Deep learning-based cloud storage resource adaptive allocation method and system

Abstract

This invention provides a deep learning-based adaptive allocation method and system for cloud storage resources, relating to the field of cloud storage technology. First, a business resource symbiotic network is constructed to reflect the dynamic balance between business demand and resource supply. Next, a deep learning model is invoked to extract network evolution characteristics. Based on these characteristics, a business resource adaptation demand sequence is derived to determine the adaptation direction and pattern of storage resources for different time periods. Then, the deep learning model dynamically matches and deduces the adaptation demand sequence with the resource supply spectrum to generate a set of elastic resource allocation schemes. Finally, a real-time allocation instruction stream is generated based on the set of elastic resource allocation schemes to drive dynamic reorganization of storage resources, achieving real-time adaptation between business demand and resource supply.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of cloud storage, in particular to a cloud storage resource adaptive allocation method and system based on deep learning. BACKGROUND

[0002] In today's digital era, cloud storage systems, as the key infrastructure for data storage and management, are widely used in various business scenarios such as online office, e-commerce, social media, etc. With the continuous development and diversification of business, the demand for cloud storage resources by different businesses presents complex and variable characteristics. On the one hand, different businesses have different data storage and processing characteristics, for example, video streaming businesses require a large amount of bandwidth and storage space to ensure smooth playback, while database businesses focus more on data read / write speed and security. On the other hand, the resource demand of the same business will also change significantly at different time periods, such as during e-commerce promotional activities, the business volume will increase significantly, and the demand for cloud storage resources will also increase accordingly.

[0003] Currently, the allocation of cloud storage resources mainly adopts static allocation or dynamic allocation based on simple rules. The static allocation method allocates fixed storage resources according to the estimated demand of the business at system initialization, which cannot adapt to the dynamic changes of business demand, and is prone to resource waste or resource shortage. Although the dynamic allocation method based on simple rules can adjust the resources to some extent according to the real-time load of the business, these rules are often set based on experience, lacking in-depth analysis and prediction of the complex relationship between business demand and resource supply, making it difficult to achieve efficient and accurate allocation of resources. Therefore, the existing cloud storage resource allocation method cannot meet the needs of the diversified and dynamic development of businesses. SUMMARY

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present application, the present application embodiment provides a cloud storage resource adaptive allocation method based on deep learning, which comprises: A business resource symbiotic network is constructed, which takes the resource consumption trajectory of various businesses and the resource supply pedigree of the cloud storage system as the core elements, and forms a network system for reflecting the dynamic balance relationship between business demand and resource supply through node association, edge weight mapping and topology evolution process; A deep learning model is called to extract multi-dimensional evolution factors of the business resource symbiotic network, generating network evolution trend characteristics, which include business node demand fluctuation trend, resource node supply capacity change and inter-node symbiotic association strength evolution core information; Based on the network evolution characteristics, a service resource adaptation requirement sequence is derived. This sequence is transformed from the interaction patterns of each node and the edge weight change characteristics in the network evolution characteristics. It is used to determine the adaptation direction and adaptation mode of various services to storage resources in different prediction periods. By using a deep learning model, the sequence of business resource adaptation needs is dynamically matched and deduced with the resource supply spectrum of the cloud storage system, generating a set of resource elastic allocation schemes. The set of resource elastic allocation schemes includes resource allocation adjustment strategies, execution paths and adaptation guarantee-related steps for different business scenarios. Based on the aforementioned resource elastic allocation scheme set, a real-time allocation instruction stream is generated, driving the storage resources in the cloud storage system to complete dynamic reorganization across nodes and prediction periods according to the instruction stream, thereby forming a real-time adaptation between business needs and resource supply.

[0005] Furthermore, embodiments of the present invention also provide a cloud storage resource adaptive allocation system based on deep learning, characterized in that it includes: A processor; a machine-readable storage medium for storing machine-executable instructions of the processor; wherein the processor is configured to execute the aforementioned deep learning-based adaptive allocation method for cloud storage resources by executing the machine-executable instructions.

[0006] In another aspect, embodiments of the present invention also provide a computer program product, the computer program product including machine-executable instructions, the machine-executable instructions being stored in a computer-readable storage medium, the processor of the deep learning-based cloud storage resource adaptive allocation system reading the machine-executable instructions from the computer-readable storage medium, the processor executing the machine-executable instructions, causing the deep learning-based cloud storage resource adaptive allocation system to execute the aforementioned deep learning-based cloud storage resource adaptive allocation method.

[0007] Based on the above, by constructing a business resource symbiotic network, and taking the resource consumption trajectory of various businesses and the resource supply spectrum of the cloud storage system as core elements, the dynamic balance between business demand and resource supply is characterized. Then, a deep learning model is used to extract multi-dimensional evolutionary factors from the business resource symbiotic network. The generated network evolutionary characteristics can accurately capture core information such as the fluctuation trend of business node demand, the change of resource node supply capacity, and the evolution of the symbiotic relationship strength between nodes. Based on the network evolutionary characteristics, the business resource adaptation demand sequence is derived. According to the interaction rules of each node and the edge weight change characteristics, the adaptation direction and adaptation mode of various businesses to storage resources in different prediction periods can be determined, thus achieving accurate prediction of business resource demand. By using a deep learning model to dynamically match and deduce the sequence of business resource adaptation needs with the resource supply spectrum of the cloud storage system, a set of resource elastic allocation schemes is generated. This set includes resource allocation adjustment strategies, execution paths, and adaptation guarantee steps for different business scenarios. Based on the set of resource elastic allocation schemes, a real-time allocation instruction stream is generated, driving the storage resources in the cloud storage system to complete dynamic reorganization across nodes and prediction periods according to the instruction stream. This enables real-time adaptation to business needs and resource supply, effectively improving the utilization rate of cloud storage resources, reducing resource waste, and enhancing the adaptability of the cloud storage system to business changes, thereby improving the stability and reliability of the system. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the execution flow of the cloud storage resource adaptive allocation method based on deep learning provided in an embodiment of the present invention.

[0009] Figure 2 This is a schematic diagram of exemplary hardware and software components of the cloud storage resource adaptive allocation system based on deep learning provided in an embodiment of the present invention. Detailed Implementation

[0010] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a deep learning-based adaptive allocation method for cloud storage resources, provided in one embodiment of the present invention. The following is a detailed description of this deep learning-based adaptive allocation method for cloud storage resources.

[0011] Step S110: Construct a business resource symbiotic network. The business resource symbiotic network takes the resource consumption trajectory of various businesses and the resource supply spectrum of the cloud storage system as core elements. Through node association, edge weight mapping and topology evolution process, it forms a network system that reflects the dynamic balance between business demand and resource supply.

[0012] This embodiment uses a cloud storage resource allocation scenario of a large internet company as an example. This scenario involves various business types, such as video-on-demand, online document collaboration, and big data analytics. The cloud storage system includes various types of resources, such as distributed file storage, object storage, and block storage. Constructing a business resource symbiotic network is the foundation for achieving adaptive allocation of cloud storage resources. By abstracting businesses and resources into nodes and establishing relationships between nodes, the dynamic interaction between business demand and resource supply can be reflected.

[0013] Step S111: Collect the resource consumption trajectory of various services throughout the entire operation cycle. The resource consumption trajectory includes the call sequence, usage duration, access frequency and demand switching path of different types of storage resources by the services, and fully record the entire interaction process between services and storage resources.

[0014] In this internet enterprise scenario, for video-on-demand (VOD) services, the collection of resource consumption trajectories needs to cover the entire operational cycle from the user initiating a video request to video data transmission, playback, pause, resume playback, and eventual completion. Regarding call timing, the start and end times of the service's calls to object storage resources need to be recorded at different time periods, such as the distribution of call times during peak and off-peak hours each day. Usage duration refers to the duration of each storage resource call, such as the time taken for a video data read operation from start to finish. Access frequency is the number of times the service calls a specific storage resource per unit of time; for popular video content, the access frequency to corresponding object storage resources will be significantly higher than for less popular content. Demand switching paths record the switching process between different storage resource types; for example, a VOD service might call distributed file storage resources to obtain metadata during the initial buffering phase, and then switch to object storage resources for video stream data transmission. Through log collection agents deployed on business servers and storage nodes, the above information is collected in real time and transmitted to the data processing center via message queues for aggregation, forming a continuous resource consumption trajectory data stream.

[0015] Step S112: Collect the resource supply spectrum of the cloud storage system. The resource supply spectrum covers the core information of all storage resources, including physical attributes, logical affiliation, performance parameters, deployment location, and available prediction period, reflecting the supply capacity boundary and service scope of each storage resource.

[0016] For this enterprise's cloud storage system, the resource supply hierarchy needs to be collected for all storage devices and logical storage units. Physical attributes include storage media type (such as solid-state drives, hard disk drives), single disk capacity, interface type, etc.; logical affiliation clearly defines the storage pool, cluster, and management domain to which the storage resource belongs, for example, a certain object storage resource belongs to the first storage cluster in East China; performance parameters include key indicators such as maximum read / write bandwidth, IOPS (input / output operations per second), and average response time; deployment location records the data center location, rack number, and network location information of the storage resource; availability prediction time periods are based on factors such as storage resource maintenance plans and expansion plans to predict the availability status for multiple future time periods. The above information is obtained by periodically polling through the storage management system's API interface and combined with hardware status data collected by the data center monitoring system to build a complete resource supply hierarchy database. This database is dynamically updated as storage devices are added or removed, performance changes, and maintenance plans are adjusted.

[0017] Step S113: Create a dedicated business node for each business and embed the core feature information from the resource consumption trajectory into the business node. The core feature information includes the business's resource demand type preference, demand fluctuation pattern, peak demand prediction period, and key content of service level requirements, so that the business node has the ability to uniquely identify the characteristics of the business demand.

[0018] In this scenario, a dedicated business node is created for the video-on-demand (VOD) service, possessing a unique business identifier. Regarding resource demand type preferences, analysis of the VOD service's resource consumption trajectory reveals that it has historically prioritized object storage for video file storage and distributed file storage for metadata storage. Therefore, object storage and distributed file storage are embedded as the primary demand type preferences for this business node in its attributes. Demand fluctuation patterns are derived from historical resource consumption data. For example, the service's resource demand is significantly higher on weekday evenings and weekends compared to other times; these periodic fluctuation patterns are extracted and embedded into the node. Peak demand prediction periods are based on historical peak occurrence patterns, combined with user growth trends and content update plans, to predict specific periods of peak demand likely to occur in the near future, such as specific time periods during holidays. Service level requirements are determined based on the importance of the business and user experience needs. VOD services typically require high availability and low access latency; therefore, the service level requirement is set to Platinum, including 99.99% availability and millisecond-level response time requirements. These core characteristics are structured and stored in the attribute fields of the business node, achieving a unique identifier for the business demand characteristics.

[0019] Step S114: Create a dedicated resource node for each type of storage resource, embed the core attribute information of the resource supply spectrum into the resource node. The core attribute information includes key content such as the storage resource's capacity, read / write speed, stability performance, expansion potential, and maintenance characteristics, so that the resource node has the ability to uniquely identify the resource supply characteristics.

[0020] For object storage resources in the cloud storage system, dedicated resource nodes are created. Regarding capacity, information such as the total available capacity, used capacity, and remaining available capacity of the object storage cluster is embedded; this information is extracted from the resource supply spectrum and updated in real time. Read / write speeds include maximum sequential read / write speeds and random read / write speeds, reflecting the storage resource's ability to handle large data transfers and frequent small data accesses. Stability performance is measured using indicators such as the frequency of failures, failure recovery time, and data consistency assurance mechanisms in historical operating data, such as the number of failures and mean time between failures (MTBF) of the object storage resource over the past year. Expansion potential is assessed based on factors such as the storage cluster's architecture design, the number of hardware expansion slots, and network bandwidth margins to determine whether capacity and performance can be expanded by adding nodes or disks, as well as the maximum scale and time required for expansion. Maintenance characteristics include maintenance window time, data backup strategies, and disaster recovery capabilities. For example, if the object storage resource is scheduled for routine maintenance every Tuesday morning, employs a three-replica backup strategy, and has cross-regional disaster recovery capabilities, the above core attribute information is integrated and embedded into the resource node, enabling it to comprehensively reflect the resource's supply characteristics.

[0021] Step S115: Calculate the symbiotic association strength between each business node and each resource node. The symbiotic association strength is based on the degree of fit between the resource demand characteristics of the business node and the supply characteristics of the resource node, the frequency of historical interactions, and the adaptation effect feedback factors. It is derived by converting each factor into a dimensionless standard score.

[0022] When calculating the symbiotic relationship strength between video-on-demand service nodes and object storage resource nodes, each influencing factor is first evaluated. Regarding fit, the matching between the resource requirement type preference of the service node and the type of resource node is compared. The high demand for object storage by the video-on-demand service is highly compatible with the type of resource node, thus receiving a higher score. The service level requirements of the service are compared with the performance parameters of the resource node (such as response time and availability). If the performance of the resource node meets or exceeds the service requirements, the fit score is correspondingly increased. Historical interaction frequency is calculated by statistically analyzing the number of times the service node calls the resource node over a past period and normalizing this number with the number of times the service calls other resource nodes, resulting in a dimensionless frequency score. Adaptation effect feedback is scored based on user experience metrics (such as video buffering rate and loading time) and service stability (such as the number of service interruptions due to storage issues) after the service uses the resource node. If the user experience is good and the service operation is stable, the adaptation effect feedback score is higher. The scores of the three factors mentioned above are assigned different weights (the weights are dynamically adjusted according to factors such as the importance of the business and the scarcity of resources), and then a weighted sum is performed to obtain a comprehensive score of the symbiotic relationship strength. The score value is between 0 and 1, and the closer it is to 1, the higher the symbiotic relationship strength.

[0023] Step S116: Establish a connection edge between business nodes and resource nodes based on the symbiotic association strength. The existence of the connection edge is directly determined by whether the symbiotic association strength reaches a preset symbiotic threshold. If the symbiotic threshold is reached, a connection edge is established; otherwise, no connection edge is established, reflecting the effective adaptation relationship between business and resources.

[0024] The preset symbiosis threshold is determined based on the overall resource utilization target and service quality requirements of the cloud storage system, and is set to 0.6 in this scenario. When the calculated symbiosis association strength between a video-on-demand service node and an object storage resource node is 0.75, which is greater than the symbiosis threshold of 0.6, an association edge is established between these two nodes. However, for a video-on-demand service node and a block storage resource node, the calculated symbiosis association strength is 0.4, which is less than the threshold of 0.6, so no association edge is established. Through this method, only resource nodes with a strong compatibility with business needs will establish connections with business nodes, ensuring that the association edge accurately reflects the effective compatibility between business and resources. The establishment of the association edge is completed in the network construction module. Once established, its information (including the identifiers of the two connected nodes) will be stored in the network topology database.

[0025] Step S117: Assign dynamic edge weights to each associated edge. The dynamic edge weights are determined by the symbiotic association strength, the service level requirements of the business node, and the performance parameters of the resource node. They are determined by normalizing each parameter to a comparable scale and are updated in real time as the business resource consumption trajectory changes and the resource supply spectrum is adjusted, reflecting the current adaptation value of the associated edge.

