Security system verification method in edge computing power secure channel

Abstract

The invention provides a security system verification method in an edge computing power secure channel, which comprises the following steps: acquiring computing power, storage resources and network condition information of each node in the edge computing power secure channel to obtain a node capability matrix which is a two-dimensional array, in which rows represent nodes and columns represent performance indexes; according to the node capability matrix, grouping the nodes by adopting a clustering algorithm, and dividing the nodes with similar computing capabilities, storage resources and network conditions into the same group; after each node receives the verification task and the evaluated delay prediction information, a time window of data synchronization is adjusted according to the delay prediction information in task metadata; and after each node completes the verification task, the verification result and the intermediate data are reported to the scheduling node, and the scheduling node compares and verifies the verification results reported by each node group to obtain a final consistent verification result.

Description

Technical Field

[0001] The present invention relates to the field of information technology, and in particular to a security system verification method in an edge computing power security channel. Background Art

[0002] When performing system verification in the edge computing power security channel, since the verification task usually requires multiple nodes to complete in collaboration, and the computing power, storage resources, and network conditions of different nodes vary greatly, a series of technical difficulties are caused. First, the communication delay problem between nodes seriously affects the efficiency of collaboration. When the network condition of a node is poor, the data transmission speed between it and other nodes will be greatly reduced, resulting in the progress of the entire verification process being dragged down. Secondly, the differences in computing power and storage resources of different nodes also bring challenges to data synchronization. Nodes with stronger computing power may complete their verification tasks faster, but they must wait for nodes with weaker computing power to complete their tasks before they can synchronize data, which leads to a decrease in resource utilization and an increase in verification time. Finally, there may be inconsistencies in the verification results generated by different nodes, which is mainly caused by differences in computing power and algorithm implementation of each node. In short, in the system verification process of the edge computing power security channel, due to the large differences in computing power, storage resources, and network conditions of different nodes, communication delays, data synchronization, consistency of verification results, and other problems frequently occur, which seriously affects the efficiency and accuracy of multi-node collaborative verification. In other words, in the current system verification of edge computing security channels, how to solve the problems caused by differences in computing power and network conditions between nodes while ensuring the accuracy of the verification results is a technical problem that needs to be solved urgently. Summary of the invention

[0003] The present invention provides a security system verification method in an edge computing power security channel, which mainly includes:

[0004] Obtain the computing power, storage resources, and network status information of each node in the edge computing power security channel to obtain the node capability matrix. The node capability matrix is ​​a two-dimensional array, in which rows represent nodes and columns represent performance indicators.

[0005] According to the node capability matrix, a clustering algorithm is used to group nodes, and nodes with similar computing capabilities, storage resources, and network conditions are divided into the same group;

[0006] Determine the task allocation plan within each node group based on the node capability matrix. By solving the multi-objective optimization model, minimize the capability difference between nodes in the group under the premise of meeting the verification time constraint to obtain the optimal task allocation plan. The multi-objective optimization model takes the node capability matrix as input and the task allocation plan as output.

[0007] According to the task allocation plan, the verification task is sent to each node. When sending the task, the communication delay that may occur during the task execution is predicted based on the node's network status. The prediction result is sent to the node together with the task as part of the task metadata.

[0008] After receiving the verification task and the evaluated delay prediction information, each node adjusts the time window for data synchronization according to the delay prediction information in the task metadata;

[0009] During the execution of the verification task, the execution progress and resource utilization of each node are monitored in real time. When it is found that the execution progress of a node lags behind the average progress by more than the preset threshold, or its resource utilization is lower than the preset threshold, some tasks are dispatched from other nodes to this node to balance the load and progress of each node.

[0010] After each node completes the verification task, it reports the verification results and intermediate data to the scheduling node. The scheduling node compares and verifies the verification results reported by each node group to obtain the final consistent verification results.

[0011] The scheduling node analyzes the final consistent verification results, generates a verification report, and feeds back the verification report and final results to each node. If the consistency index is lower than the preset threshold, the data synchronization operation is triggered. Each node synchronizes and updates the local intermediate data according to the final verification results sent by the scheduling node.

[0012] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0013] The present invention discloses a method for verifying a security system in an edge computing power security channel. The method first obtains the computing, storage and network capability information of each node and constructs a node capability matrix. Then, a clustering algorithm is used to group nodes with similar capabilities, and a multi-objective optimization model is solved for each group of nodes to obtain an optimal task allocation scheme. When issuing tasks, the present invention predicts possible communication delays and issues them as metadata, and each node dynamically adjusts the data synchronization time window accordingly. During the execution process, the present invention monitors the node progress and resource utilization in real time, and triggers dynamic scheduling to balance the load when necessary. After the verification is completed, a weighted majority voting mechanism is used to fuse the results of each node to generate a final consistent verification result and report. If the consistency is insufficient, an incremental data synchronization operation is triggered. The present invention significantly improves the execution efficiency and reliability of distributed verification tasks through accurate modeling of node capabilities, intelligent allocation of tasks, dynamic adjustment of the execution process, and consistency guarantee of results, and provides an effective solution for security verification in an edge computing environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1This is a flow chart of a security system verification method in an edge computing power security channel of the present invention.

[0015] Figure 2 A schematic diagram of a security system verification method in an edge computing power security channel of the present invention.

[0016] Figure 3 This is another schematic diagram of a security system verification method in an edge computing power security channel of the present invention. DETAILED DESCRIPTION

[0017] In order to further understand the content of the present invention, the present invention is described in detail in conjunction with the accompanying drawings and embodiments. The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are only used to explain the relevant inventions, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.

[0018] like Figure 1-3 In this embodiment, a security system verification method in an edge computing power security channel may specifically include:

[0019] Step S101, obtain the computing power, storage resources and network status information of each node in the edge computing power security channel to obtain a node capability matrix. The node capability matrix is ​​a two-dimensional array, in which rows represent nodes and columns represent performance indicators.

[0020] Obtain real-time computing power data, storage resource usage, and network status parameters of each node in the edge computing security channel. Real-time computing power data includes processor utilization, memory occupancy, and disk I / O rate. Clean and standardize the acquired data to obtain valid data. Organize the valid data to form a node capability matrix, which includes the computing power indicators, storage capacity, network bandwidth, and delay parameters of each node.

