Network cache data storage method and system based on edge computing

Abstract

The invention provides a network cache data storage method and system based on edge computing, and the method comprises the steps: collecting the position of a terminal access node, network time delay and cache request content through a region edge gateway; dividing the regional edge gateways which are high in cache request content association degree and adjacent in position into the same request cluster through a density clustering algorithm; collecting storage, load and throughput data of each edge cache node, constructing a corresponding resource demand prediction model in combination with historical cache data of the request cluster, and creating a corresponding elastic resource pool according to a prediction result output by the resource demand prediction model; analyzing operation semantics and a data dependency relationship of the cache request, and constructing a corresponding semantic conflict interception matrix; and generating a corresponding dynamic execution sequence through the semantic conflict interception matrix, and collecting target cache data corresponding to the dynamic execution sequence so as to store the target cache data into the elastic resource pool. The storage efficiency can be improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a network caching data storage method and system based on edge computing. Background Technology With the advancement of technology and the rapid development of the times, significant progress has been made in the fields of the Internet and computer technology. Specifically, network caching is a core technology for improving data access efficiency and reducing the load on the origin server, and it is indispensable in high-concurrency business. Existing cache storage systems generally adopt a multi-threaded / process parallel architecture to adapt to massive access, but the cross-reading and writing of the same cache resource by multiple execution units can easily lead to resource contention conflicts, becoming a key bottleneck restricting the stability of cache storage.

[0002] Among cache concurrency conflicts, the most serious harm comes from concurrent rebuilding following cache breakdown: when a hot cache expires, a large number of concurrent requests will simultaneously detect the cache missing and send rebuilding requests to the origin server, forming a "concurrent rebuilding storm." This can lead to a surge in the load on the origin server, service outages, and may also cause data consistency issues such as dirty reads and write overwrites of the cache.

[0003] Furthermore, existing technologies have introduced solutions such as distributed locks and cache preheating to address the issue of cache concurrency conflicts. However, in practical applications, distributed locks introduce additional latency, cache preheating cannot adapt to sudden hotspots, and as business traffic increases, the concurrency conflict problem becomes more severe, which correspondingly reduces the storage efficiency of network cache data. Summary of the Invention

[0004] Therefore, the purpose of this invention is to provide a network cache data storage method and system based on edge computing to solve the problem that the existing technology is prone to concurrency conflicts, which leads to a reduction in the efficiency of network cache data storage.

[0005] The first aspect of the present invention proposes: A network cache data storage method based on edge computing, applied to a multi-region edge computing network containing regional edge gateways, edge cache nodes, and terminal access nodes, specifically includes the following steps: The location, network latency, and cached request content of the terminal access node are collected by the regional edge gateway, and the regional edge gateways with high correlation of cached request content and close location are divided into the same request cluster by density clustering algorithm. The storage, load, and throughput data of each edge cache node are collected, and a corresponding resource demand prediction model is constructed by combining the historical cache data of the request cluster. A corresponding elastic resource pool is created based on the prediction results output by the resource demand prediction model. When the regional edge gateway receives a cache request, it parses the operation semantics and data dependencies of the cache request and constructs a corresponding semantic conflict interception matrix. When a data mutual exclusion operation request is detected, a corresponding dynamic execution sequence is generated through the semantic conflict interception matrix, and target cache data corresponding to the dynamic execution sequence is collected to store the target cache data inside the elastic resource pool.

[0006] The beneficial effects of this invention are as follows: By collecting the location, network latency, and cached request content of each node, each regional edge gateway can be divided into several request clusters. Based on this, a corresponding resource demand prediction model can be constructed according to the historical cached data of the request clusters, and a corresponding elastic resource pool can be created. Based on this, the operation semantics and data dependencies of each issued cache request can be parsed, a corresponding semantic conflict interception matrix can be constructed, and finally, the target cached data to be stored can be dynamically stored through the elastic resource pool, thereby effectively avoiding the phenomenon of concurrent conflicts and improving the storage efficiency of network cached data.

[0007] Furthermore, the step of collecting storage, load, and throughput data of each edge cache node and combining it with historical cache data of the request cluster to construct a corresponding resource demand prediction model includes: The storage, load, and throughput data of each edge cache node are integrated into corresponding node resource data, and the node resource data is spatiotemporally aligned with the historical cache data at the time granularity. Furthermore, outlier values ​​are detected and corrected using the isolated forest algorithm to generate a corresponding standard data source. Extract the resource elasticity features corresponding to each edge cache node and the spatiotemporal correlation features of each request cluster from the standard data source to construct a corresponding two-dimensional feature set; Using the two-dimensional feature set as input, an improved temporal attention LSTM model is constructed and set as the resource demand prediction model.

[0008] Furthermore, the step of creating a corresponding elastic resource pool based on the prediction results output by the resource demand prediction model includes: Extract the storage capacity and concurrency parameters from the prediction results, and synchronously collect the request fluctuation data of the regional edge gateway to generate a corresponding scenario-based demand list. With resource matching degree, transmission latency and energy consumption as optimization objectives, the node attributes of each edge cache node are virtualized into corresponding standard units through distributed proxy, and then aggregated to form a basic matrix of resource pool; Based on the scenario-based requirement list, a corresponding scaling threshold is set, and the resource pool base matrix is ​​pooled based on the scaling threshold to generate the corresponding elastic resource pool.