[0026] For the established association edges between video-on-demand service nodes and object storage resource nodes, the calculation of dynamic edge weights needs to comprehensively consider three factors. First, the symbiotic association strength (0.75) is used as the base value. Second, the service level requirement (Platinum level) of the service node corresponds to a level coefficient, for example, Platinum level corresponds to 1.2, Gold level corresponds to 1.0, and Silver level corresponds to 0.8. The performance parameters of the resource node are compared with its actual performance indicators (such as response time and throughput) to obtain a performance satisfaction coefficient. If the average response time of the resource node is better than the threshold of the business requirement, the performance satisfaction coefficient is 1.1; otherwise, the coefficient is reduced according to the degree of difference. These three values ​​(symbiotic association strength, level coefficient, and performance satisfaction coefficient) are normalized to make them fall within the same numerical range (such as 0 to 2), and then the dynamic edge weight is calculated using a geometric mean. For example, if the normalized symbiotic association strength is 0.75, the level coefficient is 1.2, and the performance satisfaction coefficient is 1.1, the dynamic edge weight is the cube root of the product of these three numbers. When the service level requirements of video-on-demand services are increased or the performance of object storage resource nodes decreases due to increased load, the dynamic edge weights will be adjusted accordingly to reflect the changes in the adaptation value of associated edges in real time.

[0027] Step S118: Construct an initial symbiotic network topology based on all business nodes, resource nodes, and associated edges. The initial topology fully presents the initial association state of business and resources, including node attributes, edge weight information, and direct association relationships between nodes.

[0028] After creating all business nodes and resource nodes, establishing associated edges, and assigning edge weights, these elements are integrated to construct the initial symbiotic network topology. In this structure, each node (business node and resource node) is identified by a unique identifier and includes its core attribute information; each associated edge connects the corresponding business node and resource node, and its dynamic edge weight is labeled. A graph database (such as Neo4j) is used to store and represent the initial topology, where nodes are vertices in the graph, associated edges are directed edges connecting vertices (from business nodes to resource nodes), and the edge attributes store dynamic edge weights. Once the initial topology is formed, it can be displayed using graph visualization tools to intuitively present the initial association state between businesses and resources. For example, a video-on-demand business node is connected to an object storage resource node via an associated edge with a specific weight, but has no connection to a block storage resource node.

[0029] Step S119: Perform time-series optimization on the initial symbiotic network topology by introducing a time dimension factor and incorporating historical change data of business resource consumption trajectories and resource supply spectrum into the topology so that the topology reflects the evolution of nodes and associated edges over time.

[0030] The introduction of the time dimension factor is achieved by adding time-series attributes to each node and associated edge in the topology. For video-on-demand service nodes, historical data on their resource consumption trajectories (such as access frequency and demand fluctuations over the past three months) are integrated into a time series and stored in the node's time attribute field. Similarly, historical performance parameter changes of object storage resource nodes (such as IOPS fluctuations and available capacity changes over the past six months) are also embedded as time series in the node attributes. The time dimension information of associated edges includes the changes in dynamic edge weights over different historical periods, such as the daily average value sequence of edge weights over the past month. Through time window sliding technology, historical data is divided into multiple continuous time segments, and the changing trends of node attributes and edge weights within each time segment are analyzed. This allows the topology to not only reflect the current association state but also show its evolution over time, such as the periodic growth pattern of demand for object storage resource nodes from video-on-demand service nodes during holidays.

[0031] Step S1110: By iteratively updating the process, continuously synchronize the real-time changes in the business resource consumption trajectory and the dynamic adjustment of the resource supply spectrum, continuously correct the dynamic edge weights of business node attributes, resource node attributes and related edges, remove invalid related edges, add valid related edges, and form a business resource symbiotic network that reflects the dynamic balance between business demand and resource supply.

[0032] The iterative update process is executed at fixed time intervals (e.g., hourly). In each iteration, real-time data streams of business resource consumption trajectories are first received. The latest resource demand changes for various services, such as video-on-demand, are analyzed, including sudden increases in access frequency or changes in demand type. Based on this, the attribute information of the corresponding business nodes is updated, such as demand fluctuation patterns and peak demand prediction periods. Simultaneously, dynamic adjustment information of the cloud storage system's resource supply spectrum is acquired, such as the completion of capacity expansion for a storage node or performance parameters improvement due to hardware upgrades, updating the attributes of the corresponding resource nodes. Next, the symbiotic association strength between all business nodes and resource nodes is recalculated. For existing association edges, their dynamic edge weights are updated based on the new symbiotic association strength, business service level requirements, and resource performance parameters. If the updated symbiotic association strength is lower than the symbiotic threshold, the association edge is deemed invalid and removed from the topology. For business node-resource node pairs that previously did not have an established association edge but whose newly calculated symbiotic association strength is higher than the threshold, a new association edge is added and assigned a dynamic edge weight. Through continuous iterative updates, the business resource symbiotic network can reflect the dynamic balance between business demand and resource supply in real time, ensuring the accuracy and timeliness of the network topology.

[0033] Step S120: Call the deep learning model to extract multi-dimensional evolution factors from the business resource symbiotic network and generate network evolution status features. The network evolution status features include core information such as the fluctuation trend of business node demand, the change of resource node supply capacity, and the evolution of the symbiotic relationship strength between nodes.

[0034] In the cloud storage resource allocation scenario of this internet company, a deep learning model is invoked to extract evolutionary factors from the constructed business resource symbiotic network. This aims to capture key evolutionary information from the complex network structure and dynamic changes. The input to the deep learning model is the current state and historical evolution data of the business resource symbiotic network. Through multi-layer processing, features that characterize the overall and local changes of the network are extracted. These features collectively constitute the network evolutionary status characteristics, comprehensively reflecting the dynamic evolution of business needs and resource supply.

[0035] Step S121: Convert the current topology data of the business resource co-existence network into structured input data that can be recognized by the deep learning model. The structured input data includes a node feature matrix, an edge weight feature matrix, and a topology correlation matrix. The node feature matrix stores the attribute information of all business nodes and resource nodes, the edge weight feature matrix stores the dynamic edge weights of all associated edges, and the topology correlation matrix stores the association relationships between nodes.

[0036] The current topology data of the business resource symbiotic network is stored in the form of a graph database. To input it into a deep learning model, a structured transformation is required. The node feature matrix is ​​constructed as follows: for video-on-demand (VOD) business nodes, their attribute information (resource demand type preference, demand fluctuation patterns, service level requirements, etc.) is quantified into numerical vectors. For example, the proportion of object storage and distributed file storage in demand type preference corresponds to different dimensions of the vector. For object storage resource nodes, their capacity, read / write speed, and stability performance are also quantified into numerical vectors. All quantized vectors of business nodes and resource nodes are arranged in rows to form the node feature matrix. The number of rows in the matrix is ​​the total number of nodes, and the number of columns is the number of attribute dimensions for each node. The edge weight feature matrix is ​​a square matrix with the number of rows and columns equal to the total number of nodes. The element values ​​in the matrix correspond to the dynamic edge weight of the edge between two nodes. If there is no edge between two nodes, the element value is 0. For example, the matrix element value corresponding to a VOD business node and an object storage resource node is the dynamic edge weight of that edge, while the element value corresponding to a block storage resource node is 0. The topological association matrix is ​​also a square matrix, where a value of 1 indicates that there is an association edge between two nodes, and a value of 0 indicates that there is no association edge. Its structure is similar to that of the edge weight feature matrix, but it only reflects the connection relationship and does not contain weight information. The above matrix is ​​constructed through the data preprocessing module and converted into a tensor format supported by deep learning models.

[0037] Step S122: Input the structured input data into the feature extraction layer of the deep learning model. The feature extraction layer uses multi-layer graph convolution operation to jointly process the node feature matrix, edge weight feature matrix and topological association matrix. Through neighborhood node feature aggregation and edge weight weighting, the local evolution features at the node level are extracted. The local evolution features reflect the changing trend of a single node and its directly related nodes.

[0038] The feature extraction layer of the deep learning model consists of three graph convolutional layers. First, the node feature matrix, edge weight feature matrix, and topological association matrix are input into the first graph convolutional layer. In this layer, for a video-on-demand (VOD) service node, its neighboring nodes are the object storage resource nodes directly associated with it. Through neighboring node feature aggregation, the feature vectors of the object storage resource nodes are combined with the feature vectors of the VOD service node itself. Edge weighting propagation then weights the feature vectors of the neighboring nodes according to the dynamic edge weights of the associated edges; the higher the weight of the neighboring node's feature, the greater its weight in the aggregation process. After processing by the first graph convolutional layer, preliminary local features are obtained. These local features are then input into the second graph convolutional layer to further aggregate features from a broader neighborhood (such as the neighboring nodes of the object storage resource node, i.e., other service nodes associated with that object storage resource node), but the weights decrease as the neighborhood distance increases. The third convolutional layer continues to perform similar operations, ultimately outputting local evolution features at the node level. These local evolution features are represented in vector form, with each vector corresponding to a node. They contain information on the changing trends of the node and its directly related nodes in terms of resource demand and supply capacity. For example, the local evolution features of a video-on-demand service node may show that its demand for object storage resources has recently increased, while the supply capacity of the object storage resource node has slightly decreased.

[0039] Step S123: The local evolution features are globally integrated across nodes and related edges through the global feature capture layer of the deep learning model. The attention mechanism is used to focus on the core nodes and important related edges that play a key role in the network evolution, and the global evolution features at the network level are extracted. The global evolution features reflect the overall change pattern of the entire business resource symbiotic network.

[0040] The global feature capture layer consists of a self-attention mechanism module and a fully connected layer. The input to the self-attention mechanism module is the local evolutionary feature vector of all nodes. For the local evolutionary feature vector of a video-on-demand service node, the self-attention mechanism calculates its similarity to the local evolutionary feature vectors of all other nodes in the network (including other service nodes and resource nodes), obtaining attention weights. Nodes corresponding to core nodes (such as video-on-demand service nodes that have a significant impact on overall resource consumption) and important related edges (such as service-resource related edges with high dynamic edge weights) receive higher attention weights. By multiplying the local evolutionary feature vector of each node with its attention weight and summing the results, global integration across nodes and related edges is achieved. The integrated result is input into the fully connected layer for nonlinear transformation to further extract higher-order features, ultimately obtaining the global evolutionary feature vector at the network level. This global evolutionary feature vector reflects the overall change pattern of the entire service-resource symbiotic network, such as the overall growth trend of resource demand in the network and the bottleneck areas of resource supply.

[0041] Step S124: Input the local evolutionary features and global evolutionary features into the feature fusion layer of the deep learning model, and use cross-dimensional splicing and element-level weighted fusion to deeply integrate the two features to generate fused evolutionary features that have both local details and global perspective.

[0042] The feature fusion layer first adjusts the dimensions of local and global evolutionary features to ensure they share the same feature dimension. For each node's local evolutionary feature vector, a cross-dimensional concatenation method is used to concatenate it with the global evolutionary feature vector along the same feature dimension, forming a higher-dimensional joint feature vector. Then, an element-wise weighted fusion is performed on the joint feature vector using a learnable weight vector. Each element in the weight vector corresponds to a dimension of the joint feature vector, and the importance of different feature dimensions is automatically learned through model training. For example, for a video-on-demand service node, the demand fluctuation pattern dimension in its local evolutionary features might be given a higher weight, while the overall resource supply trend dimension in its global evolutionary features might also be given a higher weight. After weighted fusion, a fused evolutionary feature vector is obtained for each node. This vector contains both local details of the node itself and its neighborhood, as well as global perspective information of the entire network, achieving a deep integration of local and global features.

[0043] Step S125: Decompose the fusion evolution features in multiple dimensions to extract three major categories of core evolution factors: features related to the evolution of business nodes, features related to the evolution of resource nodes, and features related to the evolution of associated edges. Each category of core evolution factors contains multiple specific evolution indicators, covering the key dimensions of network evolution.

[0044] Multi-dimensional decomposition is achieved through feature selection and clustering. First, based on the correlation analysis between each dimension in the fusion evolution feature vector and the evolution of business nodes, resource nodes, and associated edges, dimensions highly correlated with business node evolution are selected to form business node evolution-related features. Similarly, dimensions related to resource node evolution are selected to form resource node evolution-related features, and dimensions related to associated edge evolution are selected to form associated edge evolution-related features. For business node evolution-related features, specific evolution indicators may include demand type change indicators, demand scale fluctuation indicators, and demand peak indicators; resource node evolution-related features may include capacity change indicators, performance fluctuation indicators, and stability indicators; associated edge evolution-related features may include intensity change indicators and adaptation value indicators. For example, from the fusion evolution features of video-on-demand business nodes, specific evolution indicators such as the rate of change of object storage proportion in demand type preference (demand type change indicator) and the fluctuation range of demand scale per unit time (demand scale fluctuation indicator) are decomposed. These indicators collectively constitute business node evolution-related features.

[0045] Step S126: Analyze the resource consumption trajectory change pattern of each business node based on the relevant characteristics of business node evolution, and extract the demand fluctuation trend factor of business node. The demand fluctuation trend factor of business node includes the direction of change of demand type, the fluctuation range of demand scale, the frequency of demand peak occurrence, and the specific evolution information of demand duration adjustment, so as to depict the demand evolution trend of business node.

[0046] Taking video-on-demand (VOD) service nodes as an example, this paper analyzes the changing patterns of their resource consumption trajectory. The direction of demand type change is determined by comparing the proportion of various storage resources called by the business in different time periods. If the proportion of object storage resource calls has recently shown an upward trend, while the proportion of distributed file storage resource calls has decreased, then the direction of demand type change is tilting towards object storage. The fluctuation range of demand scale is obtained by calculating the difference between the maximum and minimum resource demand scale per unit time and then dividing by the average value, reflecting the drastic change in demand scale. The frequency of demand peaks is counted as the number of times the demand scale reaches the peak threshold within a set time window (e.g., one day). For example, during prime time in the evening, the resource demand of VOD services may experience multiple peaks. The adjustment of demand duration is determined by comparing the changes in the length of time that demand remains in a high-demand state in different periods. For example, the high-demand duration of VOD services has recently been extended by a certain proportion compared to the past. Integrating the above specific evolutionary information forms a demand fluctuation trend factor for business nodes, comprehensively depicting the demand evolution trend of VOD service nodes.

[0047] Step S127: Based on the evolutionary characteristics of resource nodes, analyze in detail the supply spectrum adjustment of each resource node, extract the supply capacity change factors of resource nodes, including specific evolutionary information on capacity scale changes, read / write rate fluctuations, stability adjustments and expansion potential changes, and present the supply evolution trend of resource nodes.

[0048] Taking object storage resource nodes as an example, this paper analyzes the adjustment of their supply spectrum. Capacity changes are calculated by comparing current available capacity with historical available capacity, determining the percentage increase or decrease and the rate of change. For example, due to the addition of new disks, the total capacity of this object storage resource node has increased by a certain percentage compared to the previous month. Read / write rate fluctuations are analyzed to reflect changes in actual read / write rates per unit of time, measuring the degree of fluctuation by calculating the standard deviation of the rate. If the fluctuation range of read / write rates has increased recently due to increased storage node load, this indicates improved stability. Stability adjustments are based on changes in failure frequency and failure recovery time. If, through system optimization, the mean time between failures (MTBF) of this resource node is extended and the failure recovery time is shortened, stability is improved. Changes in expansion potential assess the possibility and space for further expansion of the resource node under the existing architecture. For example, with technological advancements, the maximum cluster size supported by this object storage resource node has increased, thus increasing its expansion potential. Integrating the above information into resource node supply capacity change factors clearly presents the evolution of object storage resource node supply.

[0049] Step S128: Analyze the dynamic edge weight change trajectory of each associated edge based on the evolution-related characteristics of the associated edges, and extract the symbiotic association strength evolution factor between nodes. The symbiotic association strength evolution factor between nodes includes specific evolution information such as the magnitude of association strength change, change rate, peak occurrence time, and stable duration, showing the adaptive evolution trend of the associated edges.