[0021] Specifically, real-time computing power data, storage resource usage and network status parameters are obtained from each node in the edge computing power security channel, the collected raw data is cleaned and standardized, the 3σ principle is used to identify outliers, invalid data is identified through data integrity check, and the processed valid data is organized into a node capability matrix in the form of a two-dimensional array according to a predefined format, and the data is encrypted and transmitted using the RSA encryption algorithm to ensure data security. For each node in the node capability matrix, key performance indicators such as computing power indicators, storage capacity, network bandwidth and delay parameters are extracted, the weight coefficient of each indicator is determined by the hierarchical analysis method, and the fuzzy comprehensive evaluation method is used to quantify and score each indicator. The comprehensive performance score of the node is obtained by weighted summation, and the security level of the node is considered as a dimension of the score. When obtaining node data from the edge computing power security channel, a distributed data collector is used to collect information from multiple nodes at the same time. For example, for a network containing 100 edge nodes, each node generates 1KB of performance data per second, and the data collector collects data every 5 seconds. When using the 3σ principle to identify outliers, the mean value μ and standard deviation σ of CPU usage are calculated. If the CPU usage of a node exceeds the range of μ±3σ, it is considered an outlier. Data integrity check is verified by checksum algorithm to ensure that the data has not been tampered with during transmission. RSA encryption uses a 2048-bit key to ensure data transmission security. In node performance evaluation, the hierarchical analysis method is used to construct a judgment matrix and calculate the eigenvector to determine the weight. Assume that the weights of computing power, storage capacity, network bandwidth, and delay parameters are 0.4, 0.3, 0.2, and 0.1 respectively. In the fuzzy comprehensive evaluation, five evaluation levels are set as very poor, poor, general, good, and very good, and are assigned 1-5 points respectively. (Weight × fuzzy relationship matrix) = evaluation result vector, and the level corresponding to the maximum membership is taken as the final evaluation result. The node security level is evaluated based on factors such as encryption strength and access control strategy, and is divided into three levels: low, medium, and high, which are combined with the performance score to form a comprehensive score. The K-means clustering algorithm dynamically groups nodes, selects the initial cluster center, calculates the Euclidean distance from each node to the center, and iteratively updates the cluster center until convergence. Set two clusters, and the corresponding performance threshold is the dividing line of the cluster. For 100 nodes, after 10 iterations, 40 high-performance node groups and 60 ordinary node groups are obtained. Combine the security level 1-3 points with the performance level 1-5 points to generate a 3×5 combination matrix to achieve more detailed node classification.

[0022] Step S102: according to the node capability matrix, a clustering algorithm is used to group the nodes, and nodes with similar computing capabilities, storage resources and network conditions are grouped into the same group.

[0023] Obtain the node capability matrix, and extract the node's CPU core number, memory capacity, disk space, network bandwidth, and latency characteristic indicators from the node capability matrix. For the characteristic indicators, the minimum and maximum normalization method is used to process the characteristic indicators to obtain the standardized node feature vector. According to the standardized node feature vector, the initial cluster center is selected, and clustering is performed by calculating the Euclidean distance between nodes as the similarity measurement indicator. If the clustering result does not meet the preset conditions, the elbow rule is used to determine the optimal cluster number K; obtain the clustering result, calculate the silhouette coefficient to evaluate the clustering quality, and if the silhouette coefficient is lower than the preset threshold, adjust the cluster number K or the feature weight; wherein, the feature weight adjustment method is: increase the feature weight of the large difference between groups, and reduce the feature weight of the small difference between groups. Repeat the clustering algorithm until the silhouette coefficient is greater than the preset threshold or reaches the preset maximum number of attempts, and obtain the node grouping result that meets the conditions.

[0024] Specifically, the characteristic indicators such as the number of CPU cores, memory capacity, disk space, network bandwidth and latency of each node are extracted from the node capability matrix. The extracted characteristic data are processed by the minimum-maximum normalization method, and each eigenvalue xi is mapped to the interval [0,1]. The calculation formula is xi_normalized=(xi-min(x)) / (max(x)-min(x)), where x is the eigenvector, and the standardized node eigenvector is obtained. min(x) represents the minimum value in the eigenvector x, max(x) represents the maximum value in the eigenvector x, and xi_normalized is the new eigenvalue after normalization. The K-means++ algorithm is used to select the initial cluster center. By calculating the Euclidean distance between nodes as the similarity measurement indicator, each node is assigned to the group to which the nearest cluster center belongs. The elbow rule is used to determine the optimal number of clusters K. The intra-group sum of squares (WSS) is calculated for different K values, and the K-WSS curve is drawn. The K value corresponding to the inflection point of the curve is selected as the final number of clusters. According to the current grouping result, the cluster center of each group is recalculated and updated to the mean of the feature vectors of all nodes in the group, and then the nodes are redistributed. The maximum number of iterations is set to 100. When the moving distance of the cluster center is less than 0.001 or the maximum number of iterations is reached, the iteration is stopped to obtain a stable clustering result. The silhouette coefficient is calculated to evaluate the clustering quality. The silhouette coefficient range is [-1,1]. The larger the value, the better the clustering effect. If the silhouette coefficient is lower than the preset threshold of 0.5, it is optimized by adjusting the number of clusters K or the feature weight. The feature weight adjustment method is to increase the feature weight of the feature with large differences between groups and reduce the feature weight of the feature with small differences between groups. Repeat the clustering algorithm until the silhouette coefficient is greater than 0.5 or the preset maximum number of attempts is reached, and finally the node grouping result that meets the conditions is obtained. In practical applications, it is assumed that there are 100 edge nodes, and the capacity matrix of each node includes five indicators: the number of CPU cores, memory capacity, disk space, network bandwidth, and latency. After data extraction, each indicator is normalized to the minimum and maximum values. For example, if the original number of CPU cores of a node is 8, the minimum value is 2, and the maximum value is 16, the normalized value is (8-2) / (16-2)=0.43. All indicators are processed similarly to obtain standardized feature vectors. Then the K-means++ algorithm is used to select the initial cluster center, and the similarity between each node and the cluster center is determined by calculating the Euclidean distance. To determine the optimal number of clusters K, the elbow rule is used to calculate different K values, such as the within-group sum of squares (WSS) under 2 to 10, and the K-WSS curve is drawn. Assuming that there is an obvious inflection point when K=5, 5 is selected as the final number of clusters. The clustering iteration process is carried out, and the cluster center is updated and the nodes are redistributed in each iteration. The maximum number of iterations is set to 100, and it stops when the moving distance of the cluster center is less than 0.001 or the maximum number of iterations is reached.Assume that at the 47th iteration, the moving distance of the cluster center drops to 0.0009, then the iteration is stopped and 5 node groups are obtained. The silhouette coefficient is calculated for the clustering result. If the value obtained is 0.68, which is higher than the preset threshold of 0.5, the clustering result is accepted. If the silhouette coefficient is lower than 0.5, the feature weight is adjusted, for example, the weight of the number of CPU cores and memory capacity is increased from the original 1 to 1.2, and the clustering algorithm is re-executed. After 3 adjustments, assuming that the silhouette coefficient is increased to 0.72, the final node grouping result is obtained, which includes 5 node groups with similar performance, with 18, 22, 25, 20, and 15 nodes in each group. In this way, data standardization, clustering analysis and result optimization achieve effective grouping of edge nodes.

