, reward obtained, and next state. Samples are extracted from the buffer through a random sampling mechanism for network training, breaking the temporal correlation between samples. Temporal difference learning calculates the temporal difference error between the current Q-value and the target Q-value, and updates the network weight parameters through backpropagation using the gradient descent algorithm, gradually optimizing the network's decision-making ability. The algorithm ultimately outputs the CPU allocation ratio and memory allocation amount of the virtual machine on each server, forming a detailed table of virtual machine resource allocation parameters.

[0030] Data importance rating is performed by classifying and evaluating data based on multiple dimensions, including business criticality, data access frequency, user sensitivity, and legal compliance requirements. Critical data encompasses high-value data such as core business data, user account information, and transaction records. This type of data requires extremely high availability and security, therefore five backup copies are allocated to ensure redundancy protection. General-level data includes relatively low-value data such as system log files, temporary calculation results, and cached data, and three backup copies are allocated to strike a balance between reliability and storage cost. Load balancing calculation is performed by monitoring real-time metrics such as storage capacity utilization, network bandwidth utilization, and disk I / O load in each data center to calculate a comprehensive load score for each data center, selecting data center nodes with lower load scores as candidate storage nodes. Cross-regional distribution optimization is performed based on the geographical coordinates of each data center, using the spherical distance calculation formula to calculate the geographical distance between any two data centers, constructing a geographical distance matrix. The minimum spanning tree algorithm performs connectivity analysis on the distance matrix, generating a minimum connection topology graph between data centers, ensuring that all data centers maintain connectivity with minimal connection cost. A greedy algorithm iterates through the candidate list of storage nodes, selecting the node combination that minimizes the total transmission distance as the optimal node combination scheme while satisfying the cross-regional distribution requirement. Finally, the nodes are sorted according to transmission priority to generate a complete backup storage path scheme that includes the primary backup path and the secondary backup path.

[0031] Power Usage Effectiveness (PUE) is a core indicator of data center energy efficiency. It is calculated by dividing the total power consumption of the data center by the power consumption of IT equipment. An ideal PUE value close to 1.0 indicates that almost all electrical energy is used for IT equipment operation. Power sensors collect three-phase current and voltage data from each server in real time using current transformers and voltage sensors. The sampling frequency is typically set to multiple times per second to obtain accurate power consumption measurements. The raw power consumption data of the servers is calculated through electrical power calculation, which is the product of voltage and current multiplied by the power factor, taking into account the phase difference factor in AC circuits. The total power consumption of a data center includes multiple components, including not only the power consumption of IT equipment such as servers, but also the power consumption of air conditioning and cooling systems, lighting systems, UPS uninterruptible power supply losses, power distribution system losses, network equipment power consumption, and other auxiliary facilities. The real-time PUE monitoring value is obtained by dividing the total power consumption of the data center by the power consumption of IT equipment. When this value exceeds a preset PUE threshold, the anomaly detection algorithm automatically marks the corresponding server as having excessive power consumption. The power consumption deviation is calculated by dividing the difference between the current power consumption value and the standard power consumption value by the standard power consumption value to obtain the deviation percentage. The system sorts the servers in descending order according to the deviation percentage and generates a list of servers with excessive power consumption, sorted by the severity of the power consumption deviation. The server with the largest deviation is processed first.

[0032] The Particle Swarm Optimization (PSO) algorithm abstracts the virtual machine migration problem into a particle search process in a multi-dimensional optimization space, where each particle represents a complete virtual machine migration scheme. The particle position vector uses integer encoding, with each element corresponding to a virtual machine and the element value representing the target server number for that virtual machine. This encoding establishes a mapping relationship between the source and target servers. When initializing the particle swarm population, the algorithm uses a random number generator to assign initial positions and velocities to each particle, ensuring diversity in the initial distribution of particles in the solution space. The multi-objective fitness function comprehensively considers multiple optimization objectives, including the reduction in total energy consumption after migration, network transmission costs incurred during virtual machine migration, business impact caused by service interruption time, and resource reconfiguration overhead of the target server. Multiple objectives are merged into a single fitness value through a weighted summation. The particle velocity update mechanism combines individual historical best positions and global historical best positions, controlling the particle's search behavior through three parameters: inertia weight, individual learning factor, and global learning factor. The inertia weight controls the particle's tendency to maintain its current direction of motion, the individual learning factor controls the particle's tendency to move towards its individual best position, and the global learning factor controls the particle's tendency to move towards the global best position. Particle position updates are achieved by adding a velocity increment to the current position to obtain the new position at the next moment. After the position update, boundary constraint checks are performed to ensure that the particle position is within the valid solution space. During the iterative optimization process, the algorithm continuously evaluates the fitness value of each particle, updates the individual optimal position and the global optimal position, and converges after a preset number of iterations, outputting the global optimal particle position as the optimal migration strategy. The final generated server load redistribution table contains detailed information for each virtual machine, including the source server identifier, target server identifier, migration start time, estimated migration completion time, and post-migration resource allocation.

[0033] In one specific embodiment, the process of performing step S101 may specifically include the following steps:

[0034] The raw CPU utilization data of each server node is collected and then processed by scaling to obtain a standardized CPU utilization vector.

[0035] The memory usage data and the standardized CPU utilization vector are concatenated column-wise to obtain a multi-dimensional state matrix containing dual resource indicators.

[0036] A fitness function with the goal of minimizing energy consumption is constructed, and the multidimensional state matrix is ​​initialized and the population is generated using the coyote algorithm to obtain the initial set of weight solutions.

[0037] The social hierarchy of coyote packs is simulated, and the initial weighted solution set is optimized through collaborative search by Alpha, Beta and Omega roles to obtain the converged optimal weighted solution.

[0038] Based on the server CPU base power consumption and memory power consumption coefficient, the optimal weight solution is processed by linear combination operation to obtain the server energy consumption coefficient matrix reflecting the energy consumption characteristics of each server.