[0050] Taking the relationship between video-on-demand service nodes and object storage resource nodes as an example, this paper analyzes the trajectory of dynamic edge weight changes. The magnitude of the relationship strength change is calculated as the difference between the maximum and minimum values ​​of the dynamic edge weight over a period of time, reflecting the overall range of edge weight changes. The rate of change is calculated by dividing the difference in edge weights at adjacent time points by the time interval, indicating how quickly the edge weight changes. For example, during peak business periods, the dynamic edge weight may rise rapidly in a short time. The peak occurrence time records the specific moment when the dynamic edge weight reaches its maximum value, such as 8 PM daily. The stable duration is the length of time the dynamic edge weight remains within a certain stable range. For example, during off-peak periods, the edge weight may fluctuate within a small range, resulting in a longer stable duration. Integrating the above specific evolutionary information forms the symbiotic relationship strength evolution factor between nodes, showcasing the adaptive evolution of the relationship edge.

[0051] Step S129: Structure and integrate the business node demand fluctuation trend factor, the resource node supply capacity change factor, and the symbiotic relationship strength evolution factor between nodes, and classify and arrange them according to multiple dimensions such as time dimension, node type, and relationship strength level to form a multi-dimensional evolution factor set.

[0052] In the structured integration process, a timestamp is first added to each evolutionary factor to clarify its corresponding time information, and they are arranged in chronological order. Then, the demand fluctuation trend factors for business nodes and the supply capacity change factors for resource nodes are categorized according to node type (business node or resource node). Business node types include demand fluctuation trend factors for video-on-demand business nodes, online document collaboration business nodes, etc., while resource node types include supply capacity change factors for object storage resource nodes, block storage resource nodes, etc. For the evolutionary factors of the symbiotic association strength between nodes, they are categorized according to association strength levels (e.g., high, medium, low), with different levels corresponding to different edge weight ranges. For example, the demand fluctuation trend factors for video-on-demand business nodes, the supply capacity change factors for object storage resource nodes, and the symbiotic association strength evolutionary factors of the edges between them (assuming a high strength level) are integrated together and arranged in chronological order to form a multi-dimensional set of evolutionary factors. Each element in this set contains time, node type or association strength level, and specific evolutionary information.

[0053] Step S1210: Construct the basic structure of network evolution status features based on the set of evolutionary factors. The basic structure includes feature titles, dimension classification labels, factor display areas, and time series axis core components.

[0054] The basic structure of the network evolution status feature adopts a hierarchical data structure design. The feature title identifies the name of the network evolution status feature and the corresponding business resource symbiotic network version information; the dimension classification identifies the classification dimension of the corresponding evolution factor set, such as time dimension, node type (business node, resource node), and association strength level (high, medium, low); the factor display area accommodates the specific information of various evolution factors, displayed in partitions according to the dimension classification; the time axis runs through the entire basic structure, marking the time points or time periods corresponding to each evolution factor, realizing the time positioning of evolution information. For example, the feature title of the basic structure is "Cloud Storage Resource Symbiotic Network Evolution Status Feature_v1.0", the dimension classification includes "First Quarter of 2024" (time dimension), "Business Node - Video on Demand" (node ​​type dimension), "Association Strength - High" (association strength level dimension), etc. The factor display area is divided into different sub-areas according to these identifiers, displaying the corresponding category of evolution factors, and the time axis marks the time scale from January 1, 2024 to March 31, 2024.

[0055] Step S1211: Fill the basic structure with the specific evolution information in the set of evolutionary factors, generate a unique visualization method for each evolutionary factor, and intuitively display the evolution trend through curves, bar charts, and heat maps, generating network evolution status characteristics that include core information such as the fluctuation trend of business node demand, the change of resource node supply capacity, and the evolution of the symbiotic relationship strength between nodes.

[0056] The specific evolutionary information from the set of evolutionary factors is filled into the corresponding factor display area of ​​the basic structure. For the demand scale fluctuation amplitude evolutionary factor of video-on-demand service nodes, a line chart visualization is used, with the horizontal axis representing the time series and the vertical axis representing the fluctuation amplitude value. The rise and fall of the curve visually shows its trend over time. The capacity scale change factor of resource nodes can be represented by a bar chart, with each bar representing the capacity value of a time unit (e.g., one week). The change in capacity is reflected by the change in the height of the bars. For the evolutionary factor of the symbiotic association strength between nodes, if multiple association edges are involved, a heatmap can be used. The horizontal and vertical axes represent service nodes and resource nodes, respectively, and the color intensity of the heatmap indicates the magnitude of the association strength change. By embedding the above charts into the factor display area of ​​the basic structure through a visualization rendering engine, a complete network evolution trend feature is finally formed. This feature is presented in the form of a visualization report, which clearly includes the core information of the business node demand fluctuation trend (such as the demand fluctuation curve of video on demand business), the resource node supply capacity change (such as the capacity bar chart of object storage resources), and the evolution of the symbiotic relationship strength between nodes (such as the heat map of the relationship edge strength).

[0057] Step S130: Based on the network evolution situation characteristics, derive the service resource adaptation demand sequence. The service resource adaptation demand sequence is transformed from the interaction rules of each node and the edge weight change characteristics in the network evolution situation characteristics. It is used to determine the adaptation direction and adaptation mode of various services to storage resources in different prediction periods.

[0058] In this internet enterprise scenario, the network evolution fully reflects the dynamic evolution of business needs and resource supply. Deriving a sequence of business resource adaptation requirements based on this involves extracting the interaction patterns between business and resources, as well as the changing characteristics of associated edge weights, from this evolutionary information, and transforming them into specific adaptation requirements. Different prediction periods correspond to different stages of business demand development; therefore, the adaptation requirement sequence needs to determine the adaptation direction (e.g., what kind of resources need to be added, or the performance of existing resources need to be adjusted) and adaptation mode (e.g., resource access methods, association methods) for various businesses (such as video-on-demand, online document collaboration) regarding storage resources for each period.

[0059] Step S131: Extract the core evolution information from the network evolution status characteristics, including the fluctuation trend of business node demand, the change of resource node supply capacity, and the evolution of the symbiotic relationship strength between nodes, and establish the corresponding relationship between the evolution information and the business resource adaptation requirements.

[0060] Core evolutionary information is extracted from the visualized representation of network evolution characteristics. For video-on-demand service nodes, the direction of demand type change (leaning towards object storage) and the magnitude of demand scale fluctuations (recently increased fluctuations) are extracted from their demand fluctuation trends. For object storage resource nodes, the changes in capacity scale (continuous growth) and read / write rate fluctuations (recently decreased) are extracted from their supply capacity changes. For the evolution of the symbiotic association strength between the two, the magnitude of association strength change (overall increase) and peak occurrence time (prime evening hours) are extracted. When establishing corresponding associations, if the demand type of video-on-demand services leans towards object storage, and the capacity scale of object storage resources is increasing but the read / write rate is decreasing, the corresponding business resource adaptation requirements may include improving the read / write performance of object storage resources to match business needs. If the association strength of the association reaches its peak during prime evening hours, the adaptation requirements may include ensuring sufficient resource supply during this period to maintain a high-strength association. By constructing a decision rule base, different combinations of evolutionary information are mapped to corresponding adaptation requirement types, realizing the corresponding association between evolutionary information and adaptation requirements.

[0061] Step S132: Divide the network evolution characteristics into multiple future prediction periods in chronological order. The duration of each future prediction period is determined comprehensively based on the business demand change cycle and resource supply adjustment cycle.

[0062] The business demand change cycle is determined by analyzing the historical resource consumption patterns of various services, such as video-on-demand (VOD). For example, the demand change cycle for VOD is typically one week, exhibiting a clear intra-weekly fluctuation pattern. The resource supply adjustment cycle is determined based on the expansion and maintenance cycles of the cloud storage system; for instance, storage resource expansion planning is usually carried out on a monthly basis. Taking both cycles into account, the future forecast period is divided into short-term (e.g., the next 24 hours, with each hour as a sub-period), medium-term (the next week, with each day as a sub-period), and long-term (the next month, with each three days as a sub-period). The length of each forecast period is chosen to capture both short-term fluctuations in business demand and accommodate medium- and long-term adjustments in resource supply. For example, the hourly sub-periods of the short-term forecast period can accurately address the demand changes of VOD services at different hours within a day, while the three-day sub-periods of the long-term forecast period provide sufficient preparation time for adjustments such as storage resource expansion.

[0063] Step S133: For each future forecast period, extract the demand fluctuation trend information of all business nodes within that future forecast period to determine the direction of demand type change, the magnitude of demand fluctuation, the frequency of demand peaks, and the adjustment of demand duration for each business node within that future forecast period.

[0064] Taking a specific hourly sub-period within a short-term forecast period (e.g., 20:00-21:00 on a future day) as an example, the demand fluctuation trend information of video-on-demand service nodes is extracted. The direction of demand type change is determined by analyzing the predicted proportion of resource calls to object storage, distributed file storage, etc., within this sub-period. If the predicted proportion of object storage calls continues to rise by 5%, the direction of demand type change is a further shift towards object storage. The fluctuation amplitude of demand scale is calculated based on the difference between the maximum and minimum predicted demand scale within this sub-period, reflecting the drastic change in demand within that hour. The frequency of demand peaks predicts the number of times the demand scale reaches a preset peak threshold within this sub-period. Since 20:00-21:00 is a peak period for video-on-demand, multiple peaks may be predicted. The adjustment of demand duration compares the length of time that demand has remained in a high-demand state within this sub-period with the same period in history, predicting whether it will be extended or shortened. By extrapolating the historical evolution patterns in the network evolution characteristics to this future forecast period and combining them with the recent development plans of the business (e.g., new video upload plans), the above-mentioned demand fluctuation trend information is extracted.

[0065] Step S134: Extract the supply capacity change information of all resource nodes within the future forecast period to determine the capacity size change, read / write rate fluctuation, stability adjustment and expansion potential change of each resource node within the future forecast period.

[0066] For object storage resource nodes, during the 20:00-21:00 sub-period of the aforementioned short-term forecast period, capacity changes are predicted based on current used capacity, expected data write volume, and data deletion volume. If the expected write volume exceeds the deletion volume during this sub-period, the capacity will increase. Read / write rate fluctuations are predicted based on current load conditions and historical rate change patterns. If a large number of users are expected to access video content during this sub-period, it may lead to a certain degree of fluctuation in read / write rates. Stability adjustments are determined based on the hardware status prediction and maintenance plan of the storage nodes. If there is no planned maintenance and the hardware is in good condition during this sub-period, stability is expected to remain at a high level. Expansion potential changes are assessed to determine whether the resource nodes have the conditions for further expansion during this sub-period, such as whether there are free disk slots or available network bandwidth. If so, expansion potential exists; otherwise, it does not. The supply capacity changes of each resource node in the future forecast period are predicted comprehensively using real-time data collected by the resource monitoring system and planning data from the resource management system.

[0067] Step S135: Analyze the evolution information of the symbiotic association strength between business nodes and resource nodes within the future prediction period to determine the dynamic edge weight change trend, association strength level and adaptation value change of the association edges between nodes.

[0068] For the association edges between video-on-demand service nodes and object storage resource nodes, the dynamic edge weight change trend during the 20:00-21:00 sub-period is extrapolated by analyzing the edge weight change curves of the same period in history and combining them with current demand and supply forecasts. It is predicted that the edge weights will first rise and then fall during this period, reaching a peak around 20:30. The association strength level is determined based on the predicted dynamic edge weight range. If the peak edge weight is in the high-strength range (e.g., 0.8-1.0), the association strength level for this period is high. The adaptation value change is determined by comparing the predicted edge weights for this period with the average edge weights of the previous period (e.g., 19:00-20:00). If the predicted average is higher than the previous period, the adaptation value increases; otherwise, it decreases. For the association edges between other service nodes and resource nodes, a similar method is used to analyze their symbiotic association strength evolution information to determine the dynamic edge weight change trend, association strength level, and adaptation value change.

[0069] Step S136: Based on the fluctuation trend of business node demand and the changes in resource node supply capacity, combined with the evolution information of the symbiotic relationship strength between nodes, determine the adaptation priority of each business node to storage resources in the future prediction period. The adaptation priority is jointly determined by the urgency of business demand, the sufficiency of resource supply, and the degree of adaptation and fit between the two.

[0070] The adaptation priority is determined using a weighted scoring method. The urgency of business needs is assessed based on the fluctuation range of demand scale and the frequency of peak demand occurrences for video-on-demand service nodes. Large fluctuations and high peak frequencies indicate high urgency and a higher weight. Resource sufficiency is assessed based on changes in the capacity scale of object storage resource nodes and fluctuations in read / write speeds. Sufficient capacity but declining read / write speed fluctuations indicate moderate sufficiency and a moderate weight. Adaptation fit is assessed based on the evolution of the symbiotic relationship strength between nodes and changes in adaptation value. High relationship strength and increased adaptation value indicate high fit and a higher weight. Each of these three factors is scored (e.g., 1-5 points), and then a weighted total score is calculated according to the set weights (e.g., urgency 40%, sufficiency 30%, fit 30%). The adaptation priority is then divided into high, medium, and low levels based on the total score. For example, if the urgency score for video-on-demand services during the 20:00-21:00 sub-period is 5, the sufficiency score is 3, and the fit score is 4, the weighted total score would be 5*40%+3*30%+4*30%=4.1, and the adaptation priority would be determined as high.

[0071] Step S137: Based on the adaptation priority and the fluctuation trend of business node demand, determine the adaptation direction of each business node in the future forecast period. The adaptation direction is used to reflect the types of storage resources that the business node needs to add, adjust or maintain, and the performance requirements of the corresponding resources.

[0072] For video-on-demand service nodes with high adaptation priority, the adaptation direction should be determined based on their demand fluctuation trends (demand type shifting towards object storage, increased fluctuation in demand scale). If the current read / write speed of object storage resources fluctuates and declines, failing to meet the needs of business growth, the performance of object storage resources needs to be adjusted, such as upgrading the hardware of storage nodes or optimizing storage algorithms to improve read / write speeds. If the business demand, in addition to object storage, also shows a slight increase in demand for distributed file storage, but the current supply of such resources is relatively sufficient, it may be necessary to maintain the existing configuration of distributed file storage resources. If it is predicted that the business will introduce new high-definition video formats in the future, requiring higher capacity storage resources, and the capacity growth rate of existing object storage resources may not be able to fully match this, then it is necessary to add capacity to object storage resources. At the same time, the performance requirements of the corresponding resources should be clearly defined, such as the minimum threshold that the read / write speed of the adjusted object storage resources should reach, and the expansion potential that the newly added capacity object storage resources should possess, forming a specific description of the adaptation direction.

[0073] Step S138: Combining the changes in the supply capacity of resource nodes and the evolution of the symbiotic relationship strength between nodes, determine the adaptation mode of each business node in the future prediction period. The adaptation mode is used to reflect the association method between business nodes and resource nodes, the resource call method, and the adaptation adjustment frequency.

[0074] Step S1381: Extract key information on capacity scale changes, read / write rate fluctuations, and stability adjustments in the resource node supply capacity changes during the future prediction period, reflecting the upper limit and range of changes in the service capacity that the resource node can provide during the future prediction period.

[0075] For object storage resource nodes, during the 20:00-21:00 sub-period, capacity scale change information shows that its total capacity will increase, but the growth rate of available capacity will slow down due to increased demand; read / write rate fluctuation information indicates that its maximum read / write rate may drop to a low level during this period, but the average read / write rate can still be maintained within a certain range; stability adjustment information shows that the probability of failure is low and the stability is good. The above information comprehensively reflects the upper limit of the service capacity of the resource node during the predicted period (such as maximum available capacity and maximum read / write rate) and the range of variation of these upper limits (such as the fluctuation range of read / write rate).