[0025] Step S103, based on the node capability matrix, determine the task allocation plan within each node group, and by solving the multi-objective optimization model, minimize the capability difference between nodes in the group while satisfying the verification time constraint to obtain the optimal task allocation plan. The multi-objective optimization model takes the node capability matrix as input and the task allocation plan as output.

[0026] The objective function of the multi-objective optimization model is constructed based on the computing power, storage resources and network status data in the node capability matrix. The objective function includes two goals: minimizing the capability difference between nodes in the group and satisfying the verification time constraint. The genetic algorithm is used to solve the multi-objective optimization model, and the task allocation plan is obtained by initializing the chromosome population. The chromosome population uses integer coding to represent the nodes to which the task is assigned. The non-dominated solution set obtained by the genetic algorithm is screened, and the non-inferior solution is selected using the concept of Pareto optimality, and the weight of each sub-objective in the objective function is determined by the hierarchical analysis method. The obtained optimal task allocation plan is applied to the node group, and the optimal task allocation plan is determined by the non-inferior solution with the highest comprehensive score. Tasks are assigned to specific nodes according to the optimal task allocation plan, and the computing, storage and network resources occupied by each task are calculated, and the resource occupancy in the node capability matrix is ​​updated. The task completion time and resource usage are recorded and compared with the estimated value. If there is a difference between the actual execution and the estimated value, the task allocation weight is adjusted.

[0027] Specifically, according to the computing power, storage resources and network status data in the node capacity matrix, the objective function of the multi-objective optimization model is constructed, including minimizing the capacity difference between nodes in the group and satisfying the verification time constraint. The standard deviation is used to measure the capacity difference between nodes, and the maximum allowed verification time is set as a constraint. At the same time, the constraints of task allocation are set, such as each task can only be assigned to one node, node resources cannot be overloaded, etc. The objective function expression is f = w1*σ(capacity)+w2*max(0,t-T_max), where σ(capacity) is the standard deviation of node capacity, t is the actual verification time, T_max is the maximum allowed verification time, and w1 and w2 are the corresponding weight coefficients. max(0,t-Tmax) means that if the actual verification time t exceeds the maximum allowed verification time Tmax, the excess part will be included in the objective function; if it does not exceed, the contribution of this part is zero. The constructed multi-objective optimization model is solved by genetic algorithm, the chromosome population is initialized, and integer coding is used to indicate which node the task is assigned to. Each chromosome represents a possible task allocation scheme, and a new population is generated through operations such as crossover and mutation. The fitness function is designed to be the inverse of the objective function. The higher the fitness, the better the solution. Iterate the optimization until the preset termination condition is reached, such as reaching the maximum number of iterations of 1000 or the optimal solution has not been improved for 50 consecutive generations. The non-dominated solution set obtained by the genetic algorithm is screened, and the non-inferior solution is selected using the concept of Pareto optimality. The analytic hierarchy process is used to determine the weights of each sub-goal in the objective function, construct a judgment matrix, and calculate the eigenvector to obtain the weight vector. The comprehensive score of each non-inferior solution is calculated, and the solution with the highest comprehensive score is selected as the final task allocation solution. The optimal task allocation solution is applied to the node group and tasks are assigned to specific nodes. The computing, storage, and network resources occupied by each task are calculated, and the resource occupation in the node capacity matrix is ​​updated. The specific update process includes the remaining number of CPU cores = the original number of CPU cores - the number of cores occupied by the task, the remaining memory = the original memory - the memory occupied by the task, and the remaining bandwidth = the original bandwidth - the bandwidth occupied by the task. According to the actual execution situation, the task completion time and resource usage are recorded, compared with the estimated value, and the task allocation weight is adjusted to provide accurate node status information and optimization basis for subsequent task allocation. In practical applications, assume that a node group contains 10 nodes, and the capacity matrix of each node contains four indicators: number of CPU cores, memory capacity, disk space, and network bandwidth. When constructing a multi-objective optimization model, the weight w1 of the capacity difference is set to 0.6, the weight w2 of the verification time constraint is set to 0.4, and the maximum allowed verification time T_max is 100 seconds. When using the genetic algorithm to solve, 100 chromosomes are initialized, and each chromosome is encoded with a 10-bit integer to indicate which node the 10 tasks are assigned to. Set the crossover probability to 0.8, the mutation probability to 0.1, and the maximum number of iterations to 1000 generations.In each generation, the fitness of each chromosome is calculated, and the fitness function is 1 / (0.6*σ(capacity)+0.4*max(0,t-100)). After 950 iterations, the improvement of the optimal solution is less than 0.1% for 50 consecutive generations, and the algorithm terminates. 20 non-dominated solutions are obtained, and the weights are determined by applying the hierarchical analysis method. A 3×3 judgment matrix is ​​constructed, and the eigenvector is calculated to obtain the weight vector [0.5, 0.3, 0.2]. The comprehensive scores of the 20 non-inferior solutions are calculated, and the solution with the highest score of 89.7 is selected as the final solution. The solution is applied to the node group, such as assigning task 1 to node 3 and task 2 to node 7. The remaining resources of node 3 are updated, and the number of CPU cores is reduced from 8 to 6, the memory is reduced from 16GB to 12GB, and the bandwidth is reduced from 1Gbps to 800Mbps. The actual completion time of task 1 is recorded as 85 seconds, which is 5 seconds earlier than the estimated 90 seconds. This information is used to optimize the subsequent task allocation, such as slightly increasing the weight of node 3.

[0028] Step S104, according to the task allocation plan, the verification task is sent to each node; while sending the task, the communication delay that may occur during the task execution is predicted according to the network status of the node, and the prediction result is sent to the node together with the task as part of the task metadata.

