A Campus Network Service Management Method Based on Big Data and Distributed Storage

Abstract

This invention discloses a campus network service management method based on big data and distributed storage, comprising the following steps: S1, collecting and preprocessing data before writing it into a distributed storage unit; S2, extracting the preprocessed results within a preset time period to form a service data sequence; S3, performing state encoding on the service data sequence to construct a service feature sequence; S4, performing correlation analysis on the service feature sequence to obtain a service status sequence; S5, identifying anomalies in the service status sequence to generate a service anomaly sequence; S6, matching control parameters based on the service anomaly sequence to construct a control strategy sequence; S7, executing the control strategy sequence to obtain the network service management result. This invention, combined with the Mamba model, possesses advantages such as strong data carrying capacity, accurate anomaly identification, fast response speed, and high management stability.

Description

Technical Field

[0001] This invention relates to the field of network management technology, and in particular to a method for managing campus network services based on big data and distributed storage. Background Technology

[0002] The campus network is used to support office access, business access, resource access, and data transmission, and typically contains multiple network service nodes. As the campus expands, each network service node continuously generates access data, operational data, link data, and alarm data during operation, causing a rapid increase in the amount of data in the campus network and more frequent changes in node status. To ensure the stability of the campus network service, it is necessary to continuously monitor the operational status of each network service node and promptly adjust it when anomalies occur.

[0003] In existing technologies, most campus network service management adopts a centralized approach, aggregating data generated by various network service nodes to a management platform. The platform then assesses node status based on preset thresholds, fixed rules, or human experience, and performs resource adjustments, access control, and path switching. While this approach can meet basic management needs when the network size is small, it easily leads to problems such as high data storage pressure, low efficiency in historical data retrieval, and untimely anomaly detection as the number of network service nodes increases and data volume continues to grow.

[0004] To enhance data carrying capacity, some existing technologies have begun to employ distributed storage to disperse and store campus network data. However, these methods mostly only address data storage; during service management, status assessments still primarily rely on single-time-step data or fixed statistical results, lacking correlation analysis of resource occupancy, link stability, and service response changes over continuous time periods. Therefore, it is difficult to accurately identify abnormal service nodes. Furthermore, after anomaly identification, the control strategies generated by existing technologies are typically fixed, making it difficult to develop more targeted management results for different anomaly types.

[0005] Therefore, how to provide a campus network service management method based on big data and distributed storage is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a campus network service management method based on big data and distributed storage. This invention performs distributed storage and continuous time analysis on access data, operation data, link data, and alarm data generated by each network service node in the campus network. It uses the Mamba model to perform time-series correlation modeling of service status and adopts the Fast-MCD algorithm to identify abnormal service nodes. Then, it generates corresponding resource adjustment, access control, and path adjustment strategies based on the anomaly type, thereby realizing dynamic management of campus network services. It has the advantages of strong data carrying capacity, accurate anomaly identification, fast response speed, and high management stability.

[0007] A campus network service management method based on big data and distributed storage according to an embodiment of the present invention includes the following steps: S1. Collect access data, operation data, link data and alarm data of each network service node in the park network at each time step and preprocess them, and write the preprocessing results into the distributed storage unit. S2. Extract the preprocessing results of each network service node from the distributed storage unit according to a preset period, based on the current time step and within a fixed time period, and assemble them into a service data sequence in chronological order. S3. Perform selective state coding on the service data sequence using the Mamba model to generate the load characteristics, connectivity characteristics, and response characteristics of each network service node, and construct a service feature sequence. S4. Perform time-series correlation and context-series correlation operations on the service feature sequence in the Mamba model to determine the resource occupancy status, link stability status and service response status of each network service node, and obtain the service status sequence. S5. The Fast-MCD algorithm is used to perform anomaly identification on the service state sequence. Based on the identification results, abnormal service nodes are selected, the corresponding state anomaly types are determined, and a service anomaly sequence is generated. S6. Based on the service exception sequence, perform a matching operation in the preset policy parameter library to determine the resource adjustment parameters, access control parameters and path adjustment parameters corresponding to each state exception type, and construct the control policy sequence. S7. Based on the control strategy sequence, perform resource reallocation, access permission reconfiguration, and transmission path adjustment operations on each network service node to obtain the network service management results.

[0008] Optionally, the network service node refers to a device entity that provides data transmission, service access, resource processing, and service response in the campus network; the access data refers to user request records, access source records, and access frequency records received by the network service node within a preset time range; the operation data refers to processor utilization, storage utilization, task execution volume, and resource usage records generated by the network service node during operation; the link data refers to the bandwidth, latency, packet loss rate, and link connectivity records corresponding to the transmission links between network service nodes; and the alarm data refers to fault alarms, abnormal alarms, and threshold exceedance alarm records generated by the network service node during operation. The preprocessing includes missing field completion, outlier removal, time alignment, and unified format conversion operations. The distributed storage unit refers to multiple storage nodes used for distributed storage and collaborative reading of data corresponding to each network service node. The strategy parameter library refers to a data set used to store control parameters corresponding to different state anomaly types.

[0009] Optionally, S3 specifically includes: S31. Merge the access data, operation data, link data and alarm data in the service data sequence according to the network service node and time sequence, and construct the node data sequence of each network service node according to the time sequence. S32. Extract the number of requests, access frequency, processor utilization, storage utilization, link bandwidth, link latency, packet loss rate, alarm count and task execution volume for each time step from the node data sequence, and combine them in a preset order to form a node feature vector and construct a node feature sequence. S33. Input the node feature sequence into the Mamba model. In the Mamba model, perform selective state coding based on the node feature vector of the current time step and the node feature vector of the previous time step to obtain an effective feature sequence. In the first time step, use the node feature vector of the current time step as the node feature vector of the previous time step. S34. Based on the processor utilization rate, storage utilization rate and task execution volume in the effective feature sequence, perform resource usage correlation calculation to obtain the load feature sequence; S35. Based on the link bandwidth, link delay and packet loss rate in the effective feature sequence, perform link connectivity correlation calculation to obtain the connectivity feature sequence; S36. Based on the number of requests, access frequency, and alarm count in the effective feature sequence, perform access behavior correlation calculation to obtain the response feature sequence; S37. Combine the load characteristic sequence, connectivity characteristic sequence, and response characteristic sequence according to the network service node and time order to obtain the service characteristic sequence.