[0076] Step S1382: Extract key information on the trend of association strength change, association strength level and adaptation value change in the evolution of symbiotic association strength between nodes during the future prediction period, reflecting the stability of the association between business nodes and resource nodes and the level of adaptation value.

[0077] The trend of the association strength between the video-on-demand service node and the object storage resource node during the prediction period is an increase followed by a decrease; the association strength level is high; and the adaptation value changes with a positive increase. The above information indicates that the association between the two is relatively stable during this period, especially at peak times when the association strength is high and the adaptation value is also constantly increasing, indicating that the current association is valuable to both the business and the resources.

[0078] Step S1383: Analyze the supply capacity of resource nodes in the current forecast period and the demand fluctuation trend of business nodes in the same forecast period. If the lower limit of the supply capacity of resource nodes is higher than or equal to the upper limit of the estimated demand of business nodes, the correlation method is determined to be a stable correlation method; otherwise, the correlation method is determined to be a dynamic correlation method.

[0079] During the current forecast period (20:00-21:00), the lower limit of the supply capacity of object storage resource nodes (such as minimum available capacity and minimum read / write rate) is determined based on their supply capacity change information; the upper limit of the estimated demand of video-on-demand service nodes (such as maximum demand capacity and maximum read / write bandwidth demand) is determined based on their demand fluctuation trend information. If the minimum available capacity of object storage resources is greater than or equal to the maximum demand capacity of the video-on-demand service, and the minimum read / write rate is greater than or equal to the maximum read / write bandwidth demand of the service, then the resource nodes can stably meet the service demand, and the association method is determined to be a stable association method, that is, the service nodes and resource nodes establish a fixed and long-term association relationship. Otherwise, if the lower limit of resource supply capacity cannot fully cover the upper limit of the estimated business demand, resource shortages may occur at certain times, and the association method is determined to be a dynamic association method, that is, the association relationship is dynamically adjusted according to the real-time resource supply and demand situation.

[0080] Step S1384: Based on the comparison result of the current symbiotic association strength between the business node and the resource node and the preset strength threshold, determine the association method between the business node and the resource node. If the symbiotic association strength is higher than or equal to the strength threshold, the direct association method is adopted, and the business node directly calls the corresponding resource node; if the symbiotic association strength is lower than the strength threshold, the indirect association method is adopted, and the association between the business node and the resource node is realized through an intermediate scheduling node.

[0081] The preset strength threshold is set based on the system resource allocation strategy and business importance. For example, for core businesses like video-on-demand, the strength threshold is set to 0.7. If the current symbiotic association strength (assumed to be 0.85) between the video-on-demand service node and the object storage resource node is higher than this threshold, a direct association method is adopted. The video-on-demand service node can directly initiate resource call requests to the object storage resource node without going through an intermediate node. If the symbiotic association strength between an edge service node and a resource node is 0.6, which is lower than the threshold of 0.7, an indirect association method is adopted. The resource call request of the service node is first sent to the intermediate scheduling node, which allocates an appropriate resource node based on the resource status and forwards the request.

[0082] Step S1385: Determine the resource call method based on the change rate of the number of resource demand types and the coefficient of variation of demand size of the business nodes. If the change rate of the number of demand types is lower than the type change threshold and the coefficient of variation of demand size is lower than the size fluctuation threshold, a fixed call method is adopted, and resources are called according to a preset period and a fixed quota. If the change rate of the number of demand types is higher than or equal to the type change threshold or the coefficient of variation of demand size is higher than or equal to the size fluctuation threshold, an elastic call method is adopted, and the call volume is dynamically adjusted according to the real-time resource requests of the business nodes.

[0083] The rate of change in the number of resource demand types is calculated as the ratio of the increase or decrease in the number of resource demand types per unit time to the original number of types for video-on-demand service nodes. The type change threshold is set to a small value (e.g., 0.1). The coefficient of variation in demand size is obtained by calculating the ratio of the standard deviation to the mean of the demand size, reflecting the dispersion of the demand size. The size fluctuation threshold is set to a medium level (e.g., 0.3). If the rate of change in the number of demand types for video-on-demand service is 0.05 (below 0.1) and the coefficient of variation in demand size is 0.4 (above 0.3), since the coefficient of variation in demand size is higher than the size fluctuation threshold, an elastic call method is adopted. Under this method, the service nodes do not pre-set fixed resource quotas, but dynamically adjust the call volume to object storage resource nodes according to the real-time resource request volume. For example, the call volume is automatically increased when user access volume surges and decreased when access volume decreases. If both the rate of change in the number of demand types and the coefficient of variation in demand size are lower than the corresponding thresholds, a fixed call method is adopted, and resources are called according to a preset period (e.g., per hour) and a fixed quota (e.g., a certain amount of storage space is called per hour).

[0084] Step S1386: Combining the adjustment cycle of the resource node's supply capacity and the change cycle of the business node's demand fluctuation, determine the adaptation adjustment frequency. Compare the adjustment cycle of the resource node with a first cycle threshold, and compare the change cycle of the business node with a second cycle threshold. If the adjustment cycle of the resource node is longer than the first cycle threshold and the change cycle of the business node is longer than the second cycle threshold, then determine the adaptation adjustment frequency as the first frequency value. If the adjustment cycle of the resource node is shorter than or equal to the first cycle threshold or the change cycle of the business node is shorter than or equal to the second cycle threshold, then determine the adaptation adjustment frequency as the second frequency value. Wherein, the first frequency value is lower than the second frequency value.

[0085] The adjustment cycle of resource node supply capacity (e.g., a one-month expansion cycle for object storage resources) is compared with the first cycle threshold (e.g., two weeks), and the change cycle of business node demand fluctuations (e.g., a weekly demand cycle for video-on-demand services) is compared with the second cycle threshold (e.g., three days). If the adjustment cycle of object storage resources (one month) is longer than the first cycle threshold (two weeks), and the change cycle of video-on-demand services (one week) is longer than the second cycle threshold (three days), then the adaptation adjustment frequency is determined to be the first frequency value (e.g., once a day). If the adjustment cycle of resource nodes is shortened to one week due to emergency expansion needs (shorter than the first cycle threshold of two weeks), or the change cycle of business nodes is shortened to two days due to sudden events (shorter than the second cycle threshold of three days), then the adaptation adjustment frequency is determined to be the second frequency value (e.g., once every six hours). A first frequency value lower than a second frequency value indicates a longer adjustment interval, suitable for relatively slow changes; a second frequency value with a shorter adjustment interval is suitable for faster changes, ensuring that the adaptation mode can respond promptly to changes in business and resources.

[0086] Step S1387: Integrate the association method, resource call method, and adaptation adjustment frequency to construct the core structure of the adaptation mode, and determine the logical association and execution order among the association method, resource call method, and adaptation adjustment frequency.

[0087] The core structure of the adaptation mode is built by integrating the association methods (direct association) and resource invocation methods (elastic invocation) between video-on-demand service nodes and object storage resource nodes, as well as the adaptation adjustment frequency (a second frequency value, once every six hours). In terms of logical association, direct association is the foundation of resource invocation; service nodes can only achieve elastic invocation of resource nodes through direct association. The adaptation adjustment frequency determines the time interval for checking and adjusting the association and resource invocation methods. In terms of execution order, a direct association is first established, and then resource invocation is performed using the elastic invocation method based on this association. Every six hours (adaptation adjustment frequency), based on changes in resource supply capacity and the evolution of the symbiotic association strength, the current association and resource invocation methods are checked to see if they are still applicable. If not, corresponding adjustments are made. For example, if an adjustment check reveals an increase in latency for resource invocation under the direct association method, it may be necessary to temporarily switch to an indirect association method, optimizing the path through an intermediate scheduling node.

[0088] Step S1388: Supplement the specific execution details of the association method, determine the node interaction process, data transmission path and status synchronization process in direct or indirect association; refine the specific operation specifications of the resource call method, determine the call triggering conditions, call parameter settings and call result feedback process in fixed call or elastic call; define the specific implementation standards for the adaptation adjustment frequency, determine the adjustment time interval, adjustment trigger threshold and the status synchronization process after adjustment, and form the adaptation mode of each business node in the future prediction period.

[0089] For the direct association method, the node interaction process specifies the format for video-on-demand service nodes to send resource call requests to object storage resource nodes, and the service identifier and resource identifier that must be included in the request; the data transmission path clearly defines the network route from the service node to the resource node, prioritizing low-latency links; the state synchronization process requires the resource node to return a response message containing the processing result and resource status (such as current available capacity) to the service node after processing the call request. For the elastic call method, the call trigger condition is set to automatically initiate an additional resource call request when the real-time resource demand of the service node reaches a preset trigger threshold (such as the currently used resources reaching 80% of the quota); the call parameter settings include the upper limit of the resource amount for each call, the timeout time of the call request, etc.; the call result feedback process requires the resource node to return the allocated resource ID, available time, and other information to the service node after completing the resource allocation. For the adaptation adjustment frequency (once every six hours), the specific implementation standard defines the adjustment time interval as the hourly interval every six hours. The adjustment trigger thresholds include the dynamic edge weight change of the associated edge exceeding a preset ratio, the resource call failure rate exceeding a threshold, etc. When any trigger threshold is met, the adaptation adjustment process is initiated. The post-adjustment status synchronization process requires that the adjusted association method, resource call method, and other information be synchronized to the business resource co-existence network and resource management system to ensure that the information of each module is consistent, and finally form a complete adaptation mode description.

[0090] Step S139: Integrate the adaptation priority, adaptation direction and adaptation mode of various business nodes in each future prediction period to form the business resource adaptation requirement unit corresponding to the future prediction period.

[0091] During the 20:00-21:00 sub-period, for video-on-demand service nodes, the adaptation priority is high. The adaptation direction is to adjust the performance of object storage resources (increasing read / write speeds to a specific threshold), maintain the distributed file storage resource configuration, and add object storage resource capacity (with some expansion potential). The adaptation modes are direct association and elastic invocation, and the adaptation adjustment frequency is the second frequency value (once every six hours). Integrating the above information clarifies the specific content and interrelationships of each adaptation requirement. For example, the high adaptation priority determines that the needs of adjusting object storage performance and adding capacity for video-on-demand services should be prioritized during resource allocation, and the elastic invocation mode should match the adjusted resource performance. Simultaneously, for online document collaboration service nodes, similar adaptation priorities, directions, and modes are determined and integrated into their corresponding adaptation requirement information. The adaptation requirement information of all service nodes within this predicted period is summarized to form a business resource adaptation requirement unit for this future predicted period. This unit is a structured data set containing details of the adaptation requirements of all services within this period.

[0092] Step S1310: Arrange all business resource adaptation requirement units in the time order of the future forecast period, and mark the corresponding time range for each unit to form a business resource adaptation requirement sequence that can cover multiple future forecast periods and reflect the adaptation requirements of various businesses in different future forecast periods.

[0093] The future forecast periods are arranged chronologically as follows: 20:00-21:00 on [Date], 21:00-22:00, ..., 20:00-21:00 the following day. The business resource adaptation requirements corresponding to each period are arranged in this chronological order, with each unit labeled with its specific time range. For example, the first unit is labeled "20:00-21:00," including adaptation requirements for services such as video-on-demand and online document collaboration within that period; the second unit is labeled "21:00-22:00," including adaptation requirements for the corresponding time period. This arrangement forms a sequence of business resource adaptation requirements that covers multiple forecast periods within the next 24 hours (assuming it's divided into 24 periods). Each adaptation requirement unit reflects the storage resource adaptation requirements of various services within that period, and the units in the sequence are presented sequentially in chronological order.

[0094] Step S140: Dynamically match and deduce the sequence of business resource adaptation requirements with the resource supply spectrum of cloud storage system through a deep learning model to generate a set of resource elastic allocation schemes. The set of resource elastic allocation schemes includes resource allocation adjustment strategies, execution paths and adaptation guarantee-related steps for different business scenarios.

[0095] In internet enterprise scenarios, the business resource adaptation requirement sequence clearly defines the adaptation needs of various businesses during different forecast periods, while the cloud storage system resource supply spectrum provides information on resource supply capacity. Dynamic matching and deduction using deep learning models aims to intelligently match these needs with supply, considering resource availability, business priorities, and various constraints (such as physical limitations and deployment constraints), thereby generating a series of elastic resource allocation solutions. Each solution targets a specific business scenario (such as peak-hour scenarios for video-on-demand services or daily scenarios for online document collaboration services), including how to adjust resource allocation, the path used for allocation, and how to ensure the adaptability of the allocation process and results. Ultimately, this forms a set of solutions for the system to select and execute.

[0096] Step S141: Perform time-series alignment processing on the input layer of the deep learning model to input the business resource adaptation demand sequence and the cloud storage system resource supply spectrum, so that each future prediction period in the business resource adaptation demand sequence corresponds to the corresponding prediction period information in the resource supply spectrum.

[0097] Each future forecast period in the business resource adaptation demand sequence (e.g., 20:00-21:00 on [Date] in 2024) has a defined time range. The cloud storage system resource supply spectrum also includes forecast period information organized by time dimension (e.g., resource supply forecast for 20:00-21:00 on the same day). The matching input layer associates each period in the demand sequence with the period information of the same timestamp in the supply spectrum through timestamp comparison, ensuring a one-to-one correspondence between the two in time. For example, the adaptation demand of video-on-demand service from 20:00 to 21:00 is matched with the supply information of object storage resources and distributed file storage resources during that period. For forecast periods that exist in the demand sequence but are temporarily missing in the supply spectrum, the model will make predictions to complete the forecast based on historical data and trends in the supply spectrum; for periods that exist in the supply spectrum but are not involved in the demand sequence, they are temporarily marked as backup period information.

[0098] Step S142: The business resource adaptation requirement sequence is analyzed layer by layer through the requirement analysis sub-network of the deep learning model. The adaptation priority, adaptation direction and adaptation mode of various businesses in each future prediction period are extracted to generate a structured requirement analysis result. The requirement analysis result is classified and organized according to business type and future prediction period.

[0099] The demand parsing subnetwork comprises an embedding layer and a bidirectional LSTM layer. The embedding layer converts unstructured text descriptions in the business resource adaptation demand sequence (such as "adjust object storage resource performance" in the adaptation direction) into vector representations. The bidirectional LSTM layer performs time-series modeling on the sequence data, capturing the dependencies between adaptation demands across different prediction periods. During the layer-by-layer parsing process, each future prediction period is first identified. Then, for each type of business within that period (such as video-on-demand), its adaptation priority (high), adaptation direction (adjusting object storage performance, adding object storage capacity, etc.), and adaptation mode (direct association, elastic call, etc.) are extracted. This extracted information is organized into structured data, such as using business type as the first-level key and future prediction period as the second-level key, with corresponding values ​​being dictionaries containing adaptation priority, direction, and mode. For example, “Video on Demand” -> “20:00-21:00” -> {“Adaptation Priority”: “High”, “Adaptation Direction”: [“Adjust object storage performance to threshold A”, “Add object storage capacity B”], “Adaptation Mode”: {“Association Method”: “Direct Association”, “Invocation Method”: “Elastic Invocation”}}, forming a structured requirement parsing result.

[0100] Step S143: The supply analysis subnetwork of the deep learning model is used to analyze the resource supply spectrum of the cloud storage system layer by layer, extract the supply capacity, deployment location and availability status of various resources in each prediction period, and generate structured supply analysis results. The supply analysis results are classified and organized according to resource type and prediction period.