[0009] Furthermore, the step of parsing the operational semantics and data dependencies of the cached request and constructing the corresponding semantic conflict interception matrix includes: Collect the cache request text corresponding to the cache request, and parse the cache request text using a pre-trained BERT model to extract the operation type, data identifier and permission level, and encode them into standard semantic vectors. Using data objects as nodes and dependencies as directed edges, a data dependency graph is constructed through a graph neural network; Several target elements contained within the standard semantic vector are extracted and introduced into a time axis to construct a matrix framework. The data dependency graph and several target elements are then filled into the matrix framework to generate the semantic conflict interception matrix.

[0010] Furthermore, the step of filling the matrix frame with the data dependency graph and the target elements to generate the semantic conflict interception matrix includes: Extract the node association strength and path depth features of the data dependency graph, combine them with the permission level features in the target element, calculate the dynamic weight of each feature through entropy weight-hierarchical analysis, and perform corresponding weighted fusion to generate the corresponding fused feature vector. The system calls the mutual exclusion rule library to detect potential conflicts in the fused feature vector, generates corresponding conflict prediction coefficients, and collects the remaining resource parameters of the edge cache node. The conflict prediction coefficient and the remaining resource parameters are bound to the interior of the fused feature vector, and the bound fused feature vector is filled into the interior of the matrix frame to generate the semantic conflict interception matrix.

[0011] Furthermore, the step of generating a corresponding dynamic execution sequence through the semantic conflict interception matrix when a data mutual exclusion operation request is detected includes: A spatiotemporal feature matrix is ​​constructed by combining the semantic conflict interception matrix with the location of the terminal access node and the request timestamp. The semantic conflict type, spatial distance and time interval in the data mutual exclusion operation request are extracted and spatiotemporally clustered using the DBSCAN algorithm to generate the corresponding conflict group. The requests within the conflict group are sorted to output the corresponding basic execution sequence; The basic execution sequence is adjusted collaboratively to generate the dynamic execution sequence.

[0012] Furthermore, the step of performing collaborative fault-tolerant adjustment on the basic execution sequence to generate the dynamic execution sequence includes: Semantic conflict residual identifiers are extracted from the basic execution sequence, and combined with the spatiotemporal clustering degree of the terminal access node and the resource fluctuation coefficient of the edge cache node, a corresponding risk feature vector is constructed. The risk level of execution interruption and the probability of conflict propagation of the cached request are calculated based on the risk feature vector. The basic execution sequence is dynamically adjusted based on the execution interruption risk level and the conflict propagation probability to generate the dynamic execution sequence.

[0013] The second aspect of the present invention proposes: A network caching data storage system based on edge computing is applied to a multi-region edge computing network containing regional edge gateways, edge caching nodes, and terminal access nodes, comprising the following: The acquisition module is used to acquire the location, network latency and cached request content of the terminal access node through the regional edge gateway, and to divide the regional edge gateways with high correlation of cached request content and close location into the same request cluster through density clustering algorithm; The construction module is used to collect storage, load and throughput data of each edge cache node, combine the historical cache data of the request cluster to construct a corresponding resource demand prediction model, and create a corresponding elastic resource pool based on the prediction results output by the resource demand prediction model. The parsing module is used to parse the operation semantics and data dependencies of the cache request when the regional edge gateway receives the cache request, and construct the corresponding semantic conflict interception matrix. The storage module is used to generate a corresponding dynamic execution sequence through the semantic conflict interception matrix when a data mutual exclusion operation request is detected, and to collect the target cache data corresponding to the dynamic execution sequence so as to store the target cache data in the internal elastic resource pool.

[0014] Furthermore, the building module is specifically used for: The storage, load, and throughput data of each edge cache node are integrated into corresponding node resource data, and the node resource data is spatiotemporally aligned with the historical cache data at the time granularity. Furthermore, outlier values ​​are detected and corrected using the isolated forest algorithm to generate a corresponding standard data source. Extract the resource elasticity features corresponding to each edge cache node and the spatiotemporal correlation features of each request cluster from the standard data source to construct a corresponding two-dimensional feature set; Using the two-dimensional feature set as input, an improved temporal attention LSTM model is constructed and set as the resource demand prediction model.

[0015] Furthermore, the building module is specifically used for: Extract the storage capacity and concurrency parameters from the prediction results, and synchronously collect the request fluctuation data of the regional edge gateway to generate a corresponding scenario-based demand list. With resource matching degree, transmission latency and energy consumption as optimization objectives, the node attributes of each edge cache node are virtualized into corresponding standard units through distributed proxy, and then aggregated to form a basic matrix of resource pool; Based on the scenario-based requirement list, a corresponding scaling threshold is set, and the resource pool base matrix is ​​pooled based on the scaling threshold to generate the corresponding elastic resource pool.

[0016] Furthermore, the parsing module is specifically used for: Collect the cache request text corresponding to the cache request, and parse the cache request text using a pre-trained BERT model to extract the operation type, data identifier and permission level, and encode them into standard semantic vectors. Using data objects as nodes and dependencies as directed edges, a data dependency graph is constructed through a graph neural network; Several target elements contained within the standard semantic vector are extracted and introduced into a time axis to construct a matrix framework. The data dependency graph and several target elements are then filled into the matrix framework to generate the semantic conflict interception matrix.

[0017] Furthermore, the parsing module is specifically used for: Extract the node association strength and path depth features of the data dependency graph, combine them with the permission level features in the target element, calculate the dynamic weight of each feature through entropy weight-hierarchical analysis, and perform corresponding weighted fusion to generate the corresponding fused feature vector. The system calls the mutual exclusion rule library to detect potential conflicts in the fused feature vector, generates corresponding conflict prediction coefficients, and collects the remaining resource parameters of the edge cache node. The conflict prediction coefficient and the remaining resource parameters are bound to the interior of the fused feature vector, and the bound fused feature vector is filled into the interior of the matrix frame to generate the semantic conflict interception matrix.

