An ultra-low delay remote practical training environment based on edge computing and an optimization method thereof

Abstract

The application provides an ultra-low-delay remote practical training environment based on edge computing and an optimization method thereof, including: obtaining the requirement of a real-time collaboration mechanism, deploying a collaboration service module on an edge node, obtaining operation requests of multiple students through the collaboration service module, and sorting the operation requests of the students according to a timestamp sequence to determine the priority of each request, thereby realizing real-time synchronization of student operations; a distributed practical training environment is constructed, computing and storage resources are deployed on the edge node, local processing of the practical training service is realized, the distance and delay of data transmission are minimized, and through cooperation and data synchronization between nodes, the consistency and reliability of the practical training environment are ensured.

Description

Technical Field

[0001] This invention relates to the field of information technology, and in particular to an ultra-low latency remote training environment based on edge computing and its optimization method. Background Technology

[0002] Implementing a real-time collaborative mechanism for multiple students operating simultaneously in a low-latency remote training environment based on edge computing faces numerous technical challenges. First, edge nodes need to detect and handle conflicts between multiple student operations in real time with limited computing and storage resources, placing stringent demands on the time and space complexity of conflict detection algorithms. Second, the design of conflict resolution strategies needs to comprehensively consider the priority of operations, their temporal relationships, and their impact on training results, while ensuring the fairness and real-time nature of the solutions. Third, to support real-time intervention and guidance from teachers, a flexible and efficient operation permission management mechanism needs to be designed. This mechanism must be able to dynamically adjust the operation permissions of each subject without affecting the training experience, and ensure low-latency propagation and synchronization of permission changes. Finally, considering the potentially complex and variable network conditions in the training environment, ensuring the stability and consistency of real-time collaboration under unstable network conditions is also a pressing issue. This requires designing data synchronization and fault-tolerance mechanisms adapted to dynamic network environments to ensure the continuity of the training process and the integrity of the data. Summary of the Invention

[0003] This invention provides an ultra-low latency remote training environment based on edge computing and its optimization method, mainly including:

[0004] To meet the need for a real-time collaboration mechanism, a collaboration service module is deployed on edge nodes. This module obtains operation requests from multiple students, sorts them according to their timestamps, and determines the priority of each request, thereby achieving real-time synchronization of student operations.

[0005] If multiple students operate simultaneously, a rule-based conflict detection algorithm is used to analyze the students' operation requests and determine whether there are concurrent modifications to the same resource. If a conflict is detected, a conflict resolution strategy is triggered. The execution order of each conflicting operation is determined through a negotiation mechanism or priority sorting to ensure the consistency of the operation.

[0006] During the practical training, the operation permissions of students and teachers are dynamically adjusted through the permission management module. When a teacher needs to guide or intervene in students, the permission management module will temporarily raise the teacher's operation priority, allowing the teacher's operation request to be processed first, thereby realizing the teacher's real-time control over the student's practical training process.

[0007] Lightweight communication protocols and data synchronization mechanisms are used on edge nodes to minimize data transmission latency by compressing data packets and reducing communication frequency. At the same time, the geographical advantages of edge nodes are utilized to deploy services close to the network edge of users, shortening the data transmission path and further reducing communication latency.

[0008] To address the need for operation permission management, a role-based access control (RBAC) model is adopted. Student and teacher roles are predefined, and corresponding operation permissions are assigned to each role. During the training process, the operation permissions of users are dynamically adjusted according to their roles. Through real-time updates and rapid switching of permissions, fine-grained control over training resources is achieved.

[0009] Machine learning algorithms are introduced to optimize the system. By analyzing historical operation data and resource usage, the system can predict students' operation behavior and resource needs, perform resource scheduling and caching in advance, reduce operation response time, and dynamically adjust data transmission strategies through real-time monitoring of network conditions to ensure the smoothness of the training process.

[0010] A distributed training environment is constructed by deploying computing and storage resources on edge nodes to achieve localized processing of training services, minimize data transmission distance and latency, and ensure the consistency and reliability of the training environment through inter-node collaboration and data synchronization.

[0011] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects:

[0012] This invention discloses an ultra-low latency remote training environment based on edge computing and its optimization method. The method deploys a collaborative service module on edge nodes to acquire and sort operation requests from multiple students, achieving real-time synchronization. A rule-based conflict detection algorithm is used to handle concurrent modifications, resolving conflicts through negotiation mechanisms or priority sorting. A role-based access control model is introduced, predefining roles and permissions for students, teachers, etc., supporting dynamic adjustment of operation permissions. Furthermore, machine learning algorithms are used to analyze historical data, predict operational behaviors and resource requirements, pre-schedule resources, and optimize caching strategies. By deploying a distributed training environment on edge nodes, localized processing is achieved, reducing data transmission latency. This invention effectively solves problems related to multi-user real-time collaboration, conflict handling, and permission management, improving the performance and user experience of the training system and providing reliable technical support for remote training and teaching. Attached Figure Description

[0013] Fig. 1 This is a flowchart illustrating an ultra-low latency remote training environment based on edge computing and its optimization method, according to the present invention.

[0014] Fig. 2This is a schematic diagram of an ultra-low latency remote training environment based on edge computing and its optimization method according to the present invention.

[0015] Fig. 3 This is another schematic diagram of an ultra-low latency remote training environment based on edge computing and its optimization method according to the present invention. Detailed Implementation

[0016] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0017] like Figs. 1-3 This embodiment of an ultra-low latency remote training environment based on edge computing and its optimization method may specifically include:

[0018] S101. To meet the requirements of the real-time collaboration mechanism, a collaboration service module is deployed on the edge node. The collaboration service module obtains the operation requests of multiple students and sorts the operation requests according to the timestamp order to determine the priority of each request, thereby achieving real-time synchronization of student operations.

[0019] The system receives operation request information sent by the collaboration service module. This operation request information is sent by multiple students and includes timestamps accurate to milliseconds. The system stores this operation request information in a distributed cache, which temporarily stores multiple operation requests. Based on the operation request information in the distributed cache, a multi-version concurrency control mechanism is used for concurrent processing. The operation requests are sorted according to their chronological order using a timestamp comparison algorithm. The system obtains the sorted operation request queue and calculates the priority score for each operation request using a weighted scoring system, which includes three scoring factors: operation type, user role, and timestamp. If there are similar operation requests in the queue... Conflicting operations are identified using an operation transformation algorithm. For each conflicting operation, adjustments are made based on the principles of maintaining causal consistency and convergence to obtain an ordered sequence of operations after conflict resolution. This ordered sequence is stored in a distributed message queue and broadcast to all collaborating nodes. The system receives operation execution results from the collaborating nodes, which are generated by executing operations in the order they are received. A vector clock algorithm is used to maintain the event order of the distributed system, tracking the causal relationships of operations by recording the local logical clock of each node.