[0101] The supply parsing subnetwork employs a convolutional neural network (CNN) and an attention mechanism. The CNN layer extracts local features from the raw data of the resource supply spectrum (such as capacity, rate, location, and other multi-dimensional information); the attention mechanism focuses on supply features that play a crucial role in adapting to business needs (such as the read / write rate of object storage resources, which is of interest to video-on-demand services). Parsing is performed layer by layer, based on the prediction period and resource type (object storage, distributed file storage, block storage, etc.). For each prediction period's object storage resource, its supply capacity (total capacity, available capacity, read / write rate range), deployment location (data center A, rack B), and availability status (normal, warning, maintenance) are extracted. Similarly, other resource types are extracted in a similar manner. The generated structured supply parsing result uses resource type as the first-level key and prediction period as the second-level key, with corresponding values ​​being dictionaries containing supply capacity, deployment location, and availability status. For example, "Object Storage" -> "20:00-21:00" -> {"Supply Capacity": {"Total Capacity": C, "Available Capacity": D, "Read / Write Rate Range": [E, F]}, "Deployment Location": "Data Center A - Rack B", "Availability Status": "Normal"}.

[0102] Step S144: Input the demand analysis results and supply analysis results into the dynamic matching sub-network of the deep learning model, establish matching weights based on the adaptation priority, and perform bidirectional matching of business demand and resource supply one by one according to the future prediction period to generate preliminary matching results. The preliminary matching results are used to reflect the candidate resource set of each business in the corresponding future prediction period.

[0103] The core of the dynamic matching sub-network is a Transformer-based matching module. First, based on the adaptation priority in the demand analysis results, a matching weight is assigned to each business demand. Businesses with higher adaptation priority (such as video-on-demand) receive higher weights and are prioritized for fulfillment during the matching process. The matching module processes demands sequentially according to future prediction time periods. For each time period, the demand analysis information (adaptation direction, adaptation mode) for all businesses within that period, along with the supply analysis information (supply capacity and availability status of various resources), is input into the matching module. During the bidirectional matching process, on one hand, resources (object storage resources) that meet the type requirements are selected based on the business's adaptation direction (e.g., the need to adjust object storage performance). On the other hand, the availability status and supply capacity of the resources are used to determine whether they can meet the business's adaptation mode (e.g., the stability requirements of direct association methods). For the video-on-demand business's demand between 20:00 and 21:00, the matching module will select object storage resources with normal availability and read / write speeds that meet the adjustment requirements within that time period from the supply analysis results, as a candidate resource set. The preliminary matching results include a list of candidate resources for each business-time period, as well as a matching score for each candidate resource (calculated based on the degree of fit between the adaptation direction and the pattern).

[0104] Step S145: Conduct a feasibility analysis on the preliminary matching results. Combine the physical limitations, deployment constraints and performance bottlenecks of resource supply, eliminate candidate resources whose supply characteristics do not match the business adaptation direction, or whose supply capacity cannot support the association method and resource calling method required by the business adaptation mode, and form a feasible matching result.

[0105] The feasibility analysis is conducted from multiple dimensions. Physical constraints include whether the physical location of the storage resources is in the same region as the deployment location of the business (e.g., if the video-on-demand business mainly serves users in East China, the candidate object storage resources must be deployed in the East China data center). If a candidate resource is deployed in the North China data center, it will be eliminated due to physical location constraints. Deployment constraints involve whether the storage cluster to which the resource belongs allows cross-business calls. For example, if a certain object storage resource belongs to a dedicated cluster and is only allowed to be used by specific businesses, and the video-on-demand business is not on the allowed list, it will be eliminated. Performance bottleneck analysis checks whether the supply capacity of the candidate resources can support the requirements of the business adaptation mode. For example, the elastic call method requires resources to have rapid expansion capabilities. If the expansion potential of a candidate object storage resource has reached its limit and cannot meet the sudden increase in demand for elastic calls, it will be eliminated. After the feasibility analysis, the remaining candidate resources form a feasible matching result. For example, the feasible matching result for the video-on-demand business from 20:00 to 21:00 may only be two object storage resource nodes in the East China data center.

[0106] Step S146: Based on the feasible matching results, generate a resource allocation adjustment strategy for the needs of each business in each future forecast period. The resource allocation adjustment strategy is used to reflect the initial configuration, adjustment direction and adjustment target of resource allocation.

[0107] Taking the feasible matching results (two object storage resource nodes) of the video-on-demand service from 20:00 to 21:00 as an example, a resource allocation adjustment strategy is generated. The initial configuration is the current resource allocation of the service on these two resource nodes, such as Node 1 allocating 60% of the resources and Node 2 allocating 40%. The adjustment direction is determined according to the adaptation direction. If it is necessary to improve the overall read / write performance and add capacity, possible adjustment directions include increasing the resource allocation ratio of Node 1 (because of its better performance), reducing the ratio of Node 2, and adding some capacity from Node 1. The adjustment objectives are clear and specific indicators, such as adjusting the resource allocation ratio of Node 1 to 70%, increasing the read / write rate to G value, and adding H capacity; adjusting the resource allocation ratio of Node 2 to 30%, and maintaining the read / write rate at I value. At the same time, the strategy should also explain the basis for the adjustment, such as the performance potential and current load of Node 1, and the stability performance of Node 2, forming a complete description of the resource allocation adjustment strategy.

[0108] Step S147: Plan the execution path of the resource allocation adjustment strategy. The execution path is used to reflect the initiating node, relay node, target node and data transmission link of resource allocation, and arrange the operation steps, sequence and connection method in the allocation process.

[0109] For example, step S1471: Determine the business nodes and target resource nodes corresponding to the resource allocation adjustment strategy, use the business nodes as the initiating nodes for resource allocation, and use the resource nodes that need to be added, adjusted, or maintained as the target nodes.

[0110] The resource allocation adjustment strategy clearly states that the video-on-demand service needs to adjust the allocation ratio of object storage resource nodes 1 and 2, and increase the capacity of node 1. Therefore, the initiating node is the video-on-demand service node; the target nodes are object storage resource node 1 (which needs to have its allocation ratio adjusted and its capacity increased) and object storage resource node 2 (which needs its allocation ratio adjusted).

[0111] Step S1472: Analyze the network topology between the initiating node and the target node, and in conjunction with the resource deployment architecture of the cloud storage system, determine whether it is necessary to use a relay node for resource allocation. If the initiating node and the target node are directly connected and the transmission bandwidth meets the requirements, no relay node is set. If there is a transmission bottleneck or deployment isolation, select a resource node with stable performance and a central location as the relay node.

[0112] The initiating node (video-on-demand service node) and the target nodes (object storage node 1 and node 2) are both deployed in the same network area of ​​the East China data center. Network topology analysis shows a direct gigabit Ethernet link between the initiating node and node 1, with current bandwidth utilization at 60%, indicating sufficient bandwidth for data transmission during resource allocation. A similar direct link exists between the initiating node and node 2, with similar bandwidth conditions. Therefore, it is determined that no intermediate node is needed, and resource allocation can proceed directly.

[0113] Step S1473: Plan the data transmission link between the initiating node and the target node or via a relay node, and divide the resource allocation process into multiple consecutive operation steps, including the allocation request initiation step, resource status verification step, allocation instruction issuance step, data migration execution step, resource association update step, and allocation result confirmation step.

[0114] The data transmission links are planned as dedicated links from the initiating node (video-on-demand service node) to target node 1 and target node 2 respectively. Link selection is based on the current network load, prioritizing paths with lower load. The resource allocation process is divided into six operation steps: the allocation request initiation step, where the initiating node generates and sends a resource allocation request; the resource status verification step, where the target node checks its current status (available capacity, load, performance); the allocation instruction issuance step, where the cloud storage system's scheduling center generates specific allocation instructions based on the request and verification results; the data migration execution step, where the data migration module performs data migration between target nodes (e.g., migrating some data from node 2 to node 1 to adjust the allocation ratio); the resource association update step, where the network management module updates the association relationship and edge weights between the service node and the target node; and the allocation result confirmation step, where the verification module confirms whether the allocation has achieved its target.

[0115] Step S1474: Based on the logical dependency relationship of the operation steps, the allocation request initiation step is the first step, and the subsequent steps are advanced in the following order: status verification step, instruction issuance step, migration execution step, association update step, and result confirmation step. The execution of each step is based on the completion of the previous step.

[0116] The logical dependencies of the operation steps are as follows: Only after a dispatch request is initiated can the target node perform resource status verification (step two depends on step one); after the status verification is completed and fed back to the scheduling center, the scheduling center can issue a dispatch instruction (step three depends on step two); after the dispatch instruction is issued, the data migration module can execute data migration (step four depends on step three); after the data migration is completed, the resource association can be updated (step five depends on step four); after the association update is completed, the verification module can confirm the dispatch result (step six depends on step five). Therefore, the order is determined as: dispatch request initiation step -> resource status verification step -> dispatch instruction issuance step -> data migration execution step -> resource association update step -> dispatch result confirmation step.

[0117] Step S1475: Determine the specific execution entity for each operation step. The initiating node is responsible for executing the allocation request initiation step, the target node is responsible for executing the resource status verification step, the cloud storage system's scheduling center is responsible for executing the allocation instruction issuance step, the data migration module is responsible for executing the data migration execution step, the network management module is responsible for executing the resource association update step, and the verification module is responsible for executing the allocation result confirmation step.

[0118] The execution entities for each step are clearly defined: the allocation request initiation step is executed by the resource request module of the video-on-demand service node (initiating node); the resource status verification step is executed by the status monitoring agents of object storage resource nodes 1 and 2 (target nodes); the allocation instruction issuance step is executed by the scheduling center server of the cloud storage system; the data migration execution step is executed by the data migration service module in the system; the resource association update step is executed by the topology update component of the network management module; and the allocation result confirmation step is executed by an independent verification service module. Each execution entity communicates and interacts with other entities through preset interfaces.

[0119] Step S1476: Obtain the execution content and standards for each operation step. The allocation request initiation step needs to generate an allocation request containing complete information including business identifier, target resource identifier, allocation type and allocation parameters. The resource status verification step needs to obtain the available capacity, load level and operating status information of the target node. The allocation instruction issuance step needs to generate an allocation instruction that reflects the operation type, execution time and expected results.

[0120] The execution of the allocation request initiation step involves generating a JSON-formatted allocation request, including a business identifier (e.g., "video_on_demand_001"), a target resource identifier (e.g., "obj_store_node1", "obj_store_node2"), an allocation type (e.g., "adjust allocation ratio", "add capacity"), and allocation parameters (e.g., target ratio 70% / 30%, new capacity H). The standard is that the request information is complete and the parameter format is correct. The execution of the resource status verification step involves collecting the target node's available capacity (e.g., node 1's current available capacity J), load level (e.g., CPU utilization K%, IO utilization L%), and operating status (e.g., "normal", "warning"). The standard is that the data collection is accurate and no key indicators are omitted. The execution content of the allocation instruction issuance step is that the scheduling center generates instructions based on the request and verification results, including operation type (such as "increase allocation" or "decrease allocation"), execution time (such as "starting at 20:00:00 on 2024-X-X") and expected results (such as node 1 allocation ratio reaching 70% and rate reaching G). The standard is that the instructions are clear, executable, and consistent with the requested target.

[0121] Step S1477: Establish the connection method between steps. After each step is completed, it sends an execution completion signal and related data to the subsequent steps. After receiving the signal, the subsequent steps start execution. At the same time, establish a feedback process between steps. If the previous step fails, the subsequent steps will pause execution and trigger the exception handling process.

[0122] The connection method employs a message queue mechanism. After each step is completed, an execution completion signal is sent to a designated message queue, containing the step identifier, execution result (success / failure), and related data (such as status data for the verification step and migration progress data for the migration step). Subsequent steps obtain this signal by listening to the message queue, verify the step identifier and execution result in the signal, and then start their own execution. In the feedback process, if the previous step fails (e.g., the status check finds insufficient available capacity for node 1), an exception signal is sent. Subsequent steps (instruction issuance steps) receive this signal, pause execution, and trigger an exception handling process (e.g., notifying the scheduling center to reassess the allocation strategy or alerting the administrator).

[0123] Step S1478: Determine the data transmission method in the data migration execution steps, and select batch transmission, incremental transmission or streaming transmission method according to the data volume, real-time requirements and network conditions.

[0124] The video-on-demand service requires migrating a large amount of data (assuming M bytes), with moderate real-time requirements (to be completed within the predicted timeframe), and the current network condition is good (sufficient remaining bandwidth). Considering all factors, a batch transmission method is chosen. The data to be migrated (from node 2 to node 1) is divided into multiple data blocks and transmitted sequentially. Each data block is verified after transmission to ensure data consistency. If the data volume is small and real-time requirements are high, streaming transmission may be chosen; if most of the data already exists on the target node and only incremental changes need to be synchronized, incremental transmission will be chosen.

[0125] Step S1479: Specify the content of the association information adjustment in the resource association update step, including the creation of association edges between business nodes and target nodes, the setting of edge weights, and the synchronization of topology, so that the business resource co-existence network reflects the association status after the adjustment, forming the execution path.

[0126] In the resource association update step, the association information adjustment includes: for video-on-demand service nodes and object storage node 1, due to the increased allocation ratio, creating or strengthening the association edge between them, and setting the edge weight to a value reflecting the new association strength; for service nodes and object storage node 2, due to the decreased allocation ratio, adjusting the edge weight of their association edge to a lower value; if there are newly added resource nodes (such as node 3 introduced for increased capacity), creating an association edge between the service node and node 3 and setting an initial edge weight. After completing the association edge adjustment, the updated node association relationships are synchronized to the topology database of the service resource co-existence network to ensure that the network can accurately reflect the adjusted service and resource association status. Integrating all elements of the above execution path (nodes, links, steps, sequence, main body, connection method, etc.) forms a complete execution path description.

[0127] Step S148: Set up adaptation assurance related steps, which include real-time resource status monitoring steps, abnormal handling steps in the allocation process, and adaptation effect feedback and adjustment steps, to determine the functional responsibilities, execution flow and triggering conditions of each step.

[0128] The real-time resource status monitoring step is responsible for continuously tracking key metrics (capacity, read / write speed, load, temperature, etc.) of target resource nodes (such as object storage nodes 1 and 2). The execution process involves periodically collecting metric data via a monitoring agent deployed on the nodes and sending it to the monitoring center for analysis and visualization. The trigger condition is that the monitoring agent starts and continues to execute the process without requiring additional triggering. The exception handling step is responsible for handling exceptions that occur during resource allocation (such as data migration or relational updates), such as migration failures or errors in relational edge creation. The execution process includes exception detection (through feedback signals between steps), exception classification (temporary / permanent failures), and exception recovery (retrying the operation / switching to a backup solution). The trigger condition is receiving an exception signal from a previous step or detecting a metric exceeding a threshold. The function and responsibility of the adaptation effect feedback and adjustment step is to evaluate the adaptation effect after resource allocation (such as whether business performance has been improved and whether resource utilization has been optimized). The execution process is to collect relevant business and resource data within a preset time window after the allocation is completed (such as one hour after the allocation), compare and analyze it with the allocation target, and start the fine-tuning process if the target is not achieved. The triggering condition is that the allocation result confirmation step is completed and the adaptation effect evaluation indicators are not met.

[0129] Step S149: Integrate and link the resource allocation adjustment strategy, execution path and adaptation guarantee steps to form a resource elastic allocation plan for a single business and a single future forecast period.

[0130] The resource allocation adjustment strategy (adjusting the allocation ratio of node 1 and node 2, adding capacity to node 1), execution path (initiating node, target node, transmission link, operation steps and sequence), and adaptation and assurance steps (real-time monitoring, anomaly handling, effect feedback) for the video-on-demand service during the predicted period of 20:00-21:00 are integrated and linked. Specifically, the solution clearly defines the resource allocation adjustment strategy as the core objective, the execution path as the specific operational process for implementing the strategy, and the adaptation and assurance steps as support measures to ensure smooth execution and the achievement of the strategy objectives. For example, the solution stipulates that resource allocation is performed according to the operation steps of the execution path to achieve the adjustment objectives in the resource allocation adjustment strategy. During execution, the real-time monitoring step tracks the allocation progress and node status. If an anomaly occurs, the anomaly handling step is triggered. After allocation is completed, the adaptation effect feedback step evaluates whether the strategy objectives have been achieved, forming a complete closed-loop solution including objectives, operations, and assurance—that is, a resource elastic allocation solution for a single service and a single predicted period.

