SRv6 path selection method based on knowledge graph

Abstract

The invention provides an SRv6 path selection method based on a knowledge graph. The method comprises the following steps: step S100, constructing a network knowledge graph library and a strategy and SID mapping library; step S200, performing natural semantic analysis on the business strategy input by the user to obtain business indexes; s300, arranging an SID sequence corresponding to the service strategy according to the service index, and if the service strategy input by the user is similar to the historical strategy meeting the condition, querying the strategy and SID mapping library, and outputting an SID sequence corresponding to the similar historical strategy; and if there is no similar historical strategy, mapping the service index into a specific network path weight and a constraint condition, calculating to obtain a physical path conforming to the service index, converting each node or link in the path into a corresponding SRv6 SID, arranging into an SID sequence, and storing the SID sequence and the corresponding strategy in a strategy and SID mapping library.

Description

TECHNICAL FIELD

[0001] The application relates to the field of communication, in particular to an SRv6 path selection method based on a knowledge graph. BACKGROUND

[0002] SRv6 (Segment Routing over IPv6) is a new generation of IP carrying protocol, which uses existing IPv6 forwarding technology to realize network programmability through flexible IPv6 extension header. SRv6 simplifies the network protocol type, has good expansibility and programmability, and can meet the diversified needs of more new services. Users can find a path from a source node to a destination node according to the SRv6 protocol, and the path needs to meet certain resource requirements (such as bandwidth, delay, etc.). The patent with application number 202210205838.9 provides a computing power scheduling method, system and data center. The implementation approach of the patent is "receiving a computing power scheduling request, configuring computing power resources and setting a computing power identifier, generating a computing power-aware routing table based on the computing power identifier and the routing table, then finding a scheduling path according to the SRv6 protocol, and finally scheduling resources". However, the "computing power scheduling request" described in the patent is usually a resource specification declaration or a request pointing to a specific computing power identifier, which cannot understand and process complex, multi-dimensional and non-quantitative business objectives. This makes the system become a passive "resource carrier" rather than an active "business guarantor", and can only ensure that the "computing power is found and connected", but cannot guarantee whether the quality of the connected path can meet the real experience requirements of the upper application. SUMMARY

[0003] The purpose of the present application is to provide an SRv6 path selection method based on a knowledge graph, comprising: Step S100, constructing a network knowledge graph library and a policy and SID mapping library; Step S200, performing natural semantic analysis on the business policy input by the user to obtain a business index; Step S300, arranging a SID sequence corresponding to the business policy according to the business index, if the business policy input by the user is similar to a historical policy that meets the conditions, querying the policy and SID mapping library, and outputting a SID sequence corresponding to the similar historical policy; if there is no similar historical policy, mapping the business index into specific network path weights and constraint conditions, calculating a physical path that meets the business index, converting each node or link in the path into a corresponding SRv6 SID, arranging the SRv6 SIDs into a SID sequence, and storing the SID sequence and the corresponding policy in the policy and SID mapping library.

[0004] Further, it further comprises step S400, which specifically comprises: Step S401, when the SLA index exceeds the threshold, generating an alarm information; Step S402, the network management system parses the alarm information, extracts key identifiers, including source IP address, destination IP address, service flow ID, SID sequence, and application type, protocol, port, etc. Step S403, according to the key identifiers, locating the physical path and the carried service policy; Step S404, the network management system detects the real-time state or historical performance data of each network element along the located physical path, and identifies the faulty element; Step S405, setting the inheritance constraint condition and the exclusion constraint condition, the inheritance constraint condition being the original constraint condition of the service policy, and the exclusion constraint condition excluding the identified faulty element; Step S406, recalculating the physical path, converting each node or link in the path into the corresponding SRv6 SID, and arranging into a SID sequence.

[0005] Further, in step S403, according to the identifier type, the locating process is divided into the following three cases: (1) the alarm information contains a SID sequence, Step S4031, finding the corresponding path through the SID sequence; Step S4032, finding the carried service policy through the corresponding relationship between the path and the service index; (2) the alarm information contains source IP and destination IP, Step S4033, finding the matching service policy using the service index where the source IP and the destination IP are located; Step S4034, finding the corresponding physical path and SID sequence through the service policy; (3) the alarm information contains a service flow ID, Step S4035, finding the corresponding service policy according to the service flow ID; Step S4036, finding the corresponding physical path and SID sequence through the service policy.