[0020] For example, a collaboration service module is deployed on edge nodes to meet real-time collaboration needs. This module receives operation requests from multiple students, records a timestamp accurate to milliseconds for each request, and stores the request information in a distributed cache. A multi-version concurrency control mechanism is used to process the cached requests concurrently. A timestamp comparison algorithm sorts the requests according to their chronological order, and a weighted scoring system calculates the priority score for each request. The scoring factors include operation type, user role, and timestamp, with operation type having a weight of 0.5, user role a weight of 0.3, and timestamp a weight of 0.2. The sorted request queue is obtained, and an operation transformation algorithm identifies conflicting operations. Conflicting operations are adjusted according to the principles of maintaining causal consistency and convergence, generating an ordered sequence of operations after conflict resolution, which is then stored in a distributed message queue. The processed operation sequence is broadcast to all collaboration nodes via the distributed message queue. Each node executes the operations in the order of receiving the message. A vector clock algorithm is used to maintain the event order of the distributed system, and the causal relationship of operations is tracked by recording the local logical clock of each node. In the event of network latency or partition failure, an eventual consistency model is used to ensure eventual data synchronization. When deploying the collaboration service module on edge nodes, Docker container technology is used for rapid deployment and scaling. The collaboration service module uses a high-performance Nginx server as a front-end proxy, and Node.js for real-time communication on the back-end. When a student submits an operation request, the system uses a high-precision timestamp function, such as JavaScript's `performance.now()`, to record a timestamp accurate to the microsecond level. Request information is stored in a Redis distributed cache, with the key being the timestamp and the value being the request details in JSON format. A multi-version concurrency control mechanism uses Redis's `WATCH` command to implement optimistic locking, ensuring data consistency under high concurrency. The timestamp comparison algorithm uses quicksort, with a sorting time of no more than 50 milliseconds for 10,000 requests. In the priority scoring system, operation types are divided into three categories: add, delete, and update, assigned values ​​of 0.8, 0.6, and 0.4 respectively; user roles are divided into teacher, teaching assistant, and student, assigned values ​​of 0.9, 0.7, and 0.5 respectively; timestamps are converted to relative times, with earlier submitted requests receiving higher timestamp scores. The operation conversion algorithm uses the dOPT (Distributed Operation Conversion) algorithm, resolving conflicts by maintaining an operation history. For example, if two users are simultaneously editing the same line of text in a document, the system will decide which changes to keep or how to merge them based on priority and operation type. The processed operation sequence is stored in the RabbitMQ distributed message queue to ensure reliable message delivery and sequential consistency. In the vector clock algorithm, each node maintains a vector recording the logical clock values ​​of other nodes. When a node receives a message, it updates its own logical clock and broadcasts it to other nodes.When a network partition occurs, the system employs the AP model from the CAP theorem, sacrificing strong consistency to guarantee availability, and using an eventual consistency model to ensure eventual data synchronization. For example, once the network recovers, the system compares the vector clocks of each node, detects and resolves data conflicts, and achieves eventual data consistency.

[0021] S102. If multiple students operate simultaneously, a rule-based conflict detection algorithm is used to analyze the students' operation requests and determine whether there are concurrent modifications to the same resource. If a conflict is detected, a conflict resolution strategy is triggered. The execution order of each conflicting operation is determined through a negotiation mechanism or priority sorting to ensure the consistency of the operation.

[0022] A lightweight concurrency control mechanism is employed to manage access to shared resources. This access management includes acquiring resource status and adding requests to a waiting queue. Based on the resource status, a consistent hashing algorithm is used to map resource identifiers to corresponding resource nodes, where each resource node stores the current version number and last modified timestamp. A cosine similarity algorithm is used to calculate the conflict level between operations, based on an operation feature vector constructed from resource identifier, operation type, and timestamp. If a conflicting operation is detected, a weighted scoring system is used to calculate the overall priority of the operation, which combines operation time, user weight, and resource importance. An optimistic concurrency control mechanism is used during execution, verifying the resource status by comparing the version number recorded at the start of the operation with the current resource version number. If the versions are inconsistent, an automatic merging operation is attempted; if the automatic merging operation fails, the operation is rolled back and the priority is recalculated. After each successful operation execution, the resource version number is incremented and the last modified timestamp is updated until all conflicting operations are resolved.

[0023] For example, a lightweight concurrency control mechanism is used to manage access to shared resources. A consistent hashing algorithm maps resource identifiers to corresponding resource nodes. When a student's operation request arrives, the corresponding resource status is retrieved. If the resource is being operated on, the request is added to a waiting queue, and the current version number and last modified timestamp of the resource are recorded. Student operation requests are analyzed based on a pre-defined set of conflict rules. Operation feature vectors are constructed using key information such as resource identifier, operation type, and timestamp. The degree of conflict between operations is calculated using a cosine similarity algorithm. Specific rules include that the same paragraph in the same document cannot be edited by multiple people simultaneously, and the same row in the same data table cannot be modified by multiple people simultaneously. For detected conflicting operations, a weighted scoring system is used to calculate the comprehensive priority of the operation. Priority calculation considers factors such as operation time, user weight, and resource importance, generating a priority-ordered operation queue. Simultaneously, conflicting users are notified via a message queue, allowing them to choose to retain or abandon the operation. Operations are retrieved sequentially from the priority queue and executed. An optimistic concurrency control mechanism is employed during execution, verifying the resource status by comparing the version number recorded at the start of the operation with the current resource version number. If the versions are inconsistent, an automatic merging operation is attempted; if merging fails, the operation is rolled back and the priority is recalculated. After each successful operation, the resource version number is incremented and the last modified timestamp is updated until all conflicting operations are resolved. In practical implementation, a lightweight concurrency control mechanism can be implemented using Redis, using the SETNX command to set resource locks with an expiration time of 30 seconds to prevent deadlocks. A consistent hashing algorithm uses MD5 to map resource IDs to a 0-2^32-1 ring, selecting 32 virtual nodes to ensure even resource distribution. The resource status includes a version number, an initial value of 0, and a timestamp accurate to milliseconds. The conflict rule set is stored in a JSON-formatted configuration file, containing rules such as "conflicts in the same document paragraph" and "conflicts in data table rows." The operation feature vector consists of a resource ID, operation type (e.g., CRUD operations), and a timestamp triple. The cosine similarity algorithm calculates the cosine of the angle between two vectors, with a threshold of 0.8; values ​​greater than this are considered conflicts. In the weighted scoring system, operation time has a weight of 0.3, user weight 0.4 (e.g., teacher 1.0, student 0.6), and resource importance weight 0.3 (e.g., document 1.0, image 0.8). RabbitMQ is used for the message queue, with a TTL of 10 seconds; operations are automatically abandoned upon timeout. In optimistic concurrency control, when version numbers are inconsistent, the diff algorithm is used to calculate the difference and attempt automatic merging, with a success rate of approximately 80%. Each successful operation increments the version number and updates the timestamp. Throughout the process, critical operations such as locking, unlocking, and version updates are implemented atomically using Redis Lua scripts to ensure data consistency.