[0010] Optionally, the selective state coding operation specifically includes: The node feature vector at the current time step is compared with the node feature vector at the previous time step, and the change value of each feature item between adjacent time steps is determined. The feature items include the number of requests, access frequency, processor utilization, storage utilization, link bandwidth, link latency, packet loss rate, alarm count and task execution volume. Based on the change values ​​of each feature item, the change intensity of the node feature vector at the current time step is determined. Feature items whose absolute change value is greater than the corresponding preset change threshold are determined as update feature items, and feature items whose absolute change value is not greater than the corresponding preset change threshold are determined as maintainable feature items. Initialize the encoded feature vector of the current time step according to the number of feature terms in the feature vector of the current time step node; For updating a feature term, the corresponding feature value in the current time step is written into the corresponding feature term position in the encoded feature vector of the current time step; For the feature term to be preserved, the corresponding feature value in the previous time step is written into the corresponding feature term position in the encoded feature vector of the current time step; In the Mamba model, state encoding is performed on the encoded feature vector at the current time step, and the state encoding results of each time step are arranged in chronological order to construct an effective feature sequence.

[0011] Optionally, the resource usage association calculation, link connectivity association calculation, and access behavior association calculation specifically include: Extract the processor utilization, storage utilization, and task execution volume corresponding to each time step in the effective feature sequence; Calculate the change in processor utilization between each time step and the previous time step, calculate the change in storage utilization between each time step and the previous time step, and calculate the change in task execution volume between each time step and the previous time step. In the first time step, the processor utilization, storage utilization, and task execution volume of the current time step are used as the processor utilization, storage utilization, and task execution volume of the previous time step, respectively. The resource change vector is constructed by combining the changes in occupancy, storage, and tasks in a preset order. Arrange the resource change vectors in chronological order to obtain the load characteristic sequence; Extract the link bandwidth, link latency, and packet loss rate at each time step from the effective feature sequence; Calculate the bandwidth change between the link bandwidth at each time step and the link bandwidth at the previous time step, calculate the latency change between the link latency at each time step and the link latency at the previous time step, and calculate the packet loss change between the packet loss rate at each time step and the packet loss rate at the previous time step. In the first time step, the link bandwidth, link latency, and packet loss rate of the current time step are respectively used as the link bandwidth, link latency, and packet loss rate of the previous time step. Extract the link connectivity records between each network service node from the link data, and determine the link connectivity status between each network service node at each time step based on the link connectivity records. By combining the bandwidth change value, latency change value, packet loss change value and link connectivity status in chronological order, a connectivity feature sequence is obtained; Extract the number of requests, access frequency, and alarm count at each time step from the effective feature sequence; Calculate the change in the number of requests at each time step compared to the number of requests at the previous time step, calculate the change in the access frequency at each time step compared to the access frequency at the previous time step, and calculate the change in the number of alarms at each time step compared to the number of alarms at the previous time step. In the first time step, the number of requests, access frequency, and number of alarms at the current time step are respectively used as the number of requests, access frequency, and number of alarms at the previous time step. Extract the access source records corresponding to each time step from the access data, and classify the number of requests and access frequency according to the access source records to obtain the number of requests and access frequency of each access source in the current time step. Then compare the number of requests and access frequency of each access source in the current time step with the number of requests and access frequency of the corresponding access source in the previous time step to calculate the access change value corresponding to each access source. Extract fault alarms, abnormal alarms, and threshold exceedance alarm records corresponding to each time step from the alarm data, and classify the alarm change values ​​according to the fault alarms, abnormal alarms, and threshold exceedance alarm records; The response feature sequence is obtained by combining the request change value, frequency change value, access change value corresponding to each access source, and the classified alarm change value in chronological order.

[0012] Optionally, S4 specifically includes: S41. According to the network service node and time order, align the load features, connectivity features and response features in the service feature sequence, and combine the load features, connectivity features and response features of the same network service node at the same time step in a preset order to obtain the time step feature sequence of each network service node. S42. Input the time step feature sequence of each network service node into the Mamba model. Read the load feature, connectivity feature and response feature of the current time step in the Mamba model in sequence, and compare them with the load feature, connectivity feature and response feature of the previous time step item by item to determine the change of each feature item in the three types of features. In the first time step, use the load feature, connectivity feature and response feature of the current time step as the load feature, connectivity feature and response feature of the previous time step. S43. Based on the changes in each feature item in the load characteristics, perform time-series correlation operation on the load characteristics of multiple consecutive time steps, and perform context correlation processing on the time-series correlation results. Determine the resource occupancy status of each network service node based on the context correlation results. S44. Based on the changes in each feature item in the connectivity features, perform time-series correlation operations on the connectivity features of multiple consecutive time steps, and combine the link connectivity status to perform context correlation processing on the time-series correlation results, and determine the link stability status of each network service node based on the context correlation results. S45. Based on the changes in each feature item in the response features, perform time-series correlation operation on the response features of multiple consecutive time steps, and combine the access change value corresponding to each access source with the classified alarm change value to perform context correlation processing on the time-series correlation results, and determine the service response status of each network service node based on the context correlation results. S46. Combine the resource occupancy status, link stability status, and service response status of each network service node according to time steps to obtain the service status sequence of each network service node.

[0013] Optionally, S5 specifically includes: S51. For each network service node, extract the resource occupancy status, link stability status and service response status of each time step from the corresponding service status sequence, and combine them to construct a node status vector to obtain the node status sequence of the corresponding network service node. S52. Using the Fast-MCD algorithm, perform a center state extraction operation on the node state sequence of each network service node to determine the normal state center of each network service node. S53. Based on the normal state center, calculate the degree of state deviation of the state vector of each time step node, and filter the abnormal time steps corresponding to the abnormal service nodes according to the preset deviation threshold. S54. Classify and determine the abnormal status type of the abnormal service node at the abnormal time step, based on its resource occupancy status, link stability status, and service response status. S55. Associate and combine abnormal service nodes, abnormal time steps and abnormal status types to generate a service abnormal sequence.