[0029] A verification task to be executed is obtained from the task queue, and the verification task includes a target node identifier; network status information of the target node is collected according to the target node identifier, and the network status information includes network bandwidth utilization, packet loss rate, current load and historical communication delay data; for the historical communication delay data, an ARIMA model is used to generate a communication delay prediction value of the target node, and the communication delay prediction value represents the expected delay in the future; the communication delay prediction value is obtained, and the communication delay prediction value is encapsulated into the metadata of the verification task to obtain a complete task package, which includes a task description, execution parameters and a communication delay prediction value; the complete task package is sent to the target node through a secure channel, and the target node is triggered to start a task execution monitoring program, and the task execution monitoring program is used to monitor the execution status of the verification task; if the task execution monitoring program detects that the verification task starts to be executed, the actual communication delay of the verification task is collected, and the actual communication delay is compared with the communication delay prediction value; if the actual communication delay exceeds the preset percentage of the communication delay prediction value, the timeout time of the verification task is adjusted.

[0030] Specifically, according to the task allocation scheme, the verification tasks to be executed are extracted from the task queue, and the network status information of the target node is regularly collected using the SNMP protocol, including network bandwidth utilization, packet loss rate, current load and historical communication delay data, as input parameters for communication delay prediction. The collection frequency is set to once every 30 seconds to ensure the real-time and accuracy of the data. The ARIMA (Autoregressive Integrated Moving Average) model is used to predict the communication delay of the target node. By analyzing the periodicity and trend of historical delay data and combining the current network status, the delay prediction value within the next hour is generated. The AIC (Akaike Information Criterion) is used to determine the optimal parameters of the ARIMA model, including autoregressive terms, difference order and moving average terms. The predicted communication delay information is encapsulated into the task metadata to form a complete task package containing task description, execution parameters and predicted delay. The UUID (Universally Unique Identifier) ​​algorithm is used to generate a unique 128-bit identifier for each task package. At the same time, a task execution monitoring program is prepared, and the monitoring interval is set to 10 seconds for subsequent real-time tracking of task execution status. The task package is sent to the target node through a secure channel, and the task execution monitoring program is started. The monitoring program regularly collects the task execution status, calculates the actual communication delay, and compares it with the predicted value. If the actual delay exceeds the predicted value by 20%, the task timeout is dynamically adjusted. According to the predicted delay, the data transmission strategy is optimized, such as reducing the packet size or increasing the number of retransmissions during high delay. The comparison results of the actual delay and the predicted delay are recorded in the log for subsequent parameter optimization and prediction accuracy improvement of the ARIMA model. In practical applications, assume that there is an edge computing cluster with 10 nodes, and each node needs to perform a verification task. The system first uses the SNMP protocol to collect the network status of the node every 30 seconds, such as the bandwidth utilization of node 1 is 75%, the packet loss rate is 0.5%, and the current load is 80%. Based on the historical data of the past 24 hours, the ARIMA model is used to predict the communication delay in the next hour. Through AIC evaluation, the optimal ARIMA parameters are determined to be (2,1,1), that is, 2nd order autoregression, 1st order difference, and 1st order moving average. The model predicts that the average communication delay of node 1 in the next hour is 150ms, with a fluctuation range of ±20ms. The system then generates a UUID "550e8400-e29b-41d4-a716-446655440000" for a verification task, and encapsulates the task description, execution parameters, and predicted delay of 150ms into the task package. The task execution monitor is set to collect status data every 10 seconds. When the task is sent to node 1 through the secure channel, the monitor starts working. After 30 minutes of execution, it is found that the actual average delay is 185ms, which exceeds the predicted value by 23%. The system automatically extends the task timeout from the original 5 minutes to 6 minutes, and reduces the data packet size from 1MB to 800KB.These adjustments are recorded in the log and used to subsequently optimize the ARIMA model parameters and improve the prediction accuracy. Through this dynamic adjustment mechanism, the system successfully coped with changes in network conditions and ensured the smooth completion of the verification task.

[0031] Step S105, after receiving the verification task and the evaluated delay prediction information, each node adjusts the time window for data synchronization according to the delay prediction information in the task metadata.

[0032] Obtain the delay prediction information in the task metadata, the delay prediction information includes the predicted delay value and the preset communication delay threshold; determine whether to adjust the time window for data synchronization based on the size relationship between the predicted delay value and the preset communication delay threshold, if the predicted delay value is greater than the preset communication delay threshold, increase the time window, if the predicted delay value is less than the preset communication delay threshold, reduce the time window; use a PID controller to adjust the time window size, the proportional coefficient, integral coefficient and differential coefficient of the PID controller are preset values; update the data exchange frequency and batch size between nodes based on the adjusted time window, the data exchange frequency is inversely proportional to the time window, and the batch size is proportional to the time window; compare the actual delay with the predicted delay value, calculate the prediction error, if the prediction error exceeds the preset error threshold, trigger the time window to be adjusted again; the prediction error uses the mean absolute percentage error indicator.