[0024] S103. During the practical training process, the operation permissions of students and teachers are dynamically adjusted through the permission management module. When a teacher needs to guide or intervene in students, the permission management module will temporarily increase the teacher's operation priority, allowing the teacher's operation request to be processed first, thereby realizing the teacher's real-time control over the student's practical training process.

[0025] A role-based access control mechanism is employed for permission management. This mechanism defines basic operational permissions for teachers and students using preset permission templates, which include operation type, resource access level, and time limit attributes. Based on these basic permissions, a real-time monitoring module collects student operation data. This module uses an isolated forest algorithm to identify abnormal operation patterns, including operation frequency, resource access patterns, and operation sequence characteristics. If the isolated forest algorithm detects abnormal student operation or receives a guidance request from a teacher, a permission adjustment event is triggered. After the permission adjustment event is triggered, the permission management module calculates new permission priorities according to preset rules and updates user permission information in the permission database through a database transaction mechanism. After updating user permission information, the permission management module sends a permission change notification to the operation queue manager. Upon receiving the permission change notification, the operation queue manager reorders the queue of pending operation requests, elevating teacher operation requests to the front of the queue and prioritizing the execution of teacher operation instructions through a multi-level feedback queue scheduling algorithm.

[0026] For example, a role-based access control mechanism is used to implement permission management. Basic operational permissions for teachers and students are defined through preset permission templates. These templates include attributes such as operation type, resource access level, and time limits. Permission levels are dynamically adjusted based on the training stage and task type, achieving fine-grained permission management. A real-time update mechanism for permission templates is also implemented to ensure that permission settings are synchronized with the training progress. A real-time monitoring module collects student operation data and uses an isolated forest algorithm to identify abnormal operation patterns. Anomaly judgment criteria include operation frequency, resource access patterns, and operation sequence characteristics. If an abnormal student operation is detected or a teacher initiates a guidance request, a permission adjustment event is triggered. After the permission adjustment event is triggered, the permission management module calculates the new permission priority according to preset rules, updates the user permission information in the permission database through a database transaction mechanism to ensure the atomicity of the permission update, and sends a permission change notification to the operation queue manager, recording the specific content and objectives of the teacher's guidance operation. After receiving a permission change notification, the operation queue manager reorders the queue of pending operation requests, elevating teacher requests to the front of the queue. A multi-level feedback queue scheduling algorithm prioritizes the execution of teacher commands, enabling real-time teacher intervention in student training. A real-time feedback mechanism displays the effectiveness of the guidance and student responses to the teacher. In practice, the role-based access control mechanism uses the RBAC model, with permission templates stored in JSON format, including 10 operation types, 5 resource access levels, and a 24-hour time limit. Permission updates are set every 5 minutes, synchronized with the training progress. The real-time monitoring module collects 100 data points per second, including mouse movements, keyboard input, and resource access records. The isolated forest algorithm uses 100 decision trees, with an anomaly threshold of 0.6. Operations exceeding 10 times per second, consecutive access to more than 3 restricted resources, or operation sequences with a match rate of less than 70% with a preset pattern are considered abnormal. After a permission adjustment event is triggered, priority calculation uses a weighted scoring method: teachers start with 10 points, students with 5 points, and each guidance request adds an additional 3 points. The database transactions employ a two-phase commit protocol to ensure consistency in a distributed environment. The operation queue manager is implemented using a priority heap, with teacher requests inserted at the top of the heap. A multi-level feedback queue scheduling algorithm sets up three priority queues with time slices of 20ms, 40ms, and 60ms respectively; teacher operations are assigned to the highest priority queue by default. The real-time feedback mechanism pushes data via the WebSocket protocol, updating the teacher's display every 200ms, including operation progress, student response time, and resource status changes, to achieve effective real-time interaction.

[0027] Establish a rule-based access control system, setting the priority of teacher and student operations. The default priority of teachers is higher than that of students. When teachers and students operate on the same resource, the system checks the operation type and priority of both parties. For operations that do not conflict, they are allowed to be performed simultaneously. For conflicting operations, the operation with higher priority is allowed, while the operation with lower priority is suspended or canceled.

[0028] A permission rule database is constructed to store predefined operation types, priority levels, and role permission mappings. This database uses the MurmurHash3 algorithm for fast lookup. Based on this database, the default priority for teacher roles is set to high, and the default priority for student roles is set to medium. A real-time operation monitoring module extracts the operation type, target resource, and operation timestamp from user operation requests, which are then encapsulated into standard format data packets. These standard format data packets are pushed to a priority queue based on a red-black tree, sorted by operation timestamp and user role priority. A request time overlap detection algorithm is run on requests in the priority queue, comparing the target resource and time window to identify potential conflicting operations. If potential conflicting operations are identified, they are grouped into a conflict set. For operations in the conflict set, a priority arbitration module is invoked. This module queries the permission rule database based on the operation type and user role to obtain the operation priority. Based on the operation priority, high-priority operations are executed, while low-priority operations are marked as paused and added to a delay queue.