[0014] Optionally, S52 specifically includes: S521. For each network service node, extract the node state vector in the node state sequence in chronological order, and write the node state vector into the candidate state sample set in sequence. S522. Randomly select multiple state sample combinations from the candidate state sample set according to a preset number of samples, calculate the dispersion of resource occupancy status, link stability status and service response status in each state sample combination, and take the state sample combination with the smallest dispersion as the initial central sample set. S523. Based on the state vectors of each node in the initial central sample set, determine the central state values ​​corresponding to the resource occupancy state, link stability state, and service response state respectively, and combine the central state values ​​to obtain the initial central state. S524. Based on the initial center state, calculate the degree of deviation of each node state vector in the candidate state sample set from the initial center state, sort them in ascending order of deviation, and select the node state vectors at the top of the sorted preset number to form the updated center sample set. S525. Based on the state vectors of each node in the updated central sample set, redetermine the central state values ​​corresponding to the resource occupancy state, link stability state, and service response state, and combine the new central state values ​​to obtain the normal state center.

[0015] Optionally, S6 specifically includes: S61. Extract the resource adjustment parameters, access control parameters and path adjustment parameters corresponding to each state exception type from the policy parameter library, and establish the parameter correspondence according to the state exception type to obtain the policy parameter sequence. S62. Based on the state anomaly type in the service anomaly sequence, extract the resource adjustment parameters, access control parameters and path adjustment parameters corresponding to each anomaly service node at each anomaly time step from the policy parameter sequence to obtain the node control parameter sequence. S63. Perform parameter mapping operation on the node control parameter sequence to determine the resource reallocation content, access permission reconfiguration content and transmission path adjustment content corresponding to each abnormal service node at each abnormal time step, and combine them according to the abnormal service node and abnormal time step to construct a control strategy sequence.

[0016] Optionally, S7 specifically includes: S71. Based on the resource reallocation content in the control strategy sequence, perform resource reallocation operations on each abnormal service node to obtain a resource adjustment result sequence. S72. Based on the access permission reconfiguration content in the control strategy sequence, perform access permission reconfiguration operations on each abnormal service node to obtain the access adjustment result sequence. S73. Based on the transmission path adjustment content in the control strategy sequence, perform transmission path adjustment operations on each abnormal service node to obtain a path adjustment result sequence. S74. Combine the resource adjustment result sequence, permission adjustment result sequence, and path adjustment result sequence to obtain the network service management result.

[0017] The beneficial effects of this invention are: First, this invention preprocesses the data generated by each network service node in the park network and writes it into a distributed storage unit, thereby realizing the decentralized storage and collaborative reading of massive amounts of data in the park network. This solves the problems of high storage pressure, low data writing efficiency, and difficulty in retrieving historical data under the existing centralized management method, and improves data storage capacity and data processing efficiency.

[0018] Secondly, this invention utilizes the Mamba model to perform time-series analysis on service data within a continuous time period, and combines it with the Fast-MCD algorithm to identify abnormal service nodes and status anomaly types. Compared with existing methods that rely on single time step data or fixed thresholds for judgment, this invention can more accurately identify resource occupancy anomalies, link anomalies, and service response anomalies, thereby improving the accuracy of anomaly identification.

[0019] Finally, this invention generates corresponding resource adjustment, access control, and path adjustment strategies based on the type of abnormal status, and performs targeted regulation on abnormal service nodes, thereby improving the real-time performance, relevance, and stability of campus network service management. Attached Figure Description

[0020] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a campus network service management method based on big data and distributed storage proposed in this invention; Figure 2 This is a flowchart illustrating the service status analysis of a campus network service management method based on big data and distributed storage proposed in this invention. Figure 3 This is a flowchart illustrating the anomaly identification and control process of a campus network service management method based on big data and distributed storage proposed in this invention. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0022] refer to Figures 1-3 A campus network service management method based on big data and distributed storage includes the following steps: S1. Collect access data, operation data, link data and alarm data of each network service node in the park network at each time step and preprocess them, and write the preprocessing results into the distributed storage unit. S2. Extract the preprocessing results of each network service node from the distributed storage unit according to a preset period, based on the current time step and within a fixed time period, and assemble them into a service data sequence in chronological order. S3. Perform selective state coding on the service data sequence using the Mamba model to generate the load characteristics, connectivity characteristics, and response characteristics of each network service node, and construct a service feature sequence. S4. Perform time-series correlation and context-series correlation operations on the service feature sequence in the Mamba model to determine the resource occupancy status, link stability status and service response status of each network service node, and obtain the service status sequence. S5. The Fast-MCD algorithm is used to perform anomaly identification on the service state sequence. Based on the identification results, abnormal service nodes are selected, the corresponding state anomaly types are determined, and a service anomaly sequence is generated. S6. Based on the service exception sequence, perform a matching operation in the preset policy parameter library to determine the resource adjustment parameters, access control parameters and path adjustment parameters corresponding to each state exception type, and construct the control policy sequence. S7. Based on the control strategy sequence, perform resource reallocation, access permission reconfiguration, and transmission path adjustment operations on each network service node to obtain the network service management results.

[0023] In this embodiment, a network service node refers to a device entity that provides data transmission, service access, resource processing, and service response in the campus network. Access data refers to the user request records, access source records, and access frequency records received by the network service node within a preset time range. Operation data refers to the processor utilization rate, storage utilization rate, task execution volume, and resource usage records generated by the network service node during operation. Link data refers to the bandwidth, latency, packet loss rate, and link connectivity records corresponding to the transmission links between network service nodes. Alarm data refers to the fault alarms, abnormal alarms, and threshold exceedance alarm records generated by the network service node during operation. Preprocessing includes missing field completion, outlier removal, time alignment, and unified format conversion. Distributed storage units represent multiple storage nodes used to distribute and collaboratively retrieve data corresponding to each network service node. The policy parameter library represents a data set used to store control parameters corresponding to different state anomaly types.