[0018] Furthermore, the storage module is specifically used for: A spatiotemporal feature matrix is ​​constructed by combining the semantic conflict interception matrix with the location of the terminal access node and the request timestamp. The semantic conflict type, spatial distance and time interval in the data mutual exclusion operation request are extracted and spatiotemporally clustered using the DBSCAN algorithm to generate the corresponding conflict group. The requests within the conflict group are sorted to output the corresponding basic execution sequence; The basic execution sequence is adjusted collaboratively to generate the dynamic execution sequence.

[0019] Furthermore, the storage module is specifically used for: Semantic conflict residual identifiers are extracted from the basic execution sequence, and combined with the spatiotemporal clustering degree of the terminal access node and the resource fluctuation coefficient of the edge cache node, a corresponding risk feature vector is constructed. The risk level of execution interruption and the probability of conflict propagation of the cached request are calculated based on the risk feature vector. The basic execution sequence is dynamically adjusted based on the execution interruption risk level and the conflict propagation probability to generate the dynamic execution sequence.

[0020] The third aspect of the present invention proposes: A computer includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the network cache data storage method based on edge computing as described above.

[0021] The fourth aspect of the present invention proposes: A readable storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the network cache data storage method based on edge computing as described above.

[0022] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0023] Figure 1 A flowchart illustrating the network cache data storage method based on edge computing provided in the first embodiment of the present invention; Figure 2 This is a structural block diagram of a network cache data storage system based on edge computing provided in the third embodiment of the present invention.

[0024] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0025] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Several embodiments of the invention are illustrated in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0026] It should be noted that when a component is said to be "fixed to" another component, it can be directly on the other component or there may be an intervening component. When a component is said to be "connected to" another component, it can be directly connected to the other component or there may be an intervening component. The terms "vertical," "horizontal," "left," "right," and similar expressions used in this document are for illustrative purposes only.

[0027] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains. The terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0028] Please see Figure 1 The image shows a network cache data storage method based on edge computing provided in the first embodiment of the present invention. The network cache data storage method based on edge computing provided in this embodiment can avoid storage conflicts of network cache data and improve the storage efficiency of network cache data accordingly.

[0029] Specifically, this embodiment provides: A network cache data storage method based on edge computing, applied to a multi-region edge computing network containing regional edge gateways, edge cache nodes, and terminal access nodes, specifically includes the following steps: Step S10: Collect the location, network latency and cached request content of the terminal access node through the regional edge gateway, and divide the regional edge gateways with high correlation of cached request content and close location into the same request cluster through density clustering algorithm; It's important to note that the first step involves the regional edge gateway collecting three core parameters from terminal access nodes: location information (to determine the geographical origin of the request and avoid increased latency due to cross-regional resource scheduling), network latency (reflecting the communication quality between the terminal and the gateway; high latency necessitates prioritizing local caching), and cached request content (clarifying the data type and scale that the terminal needs to store). A density-based clustering algorithm is then used to group gateways with "highly correlated cached request content (such as requests for the same type of video data) and geographically proximate locations" into the same request cluster. Specifically, the core value of this operation is to achieve "homogenized demand" aggregation, avoiding the inefficiency of individually allocating resources to scattered requests, and making subsequent resource allocation more targeted, thus facilitating subsequent processing.

[0030] Step S20: Collect storage, load and throughput data of each edge cache node, combine the historical cache data of the request cluster to construct a corresponding resource demand prediction model, and create a corresponding elastic resource pool based on the prediction results output by the resource demand prediction model. It's important to note that the second step focuses on the resource supply side: collecting data on the storage capacity (remaining available storage space), load (number of requests currently being processed), and throughput (data transfer capacity per unit time) of the edge cache nodes. These parameters directly determine the node's service capacity. A resource demand prediction model is built by combining historical cache data from the request cluster (e.g., the cache demand of a cluster during peak weekday periods). The model outputs a forecast of cache demand for a future period (e.g., a cluster needs to add 50GB of cache space in the next hour). Based on this, an elastic resource pool is created. Specifically, the elastic resource pool dynamically integrates redundant resources from various edge cache nodes, avoiding overload or idle resources on a single node, ensuring a dynamic match between resource supply and demand, thus facilitating subsequent processing.

[0031] Step S30: When the regional edge gateway receives a cache request, it parses the operation semantics and data dependencies of the cache request and constructs a corresponding semantic conflict interception matrix. It's important to note that when the regional edge gateway receives a new cache request, it must first parse the request's operation semantics (such as operation types like "read," "write," and "modify") and data dependencies (e.g., if request A requires data X and request B also needs to modify data X, then A and B are dependent). A semantic conflict interception matrix is ​​then constructed. Specifically, the core function of this matrix is ​​to identify potential data operation conflicts in advance (such as simultaneously writing the same data), providing a basis for judging mutually exclusive requests in subsequent processing.

[0032] Step S40: When a data mutual exclusion operation request is detected, a corresponding dynamic execution sequence is generated through the semantic conflict interception matrix, and target cache data corresponding to the dynamic execution sequence is collected to store the target cache data inside the elastic resource pool.