[0039] Specifically, each server node collects raw CPU utilization data through the proc file system or performance counter interface provided by the operating system kernel. This data represents the current CPU workload status as a percentage. Maximum and minimum value scaling uses a linear transformation method to convert the raw percentage data into standardized values. The specific calculation process involves subtracting the minimum CPU utilization value from all servers from each raw CPU utilization value, and then dividing by the difference between the maximum and minimum CPU utilization values. This normalization process yields a standardized CPU utilization vector with values ​​ranging from 0 to 1. This normalization process eliminates the impact of differences in hardware performance between different servers on data analysis, ensuring that the data processed by subsequent algorithms has a uniform numerical range and comparability. Each element of the standardized CPU utilization vector corresponds to the standardized CPU utilization value of a specific server, and the vector length is equal to the total number of servers in the cloud computing environment. Memory usage data collection includes multiple indicators such as the amount of memory used, total memory capacity, and memory utilization rate for each server. This data is obtained through system calls or the memory management unit. The memory usage data and the standardized CPU utilization vector are concatenated column-wise using a matrix concatenation operation. Specifically, the standardized CPU utilization vector is used as the first column, memory utilization data as the second column, used memory data as the third column, and total memory capacity data as the fourth column. This column vector concatenation forms a multidimensional state matrix containing both resource metrics. The number of rows in this multidimensional state matrix equals the number of servers, and the number of columns equals the number of monitoring metric types. Each element in the matrix represents the state value of a specific server in a specific resource dimension.

[0040] A multidimensional state matrix establishes the relationship between CPU and memory resources. The fitness function, designed to minimize energy consumption, comprehensively considers multiple energy consumption components, including CPU, memory, and heat dissipation. The fitness function value is calculated by weighted summation to combine various energy consumption types. The coyote algorithm initializes the population by assigning weight coefficients to each resource dimension in the multidimensional state matrix, with each weight coefficient representing the contribution of that resource dimension to the overall energy consumption. The initial weighted solution set is generated using a random number generator to randomly assign initial weight values ​​to each resource dimension within a preset weight range. Each weighted solution contains two main components: a CPU weight coefficient and a memory weight coefficient. The population size determines the number of weighted solutions evolving simultaneously; a larger population size increases solution diversity but computational complexity, while a smaller population size reduces computational overhead but decreases search efficiency. Each individual solution in the initial weighted solution set represents a resource weight allocation scheme, and the energy optimization effect of each scheme is evaluated using the fitness function. As the leader of the group, the Alpha coyote is responsible for the global search. It identifies the current optimal solution by evaluating the fitness values ​​of all individuals in the initial weight solution set. The Alpha coyote's search behavior simulates the decision-making process of a leader in nature, searching for the weight combination with the minimum energy consumption in the solution space.

[0041] Beta coyotes, acting as secondary leaders, assist Alpha coyotes in local searches, refining the search direction determined by the Alpha coyotes. Beta coyotes have a relatively small search range but high search accuracy. Omega coyotes, as ordinary group members, are responsible for random searches, exploring other potential optimization directions through random walks in the solution space to prevent the algorithm from getting trapped in local optima. The collaborative search optimization process of the three types of coyotes continuously improves the quality of the weighted solutions through an iterative update mechanism. In each iteration, Alpha, Beta, and Omega coyotes update the weighted solutions according to their respective search strategies, then re-evaluate the fitness values ​​and update the social hierarchy. Convergence criteria include reaching a preset number of iterations or the fitness value improvement being less than a preset threshold. When convergence conditions are met, the algorithm stops iterating and outputs the optimal weighted solution. Server CPU basic power consumption data comes from hardware specifications or actual power consumption tests, reflecting the server's power consumption characteristics under different CPU load levels. Memory power consumption coefficients are determined based on parameters such as memory type, capacity, and operating frequency; different types of memory modules have different power consumption characteristics. The linear combination operation multiplies the CPU weight coefficient in the optimal weight solution with the server CPU base power consumption, and multiplies the memory weight coefficient with the memory power consumption coefficient. Then, the two products are added together to obtain the comprehensive energy consumption coefficient of each server.

[0042] The server energy consumption coefficient matrix represents different server nodes in its rows and different energy consumption types in its columns. Each element in the matrix represents the energy consumption coefficient value of a specific server under a specific energy consumption type. This energy consumption coefficient matrix establishes a quantitative relationship between server hardware characteristics and energy consumption performance. Taking a private cloud environment of an enterprise as an example, this environment contains 500 servers with different configurations, including 200 high-performance computing servers equipped with 32-core CPUs and 64GB of memory, 200 storage servers equipped with 16-core CPUs and 32GB of memory, and 100 general-purpose servers equipped with 8-core CPUs and 16GB of memory. During the CPU utilization collection process, it was found that the CPU utilization of high-performance computing servers was generally high and fluctuated greatly, the CPU utilization of storage servers was relatively stable but the memory utilization was high, and the resource utilization of general-purpose servers was relatively balanced. Max-min scaling processing standardizes the CPU utilization of high-performance computing servers from the raw data. The CPU utilization of storage servers and general-purpose servers also undergoes standardization transformation. This standardization process eliminates the performance differences between different server types.

[0043] During the memory usage data concatenation process, the standardized CPU utilization, memory utilization, absolute value of used memory, and total memory capacity of each server are combined column-wise to form a 500-row, 4-column multidimensional state matrix. This matrix comprehensively describes the resource usage status of the entire cloud environment. The Coyote Algorithm initializes 50 weighted solutions as the initial population. Each weighted solution contains two parameters: CPU weight and memory weight. The Alpha Coyote algorithm identifies the weight combination with the lowest overall energy consumption through fitness evaluation. The Beta Coyote algorithm performs local optimization to fine-tune the weight parameters based on this. The Omega Coyote algorithm explores other weight combinations through random search. After multiple iterations, the algorithm converges. The optimal weighted solution is multiplied by the hardware power consumption parameters of each server through linear combination operations. The CPU base power consumption of the high-performance computing server is multiplied by the CPU weight coefficient to obtain the CPU energy consumption contribution value, and the memory power consumption is multiplied by the memory weight coefficient to obtain the memory energy consumption contribution value. The two are added together to obtain the server's comprehensive energy consumption coefficient. The energy consumption coefficients of storage servers and general-purpose servers are calculated in the same way, ultimately forming a complete energy consumption coefficient matrix reflecting the energy consumption characteristics of all servers.