[0024] In this embodiment, S3 specifically includes: S31. Merge the access data, operation data, link data and alarm data in the service data sequence according to the network service node and time sequence, and construct the node data sequence of each network service node according to the time sequence. S32. Extract the number of requests, access frequency, processor utilization, storage utilization, link bandwidth, link latency, packet loss rate, alarm count and task execution volume for each time step from the node data sequence, and combine them in a preset order to form a node feature vector and construct a node feature sequence. S33. Input the node feature sequence into the Mamba model. In the Mamba model, perform selective state coding based on the node feature vector of the current time step and the node feature vector of the previous time step to obtain an effective feature sequence. In the first time step, use the node feature vector of the current time step as the node feature vector of the previous time step. S34. Based on the processor utilization rate, storage utilization rate and task execution volume in the effective feature sequence, perform resource usage correlation calculation to obtain the load feature sequence; S35. Based on the link bandwidth, link delay and packet loss rate in the effective feature sequence, perform link connectivity correlation calculation to obtain the connectivity feature sequence; S36. Based on the number of requests, access frequency, and alarm count in the effective feature sequence, perform access behavior correlation calculation to obtain the response feature sequence; S37. Combine the load characteristic sequence, connectivity characteristic sequence, and response characteristic sequence according to the network service node and time order to obtain the service characteristic sequence.

[0025] In this embodiment, the selective state coding operation specifically includes: The node feature vector at the current time step is compared with the node feature vector at the previous time step, and the change value of each feature item between adjacent time steps is determined. The feature items include the number of requests, access frequency, processor utilization, storage utilization, link bandwidth, link latency, packet loss rate, alarm count and task execution volume. Based on the change values ​​of each feature item, the change intensity of the node feature vector at the current time step is determined. Feature items whose absolute change value is greater than the corresponding preset change threshold are determined as update feature items, and feature items whose absolute change value is not greater than the corresponding preset change threshold are determined as maintainable feature items. Initialize the encoded feature vector of the current time step according to the number of feature terms in the feature vector of the current time step node; For updating a feature term, the corresponding feature value in the current time step is written into the corresponding feature term position in the encoded feature vector of the current time step; For the feature term to be preserved, the corresponding feature value in the previous time step is written into the corresponding feature term position in the encoded feature vector of the current time step; In the Mamba model, state encoding is performed on the encoded feature vector at the current time step, and the state encoding results of each time step are arranged in chronological order to construct an effective feature sequence.

[0026] In this embodiment, the resource usage association calculation, link connectivity association calculation, and access behavior association calculation specifically include: Extract the processor utilization, storage utilization, and task execution volume corresponding to each time step in the effective feature sequence; Calculate the change in processor utilization between each time step and the previous time step, calculate the change in storage utilization between each time step and the previous time step, and calculate the change in task execution volume between each time step and the previous time step. In the first time step, the processor utilization, storage utilization, and task execution volume of the current time step are used as the processor utilization, storage utilization, and task execution volume of the previous time step, respectively. The resource change vector is constructed by combining the changes in occupancy, storage, and tasks in a preset order. Arrange the resource change vectors in chronological order to obtain the load characteristic sequence; Extract the link bandwidth, link latency, and packet loss rate at each time step from the effective feature sequence; Calculate the bandwidth change between the link bandwidth at each time step and the link bandwidth at the previous time step, calculate the latency change between the link latency at each time step and the link latency at the previous time step, and calculate the packet loss change between the packet loss rate at each time step and the packet loss rate at the previous time step. In the first time step, the link bandwidth, link latency, and packet loss rate of the current time step are respectively used as the link bandwidth, link latency, and packet loss rate of the previous time step. Extract the link connectivity records between each network service node from the link data, and determine the link connectivity status between each network service node at each time step based on the link connectivity records. By combining the bandwidth change value, latency change value, packet loss change value and link connectivity status in chronological order, a connectivity feature sequence is obtained; Extract the number of requests, access frequency, and alarm count at each time step from the effective feature sequence; Calculate the change in the number of requests at each time step compared to the number of requests at the previous time step, calculate the change in the access frequency at each time step compared to the access frequency at the previous time step, and calculate the change in the number of alarms at each time step compared to the number of alarms at the previous time step. In the first time step, the number of requests, access frequency, and number of alarms at the current time step are respectively used as the number of requests, access frequency, and number of alarms at the previous time step. Extract the access source records corresponding to each time step from the access data, and classify the number of requests and access frequency according to the access source records to obtain the number of requests and access frequency of each access source in the current time step. Then compare the number of requests and access frequency of each access source in the current time step with the number of requests and access frequency of the corresponding access source in the previous time step to calculate the access change value corresponding to each access source. Extract fault alarms, abnormal alarms, and threshold exceedance alarm records corresponding to each time step from the alarm data, and classify the alarm change values ​​according to the fault alarms, abnormal alarms, and threshold exceedance alarm records; The response feature sequence is obtained by combining the request change value, frequency change value, access change value corresponding to each access source, and the classified alarm change value in chronological order.

[0027] In this embodiment, S4 specifically includes: S41. According to the network service node and time order, align the load features, connectivity features and response features in the service feature sequence, and combine the load features, connectivity features and response features of the same network service node at the same time step in a preset order to obtain the time step feature sequence of each network service node. S42. Input the time step feature sequence of each network service node into the Mamba model. Read the load feature, connectivity feature and response feature of the current time step in the Mamba model in sequence, and compare them with the load feature, connectivity feature and response feature of the previous time step item by item to determine the change of each feature item in the three types of features. In the first time step, use the load feature, connectivity feature and response feature of the current time step as the load feature, connectivity feature and response feature of the previous time step. S43. Based on the changes in each feature item in the load characteristics, perform time-series correlation operation on the load characteristics of multiple consecutive time steps, and perform context correlation processing on the time-series correlation results. Determine the resource occupancy status of each network service node based on the context correlation results. S44. Based on the changes in each feature item in the connectivity features, perform time-series correlation operations on the connectivity features of multiple consecutive time steps, and combine the link connectivity status to perform context correlation processing on the time-series correlation results, and determine the link stability status of each network service node based on the context correlation results. S45. Based on the changes in each feature item in the response features, perform time-series correlation operation on the response features of multiple consecutive time steps, and combine the access change value corresponding to each access source with the classified alarm change value to perform context correlation processing on the time-series correlation results, and determine the service response status of each network service node based on the context correlation results. S46. Combine the resource occupancy status, link stability status, and service response status of each network service node according to time steps to obtain the service status sequence of each network service node.