[0033] Specifically, JSON parsing is used to extract delay prediction information from the received task metadata to obtain the predicted delay value and the preset communication delay threshold of the current node. By comparing the size relationship between the predicted delay and the threshold, it is determined whether the time window for data synchronization needs to be adjusted. If the predicted delay is greater than the threshold, the time window is increased; if it is less than the threshold, the time window is reduced. According to the comparison result between the delay prediction information and the preset threshold, the PID (proportional-integral-differential) controller is used to dynamically adjust the time window size. The proportional coefficient Kp of the PID controller is set to 0.5, the integral coefficient Ki is set to 0.1, and the differential coefficient Kd is set to 0.05. The adjustment amplitude of the time window is calculated to make it proportional to the predicted communication delay. The upper limit of the time window is set to 1 second and the lower limit is set to 0.1 second to prevent the window from being too large or too small. The adjusted time window is applied to the data synchronization mechanism to update the data exchange frequency and batch size between nodes. The data exchange frequency is inversely proportional to the time window, and the batch size is proportional to the time window. For example, if the time window increases from 0.5 seconds to 0.8 seconds, the data exchange frequency decreases from 2 times per second to 1.25 times per second, and the batch size increases from 500KB to 800KB. Start the real-time monitoring program and record the delay in the actual synchronization process every 100 milliseconds for subsequent prediction model optimization and immediate adjustment. Periodically compare the actual delay with the predicted delay to calculate the prediction error. Use the mean absolute percentage error (MAPE) as the error indicator and set the error threshold to 10%. When the MAPE exceeds 10%, the time window is adjusted again, and the adjustment result is fed back to the central scheduler to update the global delay prediction model. At the same time, monitor the changes in network conditions. If the network throughput or packet loss rate changes by more than 20%, the time window adjustment is triggered immediately without waiting for the next periodic check. In actual applications, an edge node receives a verification task whose metadata contains the delay prediction information {"predicted_delay":150} in JSON format. The node uses the JSON parser to extract the predicted delay value of 150ms and reads the preset communication delay threshold of 100ms. Since the predicted delay is greater than the threshold, the system decides to increase the time window for data synchronization. The PID controller calculates the adjustment amount based on the current time window of 0.5 seconds and the predicted delay of 150ms. Assume Kp = 0.5, Ki = 0.1, Kd = 0.05, error e = 150-100 = 50ms, integral error Σe = 200ms (assuming previous accumulation), differential error Δe = 30ms (compared with the last time). PID output = 0.550 + 0.1200 + 0.05*30 = 45ms. The new time window is set to 0.5 + 0.045 = 0.545 seconds, which is within the limit of 0.1 to 1 second. The system then adjusted the data exchange frequency from 2 times per second to 1.83 times per second (1 / 0.545), and the batch size increased from 500KB to 545KB.The real-time monitoring program is started, and the actual delay is recorded every 100ms. After 30 seconds, the system calculates the MAPE and finds that the actual average delay is 140ms. The MAPE is (|140-150| / 150)*100%=6.67%, which is lower than the 10% threshold and does not require immediate adjustment. However, the monitoring program detects that the network throughput suddenly drops by 25%, triggering an immediate adjustment. The system runs the PID controller again, further increasing the time window to 0.7 seconds, correspondingly reducing the frequency of data exchange and increasing the batch size. These adjustment results are sent back to the central scheduler for optimizing the global delay prediction model. Through this dynamic adaptation mechanism, the system can effectively respond to changes in network conditions and maintain the efficiency and reliability of data synchronization.

[0034] Step S106, during the execution of the verification task, the execution progress and resource utilization of each node are monitored in real time; when it is found that the execution progress of a node lags behind the average progress by more than a preset threshold, or its resource utilization is lower than a preset threshold, some tasks are scheduled from other nodes to this node to balance the load and progress of each node.

[0035] Collect node execution progress and resource utilization information, calculate the average execution progress and resource utilization, and obtain the average execution progress and average resource utilization of all nodes as the benchmark value. Use the time series prediction method to analyze the benchmark value and predict the future node load. Use the moving average method to calculate the dynamic threshold, and use the dynamic threshold to compare the difference between the actual execution progress and resource utilization of each node and the benchmark value. If the node execution progress lags behind the average progress by more than the dynamic threshold or the resource utilization is lower than the dynamic threshold, the node is marked as a waiting state. For the nodes in the waiting state, select the source node from the nodes with the preset percentage of execution progress ranking. Calculate the amount of tasks that can be migrated and generate a task migration plan containing the ID of the task to be migrated, the source node and the target node information. Execute the task migration plan and use the two-phase commit protocol to ensure the transfer of task status. Pause the relevant tasks on the source node and transfer the task status and data to the target node. After the target node confirms that the reception is complete, update the task allocation information in the distributed database and resume task execution on the target node.

[0036] Specifically, a real-time monitoring mechanism is established to collect information such as the execution progress, CPU usage, memory usage, and network throughput of each node every 10 seconds. The collected data is stored in a distributed database, and the average execution progress and average resource utilization of all nodes are calculated as the benchmark value. The time series prediction algorithm is used to predict the node load in the next 30 minutes based on historical data to provide a basis for subsequent task scheduling. The dynamic threshold is calculated using the moving average method to compare the actual execution progress and resource utilization of each node with the benchmark value. If the execution progress of a node lags behind the average progress by more than the dynamic threshold, such as 20%, or the resource utilization is lower than the dynamic threshold, such as 50%, the node is marked as a state to be scheduled. At the same time, combined with the load prediction results, potential performance bottleneck nodes are identified. For nodes in the state to be scheduled, a suitable source node is selected from the top 30% of the nodes in terms of execution progress. The task splitting method is used to calculate the amount of tasks that can be migrated to ensure that the load of the source node and the target node tends to be balanced after migration. A task migration plan is generated, including the task ID to be migrated, the source node and target node information, and the estimated migration time and resource consumption. The feasibility of the migration scheme is verified and the impact on the overall system performance is evaluated. The task migration operation is performed and the two-phase commit protocol is used to ensure the atomic transfer of the task status. First, the relevant tasks on the source node are suspended and the task status and data are transferred to the target node. After the target node confirms that the reception is complete, the task allocation information in the distributed database is updated and the task execution is resumed on the target node. The execution progress and resource utilization of each node are recalculated and the load prediction model is updated. The performance of the system after migration is continuously monitored. If an abnormality occurs, the rollback mechanism is triggered to restore the task to the original node. In a distributed verification system containing 100 edge nodes, the real-time monitoring mechanism collects node data every 10 seconds. Taking node A as an example, a certain collection shows that its CPU usage is 75%, memory usage is 85%, network throughput is 200Mbps, and execution progress is 60%. The system calculates that the average execution progress of all nodes is 70% and the average CPU usage is 65%. The ARIMA model is used to predict that the CPU usage of node A will reach 90% in the next 30 minutes. The dynamic threshold calculation uses the moving average method of a 5-minute window, and the progress lag threshold is 15%, and the resource utilization lower limit threshold is 55%. Since the execution progress of node A is 20% behind (70%-60%>15%), the system marks it as a waiting-for-scheduling state. From the top 30 nodes in terms of execution progress, select node B with the lightest load as the source node. The task splitting algorithm calculates that the amount of tasks that can be migrated from B to A is 20% of the total load. Generating a migration plan includes migrating task IDs 1001-1005 from B to A, and the estimated migration time is 30 seconds. The execution of a two-phase commit includes the first phase, in which B suspends these tasks and sends the status to A; in the second phase, after A confirms that it has received the full information, it updates the task allocation information in the distributed database.After the migration is complete, the CPU usage of A drops to 70% and the execution progress increases to 65%. The system is continuously monitored for 30 minutes and the performance is found to be stable, so there is no need to trigger the rollback mechanism.