[0044] In one specific embodiment, the process of performing step S102 may specifically include the following steps:

[0045] The server energy consumption coefficient matrix is ​​converted into a state space vector, and after dimensional mapping, a multidimensional state input is obtained.

[0046] A dual-network architecture consisting of a target network and an evaluation network is constructed, and forward propagation processing is performed on the multidimensional state input to obtain the Q-value action matrix;

[0047] The CPU allocation action is selected from the Q-value action matrix based on the ε-greedy strategy, and the sample storage process is performed through the experience replay mechanism to obtain the training buffer.

[0048] Temporal difference learning is used to update the parameters of the training buffer to obtain the DDQN decision model;

[0049] Based on the virtual machine load requirements, the CPU allocation ratio and memory allocation output by the DDQN decision model are combined and processed to obtain the virtual machine resource allocation parameter table.

[0050] Specifically, the server energy consumption coefficient matrix is ​​converted into a state space vector through matrix flattening, rearranging the two-dimensional energy consumption coefficient matrix into a one-dimensional vector by row or column order. Each energy consumption coefficient value in the matrix occupies a fixed position index in the state space vector. Dimensional mapping processing uses linear or nonlinear transformation methods to convert the state space vector into an input format acceptable to the neural network. Linear transformation adjusts the vector dimension through weight matrix multiplication, while nonlinear transformation introduces nonlinear characteristics through activation functions to enhance the network's expressive power. The multidimensional state input contains the energy consumption coefficient information of each server and the current virtual machine distribution status, forming a comprehensive input vector describing the resource status of the entire cloud environment. The dual-network architecture of the DDQN algorithm includes two deep neural networks with identical structures but independent parameters: a target network and an evaluation network. The target network is responsible for calculating the target Q-value as a learning reference standard, while the evaluation network is responsible for selecting the optimal action based on the current state. During the forward propagation process, the multidimensional state input first enters the network through the input layer of the evaluation network, undergoes neuron computation and activation function processing in multiple hidden layers, and finally generates Q-values ​​corresponding to different CPU allocation actions at the output layer. The rows of the Q-value action matrix represent different virtual machines, and the columns represent different CPU allocation ratio options. Each element in the matrix represents the expected reward value for performing a specific CPU allocation action in the current state. The parameter update strategies of the target network and the evaluation network differ. The parameters of the evaluation network are updated in real time to adapt to environmental changes, while the parameters of the target network are periodically copied from the evaluation network to maintain the stability of the learning process.

[0051] The ε-greedy strategy balances exploring unknown actions with utilizing known optimal actions. When the generated random number is less than the preset ε value, the algorithm selects a random CPU allocation action for environment exploration. When the random number is greater than or equal to the ε value, the algorithm selects the CPU allocation action with the largest Q value in the Q-value action matrix for utilization. CPU allocation actions include specific operations such as assigning virtual machines to different servers, adjusting the CPU usage ratio of virtual machines, and modifying the priority of virtual machines. The experience replay mechanism establishes a fixed-capacity circular buffer to store state transition experiences. Each experience contains four key elements: current state, executed action, reward obtained, and next state. When the buffer capacity reaches its limit, new experiences overwrite the oldest experiences. Sample storage processing writes the current state transition experience into the training buffer according to a fixed format, while maintaining the timestamp and priority information of the experience to support subsequent sampling operations.

[0052] Temporal difference learning uses the Bellman equation to calculate the temporal difference error between the current Q-value and the target Q-value. The target Q-value is calculated through the target network, and the current Q-value is calculated through the evaluation network. Parameter updates use gradient descent to minimize the square of the temporal difference error, calculate the gradient of the network parameters using backpropagation, and then use an optimizer to update the weights and biases of the evaluation network. Experience samples in the training buffer are selected for network training through a random sampling mechanism. Batch training processes multiple experience samples simultaneously to improve training efficiency and stability. The DDQN decision model gradually learns the optimal CPU allocation strategy during training. After the model parameters converge, it can output the optimal virtual machine resource allocation scheme based on the current server energy consumption status.

[0053] Virtual machine load requirements encompass multiple dimensions, including CPU computing needs, memory usage, network bandwidth requirements, and storage I / O requirements. This information is derived from historical virtual machine runtime data and current task load. Parameter combination processing matches the CPU allocation ratio output by the DDQN decision model with the virtual machine's CPU needs, ensuring that the allocated CPU resources meet the virtual machine's computing requirements. Simultaneously, it compares the memory allocation with the virtual machine's memory needs and adjusts the memory allocation to meet the virtual machine's memory usage requirements. Each row in the virtual machine resource allocation parameter table corresponds to one virtual machine, and each column corresponds to a resource type. The values ​​in the table represent the specific amount of resources allocated to a particular virtual machine, including detailed parameters such as the number of CPU cores, CPU usage ratio, memory capacity, and storage space.

[0054] In one specific embodiment, the process of executing step S103 may specifically include the following steps:

[0055] The virtual machine identifier and CPU allocation ratio are extracted from the virtual machine resource allocation parameter table, and a data priority classification matrix is ​​obtained through data importance rating.

[0056] Determine the importance level of each data block in the data priority classification matrix. When the importance level is critical, allocate 5 backup copies; when the importance level is moderate, allocate 3 backup copies. This will result in a copy number allocation table.

[0057] Based on the replica quantity allocation table, load balancing calculations are performed on the storage capacity of each data center to obtain a candidate list of storage nodes;

[0058] Based on the geographical distance matrix of the data center, the candidate list of storage nodes is optimized for cross-regional distribution to obtain a backup storage path scheme;

[0059] The backup storage path scheme is associated with the virtual machine resource allocation parameters to obtain a backup node distribution configuration table containing the specific storage node addresses.