[0033] It's worth noting that when a data-mutually exclusive operation request is detected (such as two terminals simultaneously requesting to modify the same cached data), a dynamic execution sequence is generated through a semantic conflict interception matrix (clearly defining the execution order of requests) to ensure orderly operation. Simultaneously, the target cached data is collected and stored in the elastic resource pool, completing the entire cache storage process. This process integrates "precise demand perception - elastic resource provisioning - orderly conflict control," significantly improving the efficiency and reliability of edge network cache storage, thus facilitating subsequent processing.

[0034] Second Embodiment Furthermore, the step of collecting storage, load, and throughput data of each edge cache node and combining it with historical cache data of the request cluster to construct a corresponding resource demand prediction model includes: The storage, load, and throughput data of each edge cache node are integrated into corresponding node resource data, and the node resource data is spatiotemporally aligned with the historical cache data at the time granularity. Furthermore, outlier values ​​are detected and corrected using the isolated forest algorithm to generate a corresponding standard data source. Extract the resource elasticity features corresponding to each edge cache node and the spatiotemporal correlation features of each request cluster from the standard data source to construct a corresponding two-dimensional feature set; Using the two-dimensional feature set as input, an improved temporal attention LSTM model is constructed and set as the resource demand prediction model.

[0035] It's important to note that the first step involves integrating the storage, load, and throughput data of the edge cache nodes into node resource data, while simultaneously correlating it with the historical cache data of the request cluster (such as the cluster's cache request volume and data type distribution over the past week). Since edge nodes may experience data fluctuations during operation (such as abnormal data caused by sudden load spikes), the two types of data need to be spatiotemporally aligned by time granularity (e.g., 5 minutes per time slice) to ensure that node resource data and cluster cache data correspond within the same time slice. Then, the Isolation Forest algorithm is used to detect and correct outliers. Specifically, the Isolation Forest algorithm is a highly efficient anomaly detection algorithm that can quickly identify data points deviating from the normal distribution without needing labeled data, preventing outliers from causing model prediction bias. The processed data forms a standard data source, providing high-quality input for the model.

[0036] The second step involves extracting a two-dimensional feature set from standard data sources: first, the "resource elasticity characteristics" of edge cache nodes (such as the fluctuation range of storage capacity, the difference between peak and valley loads, and the adjustment response speed of throughput), reflecting the dynamic supply capacity of node resources; and second, the "spatiotemporal correlation characteristics" of request clusters (such as changes in request volume at different times and the clustering of request types in the same geographical area), reflecting the patterns of cluster caching demand. This two-dimensional feature set covers both the "supply side" and the "demand side," ensuring the comprehensiveness of the features.

[0037] Using a two-dimensional feature set as input, an improved temporal attention LSTM model is constructed and set as a resource demand prediction model. Traditional LSTM models excel at processing time-series data, but they pay equal attention to features at different time steps. The improved temporal attention mechanism allows the model to automatically focus on key time segments for demand prediction (such as request features during weekday morning rush hours), enhancing its ability to capture sudden changes in demand. This model can accurately output predictions of cache resource demand (such as storage capacity demand and concurrent request processing demand) within a preset future time period, providing a scientific basis for the creation of elastic resource pools and facilitating subsequent processing.

[0038] Furthermore, the step of creating a corresponding elastic resource pool based on the prediction results output by the resource demand prediction model includes: Extract the storage capacity and concurrency parameters from the prediction results, and synchronously collect the request fluctuation data of the regional edge gateway to generate a corresponding scenario-based demand list. With resource matching degree, transmission latency and energy consumption as optimization objectives, the node attributes of each edge cache node are virtualized into corresponding standard units through distributed proxy, and then aggregated to form a basic matrix of resource pool; Based on the scenario-based requirement list, a corresponding scaling threshold is set, and the resource pool base matrix is ​​pooled based on the scaling threshold to generate the corresponding elastic resource pool.

[0039] It should be noted that the first step involves extracting the core parameters from the resource demand forecast results. Specifically, these include storage capacity parameters (e.g., a cluster needs 60GB of available storage) and concurrency capability parameters (e.g., needing to support 300 cache requests per second). Simultaneously, request fluctuation data from the regional edge gateway is collected (e.g., the variance of request volume over the past hour, reflecting the stability of demand). These three parameters are then integrated to generate a scenario-based demand list. Specifically, this list differs from general demands and can accurately match the current cluster's caching scenario (e.g., video caching scenarios require large storage, while real-time data caching scenarios require high concurrency).

[0040] The second step, aiming for "highest resource matching, lowest transmission latency, and lowest energy consumption," uses distributed proxy technology to virtualize the node attributes (such as storage type, computing power specifications, and network bandwidth) of each edge cache node into standardized resource units. Specifically, edge cache node attributes vary significantly between different vendors; virtualization enables unified resource management and scheduling, avoiding resource waste caused by hardware heterogeneity. These virtualized resource units are then aggregated to form a resource pool matrix. Each element in the matrix represents the state of a standardized resource unit (e.g., "idle," "occupied," "pending release"), providing a clear view of the distribution of edge cache resources across the entire network.

[0041] Based on a list of scenario-based requirements, scaling thresholds are set (e.g., automatically reducing the resource pool size when storage capacity requirements are below 20GB and automatically expanding when they exceed 50GB). The resource pool base matrix is ​​then pooled based on these thresholds: resource units meeting the requirements are included in the elastic resource pool, and a dynamic resource adjustment mechanism is established (e.g., automatically adding idle resource units from the resource pool base matrix to the elastic resource pool when a sudden surge in requests is detected). The resulting elastic resource pool satisfies current caching needs while also handling future demand fluctuations, achieving efficient resource utilization for subsequent processing.