[0037] Step S107, after each node completes the verification task, it reports the verification result and intermediate data to the scheduling node; the scheduling node compares and verifies the verification results reported by each node group to obtain a final consistent verification result.

[0038] Receive the formatted data packets reported by each verification node, which contain verification results, key intermediate data, node identification and timestamp; parse the verification results and intermediate data according to the data packets, use cosine similarity to calculate the similarity of the verification results, and detect abnormal results through the local anomaly factor algorithm; for the verification nodes, use the hierarchical analysis method to construct a judgment matrix based on the historical credibility of the nodes, resource utilization and task completion time, and calculate the weight of each node; use the node weight, regard the verification results of each node as the source of evidence, calculate the basic probability distribution function, and iteratively fuse all node results; after obtaining the fused verification results, calculate the information entropy of the fusion results, generate a result credibility assessment report, store the fusion results and credibility report in the distributed database, and update the historical credibility records of the verification nodes at the same time.

[0039] Specifically, after each node completes the verification task, it packages the verification results and key intermediate data, including important parameters and timestamps in the verification process, into a JSON format data packet, reports it to the scheduling node through a secure communication protocol, and records the reporting timestamp and node identification information. The JSON data packet structure contains four main fields: verification results, intermediate data, node ID, and timestamp. After receiving the reported data packet, the scheduling node parses the JSON content and extracts the verification results and intermediate data. The cosine similarity is used to calculate the similarity of the result vector, and a threshold of 0.95 is set to judge the consistency. The cosine similarity between the results of each node and the average result is calculated to obtain the result difference. At the same time, the local anomaly factor algorithm is used to detect the verification results that are obviously deviated, and the results with anomalies exceeding the preset threshold are marked as potential anomalies. According to factors such as the historical credibility of the node, resource utilization, and completion time of the verification task, the hierarchical analysis method is used to assign weights to each node. First, a hierarchical structure is established, with node credibility, resource utilization, and task completion time as three criterion layers. Then a judgment matrix is ​​constructed, and the criteria are compared pairwise using a 9-point scale. The eigenvector of the judgment matrix is ​​calculated to obtain the weight of each criterion. Finally, the comprehensive weight is calculated according to the performance of each node under each criterion. The constructed weight matrix is ​​used to perform weighted fusion of the verification results of each node using the Dempster-Shafer evidence theory. The verification result of each node is regarded as an evidence source, and the basic probability distribution function is calculated according to the node weight. The results of all nodes are iteratively fused through the Dempster combination rule to obtain the final comprehensive verification result. The credibility of the result is calculated using information entropy, and a credibility assessment report of the result is generated. The fused verification results and credibility report are stored in a distributed database, and the historical credibility records of each node are updated at the same time. In a distributed verification system containing 50 edge nodes, after node A completes the verification task, it generates a JSON data packet. After the scheduling node receives the data packets from all nodes, it calculates the cosine similarity of the result vector. The similarity between the result of node A and the average result is 0.98, which is higher than the threshold of 0.95 and is judged to be consistent. The local anomaly factor algorithm detects that the abnormality of the result of node B is 2.5, which exceeds the preset threshold of 2.0, and marks it as a potential anomaly. The judgment matrix was constructed by the hierarchical analysis method, and the weights of node credibility, resource utilization and task completion time were 0.5, 0.3 and 0.2 respectively. The scores of node A under these three criteria were 0.9, 0.8 and 0.7, and its comprehensive weight was calculated to be 0.85. The verification results of node A [0.95, 0.87, 1.02] were taken as an evidence source using the Dempster-Shafer theory fusion results, and the basic probability distribution was m({0.95, 0.87, 1.02}) = 0.85. All node results were iteratively fused through the Dempster combination rule to obtain the final verification result [0.96, 0.89, 1.01].The information entropy is calculated and the credibility of the result is 0.92. The result and credibility report are stored in the distributed database, and the historical credibility of node A is updated from 0.88 to 0.90.

[0040] In step S108, the scheduling node analyzes the final consistent verification results, generates a verification report, and feeds back the verification report and the final results to each node; if the consistency index is lower than the preset threshold, the data synchronization operation is triggered, and each node synchronizes and updates the local intermediate data according to the final verification results sent by the scheduling node.

[0041] After receiving the multi-node verification results, the scheduling node calculates the consistency index. A verification report is generated based on the consistency index. The verification report is distributed to each node through the secure channel established by the TLS protocol. After receiving the verification report, each node returns a digital signature confirmation message containing a unique identifier. It is determined whether the consistency index is lower than the preset threshold. If it is lower than the preset threshold, the data synchronization operation is triggered. The data synchronization operation includes generating a synchronization instruction packet, which contains the operation type, data item identifier and update value; sending the synchronization instruction packet to each node. After receiving the synchronization instruction packet, each node compares the local data with the data in the instruction packet. According to the comparison result, the changed data part is updated. The checksum of the updated data is calculated to obtain the synchronization status and checksum. The synchronization status and checksum are sent to the scheduling node. The scheduling node receives the synchronization status and checksum, and determines whether it matches. If it does not match, the retry mechanism is triggered.