[0131] Step S1410: Classify and summarize all resource elastic allocation schemes according to business type and future forecast period, eliminate duplicate schemes, optimize conflicting schemes, and form a set of resource elastic allocation schemes that cover all future forecast periods and are applicable to different business scenarios.

[0132] The resource elastic allocation schemes for all services (video-on-demand, online document collaboration, big data analytics, etc.) across all future forecast periods (e.g., every hour within a 24-hour period) are categorized. First, they are grouped by service type; for example, all time-period schemes for video-on-demand are grouped into one category, and those for online document collaboration into another. Then, within each service type group, the schemes are arranged chronologically according to the future forecast period. Duplicate schemes are eliminated; for example, if two video-on-demand schemes for different time periods but with identical content are used, the most recent one is retained. Conflicting schemes are optimized; if video-on-demand and online document collaboration have allocation needs for the same object storage resource node during the same time period, and these needs conflict (e.g., both requesting an increased allocation ratio, causing the total demand to exceed resource capacity), optimization is performed based on service adaptation priority (video-on-demand has higher priority). The higher-priority service needs are prioritized, and the allocation schemes for lower-priority services are adjusted (e.g., allocating other available resource nodes to online document collaboration). The final set of resource elastic allocation schemes includes optimized solutions for different service scenarios (e.g., peak video-on-demand scenarios, daily online document scenarios) covering all future forecast periods.

[0133] Step S150: Generate a real-time allocation instruction stream based on the resource elastic allocation scheme set, drive the storage resources in the cloud storage system to complete the dynamic reorganization across nodes and prediction periods according to the instruction stream, and form a real-time adaptation between business demand and resource supply.

[0134] The resource elastic allocation scheme set provides specific strategies and paths for resource allocation. Generating a real-time allocation instruction stream transforms the above schemes into executable instructions. These instructions need to be sent to the scheduling and execution module of the cloud storage system in chronological order and priority to drive the dynamic reorganization of storage resources between different nodes (across nodes) and between different forecast periods (across forecast periods), such as data migration, capacity adjustment, and performance optimization. Ultimately, this enables real-time adaptation between business needs and resource supply, ensuring that the business's adaptation needs can be met in each forecast period.

[0135] Step S151: Extract the core content of resource allocation adjustment strategy, execution path and adaptation guarantee related steps of each resource elastic allocation scheme in the resource elastic allocation scheme set, sort the schemes according to business type and execution time, and determine the execution priority of the resource elastic allocation scheme.

[0136] From the set of resource elastic allocation solutions, the core content of each solution is extracted: the adjustment goals and key parameters in the resource allocation adjustment strategy (e.g., 70% allocation ratio for node 1), the operation steps and time schedule in the execution path (e.g., execution starts at 20:00), and the triggering conditions in the adaptation and assurance related steps (e.g., the threshold for starting exception handling). When sorting by business type, all solutions for the same business (e.g., video on demand) are grouped together; within the business group, they are sorted in ascending order by execution time (the start time of the future predicted period). The determination of execution priority comprehensively considers the adaptation priority of the business (e.g., video on demand is a high priority) and the urgency of the solution (e.g., expansion solutions that need to be executed immediately have higher priority than optimization solutions that can be postponed), assigning each solution a priority value (e.g., 1-5, with 1 being the highest). For example, the solution for video on demand from 20:00 to 21:00 has a priority of 1, and the solution for online document collaboration during the same period has a priority of 3.

[0137] Step S152: For each resource elastic allocation scheme, determine the core parameters of the allocation instruction according to the resource allocation adjustment strategy. The core parameters include business identifier, target resource identifier, allocation type, adjustment range, execution time window, and key information on expected target status.

[0138] Taking the video-on-demand service from 20:00 to 21:00 as an example, the core parameters are determined based on its resource allocation adjustment strategy. The service identifier is "video_on_demand_001"; the target resource identifiers are "obj_store_node1" and "obj_store_node2"; the allocation type is "allocation ratio adjustment" and "capacity addition"; the adjustment range is that the allocation ratio of node 1 increases from 60% to 70% (adjustment range 10%), and the allocation ratio of node 2 decreases from 40% to 30% (adjustment range -10%), with an additional capacity H for node 1; the execution time window is "2024-X-X 20:00:00-20:30:00" (reserve execution time within the predicted period); the expected target state is that the allocation ratio of node 1 is 70%, and the read / write rate is ≥G, the allocation ratio of node 2 is 30%, and the read / write rate is ≥I, with the additional capacity H of node 1 available. The above core parameters uniquely identify the target and scope of the allocation instruction.

[0139] Step S153: Determine the transmission path and execution subject of the allocation instruction based on the execution path, determine the node identifiers corresponding to the initiating end, receiving end and relay end of the allocation instruction, and specify the transmission protocol and format requirements of the allocation instruction.

[0140] The execution path explicitly identifies the initiating node for resource allocation as the video-on-demand service node (identified by "service_node_vod"), and the target nodes as object storage node 1 (identified by "storage_node_obj1") and node 2 (identified by "storage_node_obj2"), with no intermediate nodes. Therefore, the transmission path for the allocation command is directly from the initiating end "service_node_vod" to the receiving ends "storage_node_obj1" and "storage_node_obj2". The transmission protocol uses TCP to ensure reliable transmission of instructions; the format requirement is JSON, including an instruction header (instruction identifier, sending time, priority) and an instruction body (core parameters, operation type), for example: {"Instruction Identifier":"cmd_001","Sending Time":"2024-X-X19:50:00","Priority":1,"Service Identifier":"video_on_demand_001","Target Resource Identifier":["storage_node_obj1","storage_node_obj2"],"Allocation Type":"Allocation Ratio Adjustment",...}.

[0141] Step S154: In conjunction with the monitoring requirements in the adaptation and assurance related steps, add status feedback parameters to the allocation instruction to determine the resource status indicators, feedback frequency and feedback method that need to be fed back during the instruction execution process.

[0142] Real-time monitoring of resource status in the adaptation and assurance steps requires tracking metrics such as the target node's capacity, read / write rate, and load. Therefore, a status feedback parameter is added to the allocation command: the resource status metrics to be fed back include the target node's real-time available capacity, current read / write rate, CPU and IO load; the feedback frequency is once every 5 minutes; and the feedback method is for the target node to send status data to the specified monitoring center API via an HTTP POST request. For example, the status feedback parameter in the command is: "Status Feedback":{"Metrics":["Available Capacity","Read / Write Rate","CPU Load","IO Load"],"Frequency":"5 minutes / time","Method":"HTTP POST","Receiving Address":"http: / / monitor-center / api / status"}.

[0143] Step S155: Integrate the core parameters, transmission path, execution subject and status feedback parameter information to generate corresponding resource allocation instructions. Each resource allocation instruction contains a unique instruction identifier to distinguish and track the instruction execution process.

[0144] The core parameters (service identifier, target resource identifier, etc.), transmission path (initiator, receiver), execution entity (target node performs the allocation operation), and status feedback parameters (indicators, frequency, method) of the video-on-demand service allocation scheme are integrated to generate resource allocation instructions. Each instruction is assigned a unique instruction identifier (e.g., "cmd_vod_2024XXXX2000"), which includes service type, date, and time information for easy identification and tracking. The overall structure of the instruction includes the instruction identifier, basic information (sender, receiver), core parameters, execution information (transmission protocol, execution entity), status feedback requirements, and verification information (to ensure instruction integrity), forming a complete and executable resource allocation instruction.

[0145] Step S156: Arrange all resource allocation instructions according to execution time order and execution priority to form an initial instruction sequence, and perform conflict detection on the initial instruction sequence to identify instruction combinations that have resource occupation conflicts, execution time conflicts, or transmission link conflicts. Conflict detection is based on the target resource identifier, execution time window, and transmission path information in the initial instruction sequence.

[0146] All resource allocation instructions for all services are sorted in ascending order of their execution time window start time. Instructions with the same execution time window are sorted in descending order of execution priority to form an initial instruction sequence. During conflict detection, resource occupancy conflict identification checks whether the target resource identifiers of different instructions are the same and whether there is an adjustment type conflict (e.g., two instructions simultaneously requesting an increase in the allocation ratio of the same resource node); execution time conflict identification compares whether the execution time windows of the instructions overlap and cannot be executed in parallel (e.g., two time-consuming operation time windows for the same node overlap); and transmission link conflict identification checks whether the transmission paths of the instructions use the same physical link and whether the total bandwidth requirement exceeds the link capacity. For example, if the target resource identifier of the video-on-demand service instruction and the online document collaboration service instruction are both "storage_node_obj1", and their execution time windows overlap, and both request an increase in the allocation ratio, then it is determined to be a resource occupancy conflict and an execution time conflict.

[0147] Step S157: Adjust and optimize conflicting instruction combinations, divide the optimized instruction sequence into multiple consecutive instruction batches according to time slices, the duration of each time slice is determined according to the instruction execution cycle and system processing capacity, and each instruction batch contains all resource allocation instructions that need to be executed within the time slice.

[0148] The optimization of conflicting instruction combinations adopts a priority-based preemption principle. High-priority instructions (such as video-on-demand) retain their original execution time window, while low-priority instructions (such as online document collaboration) have their execution time window adjusted to after or before the conflict period, or the target resource node is adjusted. The optimized instruction sequence is divided into batches by time slices. The duration of a time slice is determined to be 1 hour based on the average execution cycle of the instructions (e.g., 30 minutes) and the system's ability to process instructions simultaneously (e.g., N instructions at a time). Each time slice corresponds to an instruction batch, containing all resource allocation instructions that need to be executed within that hour. For example, the instruction batch for the 20:00-21:00 time slice includes allocation instructions for video-on-demand services during that time period, as well as non-conflicting instructions from other services during that time period.

[0149] Step S158: Issue each batch of instructions sequentially according to time to form a real-time allocation instruction stream. The real-time allocation instruction stream is continuously transmitted to the scheduling and execution module of the cloud storage system in batches. The scheduling and execution module receives the real-time allocation instruction stream and drives the corresponding storage resources to perform dynamic reorganization operations across nodes and prediction periods according to the instruction requirements. The dynamic reorganization operations include resource allocation operations to allocate new storage resource space to business nodes, migration operations to transfer business data from one resource node to another, expansion operations to increase the available capacity of resource nodes, and release operations to reclaim the resource space occupied by business nodes.

[0150] For example, in step S1581: the scheduling execution module receives the batch of instructions in the real-time allocation instruction stream, parses each resource allocation instruction one by one according to the instruction identifier order, and extracts the core parameters of the resource allocation instruction, such as the business identifier, target resource identifier, allocation type, adjustment range, execution time window, and expected target status.

[0151] The scheduling and execution module receives batches of instructions by listening to a preset message queue and parses them one by one according to the order of the instruction identifiers (e.g., by generation time). During parsing, the service identifier is extracted from the instruction to determine the service to which it belongs, the target resource identifier is used to locate the specific storage resource node, the allocation type (e.g., allocation ratio adjustment, capacity increase) determines the operation type, the adjustment range specifies the specific scale of the operation, the execution time window specifies the time range of the operation, and the expected target state sets the indicator requirements after the operation is completed. For example, parsing the instruction for video-on-demand service, the target resource identifier is extracted as "storage_node_obj1", the allocation type is "capacity increase", the adjustment range is "H", the execution time window is 20:00-20:30, and the expected target state is that the newly added capacity "H" is available.

[0152] Step S1582: Locate the corresponding storage resource node based on the target resource identifier, and obtain the current operating status of the storage resource node through the resource status query interface of the cloud storage system, including resource utilization, load level, data storage distribution and network connection status information.

[0153] The scheduling and execution module uses the target resource identifier "storage_node_obj1" to call the cloud storage system's resource status query API (e.g., "GET / api / resources / {node_id} / status") to obtain the current running status of the node. The returned status information includes resource utilization (e.g., percentage of used capacity to total capacity), load level (CPU utilization, memory utilization, IOPS), data storage distribution (storage location of each data block, number of replicas), and network connection status (uplink / downlink bandwidth, number of connections, packet loss rate). This information is used to determine whether the target node has the conditions to execute the allocation command. If the resource utilization is too high, the new capacity operation may not be possible.

[0154] Step S1583: Determine the specific type of dynamic reorganization operation based on the allocation type. If the allocation type is resource allocation, execute the resource allocation operation; if it is resource migration, execute the migration operation; if it is resource expansion, execute the expansion operation; if it is resource release, execute the release operation.

[0155] The dynamic reorganization operation type is determined based on the allocation type field in the allocation instruction. If the allocation type is "resource allocation" (e.g., allocating initial storage resources for a new service), a resource allocation operation is performed; if it is "resource migration" (e.g., adjusting the allocation ratio of different nodes), a migration operation is performed; if it is "resource expansion" (e.g., adding node capacity), an expansion operation is performed; if it is "releasing resources" (e.g., reclaiming idle resources), a release operation is performed. For example, if the allocation type of the video-on-demand service instruction is "capacity increase," an expansion operation is performed; if the allocation type is "allocation ratio adjustment," a migration operation is performed (migrating data between nodes to change the ratio).

[0156] Step S1584: For resource allocation operations, based on the adjustment range and expected target state, allocate the corresponding resource space in the target resource node, and configure the access permissions, performance parameters and storage policies of the resource space.

[0157] During resource allocation, resource space is divided within the target resource node (e.g., object storage node 1) based on the adjustment range (e.g., allocating N bytes of capacity) and the expected target state (e.g., access latency ≤ P milliseconds). Specific operations include: reserving N bytes of contiguous storage space within the node's storage pool and marking it as dedicated to this service; configuring access permissions, granting read and write permissions to the account corresponding to the service identifier, while denying permissions to other accounts; adjusting performance parameters, such as enabling a caching mechanism for this space to reduce access latency and setting the pre-read size to Q bytes; and formulating storage strategies, such as using triple-replica storage to ensure data reliability, with a backup cycle of once daily. After configuration, the unique identifier of the resource space (e.g., "space_vod_001") and related configuration parameters are recorded.

[0158] Step S1585: Establish associations between the allocated resource spaces and the corresponding business nodes, create corresponding association edges in the business resource symbiotic network, set the dynamic edge weights of the association edges, and synchronously update the resource allocation information in the node attributes.

[0159] In the business resource co-existence network, new association edges are created between video-on-demand service nodes and target resource nodes belonging to newly allocated resource spaces. Dynamic edge weights are calculated based on the service node's adaptation priority, the resource node's supply capacity, and the degree of compatibility between the two, and these weights are set in the association edge attributes. Simultaneously, the resource allocation information in the service node attributes is updated, recording the newly allocated resource space identifier, capacity, and access path; the resource allocation information in the resource node attributes is also updated, recording the space's occupant (service identifier), occupied capacity, and allocation time, ensuring that the business resource co-existence network reflects the new resource association status in real time.

[0160] Step S1586: For the migration action, based on the adjustment range and the expected target state, determine the amount of data to be migrated and the migration direction, select the appropriate migration method, and establish a migration channel between the initiating node and the target node.