[0029] For example, a permission rule database is constructed to store predefined operation types, priority levels, and role permission mappings. The MurmurHash3 algorithm enables fast lookups. The default priority for teacher roles is set to high, and for student roles, it's set to medium. An interface for registering custom operation types and priorities is provided, supporting dynamic updates to permission rules. A real-time operation monitoring module captures user operation requests, extracts key information such as operation type, target resource, and operation timestamp, encapsulates the requests into standard-format data packets, and pushes them to a priority queue based on a red-black tree, sorting them according to operation timestamp and user role priority. A time overlap detection algorithm analyzes requests in the operation queue, identifying potential conflicting operations by comparing target resources and time windows. Conflicting operations are grouped into conflict sets, while non-conflicting operations are passed to the parallel execution module for direct processing. For operations in the conflict set, a priority arbitration module is invoked to query the permission rule database based on operation type and user role, obtain operation priorities, compare priorities, execute high-priority operations, mark low-priority operations as paused and add them to a delay queue, and implement an exponential backoff retry mechanism, automatically retrying paused operations after the high-priority operation completes. In practical implementation, the permission rule database adopts a key-value storage structure and uses the MurmurHash3 algorithm to calculate the hash value of the key, achieving a query complexity of O(1). The database initially contains 50 predefined operation types, 3 priority levels (high, medium, low), and 2 roles (teacher, student). The custom operation registration interface supports processing 100 update requests per second. The real-time operation monitoring module adopts an event-driven architecture and can process 1000 operation requests per millisecond. The data packet structure includes a 64-bit operation ID, a 32-bit user ID, a 16-bit operation type, a 64-bit timestamp, and a 256-bit resource identifier. The priority queue implemented by the red-black tree supports O(logn) insertion and deletion operations, and the queue capacity is set to 10000. The time overlap detection algorithm uses scanline technology with a time complexity of O(nlogn) and can process 1000 operation requests simultaneously. The conflict determination threshold is set to a time overlap of 50% or more. The parallel execution module adopts a work-stealing algorithm, using 8 worker threads, and the average task execution time is controlled within 10ms. The priority arbitration module uses a weighted scoring algorithm, with operation type weighting 0.6 and user role weighting 0.4, and a score range of 0-100. The delay queue is implemented based on a heap and supports up to 1000 paused operations. The exponential backoff retry mechanism has an initial waiting time of 100ms, a maximum waiting time of 10s, and a maximum number of retries of 5, helping the system to recover from temporary errors.

[0030] S104. Lightweight communication protocols and data synchronization mechanisms are adopted on edge nodes. By compressing data packets and reducing communication frequency, the latency of data transmission is minimized. At the same time, the geographical advantages of edge nodes are utilized to deploy services close to the network edge of users, shortening the data transmission path and further reducing communication latency.

[0031] The system receives connection request information based on the WebSocket protocol sent by an edge node, the connection request information containing the edge node's unique identifier; establishes a full-duplex communication connection with the edge node based on the connection request information, the full-duplex communication connection being maintained using an adaptive heartbeat mechanism; acquires the network latency data of the edge node, dynamically adjusts the heartbeat interval based on the network latency data, the initial value of the heartbeat interval being set to a preset time threshold; receives operation log data sent by the edge node, the operation log data being stored in the edge node's local cache; processes the operation log data using an operation conversion algorithm to obtain synchronization data; compresses the synchronization data, selecting a compression algorithm through an adaptive compression mechanism, the compression algorithm being determined based on the data type of the synchronization data and the network status of the edge node; acquires the geographical location information and network topology data of the edge node; constructs an edge node selection algorithm based on the geographical location information and network topology data; uses the edge node selection algorithm to select the optimal edge node from multiple edge nodes; and deploys services on the optimal edge node.

[0032] For example, a lightweight communication protocol based on WebSockets is used to achieve full-duplex communication on edge nodes. An adaptive heartbeat mechanism maintains long-term connections, with an initial heartbeat interval of 30 seconds, dynamically adjusted based on network latency (ranging from 10 to 60 seconds). The timeout period is set to twice the heartbeat interval, and automatic reconnection occurs three times. Higher transmission priority is assigned to teacher intervention operations. An operation conversion algorithm is used to implement a data synchronization mechanism, storing operation logs in a local cache. Data is categorized and prioritized, and incremental data is periodically pushed to the server at 5-second intervals. Pushes are triggered immediately if the data volume exceeds a preset threshold or if a high-priority operation exists. Transmitted data is compressed using an adaptive compression mechanism, dynamically selecting LZ4, Zstd, or Snappy algorithms based on data type and network conditions. The initial compression ratio is set to 0.6 and dynamically adjusted. Data packets are formatted using a binary protocol to reduce packet size, and differential encoding is used to transmit only the data differences from the previous synchronization. Based on geographic location information and network topology, an edge node selection algorithm is constructed. By measuring network metrics such as round-trip latency, packet loss rate, and bandwidth, the optimal edge node is selected for service deployment. A weighted round-robin algorithm is used to achieve dynamic load balancing, and a hot data caching mechanism is implemented on the edge nodes. The caching strategy adopts the LRU algorithm, and the cache capacity is dynamically adjusted according to node resources. In actual implementation, after the WebSocket connection is established, the adaptive heartbeat mechanism initially sets the interval to 30 seconds. Network latency is calculated using an exponential moving average algorithm. When the latency exceeds 100ms, the heartbeat interval decreases by 5 seconds; when it is below 50ms, it increases by 5 seconds, maintaining a range between 10-60 seconds. A priority flag of 1 is added to the WebSocket message header for teacher intervention operations, while 0 is used for normal operations. Messages with priority 1 are processed first in the sending queue. The operation transformation algorithm uses the JupiterOT system to transform concurrent operations, ensuring eventual consistency. Local caching uses LevelDB to store operation logs. Data categories include text, graphics, and control commands, sorted by priority from high to low. In the adaptive compression mechanism, text data uses the LZ4 algorithm by default, with a compression ratio of 0.6; image data uses the Zstd algorithm, with a compression ratio of 0.7; and control commands use the Snappy algorithm, with a compression ratio of 0.5. The compression ratio is dynamically adjusted by ±0.1 every 10 seconds based on compression time and network throughput. The edge node selection algorithm uses a weighted score, with round-trip latency weighted at 0.5, packet loss rate weighted at 0.3, and bandwidth weighted at 0.2, and is re-evaluated every 5 minutes. The weighted round-robin load balancing has an initial weight of 1, and the weight is adjusted based on node CPU utilization after processing every 100 requests.Hot data caching uses the LRU algorithm, with an initial capacity of 10% of memory. The cache size increases by 2% when the cache hit rate is below 80% and decreases by 1% when it is above 95%. This achieves high efficiency and stability for the system in high-concurrency, multi-user, and real-time data processing environments, ensuring timely message processing and optimizing data transmission and storage.

[0033] S105. To address the need for operation permission management, a role-based access control (RBAC) model is adopted. Student and teacher roles are predefined, and corresponding operation permissions are assigned to each role. During the training process, the operation permissions of users are dynamically adjusted according to their roles. Through real-time updates and rapid switching of permissions, fine-grained control over training resources is achieved.