[0006] Further, step S200 specifically includes: Step S210, performing word segmentation processing on the input service policy text; Step S220, analyzing the sentence structure and grammatical relationship in the input service policy text according to the segmented words, establishing a structured representation, and obtaining the service index.

[0007] Further, if the description of the service policy is in Chinese, step S210 specifically includes: Step S211, cleaning the input business policy text, using a basic tokenizer for basic tokenization; Step S212, adjusting the sliding window to scan the basic tokenization results, and performing multi-level matching on the scanned words or adjacent word combinations and professional vocabulary in the network knowledge graph library; Step S213, performing part-of-speech tagging and named entity recognition on the matched words, wherein the named entities include business types, SLA indicators, network objects, and action instructions.

[0008] Further, the multi-level matching of step S212 is in the order of confidence from high to low, which is exact matching, synonym matching, and semantic matching. (1) The exact matching process includes traversing the basic tokenization and matching with the professional vocabulary in the network knowledge graph library. If the basic tokenization is exactly the same as the professional vocabulary, the matching result is output. (2) The synonym matching process includes querying the domain synonym dictionary to obtain all synonym variants of the word, using the synonym set to match with the network knowledge graph library, and if any synonym matches the feature keyword, the matching result is output. (3) The semantic matching process includes calculating the cosine similarity between the word embedding vector and each feature description vector, sorting by similarity from high to low, and selecting the feature with similarity exceeding the threshold and the maximum.

[0009] Further, step S220 specifically includes: Step S221, establishing the dependency relationship between the tokenized words, constructing a tree structure with the action instruction entity as the root node based on the maximum spanning tree algorithm, finding a tree connecting all the words, and making the sum of the weights of all edges in the tree maximum. Step S222, selecting the root node of the tree structure as the core action instruction in the input business policy text, and mapping the nodes in the tree to specific roles according to the pre-defined semantic role framework. Step S223, converting the recognized discrete semantic roles into a structured representation.

[0010] Further, step S300 determines the similarity between the user input business policy and the historical policy library by weighted calculation of the comprehensive similarity, and the comprehensive similarity is obtained by weighted summation of the business type similarity, SLA requirement similarity, and feature similarity. The business type similarity acquisition method includes: Step S311, constructing a three-level business type ontology business feature vector; Step S312, respectively calculate the service type ontology level similarity and the feature vector similarity between the service policy input by the user and the historical policy, and weightedly sum to obtain the service type similarity; wherein the service type ontology level similarity is obtained by calculating the distance of the two service types in the ontology tree; the feature vector similarity is obtained by calculating the cosine similarity between the feature vectors of the two service types; The SLA requirement similarity obtaining method comprises: Step S321, normalize each index in the SLA requirement; Step S322, weightedly add the similarity of each index to obtain the SLA requirement similarity; The feature similarity obtaining method comprises: Step S331, for the ordered feature, map it to a numerical value and then calculate the similarity; Step S332, for the unordered feature, directly judge whether it is equal, 1 for equal and 0 for not equal.

[0011] Further, in step S300, if there is no historical policy satisfying the threshold condition, the business indexes of the service policy input by the user are parsed, and the business indexes are mapped to specific network path weights and constraint conditions, specifically comprising: Step S341, according to the pre-defined mapping rule, convert the business indexes to the constraint conditions and optimization objectives of the network path calculation; Step S342, according to the mapping rule, generate specific constraint conditions; Step S343, according to the business indexes, determine several optimization objectives of the path calculation; Step S344, assign weights to each objective and perform weighted sum method for multi-objective optimization.

[0012] Further, the specific process of calculating the physical path in step S300 comprises Step S351, abstract the network as a graph, the nodes represent network devices and the edges represent links, each link has multiple attributes; Step S352, convert the constraint conditions obtained by mapping into constraints on the path; Step S353, according to the optimization objectives, calculate a comprehensive weight for each link; Step S354, use Dijkstra algorithm to obtain the shortest path from the source to the destination; Step S355, check whether the path satisfies all the constraint conditions, if not, adjust the comprehensive weight and re-execute step S353; if yes, return the path as the SRv6-aware path; Step S356, convert each node or link in the path into the corresponding SRv6 SID and arrange it into a SID sequence.