[0060] Specifically, the virtual machine identifier in the virtual machine resource allocation parameter table uses a unique encoding method to identify each virtual machine. The identifier contains key information such as virtual machine type, creation time, and business affiliation. The CPU allocation ratio extraction process obtains the percentage of CPU resources occupied by each virtual machine by parsing the CPU-related fields in the parameter table. These values ​​reflect the computational intensity and business importance of the virtual machine. Data importance rating processing classifies the data carried by the virtual machine based on multiple evaluation dimensions, including business criticality, data access frequency, user scale, legal compliance requirements, and data recovery time targets. The rating algorithm assigns corresponding weight coefficients to each evaluation dimension and calculates the comprehensive importance score of each virtual machine's data through weighted summation. The score range is usually set to a numerical range of 0 to 100. The rows of the data priority classification matrix correspond to different virtual machines, and the columns correspond to different evaluation dimensions. The matrix element values ​​represent the score of a specific virtual machine on a specific evaluation dimension. The final importance level is obtained by summing through matrix operations.

[0061] The importance level of each data block in the data priority classification matrix is ​​determined through threshold comparison. The algorithm sets multiple thresholds to divide data into different categories based on importance level. Critical data typically refers to virtual machine data with a comprehensive importance score exceeding 80, including high-value information such as core business systems, user account information, transaction records, and compliance audit data. Loss or damage to this type of data would severely impact business operations. General-level data refers to virtual machine data with a comprehensive importance score between 40 and 80, including medium-value information such as business logs, cached data, report files, and temporary calculation results. Loss of this type of data would affect business continuity but would not cause fatal losses. Low-level data refers to virtual machine data with a comprehensive importance score below 40, including low-value information such as test data, expired files, and temporary storage. The replication allocation rule determines the backup strategy based on the data importance level. Critical data is allocated 5 backup copies to ensure extremely high data security; general-level data is allocated 3 backup copies to balance security and storage costs; and low-level data is allocated 1 backup copy or not backed up at all. The replica allocation table records the replica quantity requirement for each virtual machine. The table contains detailed information such as virtual machine identifier, importance level, replica quantity, and storage capacity requirement.

[0062] Load balancing calculations optimize storage allocation based on storage requirements in the replica allocation table and the current storage status of each data center. Storage capacity monitoring includes multiple metrics such as used storage space, available storage space, disk I / O load, and network bandwidth usage, which are obtained in real time through the storage management system's monitoring interface. The load balancing algorithm calculates a comprehensive load score for each data center, with the scoring formula taking into account factors such as storage capacity utilization, I / O load level, and network transmission latency. During the generation of the storage node candidate list, the algorithm prioritizes data centers with lower load scores as backup storage targets, while also considering whether the storage capacity is sufficient to accommodate the required backup data. Each storage node in the candidate list includes attributes such as data center identifier, available storage capacity, current load level, and geographical location information.

[0063] The data center geographic distance matrix records the geographic distance information between any two data centers. Distance calculation uses a spherical distance formula based on latitude and longitude coordinates to determine the actual geographic distance. Cross-regional distribution optimization ensures that backup copies are distributed across different geographic regions to reduce the impact of regional disasters on data security. The optimization algorithm sets a minimum geographic distance threshold requiring the distance between backup copies to exceed a preset value. During the backup storage path scheme generation process, the algorithm selects node combinations that meet the geographic distribution requirements from a candidate list of storage nodes. Optimization objectives include minimizing the total transmission distance, balancing the storage load across data centers, and meeting multiple constraints such as data sovereignty and compliance requirements. The path optimization algorithm uses a greedy strategy or dynamic programming method to find the optimal storage node combination while satisfying the constraints. The generated backup storage path scheme includes a complete plan for the primary backup path and secondary backup paths.

[0064] The association mapping process establishes a correspondence between the abstract storage nodes in the backup storage path scheme and the specific virtual machines in the virtual machine resource allocation parameter table. The mapping process determines the most suitable storage node based on factors such as the virtual machine's data importance level, storage capacity requirements, and geographical location preferences, while also considering the location of the physical server currently running the virtual machine to optimize data transfer efficiency. The generation of the backup node distribution configuration table combines virtual machine resource allocation information and the backup storage path scheme to assign a specific backup storage node address to each virtual machine, including detailed information such as the primary data center address, backup data center addresses, storage device identifiers, and access paths. The configuration table also includes operational management information such as backup scheduling policies, data synchronization frequency, and failover schemes to ensure the efficient operation of the backup system.

[0065] In one specific embodiment, the process of executing step S104 may specifically include the following steps:

[0066] The latitude and longitude coordinates of each data center are processed by spherical distance calculation to obtain the distance metric between data centers;

[0067] The distance metric is constructed into a symmetric matrix, and connectivity analysis is performed using the minimum spanning tree algorithm to obtain the region connectivity topology graph.

[0068] Based on the regional connection topology map, a preset distance threshold is set. When the distance between two data centers exceeds the threshold, they are marked as cross-regional node pairs, thus obtaining cross-regional node groups.

[0069] Based on the cross-regional node grouping, the candidate list of storage nodes is traversed using a greedy algorithm to select the node combination that minimizes the total transmission distance, thus obtaining the optimal node combination scheme.

[0070] The nodes in the optimal node combination scheme are sorted according to transmission priority to generate a backup storage path scheme that includes the primary backup path and the secondary backup path.

[0071] Specifically, the latitude and longitude coordinates of each data center are obtained through GPS positioning systems or geographic information databases. The coordinate information includes two key parameters: latitude and longitude, expressed in decimal degrees. The spherical distance calculation employs a great circle distance algorithm to account for the Earth's curvature. The calculation process first converts the latitude and longitude coordinates from degrees to radians, then applies the principles of spherical trigonometry to calculate the shortest distance between any two points. The distance calculation formula considers parameters such as the Earth's radius, the difference in latitude and longitude between the two points, and the cosine of latitude, obtaining the precise geographical distance through inverse cosine function and square root operations. The distance measurement between data centers is expressed in kilometers. The calculation results reflect the actual geographical distance between data centers, providing a quantitative basis for subsequent network transmission delay estimation and regional division.