[0028] In this embodiment, S43 specifically includes: S431. Extract the load characteristics of the same network service node in multiple consecutive time steps, arrange them in chronological order, and construct a load time series segment. S432. Based on the changes of each feature item between adjacent time steps in the load time sequence segment, perform a sequential association operation on the load feature of the current time step and the load feature of the previous time step to obtain the load association result corresponding to each time step. In the first time step, the load feature of the current time step is used as the load feature of the previous time step. S433. Perform context association operation on the load association results of multiple consecutive time steps in chronological order to determine the continuous segments, abrupt segments, and repetitive segments of load characteristics change between multiple consecutive time steps, and obtain the context association results. S434. Based on the context association results, merge and determine the load characteristics in multiple consecutive time steps to determine the resource occupancy status of each network service node in each time step.

[0029] In this embodiment, S44 specifically includes: S441. Extract the connectivity features of the same network service node in multiple consecutive time steps, arrange them in chronological order, and construct a connectivity time sequence segment. S442. Based on the changes of each feature item between adjacent time steps in the connected time sequence segment, perform a sequential association operation on the connectivity feature of the current time step and the connectivity feature of the previous time step to obtain the connectivity association result corresponding to each time step. In the first time step, the connectivity feature of the current time step is used as the connectivity feature of the previous time step. S443. Align and combine the link connectivity status with the connectivity association results at each time step to obtain the connectivity combination result; S444. Perform context association operation on the connectivity combination results of multiple consecutive time steps in chronological order to determine the continuous segments, abrupt segments, and repeated segments of connectivity features between multiple consecutive time steps, and obtain the context association results. S445. Based on the context association results, merge and determine the connectivity features in multiple consecutive time steps to determine the link stability status of each network service node in each time step.

[0030] In this embodiment, S45 specifically includes: S451. Extract the response features of the same network service node in multiple consecutive time steps, arrange them in chronological order, and construct a response time sequence segment. S452. Based on the changes of each feature item between adjacent time steps in the response time sequence segment, perform a sequential association operation on the response feature of the current time step and the response feature of the previous time step to obtain the response association result of each time step. In the first time step, the response feature of the current time step is used as the response feature of the previous time step. S453. Align and combine the access change values ​​corresponding to each access source, the classified alarm change values, and the response association results of each time step to obtain the response combination result. S454. Perform context association operation on the combined response results of multiple consecutive time steps in chronological order to determine the continuous segments, abrupt segments, and repetitive segments of the response characteristics between multiple consecutive time steps, and obtain the context association results. S455. Based on the context association results, merge and determine the response features in multiple consecutive time steps to determine the service response status of each network service node in each time step.

[0031] In this embodiment, S5 specifically includes: S51. For each network service node, extract the resource occupancy status, link stability status and service response status of each time step from the corresponding service status sequence, and combine them to construct a node status vector to obtain the node status sequence of the corresponding network service node. S52. Using the Fast-MCD algorithm, perform a center state extraction operation on the node state sequence of each network service node to determine the normal state center of each network service node. S53. Based on the normal state center, calculate the degree of state deviation of the state vector of each time step node, and filter the abnormal time steps corresponding to the abnormal service nodes according to the preset deviation threshold. S54. Classify and determine the abnormal status type of the abnormal service node at the abnormal time step, based on its resource occupancy status, link stability status, and service response status. S55. Associate and combine abnormal service nodes, abnormal time steps and abnormal status types to generate a service abnormal sequence.

[0032] In this embodiment, S52 specifically includes: S521. For each network service node, extract the node state vector in the node state sequence in chronological order, and write the node state vector into the candidate state sample set in sequence. S522. Randomly select multiple state sample combinations from the candidate state sample set according to a preset number of samples, calculate the dispersion of resource occupancy status, link stability status and service response status in each state sample combination, and take the state sample combination with the smallest dispersion as the initial central sample set. S523. Based on the state vectors of each node in the initial central sample set, determine the central state values ​​corresponding to the resource occupancy state, link stability state, and service response state respectively, and combine the central state values ​​to obtain the initial central state. S524. Based on the initial center state, calculate the degree of deviation of each node state vector in the candidate state sample set from the initial center state, sort them in ascending order of deviation, and select the node state vectors at the top of the sorted preset number to form the updated center sample set. S525. Based on the state vectors of each node in the updated central sample set, redetermine the central state values ​​corresponding to the resource occupancy state, link stability state, and service response state, and combine the new central state values ​​to obtain the normal state center.

[0033] In this embodiment, the calculation of the degree of dispersion specifically includes: For each combination of state samples, extract the resource occupancy status, link stability status, and service response status from the state vector of each node; For all resource occupancy states in the state sample combination, sort them according to their numerical values, and select the resource occupancy state in the middle position of the sort as the center value of the resource occupancy state. When the number of resource occupancy states in the state sample combination is even, select the smaller value of the two resource occupancy states in the middle position as the center value of the resource occupancy state. The center value of the link stable state and the center value of the service response state are determined in the same way. Calculate the state difference between the resource occupancy state and the center value of the resource occupancy state in the state vector of each node and summarize them to obtain the discrete value of resource occupancy. Calculate the state difference between the link stable state and the center value of the link stable state in the state vector of each node and summarize them to obtain the link stable discrete value; Calculate the state difference between the service response state and the center value of the service response state in the state vector of each node, and summarize them to obtain the discrete value of the service response. By combining the discrete values ​​of resource consumption, link stability, and service response, the degree of dispersion of the corresponding state sample combination is obtained.