[0034] The system receives user login requests carrying student or teacher role identifiers, which are sent when the user enters the training system. Based on the login request, it retrieves the corresponding role's permission matrix from the RBAC model permission management database. The permission matrix is ​​stored in a red-black tree structure, and user permission information is obtained by traversing the red-black tree nodes. If the user role has a permission inheritance relationship, the inherited permission information is merged into the user permission information. The system obtains user behavior logs and training progress data, and determines whether a permission change event has been triggered based on preset rule engine logic. If a permission change event is triggered, the permission node information in the red-black tree is updated. The system receives user requests to access training resources, which include resource identifiers and user identifiers. Based on the resource identifier, the access level of the resource is determined, and a bitwise AND operation is performed between the resource identifier and the user permission information using bitmask technology. If the bitwise AND result is not zero, the user is determined to have access rights. The system retrieves the user's hot-spot permission data from a multi-level cache. If the data is not present in the cache, it reads it from the permission management database and updates the cache. Expired cache data is evicted using the Least Recently Used algorithm, and cache consistency in a distributed environment is ensured through a two-phase commit protocol.

[0035] For example, an RBAC-based permission management database is constructed, defining student and teacher roles, designing a permission matrix to associate operation permissions with roles, and using a red-black tree to achieve fast permission query and update. A permission inheritance mechanism is also introduced to simplify the permission management structure. A real-time permission update module is developed to monitor user behavior and training progress, using a rule engine to manage complex permission change logic, triggering permission adjustment events according to preset rules, and pushing permission change messages using a publish-subscribe model to ensure timely synchronization of permission information. A verification mechanism is executed after permission changes to ensure the correctness of the changes. A fine-grained permission control mechanism is implemented, dividing training resources into different access levels, using bitmasking technology for fast permission judgment, performing real-time permission verification when users request resources, intercepting unauthorized operations, and implementing a permission conflict resolution mechanism to handle overlapping permissions for multiple roles. A permission caching system is designed, using a multi-level caching strategy to maintain hot permission data in memory, employing a least recently used algorithm for cache eviction, and using a two-phase commit protocol to ensure cache consistency in a distributed environment, improving the efficiency of permission query and update. A permission auditing module was added to record all permission change operations, including change time, operator, and change content. Audit logs are stored in a time-series database for easy analysis and backtracking. In actual implementation, the RBAC model's permission management database uses PostgreSQL, defining two basic roles: student and teacher. The permission matrix contains 100 types of operation permissions. The red-black tree implementation achieves an average time complexity of O(logn) for permission queries, supporting 100,000 concurrent queries. Permission inheritance uses a multiple inheritance mechanism with a maximum inheritance depth of 5 levels. The real-time permission update module uses the Drools rule engine, pre-setting 100 permission change rules stored in Redis, with update latency controlled within 50ms. The publish-subscribe pattern is implemented using RabbitMQ, achieving a message throughput of 100,000 messages per second. Fine-grained permission control divides resources into 10 access levels, using 32-bit integers for bitmasks and achieving millisecond-level permission judgment through bitwise AND operations. Permission conflict resolution employs a priority mechanism, with teacher roles having higher priority than student roles. The multi-level caching strategy includes two levels: L1 (process memory) and L2 (distributed cache). The L1 cache has a capacity of 1GB, and the L2 cache uses a Redis cluster with a total capacity of 50GB. The least recently used algorithm has an eviction period of 5 minutes. The two-phase commit protocol has a timeout of 3 seconds, with 3 retries on failure. The permission auditing module uses InfluxDB to store audit logs, retaining them for 180 days and recording approximately 1 million audit data entries daily. By tracking and recording each user's permission operations, it provides strong support for system security and compliance monitoring.

[0036] The system predefines user roles. When a user accesses resources or performs operations, it checks the user's current role and corresponding permission configuration. Access or operation is only allowed if the role has the corresponding permissions. A management interface is provided to allow real-time adjustment of user role assignments. Permissions are automatically updated after a role change without the need for re-login. Audit logs of user role and permission changes are recorded.

[0037] A role and permission database is constructed, which uses a relational database to store user, role, and permission information. Based on the information stored in the role and permission database, an attribute-based access control middleware is developed. This middleware intercepts user requests and retrieves user role and attribute information from a distributed cache. After retrieving the user role and attribute information, roles are assigned and adjusted through a role allocation management interface, which provides a RESTful API. Based on the role allocation and adjustment results, an audit log module is designed to record user role and permission change operations. If the audit log module detects user role and permission changes, a permission conflict detection algorithm is introduced to identify and resolve permission conflicts caused by multiple role assignments.

[0038] For example, a role-permission database is constructed, using a relational database to store user, role, and permission information. User-role and role-permission association tables are designed, and query performance is optimized through indexes. Triggers are used to implement cascading updates for role and permission changes. A permission set union operation is introduced to handle permission merging in multi-role scenarios. An attribute-based access control middleware is developed to intercept user requests, retrieve user role and attribute information from a distributed cache, evaluate whether a user has the permission to access resources or perform operations using a rule engine, and initialize the permission cache when the middleware starts. A role assignment management interface is implemented, providing a RESTful API for role assignment and adjustment. Distributed locks are used to handle concurrent operations, and message queues are used to asynchronously notify relevant services to update the user permission cache. A session management mechanism is also implemented, allowing real-time updates of permission information within a user session without requiring re-login. An audit log module is designed to record user role and permission change operations. Audit logs are stored in a time-series database, implementing a time-based log sharding strategy, using gzip compression, and an asynchronous writing mechanism to improve log recording efficiency. A permission conflict detection algorithm is introduced to identify and resolve permission conflicts that may arise from multi-role assignments. In practical implementation, the role and permission database uses PostgreSQL, with user, role, and permission tables and their related tables. B-tree indexes are used to optimize query performance, with an average query time controlled within 5ms. Triggers are implemented using PL / pgSQL, automatically updating the permission cache of relevant users when role permissions change. Multi-role permission merging uses bitmasking, with permissions represented by 32-bit integers, supporting up to 32 atomic permissions. The attribute-based access control middleware is implemented using Spring Security, and the rule engine uses Drools, with 100 default permission rules and an average rule evaluation time of 1ms. The distributed cache uses a Redis cluster with an initial capacity of 10GB, supporting 100,000 concurrent users. The role assignment management interface adopts a RESTful style, using Redisson to implement distributed locks with a lock timeout of 3 seconds. RabbitMQ is used for the message queue, with a single queue throughput of 100,000 messages / second. Session management is based on JWT, with tokens valid for 2 hours, and immediate revocation is achieved through a blacklist stored in Redis. Audit logs are stored using InfluxDB, sharded daily, and compressed using the gzip algorithm with a compression ratio of approximately 10:1. Asynchronous writes are performed in batches of 100 records each, with write latency controlled within 50ms. The permission conflict detection algorithm uses the maximum independent set algorithm from graph theory, achieving an average time complexity of O(nlogn) for conflict resolution, thus enhancing the rigor and flexibility of permission management.