[0042] Furthermore, the step of parsing the operational semantics and data dependencies of the cached request and constructing the corresponding semantic conflict interception matrix includes: Collect the cache request text corresponding to the cache request, and parse the cache request text using a pre-trained BERT model to extract the operation type, data identifier and permission level, and encode them into standard semantic vectors. Using data objects as nodes and dependencies as directed edges, a data dependency graph is constructed through a graph neural network; Several target elements contained within the standard semantic vector are extracted and introduced into a time axis to construct a matrix framework. The data dependency graph and several target elements are then filled into the matrix framework to generate the semantic conflict interception matrix.

[0043] It should be noted that the first step involves collecting the cache request text corresponding to the cache request (such as instructions sent by the terminal such as "modify cache data X" or "read cache data X"). The text content is then parsed using a pre-trained BERT model. The BERT model has powerful semantic understanding capabilities and can accurately extract the operation type (such as read, write, delete), data identifier (such as the unique ID of data X), and permission level (such as administrator level, ordinary user level, and permission level affects operation priority) of the request. This information is then encoded into a standard semantic vector. Specifically, the standardization process enables cache requests of different formats to be calculated and compared in a unified manner.

[0044] The second step involves using data objects as nodes (e.g., data X and data Y are independent nodes) and data dependencies as directed edges (e.g., request A needs to read data X before it can be processed, and request B needs to modify data X, so there is a "read-write" dependency between A and B, and a directed edge is constructed from A to B). A graph neural network (GNN) is used to construct a data dependency graph: GNN is good at handling data with related relationships, and can clearly present the complex dependency network between multiple cache requests and data objects, avoiding operational confusion caused by unclear dependencies.

[0045] The third step involves extracting target elements such as operation type, data identifier, and permission level from the standard semantic vector. These are then incorporated into a timeline to construct a matrix framework. Specifically, the timeline reflects the temporal characteristics of cache requests (e.g., the order of two write requests for the same data directly affects data accuracy). The node relationships in the data dependency graph (e.g., the dependency strength between A and B) and the corresponding target elements are filled into the matrix framework to generate a semantic conflict interception matrix. The rows and columns of the matrix represent cache requests, and the element values ​​represent the conflict risk level between two requests (e.g., "1" represents no conflict, and "5" represents severe mutual exclusion). This allows for the intuitive identification and interception of high-risk conflict requests, facilitating subsequent processing.

[0046] Furthermore, the step of filling the matrix frame with the data dependency graph and the target elements to generate the semantic conflict interception matrix includes: Extract the node association strength and path depth features of the data dependency graph, combine them with the permission level features in the target element, calculate the dynamic weight of each feature through entropy weight-hierarchical analysis, and perform corresponding weighted fusion to generate the corresponding fused feature vector. The system calls the mutual exclusion rule library to detect potential conflicts in the fused feature vector, generates corresponding conflict prediction coefficients, and collects the remaining resource parameters of the edge cache node. The conflict prediction coefficient and the remaining resource parameters are bound to the interior of the fused feature vector, and the bound fused feature vector is filled into the interior of the matrix frame to generate the semantic conflict interception matrix.

[0047] It's important to note that the first step involves extracting node association strength (e.g., the degree of dependence between request A and data X; a higher value indicates a greater impact of A's operation on X) and path depth features from the data dependency graph (e.g., request A requires three steps: "read X - process - write Y," with a path depth of 3; greater depth indicates more complex operation). This is combined with permission level features from the target elements (e.g., administrator permission level is 10, ordinary user is 5). The dynamic weights of each feature are calculated using the entropy-weighted analytic hierarchy process (AHP): the entropy-weighted method objectively determines weights based on data distribution (e.g., if permission conflicts are frequent during a certain period, the permission level weight is automatically increased), while the AHP combines expert experience for subjective adjustments (e.g., the association strength weight of core business data needs to be increased). This combination ensures the scientific nature of the weight allocation. Finally, each feature is weighted and fused with its corresponding weight to generate a fused feature vector. Specifically, this vector comprehensively reflects the conflict-related characteristics of cached requests, providing a more complete picture than a single feature.

[0048] The second step involves invoking a pre-defined mutual exclusion rule base (such as rules like "two write operations on the same data are mutually exclusive" and "high-privilege reads can interrupt low-privilege writes") to detect potential conflicts in the fused feature vector (e.g., if two requests are both "write data X" and have the same permission level, the mutual exclusion rule is triggered), and generating a conflict prediction coefficient (e.g., "0.9" represents a high probability of conflict, and "0.1" represents no risk of conflict). Simultaneously, the remaining resource parameters of the edge cache nodes (such as remaining storage and idle computing power) are collected. Specifically, when node resources are sufficient, the impact of conflicts can be reduced through parallel processing; when resources are scarce, the risk of conflict increases, therefore this parameter needs to be included in the conflict assessment system.

[0049] The third step involves binding the conflict prediction coefficient and remaining resource parameters to the fused feature vector, making the vector simultaneously contain "conflict risk" and "processing capacity" information. The bound fused feature vector is then filled into the cells of the matrix frame according to the request sequence and data identifier, generating the final semantic conflict interception matrix. Each element in the matrix not only reflects the conflict risk but also indicates whether the current node has the ability to handle the conflict, providing a more accurate decision-making basis for subsequently generating dynamic execution sequences, thus facilitating subsequent processing.