[0042] Specifically, the scheduling node conducts an in-depth analysis of the final consistent verification results and calculates the consistency index. The standard deviation is used to calculate the result deviation between nodes. The standard deviation of the result deviation is less than 0.1, which is considered highly consistent. The coefficient of variation is used to evaluate the consistency of the verification time. The coefficient of variation is less than 0.2, indicating good time consistency. The resource utilization efficiency is measured by calculating the mean of the CPU utilization of each node. The mean above 80% is considered high efficiency. A detailed verification report containing data tables and line charts is generated based on these quantitative indicators. The generated verification report and the final result are distributed to each node through the secure channel established by the TLS protocol. During the distribution process, a unique identifier is assigned to each data packet. After receiving it, the node needs to return a digital signature confirmation information containing the identifier. The scheduling node records the distribution timestamp and the reception confirmation information to ensure that all nodes receive the same results and reports. If the confirmation is not received within the preset timeout, the retransmission mechanism is triggered. The calculated consistency index is compared with the preset threshold. If it is lower than the threshold, the data synchronization operation is triggered. By comparing the final result with the original result of each node, the data items with significant differences are identified. A synchronization instruction packet containing the final verification result and the identifier of the intermediate data item that needs to be updated is generated. The synchronization instruction package adopts JSON format, which contains three main fields: operation type, data item identifier, and update value. After receiving the synchronization instruction package, each node uses the Merkle tree algorithm to quickly compare the local data with the data in the instruction package. According to the comparison results, only the changed data part is transmitted and updated to achieve incremental synchronization. After the update is completed, the node uses the SHA-256 algorithm to calculate the checksum of the updated data and report the synchronization status and checksum to the scheduling node. If the checksum does not match or an error occurs during the synchronization process, the retry mechanism is triggered, and a maximum of 3 retries are made. After 3 failures, the scheduling node marks the node as abnormal and notifies the administrator for manual intervention. In a distributed verification system containing 100 edge nodes, after the scheduling node receives the final verification results [0.95, 1.02, 0.98], the standard deviation of the result deviation between the calculation nodes is 0.029, which is less than the 0.1 threshold, indicating high consistency. The coefficient of variation of the verification time is 0.15, which is less than the 0.2 threshold, showing good time consistency. The average CPU utilization is 85%, which is higher than the 80% threshold, reflecting high efficiency. The scheduling node generates a verification report containing these indicators and distributes it to each node through a secure channel using the TLS1.3 protocol. Each data packet is assigned a unique identifier, such as "VR20240327001". After receiving the data packet, node A signs the identifier with a private key and returns a confirmation message. The scheduling node receives confirmation from 98 nodes within 30 seconds and retransmits data to the two unresponsive nodes. The comparison finds that the consistency indicator 0.89 is lower than the preset threshold of 0.95, triggering a data synchronization operation.By comparing, it is found that there is a significant difference between the original result [0.94, 1.05, 0.97] of node B and the final result, and a synchronization instruction package in JSON format is generated:

[0043] {"operation":"update","dataItemId":"B001","newValue":[0.95,1.02,0.98]}. After receiving the instruction packet, node B uses a Merkle tree with a depth of 4 to compare data and identify the leaf nodes that need to be updated. After completing the incremental update, node B calculates the SHA-256 checksum "8a7b5c3d..." and reports it to the scheduling node. The scheduling node verifies that the checksum matches and confirms that the synchronization is successful.

[0044] The above embodiment is only one of the preferred implementation modes of the present invention and should not be used to limit the protection scope of the present invention. Any changes or modifications that are made to the main design concept and spirit of the present invention and have no substantive significance, and the technical problems they solve are still consistent with the present invention, should be included in the protection scope of the present invention.

Claims

1. A security system verification method in an edge computing power security channel, characterized in that: The method comprises: Obtain the computing power, storage resources, and network status information of each node in the edge computing power security channel to obtain the node capability matrix. The node capability matrix is ​​a two-dimensional array, in which rows represent nodes and columns represent performance indicators. According to the node capability matrix, a clustering algorithm is used to group nodes, and nodes with similar computing capabilities, storage resources, and network conditions are divided into the same group; Determine the task allocation plan within each node group based on the node capability matrix. By solving the multi-objective optimization model, minimize the capability difference between nodes in the group under the premise of meeting the verification time constraint to obtain the optimal task allocation plan. The multi-objective optimization model takes the node capability matrix as input and the task allocation plan as output. According to the task allocation plan, the verification task is sent to each node. When sending the task, the communication delay that may occur during the task execution is predicted based on the node's network status. The prediction result is sent to the node together with the task as part of the task metadata. After receiving the verification task and the evaluated delay prediction information, each node adjusts the time window for data synchronization according to the delay prediction information in the task metadata; During the execution of the verification task, the execution progress and resource utilization of each node are monitored in real time. When it is found that the execution progress of a node lags behind the average progress by more than the preset threshold, or its resource utilization is lower than the preset threshold, some tasks are dispatched from other nodes to this node to balance the load and progress of each node. After each node completes the verification task, it reports the verification results and intermediate data to the scheduling node. The scheduling node compares and verifies the verification results reported by each node group to obtain the final consistent verification results. The scheduling node analyzes the final consistent verification results, generates a verification report, and feeds back the verification report and final results to each node. If the consistency index is lower than the preset threshold, the data synchronization operation is triggered. Each node synchronizes and updates the local intermediate data according to the final verification results sent by the scheduling node.

2. The method according to claim 1, characterized in that The computing power, storage resources and network status information of each node in the edge computing power secure channel are obtained to obtain a node capability matrix, which is a two-dimensional array, in which rows represent nodes and columns represent performance indicators, including: Obtain real-time computing power data, storage resource usage, and network status parameters of each node in the edge computing power security channel. Real-time computing power data includes processor utilization, memory occupancy, and disk I / 0 rate; Clean and standardize the acquired data to obtain valid data; A node capability matrix is ​​formed based on the effective data organization, which includes the computing capability indicators, storage capacity, network bandwidth and delay parameters of each node.

3. The method according to claim 1, characterized in that The node capacity matrix is ​​used to group nodes using a clustering algorithm, and nodes with similar computing capabilities, storage resources, and network conditions are grouped into the same group, including: Obtain the node capability matrix, and extract the node's CPU core count, memory capacity, disk space, network bandwidth, and latency characteristic indicators from the node capability matrix; For the characteristic index, the minimum and maximum normalization method is used to process the characteristic index to obtain the standardized node characteristic vector; According to the standardized node feature vector, the initial cluster center is selected, and clustering is performed by calculating the Euclidean distance between nodes as a similarity measurement indicator; If the clustering result does not meet the preset conditions, the elbow rule is used to determine the optimal number of clusters K; Get the clustering results, calculate the silhouette coefficient to evaluate the clustering quality, and if the silhouette coefficient is lower than the preset threshold, adjust the number of clusters K or feature weights; Among them, the feature weight adjustment method is: increase the feature weight of the large difference between groups, and reduce the feature weight of the small difference between groups: The clustering algorithm is repeatedly executed until the silhouette coefficient is greater than a preset threshold or the preset maximum number of attempts is reached, and a node grouping result that meets the conditions is obtained.