[0161] During the migration, the video-on-demand service allocation ratio between Node 1 and Node 2 is adjusted from 60%:40% to 70%:30%. Based on this, the amount of data to be migrated from Node 2 to Node 1 (e.g., 0 bytes) is calculated, and the migration direction is from Node 2 to Node 1. A batch transmission method is selected based on the data volume (0 bytes), real-time requirements (to be completed within 30 minutes), and network conditions (sufficient bandwidth). When establishing the migration channel, a dedicated TCP connection is created between Node 2 and Node 1. The channel's bandwidth limit (not exceeding 50% of the total bandwidth to avoid affecting other services), timeout period (300 seconds), and data verification mechanism (MD5 hash check) are configured to ensure a stable and reliable migration process.

[0162] Step S1587: Transmit data from the initiating node to the target node through the migration channel. After the data transmission is completed, update the association between the business node and the resource node, delete the association edge with the original resource node in the business resource symbiotic network, create the association edge with the target resource node, and synchronously update the relevant node attributes and edge weights.

[0163] During data transmission, Node 2 divides the O-byte data to be migrated into blocks of size R and sends them to Node 1 block by block through the migration channel. Each data block is verified after transmission. Upon successful verification, Node 1 confirms receipt and sends feedback to Node 2, which then sends the next block. After all data transmissions are complete, the association between service nodes and resource nodes is updated: in the service-resource co-existence network, the original association edges between the video-on-demand service node and Node 2 are deleted (or their edge weights are adjusted to reflect the new allocation ratio), and association edges with Node 1 are created or strengthened, with new dynamic edge weights set (based on the adjusted allocation ratio and node performance). The resource allocation ratio information in the service node attributes and the used capacity, load, and other information in the resource node attributes are updated synchronously to ensure that the network topology is consistent with the actual resource allocation.

[0164] Step S1588: For expansion operations, based on the adjustment range and expected target status, expand the storage capacity or performance limit of the target resource node and adjust the supply spectrum information of the resource node.

[0165] During the expansion operation, the adjustment amount is to add H capacity to the target resource node 1 for the video-on-demand service. The expected target state is that the total capacity of node 1 increases by H and the read / write speed increases to G. The method for expanding storage capacity is to add new physical disks (such as adding P blocks of solid-state drives with a capacity of Q) to the storage cluster where node 1 resides, adding them to the storage pool of node 1, so that the total capacity increases by P*Q (should be ≥ H). If it is necessary to improve the performance ceiling, the network interface card of node 1 may be upgraded simultaneously (e.g., from gigabit to 10 gigabit) to improve read / write speed. The resource node supply hierarchy information is adjusted, updating parameters such as the total capacity, available capacity, and maximum read / write speed of node 1 to reflect the supply capacity after the expansion.

[0166] Step S1589: Update the attribute information of the resource node in the business resource symbiotic network, adjust the dynamic edge weights of the associated edges with the associated business nodes, and reflect the changes in the adaptation value after expansion.

[0167] After expansion, the attribute information (capacity, read / write speed, etc.) of target resource node 1 is updated in the business resource co-existence network. Due to the increased capacity and improved performance of node 1, its compatibility with video-on-demand service nodes improves. Therefore, the dynamic edge weights of the associated edges between them are adjusted to increase their values ​​(reflecting the improved compatibility value). Simultaneously, the associated edges between node 1 and other related business nodes are checked. If expansion also improves the compatibility with other services, the edge weights are adjusted accordingly; if the matching degree between the demand of other services and the supply characteristics of node 1 decreases, the edge weights may be reduced to ensure that the edge weights accurately reflect the changes in compatibility value after expansion.

[0168] Step S15810: For the release operation, based on the adjustment range and the expected target state, recover part or all of the resources occupied by the business node, release the corresponding resource space, and update the resource occupancy status and supply spectrum information of the target resource node.

[0169] During the release operation, the adjustment involves reclaiming a portion of the video-on-demand service's resources on node 2 (reducing the allocation ratio from 40% to 30%), with the expected target state being the release of corresponding resource space on node 2. Specific operations include: identifying the resource space (such as data block sets) occupied by the service node on node 2, confirming that this data has been migrated to other nodes (such as node 1) or is no longer needed; deleting or marking these data blocks as reclaimable, initiating the storage system's garbage collection mechanism to release the corresponding physical storage space; updating the resource occupancy status of the target resource node 2, reducing used capacity and increasing available capacity; adjusting its supply hierarchy information, updating the total available capacity and resource allocation to reflect the supply capacity after resource release.

[0170] Step S15811: Delete or adjust the corresponding associated edges in the business resource co-existence network, and synchronously update the attribute information of business nodes and resource nodes so that the network topology reflects the associated status after resource release.

[0171] After resource release, in the business resource co-existence network, the dynamic edge weights of the association edges between the video-on-demand service node and resource node 2 are adjusted according to the new allocation ratio (30%), making them lower than the edge weights before expansion. If the proportion of released resources is large (e.g., decreasing from 50% to 10%), the association strength level may decrease. The resource allocation list in the service node attributes is updated synchronously, removing or reducing resource dependency descriptions for node 2; the attribute information of resource node 2 is updated, recording the available capacity, load, and other statuses after resource release. Through these updates, the topology of the business resource co-existence network accurately reflects the changes in the association status between services and resources after resource release.

[0172] Step S15812: After all dynamic reorganization operations are completed, verify whether the state of the target resource node has reached the expected target state, and feed back the operation execution results and the final state of the target resource node to the instruction tracing module for recording and monitoring instruction execution.

[0173] After a dynamic reorganization operation (such as migration or expansion) is completed, the verification module collects the actual status data of the target resource nodes and compares it with the expected target status in the instruction. For example, for the expansion operation of video-on-demand services, it verifies whether the total capacity of node 1 reaches the expected value (original capacity + H) and whether the read / write rate is ≥ G; for the allocation ratio adjustment, it verifies whether the actual allocation ratio of node 1 and node 2 is close to 70%:30%. If all indicators meet the standards, the operation is considered successful; if some indicators fail to meet the standards, the operation is considered partially successful or failed, and the reasons are analyzed (such as expansion hardware failure or migration data loss). The operation execution result (success / failure / partial success), the final status data of the target resource nodes (capacity, rate, load, etc.), and the operation process log are sent to the instruction tracing module through a message queue. The instruction tracing module stores the above information in the instruction execution database for subsequent auditing, analysis, and troubleshooting.

[0174] Step S159: During the execution of the instruction, resource status data is collected in real time according to the requirements of the status feedback parameters and fed back to the business resource co-existence network update module for synchronously updating the network topology.

[0175] During the execution of resource allocation instructions, target resource nodes (such as node 1 and node 2) collect resource status indicators (available capacity, read / write rate, etc.) every 5 minutes according to the status feedback parameters required in the instructions, and send them to the monitoring center via HTTP POST requests. The business resource symbiotic network update module subscribes to the status data of the monitoring center. After receiving this real-time feedback data, it dynamically updates the attribute information of the corresponding nodes in the network topology (such as current available capacity, real-time read / write rate) and the dynamic edge weights of associated edges (if the status change affects the adaptation value). For example, if the available capacity of node 1 gradually increases during the expansion operation, the update module reflects this change in the node attributes in real time, enabling the business resource symbiotic network to reflect the progress and current status of resource allocation in real time.

[0176] Step S1510: Continuously repeat the above steps of generating, issuing, executing and providing feedback instructions to achieve real-time adaptation between business needs and resource supply, so that cloud storage resources always maintain a dynamic match with the sequence of business resource adaptation needs.

[0177] The resource allocation instructions are repeatedly executed at fixed intervals (e.g., hourly), including generation (based on the latest business resource adaptation demand sequence and resource supply spectrum), distribution (sent in batches to the scheduling execution module), execution (driven by the scheduling execution module to perform dynamic reorganization operations), and feedback (updating the business resource symbiotic network with status data). Within each cycle, resource allocation strategies and instructions are dynamically adjusted based on the latest changes in business demand and the real-time status of resource supply to ensure that cloud storage resources can continuously meet the adaptation needs of the business in different forecast periods. For example, if the demand for video-on-demand services fluctuates in subsequent periods, the business resource adaptation demand sequence will be updated, thereby driving the generation and execution of new elastic resource allocation schemes, causing resource supply to be adjusted accordingly. This results in continuous, real-time adaptation between business demand and resource supply, ultimately achieving adaptive allocation of cloud storage resources.

[0178] In one exemplary embodiment, a deep learning-based adaptive allocation system for cloud storage resources is provided. This system can be a terminal, server, etc., and its internal structure diagram can be as follows: Figure 2As shown, this deep learning-based adaptive cloud storage resource allocation system includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The input / output interface is used for exchanging information between the processor and external devices. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, near-field communication, or other technologies. When the computer program is executed by the processor, it implements a deep learning-based adaptive cloud storage resource allocation method. The display unit is used to form a visually visible image and can be a display screen, projection device, or virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device can be a touch layer covering the display screen, or a button, trackball, or touchpad set on the shell of a cloud storage resource adaptive allocation system based on deep learning, or an external keyboard, touchpad, or mouse, etc.

[0179] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

Claims

1. A cloud storage resource adaptive allocation method based on deep learning, characterized in that, The method includes: A business resource symbiotic network is constructed, which takes the resource consumption trajectory of various businesses and the resource supply spectrum of the cloud storage system as core elements. Through node association, edge weight mapping and topology evolution process, a network system is formed to reflect the dynamic balance between business demand and resource supply. A deep learning model is invoked to extract multi-dimensional evolutionary factors from the business resource symbiotic network, generating network evolutionary status features. These network evolutionary status features include core information such as the fluctuation trend of business node demand, changes in resource node supply capacity, and the evolution of the symbiotic relationship strength between nodes. Based on the network evolution characteristics, a service resource adaptation requirement sequence is derived. This sequence is transformed from the interaction patterns of each node and the edge weight change characteristics in the network evolution characteristics. It is used to determine the adaptation direction and adaptation mode of various services to storage resources in different prediction periods. By using a deep learning model, the sequence of business resource adaptation needs is dynamically matched and deduced with the resource supply spectrum of the cloud storage system, generating a set of resource elastic allocation schemes. The set of resource elastic allocation schemes includes resource allocation adjustment strategies, execution paths and adaptation guarantee-related steps for different business scenarios. Based on the aforementioned resource elastic allocation scheme set, a real-time allocation instruction stream is generated, driving the storage resources in the cloud storage system to complete dynamic reorganization across nodes and prediction periods according to the instruction stream, thereby forming a real-time adaptation between business needs and resource supply.

2. The cloud storage resource adaptive allocation method based on deep learning according to claim 1, characterized in that, The construction of the business resource symbiotic network includes: Collect resource consumption trajectories of various services throughout the entire operation cycle. The resource consumption trajectory includes the call sequence, usage duration, access frequency and continuous change information of different types of storage resources by the services, and fully records the entire interaction process between services and storage resources. The resource supply spectrum of the cloud storage system is collected. The resource supply spectrum covers the core information of all storage resources, including physical attributes, logical affiliation, performance parameters, deployment location and available prediction period, reflecting the supply capacity boundary and service scope of each storage resource. Create a dedicated business node for each business and embed the core feature information from the resource consumption trajectory into the business node. The core feature information includes the business's resource demand type preference, demand fluctuation pattern, peak demand prediction period and service level requirements, so that the business node has the ability to uniquely identify the characteristics of the business demand. Create a dedicated resource node for each type of storage resource, and embed the core attribute information of the resource supply spectrum into the resource node. The core attribute information includes key content such as the storage resource's capacity, read and write speed, stability performance, expansion potential and maintenance characteristics, so that the resource node has the ability to uniquely identify the resource supply characteristics. The symbiotic association strength between each business node and each resource node is calculated. The symbiotic association strength is based on the degree of fit between the resource demand characteristics of the business node and the supply characteristics of the resource node, the frequency of historical interactions, and the adaptation effect feedback factors. It is derived by converting each factor into a dimensionless standard score. The association edges between business nodes and resource nodes are established based on the strength of the symbiotic association. The existence of the association edge is directly determined by whether the strength of the symbiotic association reaches a preset symbiotic threshold. If the symbiotic threshold is reached, the association edge is established; otherwise, the association edge is not established, reflecting the effective adaptation relationship between business and resources. Each associated edge is assigned a dynamic edge weight, which is determined by the symbiotic association strength, the service level requirements of the business node, and the performance parameters of the resource node. The dynamic edge weight is determined by normalizing each parameter to a comparable scale and is updated in real time as the business resource consumption trajectory changes and the resource supply spectrum is adjusted, reflecting the current adaptation value of the associated edge. An initial symbiotic network topology is constructed based on all business nodes, resource nodes, and associated edges. The initial topology fully presents the initial association state of business and resources, including node attributes, edge weight information, and direct association relationships between nodes. The initial symbiotic network topology is optimized by time-series optimization, introducing a time dimension factor and incorporating historical change data of business resource consumption trajectory and resource supply spectrum into the topology, so that the topology reflects the evolution of nodes and associated edges over time. By iteratively updating the process, we continuously synchronize the real-time changes in the business resource consumption trajectory and the dynamic adjustment of the resource supply spectrum, constantly correct the dynamic edge weights of business node attributes, resource node attributes and related edges, remove invalid related edges and add valid related edges, forming a business resource symbiotic network that reflects the dynamic balance between business demand and resource supply.

3. The cloud storage resource adaptive allocation method based on deep learning according to claim 1, characterized in that, The process involves using a deep learning model to extract multi-dimensional evolutionary factors from the business resource symbiotic network, generating network evolutionary status features, including: The current topology data of the business resource symbiotic network is converted into structured input data that can be recognized by the deep learning model. The structured input data includes a node feature matrix, an edge weight feature matrix, and a topological correlation matrix. The node feature matrix stores the attribute information of all business nodes and resource nodes, the edge weight feature matrix stores the dynamic edge weights of all associated edges, and the topological correlation matrix stores the association relationships between nodes. The structured input data is input into the feature extraction layer of the deep learning model. The feature extraction layer uses multi-layer graph convolution operation to jointly process the node feature matrix, edge weight feature matrix and topological association matrix. Through neighborhood node feature aggregation and edge weight weighting, local evolution features at the node level are extracted. The local evolution features reflect the changing trend of a single node and its directly related nodes. By using the global feature capture layer of the deep learning model, local evolutionary features are globally integrated across nodes and related edges. Attention mechanisms are used to focus on core nodes and important related edges that play a key role in network evolution, and global evolutionary features at the network level are extracted. These global evolutionary features reflect the overall change pattern of the entire business resource symbiotic network. Local and global evolutionary features are input into the feature fusion layer of the deep learning model. The two features are deeply integrated by cross-dimensional splicing and element-level weighted fusion to generate fused evolutionary features that have both local details and global perspective. The fusion evolution characteristics are decomposed in multiple dimensions, and three major categories of core evolution factors are decomposed: business node evolution-related characteristics, resource node evolution-related characteristics, and associated edge evolution-related characteristics. Each category of core evolution factors contains multiple specific evolution indicators, covering the key dimensions of network evolution. Based on the evolutionary characteristics of business nodes, we analyze the resource consumption trajectory change patterns of each business node, extract the demand fluctuation trend factor of business nodes, and include the direction of change of demand type, the fluctuation range of demand scale, the frequency of demand peak occurrence, and the specific evolutionary information of demand duration adjustment, thus depicting the demand evolution trend of business nodes. Based on the evolutionary characteristics of resource nodes, a detailed analysis of the supply spectrum adjustment of each resource node is conducted, and the factors of change in the supply capacity of resource nodes are extracted. The factors of change in the supply capacity of resource nodes include specific evolutionary information on changes in capacity scale, fluctuations in read and write rates, stability adjustments, and changes in expansion potential, presenting the supply evolution trend of resource nodes. Based on the evolutionary characteristics of associated edges, the dynamic edge weight change trajectory of each associated edge is analyzed, and the symbiotic association strength evolution factor between nodes is extracted. The symbiotic association strength evolution factor between nodes includes specific evolutionary information such as the magnitude of association strength change, the rate of change, the time of peak occurrence, and the duration of stable stability, thus showing the adaptive evolution trend of associated edges. The factors of business node demand fluctuation trend, resource node supply capacity change, and inter-node symbiotic relationship strength evolution are structured and integrated, and classified and arranged according to multiple dimensions such as time dimension, node type, and relationship strength level to form a multi-dimensional evolution factor set. The basic structure for constructing network evolutionary status features is based on the set of evolutionary factors. The basic structure includes feature titles, dimension classification labels, factor display areas, and time series axis core components. The specific evolutionary information from the set of evolutionary factors is populated into the basic structure, and a unique visualization method is generated for each evolutionary factor. The evolutionary trend is displayed intuitively through curves, bar charts, and heat maps, generating network evolutionary status characteristics that include core information such as the fluctuation trend of business node demand, changes in the supply capacity of resource nodes, and the evolution of the symbiotic relationship strength between nodes.