[0072] The distance metric is constructed using a symmetric matrix in square form, where the number of rows and columns equals the total number of data centers. The element in the i-th row and j-th column represents the distance between the i-th and j-th data centers. The symmetric nature of the matrix ensures that the distance matrix satisfies the property that the distance from data center A to data center B is equal to the distance from data center B to data center A. All diagonal elements are zero, indicating that the distance from a data center to itself is zero. The minimum spanning tree algorithm uses either Kruskal's algorithm or Prim's algorithm to process the distance matrix. The algorithm aims to find a tree-like connection structure that connects all data centers with the minimum total distance. Connectivity analysis ensures that all data centers maintain network connectivity with minimal connection cost using the minimum spanning tree algorithm. The generated regional connection topology graph describes the optimal connections between data centers in graph theory form. Nodes in the graph represent data centers, edges represent connections, and edge weights represent connection distances.

[0073] The preset distance threshold in the regional connectivity topology graph is determined based on business needs and geographical distribution characteristics. The threshold setting considers multiple factors such as data transmission latency, network bandwidth costs, and disaster recovery requirements. The cross-region node pair marking process traverses all edges in the topology graph. When the weight of an edge, i.e., the distance between two data centers, exceeds the preset threshold, the algorithm marks this pair of data centers as a cross-region node pair, indicating that they are located in different geographical regions. Cross-region node grouping uses the connected component algorithm from graph theory, grouping data centers with distances less than the threshold into the same regional group, and dividing data centers with distances greater than the threshold into different regional groups. Data centers within each regional group are geographically concentrated, while data centers between regional groups are geographically dispersed. The grouping results generate multiple data center sets, each representing a geographical region. Data centers within a set can be connected via a high-speed network, while data centers between sets require long-distance network connections.

[0074] The greedy algorithm iterates through the data to select the optimal storage node from a cross-regional node grouping and candidate storage node list. The algorithm aims to minimize the total transmission distance while meeting backup distribution requirements. The traversal process selects one or more storage nodes from each region group as backup targets, based on criteria such as sufficient storage capacity, low current load, and good network connectivity. The greedy strategy employs locally optimal decision-making at each selection, prioritizing the storage node with the shortest transmission distance that meets the constraints. This iterative selection process gradually builds a complete node combination. The total transmission distance calculation includes the sum of the distances from the primary data center to each backup data center, as well as the transmission distance for data synchronization between backup data centers. The algorithm seeks the node combination that minimizes this sum. The optimal node combination scheme includes a complete configuration of the primary storage node and multiple backup storage nodes, with each node having a clearly defined role and storage responsibility.

[0075] The transmission priority ranking process prioritizes each node in the optimal node combination scheme based on multiple evaluation metrics, including geographical distance, network bandwidth, storage performance, historical reliability, and maintenance costs. The primary backup path points to the backup storage node with the highest transmission priority. This node typically has the best network connection quality and the highest reliability, serving as the primary data backup target. Secondary backup paths point to multiple backup storage nodes with the next highest transmission priority. These nodes serve as backup data targets, providing redundancy protection in the event of a failure in the primary backup path. The backup storage path scheme organizes each backup path in a hierarchical structure, including detailed configuration information such as path priority, transmission protocol, bandwidth allocation, and failover strategies, ensuring efficient execution of data backup operations and rapid recovery in case of failure.

[0076] In one specific embodiment, the process of executing step S105 may specifically include the following steps:

[0077] Extract server node identifiers and storage load distribution from the backup node distribution configuration table, and collect real-time current and voltage data of each server through power consumption sensors to obtain the raw data set of server power consumption.

[0078] The raw server power consumption dataset is summed with the power consumption of the cooling system and the power consumption of the lighting system to obtain the total power consumption value of the data center.

[0079] The ratio of the total power consumption of the data center to the power consumption of IT equipment is calculated to obtain the real-time PUE monitoring value.

[0080] Determine whether the real-time PUE monitoring value exceeds the preset PUE threshold. When it exceeds the threshold, mark the corresponding server as exceeding the standard. Then, filter the servers through the anomaly detection algorithm to obtain a set of servers with abnormal power consumption.

[0081] The servers in the abnormal power consumption server set are sorted in descending order according to the degree of power consumption deviation, resulting in a list of servers with excessive power consumption, sorted by the severity of the power consumption exceedance.

[0082] Specifically, the server node identifiers in the backup node distribution configuration table adopt a hierarchical coding structure, including information such as data center number, rack number, and server location number. Each identifier uniquely corresponds to a physical server. Storage load distribution information records storage-related metrics such as the amount of backup data currently carried by each server, storage device utilization, and disk I / O load. This data is updated periodically through the storage management system's monitoring interface. Power consumption sensors include two types of devices: current transformers and voltage sensors. Current transformers measure the AC current flowing through the server's power lines using the principle of electromagnetic induction, while voltage sensors measure the voltage amplitude of the server's power supply lines using a voltage divider circuit. During real-time current and voltage data acquisition, the sensors sample at a high frequency, typically multiple times per second, to capture instantaneous changes in power consumption. The acquired analog signals are converted into digital signals by an analog-to-digital converter for subsequent processing. The raw server power consumption dataset converts the current and voltage data into power consumption values ​​using an electrical power calculation formula. The calculation process considers power factor correction in the AC circuit to obtain the instantaneous and average power consumption data for each server.

[0083] Cooling system power consumption includes the total power consumption of cooling equipment such as air conditioning units, fans, cooling towers, and water pumps. The power consumption of these devices is collected in real time by dedicated energy monitoring equipment. Cooling system power consumption is closely related to the heat load of the data center. Increased server power consumption leads to increased heat dissipation demands, and the cooling system power consumption increases accordingly to maintain a suitable operating temperature. Lighting system power consumption includes the power consumption of all lighting equipment in the computer room, such as LED lights, emergency lighting, and security lighting. Lighting power consumption is relatively stable but varies at different times. The cumulative calculation process sums the power consumption values ​​of all servers in the original server power consumption dataset to obtain the total power consumption of IT equipment. Then, the total power consumption of IT equipment is added to the power consumption of auxiliary facilities such as cooling system power consumption, lighting system power consumption, UPS losses, and power distribution losses to obtain the total power consumption value reflecting the overall energy consumption of the data center. The total power consumption value of the data center is expressed in kilowatts. This value fluctuates dynamically with changes in business load and is an important indicator for evaluating the energy efficiency of the data center.