[0039] S106. Machine learning algorithms are introduced to optimize the system. By analyzing historical operation data and resource usage, the system can predict students' operation behavior and resource needs, perform resource scheduling and caching in advance, reduce operation response time, and dynamically adjust data transmission strategies through real-time monitoring of network conditions to ensure the smoothness of the training process.

[0040] The system acquires system logs and monitoring data, which include user operation sequences, resource usage, and network status information. Based on these logs and data, it performs data cleaning and feature engineering to obtain a standardized training dataset, which includes time-series features, user behavior patterns, and resource utilization. Using this training dataset, it trains an ensemble learning model based on a long short-term memory network and a decision tree, predicting the user's next action. Based on the predicted user action and the current system load, it executes an intelligent resource scheduling algorithm, which uses a genetic algorithm to dynamically allocate computing resources and preload data. It determines network quality indicators by real-time monitoring of round-trip time, packet loss rate, and bandwidth. If the network quality indicators are below a preset threshold, it executes an adaptive data transmission mechanism, which dynamically adjusts the data compression rate and transmission protocol.

[0041] For example, a data acquisition and preprocessing module is constructed to extract user operation sequences, resource usage, and network status information from system logs and monitoring data. A standardized training dataset is generated through data cleaning and feature engineering, with features including time-series characteristics, user behavior patterns, and resource utilization. An ensemble learning model based on Long Short-Term Memory (LSTM) networks and decision trees is developed. This model is trained using historical operation sequences to predict the user's next possible operation in real time, and the prediction results are passed to the resource scheduling module. An intelligent resource scheduling algorithm is implemented. Based on the operation prediction results and the current system load, a genetic algorithm is used to dynamically allocate computing resources and preload potentially needed data. A Least Recently Used (LRU) algorithm is used to manage the cache, optimizing resource utilization efficiency, and the potential impact of resource scheduling results on network transmission is analyzed. An adaptive data transmission mechanism is designed. By monitoring network quality indicators such as round-trip time, packet loss rate, and bandwidth in real time, the data compression rate and transmission protocol are dynamically adjusted. Under bandwidth constraints, critical data is prioritized for transmission to ensure the responsiveness of core functions. Simultaneously, a user experience monitoring module is introduced to collect indicators such as user operation latency and page load time, evaluating system performance in real time as feedback for optimization. In practical implementation, the data acquisition module collects 100 data points per second, including user clicks, resource utilization, and network latency. Feature engineering extracts 20 time-series features, such as operation frequency and resource usage trends, and 10 user behavior pattern features, such as operation sequence similarity. The LSTM network contains 3 layers, each with 128 neurons, and is combined with a random forest to form an ensemble model, achieving a prediction accuracy of 85%. A genetic algorithm is used for resource scheduling, with a population size of 100, 50 generations, a crossover rate of 0.8, a mutation rate of 0.1, and a fitness function that comprehensively considers resource utilization and response time. The LRU cache capacity is set to 20% of the total memory, and the cache hit rate is increased to 90%. Network quality monitoring samples every 100ms, with an RTT threshold of 200ms, a packet loss rate threshold of 1%, and a bandwidth threshold of 1Mbps. When network quality falls below the threshold, the compression ratio is increased from the default 0.7 to 0.9, and the transmission protocol is switched from TCP to QUIC. User experience monitoring collects operation latency, targeting...

[0042] <100ms, and page load time, target <3s, to ensure the system can provide a high-quality user experience.

[0043] S107. Construct a distributed training environment by deploying computing and storage resources on edge nodes to achieve localized processing of training services, minimize data transmission distance and latency, and ensure the consistency and reliability of the training environment through inter-node collaboration and data synchronization.

[0044] The system receives a distribution request carrying user request information, generated when a user initiates a training service. Based on the distribution request, a weighted round-robin algorithm is used to determine the nearest edge node, which is used to deploy the containerized training service. The system obtains the local storage information of the edge node and determines whether cached data matching the user request information exists in the local storage information. If the cached data exists, it is retrieved from the local storage information. If the cached data does not exist, a consistent hashing algorithm is used to determine the data shard location, and the node status information is obtained using the Raft consensus algorithm. The system obtains the required data based on the node status information and the data shard location, and uses a vector clock algorithm to determine if there are any conflicts in the required data. If there are conflicts, the conflicts are resolved through incremental synchronization. If there are no conflicts, the required data is obtained through batch transmission. The required data is encrypted using the TLS encryption communication protocol, and user access permissions are determined according to a preset role-based access control policy. If the user access permissions are valid, the encrypted required data is sent to the user.

[0045] For example, a distributed architecture is designed, employing containerization technology to deploy training services. A weighted round-robin algorithm distributes user requests to the nearest edge nodes, enabling dynamic allocation and elastic scaling of computing resources. The distributed storage system Ceph is used for data persistence. Localized processing is implemented at edge nodes, caching frequently used data and computing tasks in local storage. A Squid caching proxy optimizes data access paths, reducing cross-node data transfer, and establishing a node-to-node sharing mechanism for local processing results. An inter-node collaboration mechanism is constructed, using a consistent hashing algorithm for data sharding and replica management, and the Raft consensus algorithm for node state synchronization and fault detection, ensuring high system availability. A data synchronization module is developed, using a vector clock algorithm to resolve data conflicts, optimizing data transfer efficiency through incremental synchronization and batch transmission, periodically performing Merkle tree-based data consistency checks to ensure data consistency in the distributed environment, and using the distributed state storage etcd to synchronize the training environment state. A security mechanism is introduced, implementing TLS encrypted communication and role-based access control between edge nodes. In practice, the distributed architecture uses Kubernetes for container orchestration, deploying 100 edge nodes, each configured with 8 CPU cores and 32GB of memory. The weighted round-robin algorithm dynamically adjusts weights based on node load, updating every 5 seconds. The Ceph storage cluster is configured with 10 OSD nodes, each with a 20TB capacity, implementing a 3-replica strategy. Local caching uses Redis, with a 10GB memory limit and an LRU eviction policy. The Squid proxy server cache size is set to 100GB, with a refresh interval of 30 minutes. Consistent hashing uses MD5, with the number of virtual nodes being 100 times the number of actual nodes. The Raft algorithm election timeout is set to a random value of 150-300ms, and the heartbeat interval is 50ms. The vector clock uses 64-bit integers, with a maximum deviation tolerance of 10. Incremental synchronization uses a binary diff algorithm, with a batch transmission threshold of 1000 records. The Merkle tree depth is set to 16, and the verification period is 1 hour. The etcd cluster deploys 5 nodes, achieving cross-region replication across 3 data centers. TLS encryption uses the ECC algorithm with a 256-bit key length. Role-based access control defines three roles: administrator, teacher, and student, with a total of 20 permissions. By binding permissions to roles instead of directly to users, permission management can be simplified.