4. The method according to claim 1, characterized in that: The task allocation scheme within each node group is determined based on the node capability matrix, and the capability difference between nodes within the group is minimized under the premise of satisfying the verification time constraint by solving the multi-objective optimization model to obtain the optimal task allocation scheme. The multi-objective optimization model takes the node capability matrix as input and the task allocation scheme as output, and includes: The objective function of the multi-objective optimization model is constructed based on the computing power, storage resources and network status data in the node capability matrix. The objective function includes minimizing the capability difference between nodes in the group and satisfying the verification time constraint. Genetic algorithm is used to solve the multi-objective optimization model, and the task allocation scheme is obtained by initializing the chromosome population. The chromosome population uses integer coding to represent the nodes of task allocation. The non-dominated solution set obtained by the genetic algorithm is screened, the non-inferior solution is selected using the concept of Pareto optimality, and the weight of each sub-objective in the objective function is determined by the hierarchical analysis method; Apply the obtained optimal task allocation scheme to the node group. The optimal task allocation scheme is determined by the non-inferior solution with the highest comprehensive score; Assign tasks to specific nodes according to the optimal task allocation plan, calculate the computing, storage, and network resources occupied by each task, and update the resource occupancy in the node capacity matrix; Record the task completion time and resource usage, compare them with the estimated value, and adjust the task allocation weight if there is a difference between the actual execution and the estimated value.

5. The method according to claim 1, characterized in that According to the task allocation scheme, the verification task is sent to each node. When sending the task, the communication delay that may occur during the task execution is predicted according to the network status of the node, and the prediction result is sent to the node together with the task as part of the task metadata, including: Obtain the verification task to be executed from the task queue, where the verification task contains the target node identifier; Collect the network status information of the target node according to the target node identifier, the network status information includes network bandwidth utilization, packet loss rate, current load and historical communication delay data; Based on the historical communication delay data, the ARIMA model is used to generate the communication delay prediction value of the target node. The communication delay prediction value represents the expected delay in the future. Obtain the communication delay prediction value, encapsulate the communication delay prediction value into the metadata of the verification task, and obtain a complete task package, which includes the task description, execution parameters and communication delay prediction value; Send the complete task package to the target node through a secure channel, triggering the target node to start the task execution monitoring program, which is used to monitor the execution status of the verification task; If the task execution monitoring program detects that the verification task starts to be executed, the actual communication delay of the verification task is collected, and the actual communication delay is compared with the predicted communication delay value; If the actual communication delay exceeds the preset percentage of the communication delay prediction value, the timeout period of the verification task is adjusted.

6. The method according to claim 1, characterized in that After receiving the verification task and the evaluated delay prediction information, each node adjusts the time window for data synchronization according to the delay prediction information in the task metadata, including: Obtaining delay prediction information in task metadata, the delay prediction information including a predicted delay value and a preset communication delay threshold; Determine whether to adjust the time window for data synchronization based on the relationship between the predicted delay value and the preset communication delay threshold. If the predicted delay value is greater than the preset communication delay threshold, increase the time window. If the predicted delay value is less than the preset communication delay threshold, decrease the time window. The PID controller is used to adjust the time window size, and the proportional coefficient, integral coefficient and differential coefficient of the PID controller are preset values ​​respectively; Update the data exchange frequency and batch size between nodes according to the adjusted time window. The data exchange frequency is inversely proportional to the time window, and the batch size is proportional to the time window. Compare the actual delay with the predicted delay value and calculate the prediction error. If the prediction error exceeds the preset error threshold, the time window is adjusted again. The forecast error uses the mean absolute percentage error indicator.

7. The method according to claim 1, characterized in that During the execution of the verification task, the execution progress and resource utilization of each node are monitored in real time. When it is found that the execution progress of a certain node lags behind the average progress by more than a preset threshold, or its resource utilization is lower than a preset threshold, some tasks are dispatched from other nodes to the node to balance the load and progress of each node, including: Collect node execution progress and resource utilization information, calculate the average execution progress and resource utilization, and obtain the average execution progress and average resource utilization of all nodes as benchmark values; Use time series forecasting to analyze benchmark values ​​and predict future node loads; The dynamic threshold is calculated by the moving average method, and the actual execution progress and resource utilization of each node are compared with the baseline value through the dynamic threshold; If the node execution progress lags behind the average progress by more than the dynamic threshold or the resource utilization rate is lower than the dynamic threshold, the node is marked as pending scheduling; For nodes in the waiting-to-be-scheduled state, a source node is selected from the nodes with the highest execution progress ranking in a preset percentage; Calculate the amount of tasks that can be migrated and generate a task migration plan that includes the ID of the task to be migrated, the source node, and the target node information; Execute the task migration plan and use the two-phase commit protocol to ensure the transfer of task status; Pause the relevant tasks on the source node and transfer the task status and data to the target node; After the target node confirms that it has been received completely, it updates the task allocation information in the distributed database and resumes task execution on the target node.

8. The method according to claim 1, characterized in that After completing the verification task, each node reports the verification result and intermediate data to the scheduling node. The scheduling node compares and verifies the verification results reported by each node group to obtain the final consistent verification result, including: Receive formatted data packets reported by each verification node, which include verification results, key intermediate data, node identification and timestamp; According to the data packet analysis verification results and intermediate data, the cosine similarity is used to calculate the similarity of the verification results, and the local anomaly factor algorithm is used to detect abnormal results; For verification nodes, based on the node's historical credibility, resource utilization, and task completion time, the analytic hierarchy process is used to construct a judgment matrix and calculate the weight of each node; Using the node weight, the verification results of each node are regarded as the source of evidence, the basic probability distribution function is calculated, and the results of all nodes are iteratively integrated; After obtaining the fused verification results, the information entropy of the fused results is calculated, a credibility assessment report of the results is generated, the fusion results and the credibility report are stored in the distributed database, and the historical credibility records of the verification nodes are updated at the same time.

9. The method according to claim 1, characterized in that: The scheduling node analyzes the final consistent verification results, generates a verification report, and feeds back the verification report and the final results to each node. If the consistency index is lower than the preset threshold, the data synchronization operation is triggered. Each node synchronizes and updates the local intermediate data according to the final verification result sent by the scheduling node, including: After receiving the multi-node verification results, the scheduling node calculates the consistency index; Generate verification reports based on consistency indicators; The verification report is distributed to each node through the secure channel established by the TLS protocol. After receiving the verification report, each node returns a digital signature confirmation message containing a unique identifier; Determine whether the consistency index is lower than the preset threshold. If it is lower than the preset threshold, trigger the data synchronization operation; The data synchronization operation includes generating a synchronization instruction packet, the synchronization instruction packet including an operation type and a data item identifier and an update value; Send synchronization instruction packets to each node; After receiving the synchronization instruction packet, each node compares the local data with the data in the instruction packet; According to the comparison results, update the changed data part; Calculate the checksum of the updated data and obtain the synchronization status and checksum; Send the synchronization status and checksum to the scheduling node; The scheduling node receives the synchronization status and checksum to determine whether they match. If not, a retry mechanism is triggered.

Images