[0084] PUE stands for Power Usage Effectiveness, representing the energy efficiency of a data center. An ideal PUE value close to 1.0 indicates that almost all electrical energy is used for IT equipment operation. The real-time PUE monitoring value is obtained by dividing the total power consumption of the data center by the power consumption of the IT equipment. The formula is: PUE = Total power consumption of the data center divided by the power consumption of the IT equipment. Real-time PUE monitoring values ​​typically vary between 1.2 and 2.5. The closer the value is to 1.0, the higher the energy utilization efficiency; a higher value indicates that auxiliary facilities consume too much power. Real-time calculation of PUE monitoring values ​​provides data center operations personnel with immediate feedback on energy efficiency status, helping to identify energy consumption anomalies and optimization opportunities.

[0085] The preset PUE threshold is determined based on the data center's design standards and operational goals, typically set between 1.5 and 2.0. Exceeding this threshold indicates low energy efficiency and requires optimization measures. The judgment process checks whether the real-time PUE monitoring value exceeds the preset threshold through comparative calculations. When the monitored value exceeds the threshold, the algorithm triggers an anomaly alarm process. The server exceeding the standard status marking process identifies the specific server causing the PUE value to exceed the standard, and determines the responsibility for exceeding the standard by analyzing the power consumption contribution of each server. The anomaly detection algorithm uses statistical methods to identify servers with abnormal power consumption, including the 3-sigma rule based on the mean standard deviation, time series analysis based on historical data, and anomaly pattern recognition based on machine learning. The screening process filters out devices with abnormal power consumption from all servers. The abnormal power consumption server set includes all server nodes whose power consumption significantly deviates from the normal range, providing target objects for subsequent load adjustments.

[0086] Power consumption deviation is calculated by comparing the current power consumption of each server with its historical average power consumption or theoretical standard power consumption. The deviation is expressed as a percentage, indicating the extent to which power consumption exceeds the normal range. The server list is sorted in descending order of power consumption deviation, with the server with the largest deviation at the top and the servers with smaller deviations at the bottom. The list of servers exceeding power consumption limits provides maintenance personnel with a priority order for handling these servers. Prioritizing the servers with the largest deviations will most effectively reduce the overall PUE value. The list includes detailed information such as server identifier, current power consumption, standard power consumption, deviation level, and duration of exceeding the limit.

[0087] In one specific embodiment, the process of executing step S106 may specifically include the following steps:

[0088] Extract the virtual machine distribution information of the servers that exceed the power consumption limit from the list of servers that exceed the limit, and quantify the resource consumption of each virtual machine through a load statistics algorithm to obtain a set of virtual machine migration candidates.

[0089] Initialize the particle swarm population by encoding each migration scheme in the virtual machine migration candidate set into a particle position vector and obtaining the initial particle population through random generation.

[0090] A multi-objective fitness function with the goals of reducing energy consumption and migration costs is constructed to evaluate the fitness of each particle in the initial particle swarm, resulting in a particle fitness score matrix.

[0091] Based on the particle velocity update and position update mechanism, the particle fitness score matrix is ​​iteratively optimized, and the globally optimal particle position is selected as the optimal migration strategy to obtain the virtual machine migration path configuration.

[0092] The source server and target server in the virtual machine migration path configuration are mapped and associated to generate a server load redistribution table containing migration time series and resource reallocation.

[0093] Specifically, the virtual machine distribution information in the list of servers exceeding power consumption limits includes basic information such as the number of virtual machines running on each server, virtual machine identifiers, virtual machine types, and creation times. The virtual machine distribution information extraction process queries the virtual machine list of each server exceeding power consumption limits through the API interface of the virtualization management platform to obtain the mapping relationship between virtual machines and physical servers. Load statistics algorithms perform real-time monitoring and historical statistical analysis of virtual machine CPU utilization, memory usage, disk I / O rate, and network bandwidth consumption, quantifying the resource consumption characteristics of each virtual machine by calculating statistical indicators such as averages, peak values, and fluctuation amplitudes. Resource usage quantification uses a weighted scoring method to unify different types of resource consumption into comparable values: CPU usage is multiplied by a CPU weight coefficient, memory usage by a memory weight coefficient, and storage and network usage by their respective weight coefficients, ultimately obtaining a comprehensive resource usage score for each virtual machine. The virtual machine migration candidate set includes all virtual machines running on servers exceeding power consumption limits that are suitable for migration. Virtual machines performing critical tasks, having hardware dependencies, or configured with migration restrictions are excluded from the set to ensure the feasibility and security of the migration operation.

[0094] The initialization process of the Particle Swarm Optimization (PSO) algorithm sets the population size parameter to determine the number of particles evolving simultaneously, typically set to 30 to 100 particles to balance search capability and computational complexity. Particle position vectors are encoded using integers to represent virtual machine migration schemes. The vector length equals the number of virtual machines in the candidate migration set, and each element in the vector represents the target server number for the corresponding virtual machine. Random generation assigns an initial position and initial velocity to each particle. The initial position is randomly selected from the valid server numbers by a random number generator for each virtual machine. Each element of the initial velocity vector is set to a small random value to control the initial movement amplitude of the particle. The initial particle swarm forms a solution set containing multiple different migration schemes. Each particle represents a complete virtual machine allocation strategy, and the diversity of the swarm ensures that the algorithm can explore different regions in the solution space to find the global optimum.

[0095] The multi-objective fitness function design comprehensively considers two conflicting optimization objectives: energy consumption reduction and migration cost. The energy consumption reduction objective is quantified by calculating the power consumption difference before and after migration, while the migration cost objective includes multiple cost elements such as network transmission time, service interruption time, and resource reconfiguration overhead. The fitness function uses a weighted summation method to merge the two objectives into a single evaluation index. The weight coefficients are determined based on business priority and operational strategy, typically setting the energy consumption weight to 0.6 to 0.7 and the migration cost weight to 0.3 to 0.4. During the fitness evaluation process, the algorithm calculates the objective function value of the migration scheme corresponding to each particle's position vector. The evaluation process needs to consider constraints such as the remaining capacity of the target server, network connection quality, and compatibility constraints. The particle fitness score matrix records the fitness value of each particle at its current position. Rows in the matrix correspond to different particles, and columns contain detailed evaluation results such as fitness score, energy consumption reduction, and migration cost, providing a decision-making basis for subsequent particle updates.