[0046] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. An edge computing-based ultra-low latency remote practical training environment and an optimization method thereof, characterized in that, The method comprises: obtaining the demand of real-time collaboration mechanism, obtaining the operation requests of multiple students through the edge node, and sorting the operation requests of students in timestamp order to determine the priority of each request; if multiple students operate simultaneously, the operation requests of students are analyzed to determine whether there is concurrent modification of the same resource, if a conflict is detected, a conflict resolution strategy is triggered, and the execution order of each conflict operation is determined through negotiation mechanism or priority sorting, including: using a lightweight concurrency control mechanism to manage access to shared resources, the access management includes obtaining resource state and adding request to waiting queue; according to the resource state, the resource identifier is mapped to the corresponding resource node through consistent hashing algorithm, and the resource node is used to store the current version number and last modification timestamp of the resource; the conflict degree between operations is calculated by using cosine similarity algorithm, and the conflict degree is based on the operation feature vector constructed by resource identifier, operation type and timestamp; if a conflict operation is detected, the comprehensive priority of the operation is calculated by using a weighted scoring system, the comprehensive priority combines operation time, user weight and resource importance factor; the optimistic concurrency control mechanism is used in the execution process, the optimistic concurrency control mechanism verifies the resource state by comparing the version number recorded at the beginning of the operation with the current resource version number; if the versions are inconsistent, the operation is attempted to be automatically merged, and the operation is rolled back and the priority is recalculated when the automatic merging operation fails; the resource version number is incremented and the last modification timestamp is updated after each successful operation, until all conflict operations are processed; the operation permissions of students and teachers are dynamically adjusted during the practical training, wherein when the teacher needs to guide or intervene the students, the operation priority of the teacher is temporarily improved, and the operation request of the teacher is allowed to be processed first; a communication protocol and data synchronization mechanism are used on the edge node to minimize the data transmission delay while taking advantage of the geographical location of the edge node, deploy services near the network edge of the user, shorten the data transmission path, and further reduce the communication delay, including: receiving the connection request information based on websocket protocol sent by the edge node, the connection request information contains the unique identifier of the edge node; according to the connection request information, a full-duplex communication connection with the edge node is established, and the full-duplex communication connection uses an adaptive heartbeat mechanism to maintain; obtaining network delay data of the edge node, dynamically adjusting heartbeat interval according to the network delay data, and setting the initial value of the heartbeat interval as a preset time threshold; receiving operation log data sent by the edge node, the operation log data is stored in the local cache of the edge node; the operation log data is processed by using an operation conversion algorithm to obtain synchronization data; the synchronization data is compressed by using an adaptive compression mechanism to select a compression algorithm, and the compression algorithm is determined according to the data type of the synchronization data and the network status of the edge node; obtaining geographical location information and network topology structure data of the edge node;According to the geographic location information and the network topology data, an edge node selection algorithm is constructed; the optimal edge node is selected from multiple edge nodes by using the edge node selection algorithm; the service is deployed on the optimal edge node; for the requirement of operation authority management, a role-based access control model is used to predefine the roles of students and teachers, and each role is assigned with corresponding operation authority, in the operation process, the operation authority of the user is dynamically adjusted according to the role of the user, and the real-time update and rapid switching of the authority are realized; through the analysis of the historical operation data and the resource usage, the operation behavior and the resource demand of the student are predicted, the resource scheduling and caching are performed in advance, the operation response time is reduced, and the data transmission strategy is dynamically adjusted through real-time monitoring of the network condition, including: obtaining system logs and monitoring data, the system logs and monitoring data include user operation sequence, resource usage and network state information; the standard training data set is obtained through data cleaning and feature engineering according to the system logs and monitoring data, the training data set includes time series features, user behavior patterns and resource utilization; the integrated learning model based on long short-term memory network and decision tree is trained by using the training data set, and the next operation of the user is predicted through the integrated learning model; according to the predicted user operation and the current system load condition, the intelligent resource scheduling algorithm is executed, the intelligent resource scheduling algorithm dynamically allocates computing resources and preloads data by using the genetic algorithm; the network quality index is determined by real-time monitoring of round-trip time, packet loss rate and bandwidth; if the network quality index is lower than the preset threshold, the adaptive data transmission mechanism is executed, the adaptive data transmission mechanism dynamically adjusts the data compression rate and the transmission protocol; a distributed operation environment is constructed, the computing and storage resources are deployed on the edge node, the localization processing of the operation service is realized, the distance and the delay of data transmission are minimized, and the cooperation and data synchronization among nodes are realized.

2. The method of claim 1, wherein, The requirement of obtaining real-time collaboration mechanism, obtaining operation requests of multiple students through edge nodes, and sorting the operation requests of students according to timestamp order to determine the priority of each request, includes: receiving operation request information sent by the collaboration service module, the operation request information is sent by multiple students and contains a timestamp accurate to milliseconds; store the operation request information into the distributed cache, the distributed cache is used for temporarily storing multiple operation requests; according to the operation request information in the distributed cache, concurrent processing is carried out by using multi-version concurrency control mechanism, and the operation requests are sorted according to time sequence by using timestamp comparison algorithm; obtain the sorted operation request queue, calculate the priority score of each operation request by using weighted scoring system, the weighted scoring system includes three scoring factors of operation type, user role and timestamp; if there are mutually conflicting operations in the operation request queue, the mutually conflicting operations are identified by operation conversion algorithm; for the mutually conflicting operations, the principle of maintaining causal consistency and convergence is adopted for adjustment, and the ordered operation sequence after conflict processing is obtained; store the ordered operation sequence into the distributed message queue, and broadcast the ordered operation sequence to all collaboration nodes through the distributed message queue; receive the operation execution result sent by the collaboration node, the operation execution result is generated by the collaboration node after executing the operation according to the receiving order; The vector clock algorithm is used to maintain the event order of the distributed system, and the vector clock algorithm tracks the causal relationship of operations by recording the local logical clock of each node.