[0034] In this embodiment, S6 specifically includes: S61. Extract the resource adjustment parameters, access control parameters and path adjustment parameters corresponding to each state exception type from the policy parameter library, and establish the parameter correspondence according to the state exception type to obtain the policy parameter sequence. S62. Based on the state anomaly type in the service anomaly sequence, extract the resource adjustment parameters, access control parameters and path adjustment parameters corresponding to each anomaly service node at each anomaly time step from the policy parameter sequence to obtain the node control parameter sequence. S63. Perform parameter mapping operation on the node control parameter sequence to determine the resource reallocation content, access permission reconfiguration content and transmission path adjustment content corresponding to each abnormal service node at each abnormal time step, and combine them according to the abnormal service node and abnormal time step to construct a control strategy sequence.

[0035] In this embodiment, S7 specifically includes: S71. Based on the resource reallocation content in the control strategy sequence, perform resource reallocation operations on each abnormal service node to obtain a resource adjustment result sequence. S72. Based on the access permission reconfiguration content in the control strategy sequence, perform access permission reconfiguration operations on each abnormal service node to obtain the access adjustment result sequence. S73. Based on the transmission path adjustment content in the control strategy sequence, perform transmission path adjustment operations on each abnormal service node to obtain a path adjustment result sequence. S74. Combine the resource adjustment result sequence, permission adjustment result sequence, and path adjustment result sequence to obtain the network service management result.

[0036] Example 1: To verify the feasibility of this invention in practice, it was applied to a large-scale campus network service environment. This campus network has multiple network service nodes, each responsible for office access, data storage, business access, video transmission, identity authentication, and link forwarding functions. As the number of terminals within the campus continues to increase, these network service nodes continuously generate access data, operational data, link data, and alarm data during operation. Because the original management method used centralized storage and fixed threshold judgment, as the data volume increased, problems gradually emerged, including slow historical data retrieval, inability to detect some anomalies in a timely manner, delayed link switching, and unstable service response.

[0037] Under the existing management method, when the processor utilization of a network service node exceeds a preset threshold, the management platform directly performs resource expansion for that network service node; when the link latency exceeds a preset threshold, the management platform directly switches the link; and when the access frequency increases, it directly restricts some access permissions. Although this method can complete basic management, it often leads to misjudgments because it only relies on data from a single time step and does not consider the changes in resource utilization, link changes, and access behavior over a continuous period of time. Some network service nodes may have high instantaneous access volumes, but they will automatically recover in subsequent time steps; some links may experience short-term increases in latency, but this does not actually affect service transmission; and some network service nodes may have problems with continuously increasing resource utilization, deteriorating link quality, and increased alarm frequency over multiple consecutive time steps, but these problems are not identified in time because a single indicator has not yet reached the threshold.

[0038] This invention first preprocesses the access data, operational data, link data, and alarm data generated by each network service node and writes them into a distributed storage unit. The data from each network service node is distributed across multiple storage nodes. During subsequent service management, data within a fixed-length preceding time period is extracted from the distributed storage unit based on the current time step to construct a service data sequence. Then, the Mamba model is used to selectively encode the service data sequence, and correlation analysis is performed on changes in processor utilization, storage utilization, link bandwidth, link latency, access frequency, and alarm count over consecutive time periods to obtain the resource utilization status, link stability status, and service response status of each network service node. Furthermore, the Fast-MCD algorithm is used to identify abnormal service nodes in the service status sequence and determine the anomaly type.

[0039] During a single operation, a network service node exhibited a continuous increase in processor utilization, a gradual increase in link latency, and an abnormally concentrated source of access over multiple consecutive time steps. The original management method only detected that the processor utilization did not exceed a threshold for a short period, thus failing to trigger control measures. However, this invention, through analysis of the service status over consecutive time periods, identified abnormal resource utilization and service response of the network service node. It then matched corresponding resource adjustment parameters and access control parameters from the policy parameter library, performing resource reallocation and access permission reconfiguration on the network service node. After the adjustments were completed, the processor utilization of the network service node gradually decreased, and the service response returned to stability.

[0040] Another network service node experienced a decrease in link bandwidth, an increase in latency, and a continuously rising packet loss rate during operation. The original approach only switched links after the latency reached a threshold, resulting in prolonged service interruptions. This invention, however, identifies abnormal link trends in advance based on changes in the link's stable state over a continuous time period and adjusts path parameters accordingly to complete transmission path switching ahead of time, thereby ensuring continuous service operation.

[0041] To further verify the effectiveness of the present invention, the operational results of using the original management method and adopting the present invention were compared, and the following data were obtained: Table 1 Comparison of Network Service Management Effectiveness in the Park

[0042] As shown in Table 1, under the same number of network service nodes and the same data scale, this invention significantly reduces the historical data retrieval time and data write failure rate through distributed storage units, indicating that this invention can effectively solve the performance degradation problem caused by centralized storage of massive data in campus networks. Simultaneously, this invention utilizes the Mamba model to analyze service status over continuous time periods and combines it with the Fast-MCD algorithm for anomaly identification, increasing the accuracy of anomaly service node identification from 78.4% to 95.8%, with a significant decrease in both false negative and false positive rates. This demonstrates that this invention can more accurately detect anomalies in resource usage, link anomalies, and service response anomalies.

[0043] Furthermore, this invention can identify link anomalies before they fully develop and perform resource reallocation and path adjustment in advance according to control strategies. Therefore, the average anomaly detection time, service interruption duration, and path adjustment completion time are all significantly reduced. Ultimately, the service availability of the campus network increased from 92.6% to 99.1%, and network resource utilization increased from 68.5% to 86.7%, verifying the excellent application effect of this invention in complex campus network environments.