[0096] The particle velocity update mechanism adopts the velocity update formula of the standard particle swarm optimization algorithm. The formula includes three components: inertia, cognition, and social. The inertia term keeps the particle in its current direction of motion, the cognition term moves the particle towards its historical best position, and the social term moves the particle towards the global best position of the swarm. The position update mechanism adds the updated velocity to the current position to obtain the particle's new position. After the position update, boundary constraint checks are performed to ensure that the particle position is within the effective solution space. During the iterative optimization process, the algorithm repeatedly executes fitness evaluation, velocity update, position update, and global best update steps. After each iteration, the individual best position and the global best position are updated. The iterative process continues until the convergence condition is met or the maximum number of iterations is reached. The global best particle position is selected based on the optimal fitness value in the fitness score matrix. The corresponding particle position vector represents the best virtual machine migration strategy found in the current population. The virtual machine migration path configuration includes detailed information such as the source server, target server, migration priority, and estimated migration time for each virtual machine. The configuration scheme ensures the orderly execution of migration operations and avoids resource conflicts.

[0097] The mapping and association process establishes a detailed correspondence between the source and target servers in the virtual machine migration path configuration. This association information includes technical elements such as server hardware configuration, network connection topology, storage sharing relationships, and virtualization platform compatibility. The migration time series calculates the start and estimated completion times for each virtual machine based on migration priority, data transfer volume, and network bandwidth capacity. The sequence arrangement avoids network congestion caused by migrating too many virtual machines simultaneously. Resource reallocation calculates the new resource allocation status of the target server after receiving the migrated virtual machines, including adjustments to CPU allocation ratios, memory allocation, storage space allocation, and network bandwidth allocation. The server load redistribution table records the load changes of all servers involved in the migration operation in tabular form. The table includes complete information such as server identifier, pre-migration load status, post-migration load status, load change amount, and new virtual machine distribution, providing detailed guidance for the execution and monitoring of the migration operation.

[0098] The resource management method for green cloud computing in the embodiments of this application has been described above. The resource management system for green cloud computing in the embodiments of this application is described below. Please refer to [link / reference]. Figure 2 One embodiment of the resource management system for green cloud computing in this application includes:

[0099] The processing module is used to perform weighted optimization processing on CPU utilization data and memory usage data using the coyote optimization algorithm to obtain the server energy consumption coefficient matrix.

[0100] The decision module is used to perform DDQN decision processing on the virtual machine CPU allocation ratio based on the server energy consumption coefficient matrix to obtain the virtual machine resource allocation parameter table.

[0101] The adjustment module is used to adjust the storage location of the data backup copy in the virtual machine resource allocation parameter table to obtain the backup node distribution configuration table.

[0102] The monitoring module is used to perform PUE monitoring on the real-time power consumption value of each server according to the backup node distribution configuration table, and obtain a list of servers with excessive power consumption.

[0103] The calculation module is used to perform virtual machine migration path calculation on the list of servers with excessive power consumption using the particle swarm optimization algorithm, and obtain a server load redistribution table.

[0104] above Figure 2 The resource management system for green cloud computing in this embodiment of the invention will be described in detail from the perspective of modular functional entities. The resource management device for green cloud computing in this embodiment of the invention will be described in detail from the perspective of hardware processing.

[0105] Reference Figure 3 This invention also provides a resource management device for green cloud computing, which can be a server, and its internal structure can be as follows: Figure 3 As shown, this resource management device for green cloud computing includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor in this computer design provides computing and control capabilities. The memory of this resource management device for green cloud computing includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database of this resource management device for green cloud computing is used to store the data corresponding to this embodiment. The network interface of this resource management device for green cloud computing is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0106] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the resource management device for green cloud computing on which the present invention is applied.

[0107] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of the resource management method for green cloud computing.

[0108] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0109] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a resource management device (which may be a personal computer, server, or network device, etc.) for green cloud computing to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0110] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A resource management method for green cloud computing, characterized in that, The method includes: The coyote optimization algorithm is used to perform weighted optimization on CPU utilization data and memory usage data to obtain a server energy consumption coefficient matrix. Based on this matrix, the DDQN decision-making process is applied to the virtual machine CPU allocation ratio to obtain a virtual machine resource allocation parameter table. The virtual machine resource allocation parameter table is then used to adjust the data backup copy storage location to obtain a backup node distribution configuration table. Based on this table, the real-time power consumption value of each server is monitored using PUE to obtain a list of servers exceeding power consumption limits. Finally, the particle swarm optimization algorithm is used to calculate virtual machine migration paths from this list of servers exceeding power consumption limits to obtain a server load redistribution table. The process of adjusting the data backup copy storage location based on the virtual machine resource allocation parameter table to obtain a backup node distribution configuration table includes: extracting virtual machine identifiers and CPU allocation ratios from the virtual machine resource allocation parameter table; obtaining a data priority classification matrix through data importance rating; determining the importance level of each data block in the data priority classification matrix; allocating 5 backup copies when the importance level is critical, allocating 3 backup copies when the importance level is general, and allocating 1 backup copy or no backup for low-level data to obtain a copy number allocation table; performing load balancing calculations on the storage capacity of each data center based on the copy number allocation table to obtain a storage node candidate list; optimizing the cross-regional distribution of the storage node candidate list according to the data center geographical location distance matrix to obtain a backup storage path scheme; and associating the backup storage path scheme with the virtual machine resource allocation parameters to obtain a backup node distribution configuration table containing specific storage node addresses.