[0050] Furthermore, the step of generating a corresponding dynamic execution sequence through the semantic conflict interception matrix when a data mutual exclusion operation request is detected includes: A spatiotemporal feature matrix is ​​constructed by combining the semantic conflict interception matrix with the location of the terminal access node and the request timestamp. The semantic conflict type, spatial distance and time interval in the data mutual exclusion operation request are extracted and spatiotemporally clustered using the DBSCAN algorithm to generate the corresponding conflict group. The requests within the conflict group are sorted to output the corresponding basic execution sequence; The basic execution sequence is adjusted collaboratively to generate the dynamic execution sequence.

[0051] It should be noted that, in the first step, a spatiotemporal feature matrix is ​​constructed by combining the semantic conflict interception matrix (providing conflict risk information), the location of the terminal access node (reflecting the geographical distribution of requests; requests with adjacent locations can be prioritized for scheduling to the same edge node), and the request timestamp (reflecting the order of requests). Specifically, this matrix simultaneously covers three dimensions: "semantic conflict," "spatial distribution," and "temporal order," ensuring comprehensive consideration for scheduling. The semantic conflict type (such as "read-write" conflict, "write-write" conflict), spatial distance (distance between the requesting terminal and the edge node), and time interval (time difference between two requests) are extracted from data mutual exclusion operation requests. Spatiotemporal clustering is then performed using the DBSCAN algorithm: the DBSCAN algorithm does not require a preset number of clusters and can automatically cluster requests that are spatiotemporally adjacent and have the same conflict type into conflict groups (such as "two write requests sent within 10 seconds in the same area" as a group), facilitating batch processing of conflicts.

[0052] The second step is to sort the requests within the conflict group to generate a basic execution sequence. The sorting criteria include request timestamps (first-come, first-served), permission levels (higher permissions take precedence), and business priorities (e.g., core business requests take precedence). Specifically, for example, a write request sent by an administrator at time t1 has higher priority than a write request sent by a regular user at time t2 (t2>t1), and the administrator's request will appear first after sorting. This basic execution sequence solves the problem of the execution order of requests within the conflict group, but it does not consider unexpected situations during execution (such as a sudden drop in node resources during the execution of a request).

[0053] The third step is to perform collaborative fault-tolerant adjustments on the basic execution sequence: by simulating the execution sequence, we detect potential issues such as execution interruptions (e.g., insufficient resources causing request failures) and conflict propagation (e.g., an error in the execution of a request causing related requests to become abnormal). We then optimize the sequence (e.g., adjust requests with high resource requirements to be executed during periods with sufficient resources) and finally generate a dynamic execution sequence. Specifically, this sequence ensures the orderly execution of conflicting requests and also has the fault tolerance capability to deal with unexpected situations.

[0054] Furthermore, the step of performing collaborative fault-tolerant adjustment on the basic execution sequence to generate the dynamic execution sequence includes: Semantic conflict residual identifiers are extracted from the basic execution sequence, and combined with the spatiotemporal clustering degree of the terminal access node and the resource fluctuation coefficient of the edge cache node, a corresponding risk feature vector is constructed. The risk level of execution interruption and the probability of conflict propagation of the cached request are calculated based on the risk feature vector. The basic execution sequence is dynamically adjusted based on the execution interruption risk level and the conflict propagation probability to generate the dynamic execution sequence.

[0055] It should be noted that, in the first step, semantic conflict residual markers (such as minor conflict risks that still exist after sorting, such as "read-write" request intervals being too short) are extracted from the basic execution sequence. Combined with the spatiotemporal clustering of terminal access nodes (such as the number of terminals sending requests simultaneously in a certain area; high clustering indicates high request concurrency pressure) and the resource fluctuation coefficient of edge cache nodes (such as the load fluctuation amplitude in the past 5 minutes; a large coefficient indicates unstable resource status), a risk feature vector is constructed. Specifically, this vector quantifies the various risks existing in the execution sequence, providing a basis for risk assessment.

[0056] The second step is to calculate two core indicators based on the risk feature vector: execution interruption risk level (e.g., "high", "medium", "low", calculated based on resource fluctuation coefficient and requested resource demand) and conflict propagation probability (e.g., "15%", calculated based on semantic conflict residual identifier and request correlation). Specifically, the execution interruption risk level reflects the execution reliability of a single request, and the conflict propagation probability reflects the impact range of a failed request on other requests.

[0057] The third step involves dynamically adjusting the basic execution sequence based on these two metrics: for requests with a "high" risk of execution interruption, they are rescheduled to periods with sufficient edge node resources, and resources are reserved in advance; for requests with a "high" probability of conflict propagation, the execution interval is increased (e.g., extending the interval between two related requests from 1 second to 3 seconds), and backup execution nodes are set up. The adjusted sequence addresses the potential risks of the basic sequence while adapting to the dynamic changes in the edge network, ultimately generating a reliable dynamic execution sequence that ensures the orderly execution of mutually exclusive requests and the secure storage of target cached data in the elastic resource pool for subsequent processing.