[0044] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A campus network service management method based on big data and distributed storage, characterized in that, Includes the following steps: S1. Collect access data, operation data, link data and alarm data of each network service node in the park network at each time step and preprocess them, and write the preprocessing results into the distributed storage unit. S2. Extract the preprocessing results of each network service node from the distributed storage unit according to a preset period, based on the current time step and within a fixed time period, and assemble them into a service data sequence in chronological order. S3. Perform selective state coding on the service data sequence using the Mamba model to generate the load characteristics, connectivity characteristics, and response characteristics of each network service node, and construct a service feature sequence. S4. Perform time-series correlation and context-series correlation operations on the service feature sequence in the Mamba model to determine the resource occupancy status, link stability status and service response status of each network service node, and obtain the service status sequence. S5. The Fast-MCD algorithm is used to perform anomaly identification on the service state sequence. Based on the identification results, abnormal service nodes are selected, the corresponding state anomaly types are determined, and a service anomaly sequence is generated. S6. Based on the service exception sequence, perform a matching operation in the preset policy parameter library to determine the resource adjustment parameters, access control parameters and path adjustment parameters corresponding to each state exception type, and construct the control policy sequence. S7. Based on the control strategy sequence, perform resource reallocation, access permission reconfiguration, and transmission path adjustment operations on each network service node to obtain the network service management results.

2. The campus network service management method based on big data and distributed storage according to claim 1, characterized in that, The network service node refers to the device entity that provides data transmission, service access, resource processing, and service response in the campus network. The access data refers to the user request records, access source records, and access frequency records received by the network service node within a preset time range. The operation data refers to the processor utilization rate, storage utilization rate, task execution volume, and resource usage records generated by the network service node during operation. The link data refers to the bandwidth, latency, packet loss rate, and link connectivity records corresponding to the transmission links between network service nodes. The alarm data refers to the fault alarms, abnormal alarms, and threshold exceedance alarm records generated by the network service node during operation. The preprocessing includes missing field completion, outlier removal, time alignment, and unified format conversion operations. The distributed storage unit refers to multiple storage nodes used for distributed storage and collaborative reading of data corresponding to each network service node. The strategy parameter library refers to a data set used to store control parameters corresponding to different state anomaly types.

3. The campus network service management method based on big data and distributed storage according to claim 1, characterized in that, S3 specifically includes: S31. Merge the access data, operation data, link data and alarm data in the service data sequence according to the network service node and time sequence, and construct the node data sequence of each network service node according to the time sequence. S32. Extract the number of requests, access frequency, processor utilization, storage utilization, link bandwidth, link latency, packet loss rate, alarm count and task execution volume for each time step from the node data sequence, and combine them in a preset order to form a node feature vector and construct a node feature sequence. S33. Input the node feature sequence into the Mamba model. In the Mamba model, perform selective state coding based on the node feature vector of the current time step and the node feature vector of the previous time step to obtain an effective feature sequence. In the first time step, use the node feature vector of the current time step as the node feature vector of the previous time step. S34. Based on the processor utilization rate, storage utilization rate and task execution volume in the effective feature sequence, perform resource usage correlation calculation to obtain the load feature sequence; S35. Based on the link bandwidth, link delay and packet loss rate in the effective feature sequence, perform link connectivity correlation calculation to obtain the connectivity feature sequence; S36. Based on the number of requests, access frequency, and alarm count in the effective feature sequence, perform access behavior correlation calculation to obtain the response feature sequence; S37. Combine the load characteristic sequence, connectivity characteristic sequence, and response characteristic sequence according to the network service node and time order to obtain the service characteristic sequence.

4. The campus network service management method based on big data and distributed storage according to claim 3, characterized in that, The selective state coding operation specifically includes: The node feature vector at the current time step is compared with the node feature vector at the previous time step, and the change value of each feature item between adjacent time steps is determined. The feature items include the number of requests, access frequency, processor utilization, storage utilization, link bandwidth, link latency, packet loss rate, alarm count and task execution volume. Based on the change values ​​of each feature item, the change intensity of the node feature vector at the current time step is determined. Feature items whose absolute change value is greater than the corresponding preset change threshold are determined as update feature items, and feature items whose absolute change value is not greater than the corresponding preset change threshold are determined as maintainable feature items. Initialize the encoded feature vector of the current time step according to the number of feature terms in the feature vector of the current time step node; For updating a feature term, the corresponding feature value in the current time step is written into the corresponding feature term position in the encoded feature vector of the current time step; For the feature term to be preserved, the corresponding feature value in the previous time step is written into the corresponding feature term position in the encoded feature vector of the current time step; In the Mamba model, state encoding is performed on the encoded feature vector at the current time step, and the state encoding results of each time step are arranged in chronological order to construct an effective feature sequence.

5. A campus network service management method based on big data and distributed storage according to claim 3, characterized in that, The resource usage association calculation, link connectivity association calculation, and access behavior association calculation specifically include: Extract the processor utilization, storage utilization, and task execution volume corresponding to each time step in the effective feature sequence; Calculate the change in processor utilization between each time step and the previous time step, calculate the change in storage utilization between each time step and the previous time step, and calculate the change in task execution volume between each time step and the previous time step. In the first time step, the processor utilization, storage utilization, and task execution volume of the current time step are used as the processor utilization, storage utilization, and task execution volume of the previous time step, respectively. The resource change vector is constructed by combining the changes in occupancy, storage, and tasks in a preset order. Arrange the resource change vectors in chronological order to obtain the load characteristic sequence; Extract the link bandwidth, link latency, and packet loss rate at each time step from the effective feature sequence; Calculate the bandwidth change between the link bandwidth at each time step and the link bandwidth at the previous time step, calculate the latency change between the link latency at each time step and the link latency at the previous time step, and calculate the packet loss change between the packet loss rate at each time step and the packet loss rate at the previous time step. In the first time step, the link bandwidth, link latency, and packet loss rate of the current time step are respectively used as the link bandwidth, link latency, and packet loss rate of the previous time step. Extract the link connectivity records between each network service node from the link data, and determine the link connectivity status between each network service node at each time step based on the link connectivity records. By combining the bandwidth change value, latency change value, packet loss change value and link connectivity status in chronological order, a connectivity feature sequence is obtained; Extract the number of requests, access frequency, and alarm count at each time step from the effective feature sequence; Calculate the change in the number of requests at each time step compared to the number of requests at the previous time step, calculate the change in the access frequency at each time step compared to the access frequency at the previous time step, and calculate the change in the number of alarms at each time step compared to the number of alarms at the previous time step. In the first time step, the number of requests, access frequency, and number of alarms at the current time step are respectively used as the number of requests, access frequency, and number of alarms at the previous time step. Extract the access source records corresponding to each time step from the access data, and classify the number of requests and access frequency according to the access source records to obtain the number of requests and access frequency of each access source in the current time step. Then compare the number of requests and access frequency of each access source in the current time step with the number of requests and access frequency of the corresponding access source in the previous time step to calculate the access change value corresponding to each access source. Extract fault alarms, abnormal alarms, and threshold exceedance alarm records corresponding to each time step from the alarm data, and classify the alarm change values ​​according to the fault alarms, abnormal alarms, and threshold exceedance alarm records; The response feature sequence is obtained by combining the request change value, frequency change value, access change value corresponding to each access source, and the classified alarm change value in chronological order.