3. The method of claim 1, wherein, The operation permission of the student and the teacher is dynamically adjusted in the process of practical training, and when the teacher needs to guide or intervene the student, the operation priority of the teacher is temporarily improved, and the operation request of the teacher is allowed to be processed preferentially, including: the role-based access control mechanism is used to realize permission management, the access control mechanism defines the basic operation permission of the teacher and the student through a preset permission template, and the preset permission template includes operation type, resource access level and time limit attribute; according to the basic operation permission, student operation data is collected by using a real-time monitoring module, the real-time monitoring module identifies abnormal operation mode by using an isolated forest algorithm, and the abnormal operation mode includes operation frequency, resource access mode and operation sequence characteristics; If the isolated forest algorithm detects that the student operation is abnormal or receives a guidance request initiated by the teacher, a permission adjustment event is triggered; After the permission adjustment event is triggered, the permission management module calculates a new permission priority according to a preset rule, and the permission management module updates the user permission information in the permission database through a database transaction mechanism; The permission management module updates the user permission information and sends a permission change notification to an operation queue manager; the operation queue manager reorders the operation request queue to be processed after receiving the permission change notification, and the operation queue manager promotes the operation request of the teacher to the front of the queue and executes the operation instruction of the teacher in priority through a multi-level feedback queue scheduling algorithm; further comprising: establishing a rule-based permission management system, setting the priority of the teacher and student operations, the default priority of the teacher being higher than that of the student, when the teacher and the student operate on the same resource, the system checks the operation type and priority of both parties, and the operation without conflict is directly allowed to be performed simultaneously; for the conflicting operation, the operation with high priority is allowed, and the operation with low priority is suspended or cancelled.

4. The method of claim 3, wherein, The rule-based permission management system is established, and the priority of the teacher and student operations is set, the default priority of the teacher being higher than that of the student, when the teacher and the student operate on the same resource, the system checks the operation type and priority of both parties, and the operation without conflict is directly allowed to be performed simultaneously; for the conflicting operation, the operation with high priority is allowed, and the operation with low priority is suspended or cancelled, including: constructing a permission rule database to store the pre-defined operation type, priority level and role permission mapping relationship, the permission rule database is realized by MurmurHash3 algorithm for fast query; according to the permission rule database, the default priority of the teacher role is set to high, and the default priority of the student role is set to medium; A real-time operation monitoring module is adopted to extract the operation type, target resource and operation timestamp from the operation request of the user, and the operation request is packaged into a standard format data packet; The standard format data packet is pushed to a priority queue based on a red-black tree, and the priority queue is sorted according to the operation timestamp and user role priority; a request running time overlap detection algorithm is run on the priority queue to identify potential conflict operations by comparing the target resource and time window of the operation; if a potential conflict operation is identified, the conflict operation is combined into a conflict set; for the operation in the conflict set, a priority arbitration module is called, which queries the permission rule database according to the operation type and user role to obtain the operation priority; According to the operation priority, the high-priority operation is executed, the low-priority operation is marked as a suspended state and added to a delay queue.

5. The method of claim 1, wherein, According to the demand for operation permission management, a role-based access control model is adopted to predefine the roles of students and teachers and assign corresponding operation permissions to each role, and in the practical training process, the operation permission of the user is dynamically adjusted according to the role of the user, and the real-time update and fast switching of the permission are realized, including: receiving a user login request carrying a student or teacher role identifier, the user login request being sent when the user enters the practical training system; obtaining the permission matrix of the corresponding role from the RBAC model permission management database according to the user login request; the permission matrix is stored in a red-black tree structure, and the user permission information is obtained by traversing the red-black tree node; If the user role exists a permission inheritance relationship, the inherited permission information is merged into the user permission information; Obtain user behavior logs and practical training progress data, and determine whether a permission change event is triggered according to a preset rule engine logic; If the permission change event is triggered, update the permission node information in the red-black tree; Receive a request of a user to access practical training resources, the request containing a resource identifier and a user identifier; Determine the access level of the resource according to the resource identifier, and perform a bitwise AND operation between the user permission information and the access level by using a bitmask technology; if the result of the bitwise AND operation is not zero, it is determined that the user has access permission; obtain the hot permission data of the user from a multi-level cache, and if the cache does not exist, read and update the cache from a permission management database; Use the least recently used algorithm to eliminate expired cache data, and use the two-phase commit protocol to ensure cache consistency in a distributed environment; further comprising: predefining user roles in the system, checking the current role of the user and the corresponding permission configuration when the user accesses resources or performs operations, and allowing access or operation only when the role has the corresponding permission, providing a management interface to allow real-time adjustment of the role allocation of the user, automatically updating the permission after the role is changed, without the need to re-login, and recording the audit logs of the user role and permission change.

6. The method of claim 5, wherein, The pre-defined user roles in the system, when the user accesses resources or performs operations, check the current role of the user and the corresponding permission configuration, and only allow access or operation when the role has the corresponding permission, provide a management interface to allow real-time adjustment of the role allocation of the user, automatically update the permission after the role is changed, without the need to re-login, and record the audit logs of the user role and permission change, including: constructing a role permission database, the role permission database using a relational database to store user, role and permission information; developing an attribute-based access control middleware based on the information stored in the role permission database, the access control middleware intercepting user requests and obtaining user role and attribute information from a distributed cache; after obtaining the user role and attribute information, performing role allocation and adjustment through a role allocation management interface, the role allocation management interface providing RESTful API; According to the role allocation and adjustment result, design an audit log module, the audit log module records the user role and permission change operation; If the audit log module detects a user role and permission change, introduce a permission conflict detection algorithm, the permission conflict detection algorithm identifies and solves the permission conflict problem caused by multi-role allocation.

7. The method of claim 1, wherein, The distributed practical training environment is constructed, computing and storage resources are deployed on edge nodes, local processing of practical training services is realized, the distance and latency of data transmission are minimized, and through cooperation and data synchronization between nodes, including: receiving a distribution request carrying user request information, the distribution request being generated when the user initiates a practical training service; determining the nearest edge node according to the distribution request using a weighted round robin algorithm, the edge node being used to deploy containerized practical training services; obtaining local storage information of the edge node, and judging whether there is cache data matched with the user request information in the local storage information; if the cache data exists, obtaining the cache data from the local storage information; if the cache data does not exist, determining a data shard location using a consistent hashing algorithm, and obtaining node state information through a Raft consensus algorithm; obtaining required data according to the node state information and the data shard location, judging whether the required data has conflicts using a vector clock algorithm; if the required data has conflicts, solving the conflicts through an incremental synchronization method; if the required data does not have conflicts, obtaining the required data through a batch transmission method; performing encryption processing on the required data using a TLS encryption communication protocol, judging user access authority according to a preset role access control policy; if the user access authority is legal, sending the required data after encryption processing to the user.

Images