[0058] Please see Figure 2 The third embodiment of the present invention provides: A network caching data storage system based on edge computing is applied to a multi-region edge computing network containing regional edge gateways, edge caching nodes, and terminal access nodes, comprising the following: The acquisition module is used to acquire the location, network latency and cached request content of the terminal access node through the regional edge gateway, and to divide the regional edge gateways with high correlation of cached request content and close location into the same request cluster through density clustering algorithm; The construction module is used to collect storage, load and throughput data of each edge cache node, combine the historical cache data of the request cluster to construct a corresponding resource demand prediction model, and create a corresponding elastic resource pool based on the prediction results output by the resource demand prediction model. The parsing module is used to parse the operation semantics and data dependencies of the cache request when the regional edge gateway receives the cache request, and construct the corresponding semantic conflict interception matrix. The storage module is used to generate a corresponding dynamic execution sequence through the semantic conflict interception matrix when a data mutual exclusion operation request is detected, and to collect the target cache data corresponding to the dynamic execution sequence so as to store the target cache data in the internal elastic resource pool.

[0059] Furthermore, the building module is specifically used for: The storage, load, and throughput data of each edge cache node are integrated into corresponding node resource data, and the node resource data is spatiotemporally aligned with the historical cache data at the time granularity. Furthermore, outlier values ​​are detected and corrected using the isolated forest algorithm to generate a corresponding standard data source. Extract the resource elasticity features corresponding to each edge cache node and the spatiotemporal correlation features of each request cluster from the standard data source to construct a corresponding two-dimensional feature set; Using the two-dimensional feature set as input, an improved temporal attention LSTM model is constructed and set as the resource demand prediction model.

[0060] Furthermore, the building module is specifically used for: Extract the storage capacity and concurrency parameters from the prediction results, and synchronously collect the request fluctuation data of the regional edge gateway to generate a corresponding scenario-based demand list. With resource matching degree, transmission latency and energy consumption as optimization objectives, the node attributes of each edge cache node are virtualized into corresponding standard units through distributed proxy, and then aggregated to form a basic matrix of resource pool; Based on the scenario-based requirement list, a corresponding scaling threshold is set, and the resource pool base matrix is ​​pooled based on the scaling threshold to generate the corresponding elastic resource pool.

[0061] Furthermore, the parsing module is specifically used for: Collect the cache request text corresponding to the cache request, and parse the cache request text using a pre-trained BERT model to extract the operation type, data identifier and permission level, and encode them into standard semantic vectors. Using data objects as nodes and dependencies as directed edges, a data dependency graph is constructed through a graph neural network; Several target elements contained within the standard semantic vector are extracted and introduced into a time axis to construct a matrix framework. The data dependency graph and several target elements are then filled into the matrix framework to generate the semantic conflict interception matrix.

[0062] Furthermore, the parsing module is specifically used for: Extract the node association strength and path depth features of the data dependency graph, combine them with the permission level features in the target element, calculate the dynamic weight of each feature through entropy weight-hierarchical analysis, and perform corresponding weighted fusion to generate the corresponding fused feature vector. The system calls the mutual exclusion rule library to detect potential conflicts in the fused feature vector, generates corresponding conflict prediction coefficients, and collects the remaining resource parameters of the edge cache node. The conflict prediction coefficient and the remaining resource parameters are bound to the interior of the fused feature vector, and the bound fused feature vector is filled into the interior of the matrix frame to generate the semantic conflict interception matrix.

[0063] Furthermore, the storage module is specifically used for: A spatiotemporal feature matrix is ​​constructed by combining the semantic conflict interception matrix with the location of the terminal access node and the request timestamp. The semantic conflict type, spatial distance and time interval in the data mutual exclusion operation request are extracted and spatiotemporally clustered using the DBSCAN algorithm to generate the corresponding conflict group. The requests within the conflict group are sorted to output the corresponding basic execution sequence; The basic execution sequence is adjusted collaboratively to generate the dynamic execution sequence.

[0064] Furthermore, the storage module is specifically used for: Semantic conflict residual identifiers are extracted from the basic execution sequence, and combined with the spatiotemporal clustering degree of the terminal access node and the resource fluctuation coefficient of the edge cache node, a corresponding risk feature vector is constructed. The risk level of execution interruption and the probability of conflict propagation of the cached request are calculated based on the risk feature vector. The basic execution sequence is dynamically adjusted based on the execution interruption risk level and the conflict propagation probability to generate the dynamic execution sequence.

[0065] The fourth embodiment of the present invention provides a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the network cache data storage method based on edge computing as described above.

[0066] The fifth embodiment of the present invention provides a readable storage medium on which a computer program is stored, wherein the program, when executed by a processor, implements the network cache data storage method based on edge computing as described above.

[0067] In summary, the network cache data storage method and system based on edge computing provided in the above embodiments of the present invention can avoid conflicts in network cache data during storage, thereby improving the storage efficiency of network cache data.

[0068] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can reside in the same processor; or the above modules can be located in different processors in any combination.

[0069] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0070] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0071] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0072] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0073] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A network cache data storage method based on edge computing, applied to a multi-region edge computing network containing regional edge gateways, edge cache nodes, and terminal access nodes, characterized in that, Includes the following steps: The location, network latency, and cached request content of the terminal access node are collected by the regional edge gateway, and the regional edge gateways with high correlation of cached request content and close location are divided into the same request cluster by density clustering algorithm. The storage, load, and throughput data of each edge cache node are collected, and a corresponding resource demand prediction model is constructed by combining the historical cache data of the request cluster. A corresponding elastic resource pool is created based on the prediction results output by the resource demand prediction model. When the regional edge gateway receives a cache request, it parses the operation semantics and data dependencies of the cache request and constructs a corresponding semantic conflict interception matrix. When a data mutual exclusion operation request is detected, a corresponding dynamic execution sequence is generated through the semantic conflict interception matrix, and target cache data corresponding to the dynamic execution sequence is collected to store the target cache data inside the elastic resource pool.