6. The campus network service management method based on big data and distributed storage according to claim 1, characterized in that, S4 specifically includes: S41. According to the network service node and time order, align the load features, connectivity features and response features in the service feature sequence, and combine the load features, connectivity features and response features of the same network service node at the same time step in a preset order to obtain the time step feature sequence of each network service node. S42. Input the time step feature sequence of each network service node into the Mamba model. Read the load feature, connectivity feature and response feature of the current time step in the Mamba model in sequence, and compare them with the load feature, connectivity feature and response feature of the previous time step item by item to determine the change of each feature item in the three types of features. In the first time step, use the load feature, connectivity feature and response feature of the current time step as the load feature, connectivity feature and response feature of the previous time step. S43. Based on the changes in each feature item in the load characteristics, perform time-series correlation operation on the load characteristics of multiple consecutive time steps, and perform context correlation processing on the time-series correlation results. Determine the resource occupancy status of each network service node based on the context correlation results. S44. Based on the changes in each feature item in the connectivity features, perform time-series correlation operations on the connectivity features of multiple consecutive time steps, and combine the link connectivity status to perform context correlation processing on the time-series correlation results, and determine the link stability status of each network service node based on the context correlation results. S45. Based on the changes in each feature item in the response features, perform time-series correlation operation on the response features of multiple consecutive time steps, and combine the access change value corresponding to each access source with the classified alarm change value to perform context correlation processing on the time-series correlation results, and determine the service response status of each network service node based on the context correlation results. S46. Combine the resource occupancy status, link stability status, and service response status of each network service node according to time steps to obtain the service status sequence of each network service node.

7. The campus network service management method based on big data and distributed storage according to claim 1, characterized in that, S5 specifically includes: S51. For each network service node, extract the resource occupancy status, link stability status and service response status of each time step from the corresponding service status sequence, and combine them to construct a node status vector to obtain the node status sequence of the corresponding network service node. S52. Using the Fast-MCD algorithm, perform a center state extraction operation on the node state sequence of each network service node to determine the normal state center of each network service node. S53. Based on the normal state center, calculate the degree of state deviation of the state vector of each time step node, and filter the abnormal time steps corresponding to the abnormal service nodes according to the preset deviation threshold. S54. Classify and determine the abnormal status type of the abnormal service node at the abnormal time step, based on its resource occupancy status, link stability status, and service response status. S55. Associate and combine abnormal service nodes, abnormal time steps and abnormal status types to generate a service abnormal sequence.

8. A campus network service management method based on big data and distributed storage according to claim 7, characterized in that, S52 specifically includes: S521. For each network service node, extract the node state vector in the node state sequence in chronological order, and write the node state vector into the candidate state sample set in sequence. S522. Randomly select multiple state sample combinations from the candidate state sample set according to a preset number of samples, calculate the dispersion of resource occupancy status, link stability status and service response status in each state sample combination, and take the state sample combination with the smallest dispersion as the initial central sample set. S523. Based on the state vectors of each node in the initial central sample set, determine the central state values ​​corresponding to the resource occupancy state, link stability state, and service response state respectively, and combine the central state values ​​to obtain the initial central state. S524. Based on the initial center state, calculate the degree of deviation of each node state vector in the candidate state sample set from the initial center state, sort them in ascending order of deviation, and select the node state vectors at the top of the sorted preset number to form the updated center sample set. S525. Based on the state vectors of each node in the updated central sample set, redetermine the central state values ​​corresponding to the resource occupancy state, link stability state, and service response state, and combine the new central state values ​​to obtain the normal state center.

9. A campus network service management method based on big data and distributed storage according to claim 1, characterized in that, S6 specifically includes: S61. Extract the resource adjustment parameters, access control parameters and path adjustment parameters corresponding to each state exception type from the policy parameter library, and establish the parameter correspondence according to the state exception type to obtain the policy parameter sequence. S62. Based on the state anomaly type in the service anomaly sequence, extract the resource adjustment parameters, access control parameters and path adjustment parameters corresponding to each anomaly service node at each anomaly time step from the policy parameter sequence to obtain the node control parameter sequence. S63. Perform parameter mapping operation on the node control parameter sequence to determine the resource reallocation content, access permission reconfiguration content and transmission path adjustment content corresponding to each abnormal service node at each abnormal time step, and combine them according to the abnormal service node and abnormal time step to construct a control strategy sequence.

10. A campus network service management method based on big data and distributed storage according to claim 1, characterized in that, Specifically, S7 includes: S71. Based on the resource reallocation content in the control strategy sequence, perform resource reallocation operations on each abnormal service node to obtain a resource adjustment result sequence. S72. Based on the access permission reconfiguration content in the control strategy sequence, perform access permission reconfiguration operations on each abnormal service node to obtain the access adjustment result sequence. S73. Based on the transmission path adjustment content in the control strategy sequence, perform transmission path adjustment operations on each abnormal service node to obtain a path adjustment result sequence. S74. Combine the resource adjustment result sequence, permission adjustment result sequence, and path adjustment result sequence to obtain the network service management result.

Images