4. The cloud storage resource adaptive allocation method based on deep learning according to claim 3, characterized in that, The analysis of the resource consumption trajectory changes of each business node, based on the evolutionary characteristics of business nodes, and the extraction of demand fluctuation trend factors for business nodes, includes: Extract the complete resource consumption trajectory corresponding to each business node, and divide the resource consumption trajectory into multiple consecutive analysis and prediction periods in chronological order, with each analysis and prediction period having a consistent duration. For each analysis and prediction period, extract the resource demand types of business nodes within that analysis and prediction period, determine the types of storage resources called by the business within that analysis and prediction period and the proportion of each type of resource called, and form the demand type distribution characteristics within that analysis and prediction period. By comparing the distribution characteristics of demand types in adjacent analysis and forecast periods, we can identify the addition, reduction, and proportion adjustment of demand types and determine the direction of demand type changes at business nodes. For each type of resource, calculate the demand scale of the business node for that type of resource within each analysis and forecast period. The demand scale is derived from the call volume of that type of resource. For each type of resource, compare the demand scale with that of adjacent analysis and forecast periods, calculate the demand scale difference and the percentage of the difference, determine the fluctuation range of the demand scale of that type of resource, and quantify the degree of change in the business demand for that type of resource. For each type of resource, identify the time point when the demand for that type of resource reaches or exceeds the preset peak threshold within each analysis and prediction period, count the number of peak time points within each analysis and prediction period, and obtain the frequency of peak demand for that type of resource, reflecting the peak density of business demand for that type of resource. For each type of resource, calculate the length of time during which the demand for that type of resource remains within the high demand threshold range in each analysis and forecast period to obtain the demand duration for that type of resource. Compare the demand duration in adjacent analysis and forecast periods to determine the adjustment of the demand duration for that type of resource. By integrating information on the direction of demand type changes, the fluctuation range of demand scale for each type of resource, the frequency of peak demand for each type of resource, and the adjustment of demand duration for each type of resource, the basic structure of the demand fluctuation trend factor for business nodes is constructed. Each component is assigned a corresponding time series label, and the analysis and prediction period corresponding to each component information is determined. This allows the demand fluctuation trend factor for business nodes to reflect the evolutionary pattern in the time dimension. In addition, the information of each component is standardized and described to form the demand fluctuation trend factor for business nodes.

5. The cloud storage resource adaptive allocation method based on deep learning according to claim 1, characterized in that, The process of deriving a service resource adaptation requirement sequence based on the network evolution characteristics, wherein the service resource adaptation requirement sequence is derived from the interaction patterns of each node and the edge weight change characteristics in the network evolution characteristics, is used to determine the adaptation direction and adaptation mode of various services for storage resources in different prediction periods, including: Extract core evolutionary information from the network evolutionary situation characteristics, including the fluctuation trend of business node demand, the change of resource node supply capacity, and the evolution of the symbiotic relationship strength between nodes, and establish a corresponding relationship between evolutionary information and business resource adaptation requirements; The network evolution characteristics are divided into multiple future prediction periods in chronological order. The duration of each future prediction period is determined comprehensively based on the business demand change cycle and resource supply adjustment cycle. For each future forecast period, extract the demand fluctuation trend information of all business nodes within that future forecast period to determine the direction of demand type change, the magnitude of demand fluctuation, the frequency of demand peaks, and the adjustment of demand duration for each business node within that future forecast period. Extract information on the supply capacity changes of all resource nodes within the future forecast period to determine the capacity size changes, read / write rate fluctuations, stability adjustments, and expansion potential changes of each resource node within the future forecast period. Analyze the evolution of the symbiotic relationship strength between business nodes and resource nodes during the future forecast period to determine the dynamic edge weight change trend, relationship strength level, and adaptation value change of the relationship edges between nodes; Based on the fluctuation trend of business node demand and the changes in resource node supply capacity, combined with the evolution information of the symbiotic relationship strength between nodes, the adaptation priority of each business node to storage resources in the future forecast period is determined. The adaptation priority is jointly determined by the urgency of business demand, the sufficiency of resource supply, and the degree of adaptation and fit between the two. Based on the adaptation priority and the fluctuation trend of business node demand, the adaptation direction of each business node in the future forecast period is determined. The adaptation direction is used to reflect the types of storage resources that the business node needs to add, adjust or maintain, and the corresponding performance requirements of the resources. By combining the changes in the supply capacity of resource nodes and the evolution of the symbiotic relationship strength between nodes, the adaptation mode of each business node in the future prediction period is determined. The adaptation mode is used to reflect the association method between business nodes and resource nodes, the resource call method, and the adaptation adjustment frequency. The adaptation priority, adaptation direction and adaptation mode of various business nodes in each future forecast period are integrated to form the business resource adaptation requirement unit corresponding to that future forecast period. All business resource adaptation requirement units are arranged in chronological order according to the future forecast period. Each unit is marked with a corresponding time range, forming a business resource adaptation requirement sequence that can cover multiple future forecast periods and reflect the adaptation requirements of various businesses in different future forecast periods.

6. The cloud storage resource adaptive allocation method based on deep learning according to claim 5, characterized in that, The process of combining changes in resource node supply capacity and the evolution of symbiotic relationships between nodes to determine the adaptation mode of each business node within the predicted future period includes: Extract key information on capacity changes, read / write rate fluctuations, and stability adjustments from the changes in the supply capacity of resource nodes during the future forecast period, reflecting the upper limit and range of changes in the service capacity that resource nodes can provide during the future forecast period. Extract key information on the trend of association strength change, association strength level and adaptation value change in the evolution of symbiotic association strength between nodes within the future prediction period, reflecting the stability of the association between business nodes and resource nodes and the level of adaptation value. Analyze the supply capacity of resource nodes in the current forecast period and the demand fluctuation trend of business nodes in the same forecast period. If the lower limit of the supply capacity of resource nodes is higher than or equal to the upper limit of the estimated demand of business nodes, the correlation method is determined to be a stable correlation method; otherwise, the correlation method is determined to be a dynamic correlation method. Based on the comparison between the current symbiotic association strength of the association edge between the business node and the resource node and the preset strength threshold, the association method between the business node and the resource node is determined. If the symbiotic association strength is higher than or equal to the strength threshold, the direct association method is adopted, and the business node directly calls the corresponding resource node; if the symbiotic association strength is lower than the strength threshold, the indirect association method is adopted, and the association between the business node and the resource node is realized through an intermediate scheduling node. Based on the rate of change of the number of resource demand types and the coefficient of variation of demand size of business nodes, the resource allocation method is determined. If the rate of change of the number of demand types is lower than the type change threshold and the coefficient of variation of demand size is lower than the scale fluctuation threshold, a fixed allocation method is adopted, and resources are allocated according to a preset period and a fixed quota. If the rate of change of the number of demand types is higher than or equal to the type change threshold or the coefficient of variation of demand size is higher than or equal to the scale fluctuation threshold, an elastic allocation method is adopted, and the allocation amount is dynamically adjusted according to the real-time resource requests of business nodes. By combining the adjustment cycle of resource node supply capacity and the change cycle of business node demand fluctuations, an adaptation adjustment frequency is determined. The adjustment cycle of the resource node is compared with a first cycle threshold, and the change cycle of the business node is compared with a second cycle threshold. If the adjustment cycle of the resource node is longer than the first cycle threshold and the change cycle of the business node is longer than the second cycle threshold, then the adaptation adjustment frequency is determined as the first frequency value. If the adjustment cycle of the resource node is shorter than or equal to the first cycle threshold or the change cycle of the business node is shorter than or equal to the second cycle threshold, then the adaptation adjustment frequency is determined as the second frequency value. Wherein, the first frequency value is lower than the second frequency value. By integrating association methods, resource call methods, and adaptation adjustment frequencies, the core structure of the adaptation mode is constructed, and the logical associations and execution order among the association methods, resource call methods, and adaptation adjustment frequencies are determined. The specific execution details of the association method are supplemented, and the node interaction process, data transmission path and status synchronization process in direct or indirect association are determined; the specific operation specifications of the resource call method are refined, and the call triggering conditions, call parameter settings and call result feedback process in fixed call or elastic call are determined; the specific implementation standards of the adaptation adjustment frequency are defined, and the adjustment time interval, adjustment trigger threshold and the status synchronization process after adjustment are determined, forming the adaptation mode of each business node in the future prediction period.

7. The cloud storage resource adaptive allocation method based on deep learning according to claim 1, characterized in that, The process involves using a deep learning model to dynamically match and extrapolate the sequence of business resource adaptation needs with the resource supply spectrum of the cloud storage system, generating a set of resource elastic allocation solutions, including: The business resource adaptation demand sequence and the cloud storage system resource supply spectrum are input into the matching input layer of the deep learning model for time alignment, so that each future prediction period in the business resource adaptation demand sequence corresponds to the corresponding prediction period information in the resource supply spectrum. The demand parsing subnetwork of the deep learning model is used to parse the sequence of business resource adaptation requirements layer by layer, extract the adaptation priority, adaptation direction and adaptation mode of various businesses in each future prediction period, and generate structured demand parsing results. The demand parsing results are classified and organized according to business type and future prediction period. The supply analysis subnetwork of the deep learning model is used to analyze the resource supply spectrum of the cloud storage system layer by layer, extract the supply capacity, deployment location and availability status of various resources in each prediction period, and generate structured supply analysis results. The supply analysis results are classified and organized according to resource type and prediction period. The demand analysis results and supply analysis results are input into the dynamic matching sub-network of the deep learning model. Matching weights are established based on the adaptation priority. The business demand and resource supply are matched bidirectionally according to the future prediction period to generate preliminary matching results. The preliminary matching results are used to reflect the candidate resource set of each business in the corresponding future prediction period. A feasibility analysis is conducted on the preliminary matching results. Based on the physical limitations, deployment constraints, and performance bottlenecks of resource supply, candidate resources whose supply characteristics do not match the business adaptation direction, or whose supply capacity cannot support the association method and resource calling method required by the business adaptation mode, are eliminated to form a feasible matching result. Based on the feasible matching results, a resource allocation adjustment strategy is generated for the needs of each business in each future forecast period. The resource allocation adjustment strategy is used to reflect the initial configuration, adjustment direction and adjustment target of resource allocation. The execution path of the resource allocation adjustment strategy is planned. The execution path is used to reflect the initiating node, relay node, target node and data transmission link of resource allocation, and to arrange the operation steps, sequence and connection method in the allocation process. The adaptation and assurance related steps include real-time resource status monitoring steps, abnormal handling steps in the allocation process, and adaptation effect feedback and adjustment steps, so as to determine the functional responsibilities, execution flow and triggering conditions of each step. By linking and integrating resource allocation adjustment strategies, execution paths, and adaptation and assurance steps, a flexible resource allocation plan for a single business and a single future forecast period is formed. All resource elastic allocation solutions are categorized and summarized according to business type and future forecast period. Duplicate solutions are eliminated and conflicting solutions are optimized to form a set of resource elastic allocation solutions that cover all future forecast periods and are tailored to different business scenarios.

8. The cloud storage resource adaptive allocation method based on deep learning according to claim 1, characterized in that, The process of generating a real-time allocation command stream based on the resource elastic allocation scheme set drives the storage resources in the cloud storage system to dynamically reorganize across nodes and prediction periods according to the command stream, thereby achieving real-time adaptation between business needs and resource supply, including: Extract the core content of resource allocation adjustment strategy, execution path and adaptation guarantee steps of each resource elastic allocation scheme in the resource elastic allocation scheme set, sort the schemes according to business type and execution time, and determine the execution priority of the resource elastic allocation scheme; For each resource elastic allocation scheme, the core parameters of the allocation instruction are determined according to the resource allocation adjustment strategy. The core parameters include business identifier, target resource identifier, allocation type, adjustment range, execution time window, and key information on expected target status. Based on the execution path, the transmission path and execution subject of the allocation instruction are determined, the node identifiers corresponding to the initiating end, receiving end and relay end of the allocation instruction are determined, and the transmission protocol and format requirements of the allocation instruction are specified. In conjunction with the monitoring requirements in the adaptation and assurance steps, status feedback parameters are added to the allocation instructions to determine the resource status indicators, feedback frequency, and feedback methods that need to be fed back during the execution of the instructions. By integrating core parameters, transmission paths, execution entities, and status feedback parameters, a winning resource allocation instruction is generated. Each resource allocation instruction contains a unique instruction identifier, which is used to distinguish and track the instruction execution process. All resource allocation instructions are arranged according to execution time order and execution priority to form an initial instruction sequence. Conflict detection is performed on the initial instruction sequence to identify instruction combinations that have resource occupation conflicts, execution time conflicts, or transmission link conflicts. Conflict detection is based on the target resource identifier, execution time window, and transmission path information in the initial instruction sequence. Conflicting instruction combinations are adjusted and optimized. The optimized instruction sequence is divided into multiple consecutive instruction batches according to time slices. The duration of each time slice is determined based on the instruction execution cycle and system processing capacity. Each instruction batch contains all resource allocation instructions that need to be executed within that time slice. Each batch of instructions is issued sequentially in chronological order to form a real-time allocation instruction stream. The real-time allocation instruction stream is continuously transmitted to the scheduling and execution module of the cloud storage system in batches. The scheduling and execution module receives the real-time allocation instruction stream and drives the corresponding storage resources to perform dynamic reorganization operations across nodes and prediction periods according to the instruction requirements. The dynamic reorganization operations include resource allocation operations to allocate new storage resource space to business nodes, migration operations to transfer business data from one resource node to another, expansion operations to increase the available capacity of resource nodes, and release operations to reclaim the resource space occupied by business nodes. During instruction execution, resource status data is collected in real time according to the requirements of status feedback parameters and fed back to the business resource co-existence network update module for synchronously updating the network topology. By continuously repeating the above steps of generating, issuing, executing, and providing feedback instructions, real-time adaptation between business needs and resource supply is achieved, ensuring that cloud storage resources always maintain a dynamic match with the sequence of business resource adaptation needs.

9. A cloud storage resource adaptive allocation system based on deep learning, characterized in that, include: processor; A machine-readable storage medium for storing machine-executable instructions of the processor; The processor is configured to execute the deep learning-based cloud storage resource adaptive allocation method according to any one of claims 1 to 8 by executing the machine-executable instructions.

10. A computer program product, characterized in that, The computer program product includes machine-executable instructions stored in a computer-readable storage medium. The processor of the deep learning-based cloud storage resource adaptive allocation system reads the machine-executable instructions from the computer-readable storage medium and executes the machine-executable instructions, causing the deep learning-based cloud storage resource adaptive allocation system to perform the deep learning-based cloud storage resource adaptive allocation method as described in any one of claims 1 to 8.

Images