2. The network cache data storage method based on edge computing according to claim 1, characterized in that, The step of collecting storage, load, and throughput data of each edge cache node and combining it with historical cache data of the request cluster to construct a corresponding resource demand prediction model includes: The storage, load, and throughput data of each edge cache node are integrated into corresponding node resource data, and the node resource data is spatiotemporally aligned with the historical cache data at the time granularity. Furthermore, outlier values ​​are detected and corrected using the isolated forest algorithm to generate a corresponding standard data source. Extract the resource elasticity features corresponding to each edge cache node and the spatiotemporal correlation features of each request cluster from the standard data source to construct a corresponding two-dimensional feature set; Using the two-dimensional feature set as input, an improved temporal attention LSTM model is constructed and set as the resource demand prediction model.

3. The network cache data storage method based on edge computing according to claim 1, characterized in that, The step of creating a corresponding elastic resource pool based on the prediction results output by the resource demand prediction model includes: Extract the storage capacity and concurrency parameters from the prediction results, and synchronously collect the request fluctuation data of the regional edge gateway to generate a corresponding scenario-based demand list. With resource matching degree, transmission latency and energy consumption as optimization objectives, the node attributes of each edge cache node are virtualized into corresponding standard units through distributed proxy, and then aggregated to form a basic matrix of resource pool; Based on the scenario-based requirement list, a corresponding scaling threshold is set, and the resource pool base matrix is ​​pooled based on the scaling threshold to generate the corresponding elastic resource pool.

4. The network cache data storage method based on edge computing according to claim 1, characterized in that, The step of parsing the operation semantics and data dependencies of the cached request and constructing the corresponding semantic conflict interception matrix includes: Collect the cache request text corresponding to the cache request, and parse the cache request text using a pre-trained BERT model to extract the operation type, data identifier and permission level, and encode them into standard semantic vectors. Using data objects as nodes and dependencies as directed edges, a data dependency graph is constructed through a graph neural network; Several target elements contained within the standard semantic vector are extracted, and a time axis is introduced to construct a matrix framework. The data dependency graph and several target elements are then filled into the matrix framework to generate the semantic conflict interception matrix.

5. The network cache data storage method based on edge computing according to claim 4, characterized in that, The step of filling the data dependency graph and several target elements into the matrix frame to generate the semantic conflict interception matrix includes: Extract the node association strength and path depth features of the data dependency graph, combine them with the permission level features in the target element, calculate the dynamic weight of each feature through entropy weight-hierarchical analysis, and perform corresponding weighted fusion to generate the corresponding fusion feature vector. The system calls the mutual exclusion rule library to detect potential conflicts in the fused feature vector, generates corresponding conflict prediction coefficients, and collects the remaining resource parameters of the edge cache node. The conflict prediction coefficient and the remaining resource parameters are bound to the interior of the fused feature vector, and the bound fused feature vector is filled into the interior of the matrix frame to generate the semantic conflict interception matrix.

6. The network cache data storage method based on edge computing according to claim 1, characterized in that, The step of generating a corresponding dynamic execution sequence through the semantic conflict interception matrix when a data mutual exclusion operation request is detected includes: A spatiotemporal feature matrix is ​​constructed by combining the semantic conflict interception matrix with the location of the terminal access node and the request timestamp. The semantic conflict type, spatial distance and time interval in the data mutual exclusion operation request are extracted and spatiotemporally clustered using the DBSCAN algorithm to generate the corresponding conflict group. The requests within the conflict group are sorted to output the corresponding basic execution sequence; The basic execution sequence is adjusted collaboratively to generate the dynamic execution sequence.

7. The network cache data storage method based on edge computing according to claim 6, characterized in that, The step of performing collaborative fault-tolerant adjustment on the basic execution sequence to generate the dynamic execution sequence includes: Semantic conflict residual identifiers are extracted from the basic execution sequence, and combined with the spatiotemporal clustering degree of the terminal access node and the resource fluctuation coefficient of the edge cache node, a corresponding risk feature vector is constructed. The risk level of execution interruption and the probability of conflict propagation of the cached request are calculated based on the risk feature vector. The basic execution sequence is dynamically adjusted based on the execution interruption risk level and the conflict propagation probability to generate the dynamic execution sequence.

8. A network caching data storage system based on edge computing, applied to a multi-regional edge computing network containing regional edge gateways, edge caching nodes, and terminal access nodes, characterized in that, Includes the following: The acquisition module is used to acquire the location, network latency and cached request content of the terminal access node through the regional edge gateway, and to divide the regional edge gateways with high correlation of cached request content and close location into the same request cluster through density clustering algorithm; The construction module is used to collect storage, load and throughput data of each edge cache node, combine the historical cache data of the request cluster to construct a corresponding resource demand prediction model, and create a corresponding elastic resource pool based on the prediction results output by the resource demand prediction model. The parsing module is used to parse the operation semantics and data dependencies of the cache request when the regional edge gateway receives the cache request, and construct the corresponding semantic conflict interception matrix. The storage module is used to generate a corresponding dynamic execution sequence through the semantic conflict interception matrix when a data mutual exclusion operation request is detected, and to collect the target cache data corresponding to the dynamic execution sequence so as to store the target cache data in the internal elastic resource pool.

9. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the network cache data storage method based on edge computing as described in any one of claims 1 to 7.

10. A readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the edge computing-based network cache data storage method as described in any one of claims 1 to 7.

Images