2. The resource management method for green cloud computing according to claim 1, characterized in that, The process of weighting and optimizing CPU utilization data and memory usage data using the coyote optimization algorithm to obtain a server energy consumption coefficient matrix includes: The raw CPU utilization data of each server node is collected and then processed by scaling to obtain a standardized CPU utilization vector. The memory usage data is concatenated with the standardized CPU utilization vector column by column to obtain a multi-dimensional state matrix containing dual resource indicators. A fitness function with the goal of minimizing energy consumption is constructed, and the multidimensional state matrix is ​​initialized and the population is generated using the coyote algorithm to obtain the initial weight solution set. The social hierarchy of coyote packs is simulated, and the initial weighted solution set is optimized by collaborative search through the division of roles among Alpha, Beta, and Omega to obtain the converged optimal weighted solution. Based on the server CPU base power consumption and memory power consumption coefficient, the optimal weight solution is processed by linear combination operation to obtain the server energy consumption coefficient matrix reflecting the energy consumption characteristics of each server.

3. The resource management method for green cloud computing according to claim 1, characterized in that, The step of performing DDQN decision processing on the virtual machine CPU allocation ratio based on the server energy consumption coefficient matrix to obtain a virtual machine resource allocation parameter table includes: The server energy consumption coefficient matrix is ​​converted into a state space vector, and after dimensional mapping processing, a multidimensional state input is obtained. A dual-network architecture comprising a target network and an evaluation network is constructed, and the multidimensional state input is processed by forward propagation to obtain the Q-value action matrix; The CPU allocation action is selected from the Q-value action matrix based on the ε-greedy strategy, and the sample storage process is performed through the experience replay mechanism to obtain the training buffer. The training buffer is updated using temporal difference learning to obtain the DDQN decision model. Based on the virtual machine load requirements, the CPU allocation ratio and memory allocation output by the DDQN decision model are combined and processed to obtain the virtual machine resource allocation parameter table.

4. The resource management method for green cloud computing according to claim 1, characterized in that, The step of performing cross-regional distribution optimization processing on the candidate list of storage nodes based on the geographical location distance matrix of the data center to obtain a backup storage path scheme includes: The latitude and longitude coordinates of each data center are processed by spherical distance calculation to obtain the distance metric between data centers; The distance metric is constructed into a symmetric matrix, and connectivity analysis is performed using the minimum spanning tree algorithm to obtain the region connectivity topology map. Based on the regional connection topology map, a preset distance threshold is set. When the distance between two data centers exceeds the threshold, they are marked as cross-regional node pairs, thus obtaining cross-regional node groups. Based on the cross-regional node grouping, the storage node candidate list is traversed using a greedy algorithm to select the node combination that minimizes the total transmission distance, thus obtaining the optimal node combination scheme. The nodes in the optimal node combination scheme are sorted according to transmission priority to generate a backup storage path scheme that includes a primary backup path and a secondary backup path.

5. The resource management method for green cloud computing according to claim 1, characterized in that, The process involves monitoring the real-time power consumption of each server based on the backup node distribution configuration table to obtain a list of servers with excessive power consumption, including: Extract server node identifiers and storage load distribution from the backup node distribution configuration table, and collect real-time current and voltage data of each server through power consumption sensors to obtain the raw data set of server power consumption. The original server power consumption dataset is summed with the power consumption of the cooling system and the power consumption of the lighting system to obtain the total power consumption value of the data center. The ratio of the total power consumption of the data center to the power consumption of the IT equipment is calculated to obtain the real-time PUE monitoring value; Determine whether the real-time PUE monitoring value exceeds the preset PUE threshold. If it exceeds the threshold, mark the corresponding server as exceeding the standard. Then, filter the server through the anomaly detection algorithm to obtain a set of servers with abnormal power consumption. The servers in the abnormal power consumption server set are sorted in descending order according to the degree of power consumption deviation, resulting in a list of servers with excessive power consumption sorted by the severity of power consumption exceeding the standard.

6. The resource management method for green cloud computing according to claim 1, characterized in that, The step of calculating virtual machine migration paths for the list of servers exceeding power consumption limits using the particle swarm optimization algorithm to obtain a server load redistribution table includes: Extract the virtual machine distribution information of the servers exceeding the power consumption limit from the list of servers exceeding the power consumption limit, and quantify the resource consumption of each virtual machine through a load statistics algorithm to obtain a virtual machine migration candidate set. Initialize the particle swarm population by encoding each migration scheme in the virtual machine migration candidate set into a particle position vector and obtaining the initial particle swarm through random generation. A multi-objective fitness function with the goals of reducing energy consumption and migration costs is constructed, and the fitness of each particle in the initial particle swarm is evaluated to obtain a particle fitness score matrix. Based on the particle velocity update and position update mechanism, the particle fitness score matrix is ​​iteratively optimized, and the globally optimal particle position is selected as the optimal migration strategy to obtain the virtual machine migration path configuration. The source server and target server in the virtual machine migration path configuration are mapped and associated to generate a server load redistribution table containing migration time series and resource reallocation.

7. A resource management system for green cloud computing, characterized in that, A resource management method for implementing the green cloud computing-oriented approach as described in any one of claims 1-6, wherein the green cloud computing-oriented resource management system comprises: The processing module is used to perform weighted optimization processing on CPU utilization data and memory usage data using the coyote optimization algorithm to obtain the server energy consumption coefficient matrix. The decision module is used to perform DDQN decision processing on the virtual machine CPU allocation ratio based on the server energy consumption coefficient matrix to obtain the virtual machine resource allocation parameter table. The adjustment module is used to adjust the storage location of the data backup copy in the virtual machine resource allocation parameter table to obtain the backup node distribution configuration table. The monitoring module is used to perform PUE monitoring on the real-time power consumption value of each server according to the backup node distribution configuration table, and obtain a list of servers with excessive power consumption. The calculation module is used to perform virtual machine migration path calculation on the list of servers with excessive power consumption using the particle swarm optimization algorithm, and obtain a server load redistribution table.

8. A resource management device for green cloud computing, characterized in that, It includes a memory and a processor, the memory storing a computer program that can run on the processor, and the processor executing the computer program to implement the resource management method for green cloud computing as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it causes the processor to perform the resource management method for green cloud computing as described in any one of claims 1 to 6.

Images