[0013] Further, the specific process of converting each node or link in the path into the corresponding SRv6 SID in step S356 includes: Step S3561, collect the topology information of the network through the network controller, and obtain the SRv6 function supported by each node; Step S3562, determine the SRv6 function to be used according to the service index; Step S3563, filter the original topology according to the SRv6 constraint, retain the nodes and links supporting the required SRv6 function, and establish an SRv6 capability constraint graph; Step S3564, use the constrained shortest path algorithm to calculate the path from the source node to the destination node on the SRv6 capability constraint graph; Step S3565, according to the calculated physical path, assign appropriate SIDs to each node in the path to form a SID sequence; Step S3566, verify whether the generated SID sequence meets the service index; Step S3567, issue the generated SID sequence to the source node, and the source node will use the SID sequence to encapsulate the data packet.

[0014] Compared with the prior art, the present application has the following advantages: (1) the present application uses "service policy" and decomposes it into multi-dimensional "service index" through natural semantic analysis, so that the system can directly understand the service intention; (2) the present application maps the service index into the dynamic constraint and optimization target of the network path, and performs weighted multi-objective optimization calculation, and the generated SID sequence is accurate and reliable; (3) when the SLA index exceeds the constraint condition, the present application identifies the path through the key identifier and detects the defective elements on the identified path, and automatically triggers the path re-optimization process.

[0015] The present application will be further described below in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 The present application is a method flowchart. DETAILED DESCRIPTION

[0017] In combination Figure 1 A SRv6 path selection method based on a knowledge graph, comprising: Step S100, the network management system (NMS) constructs a network knowledge graph library and a policy and SID mapping library; Step S200, the network management system performs natural semantic analysis on the service policy input by the service user to obtain the service index; Step S300, the network management system compiles the SID sequence corresponding to the service policy according to the service index; if the service policy input by the service user is similar to the historical policy that meets the condition, the policy and SID mapping library is queried, and the SID sequence corresponding to the similar historical policy is output; if there is no similar historical policy, the service index is mapped to specific network path weight and constraint condition, a physical path meeting the service index is calculated, each node or link in the path is converted into a corresponding SRv6 SID, a SID sequence is compiled, and the SID sequence and the corresponding policy are stored in the policy and SID mapping library.

[0018] In step S100, professional terms in the field of network services are collected, including service types, SLA requirements, network objects, etc., to build a network knowledge graph library. The service types are, for example, "video conference" and "VoIP", the SLA requirements are, for example, "low delay" and "high bandwidth", and the network objects are, for example, "dedicated line" and "tunnel".

[0019] The policy and SID mapping library in step S100 is a structured database or knowledge base, which functions to store user service policy descriptions, service indexes obtained after the service policy is parsed, and the mapping relationship between the SRv6 SID sequences calculated and verified for the implementation of the policy. The mapping key is usually the hash value of the policy content or a unique policy ID, which is used for fast retrieval and matching; the service policy description is the natural language policy input by the user or the standardized policy template; the service index is the bridge connecting the policy and the underlying network, which converts the fuzzy policy into precise and quantifiable network performance parameters through natural semantic parsing, for example, "low delay → latency < 50 ms", "high reliability → packet loss rate < 0.1%", and "large bandwidth → guaranteed bandwidth > 100 Mbps"; and the SID sequence is an ordered set of specific SRv6 SIDs verified by path calculation, which can meet the above service indexes, and each SID represents an operation to be performed by a network node (End SID) or a link (Adj SID).

[0020] The specific process of step S200 includes: Step S210, the input service policy text is subjected to word segmentation processing; Step S220, the sentence structure and grammatical relationship in the input service policy text are analyzed according to the segmented words, a structured representation is established, and the service index is obtained.

[0021] In step S210, the description form of the service policy can be Chinese or English, and if it is English, it can be parsed through the CLIP text encoder. This embodiment mainly describes the word segmentation processing process when the service policy is in Chinese form, which includes: Step S211, cleaning the input business policy text, removing irrelevant characters, standardizing expressions, and using a basic tokenizer for basic tokenization; Step S212, adjusting the sliding window to scan the basic tokenization results, and performing multi-level matching of the scanned words or adjacent word combinations with professional vocabulary in the network knowledge graph library. The multi-level matching is in the order of confidence from high to low, including exact matching, synonym matching, and semantic matching. Step S213, performing part-of-speech tagging and named entity recognition on the matched words. The named entities are shown in the following table.

[0022] Table 1: Named Entity Recognition Example Table Entity Type Identification Pattern Example Business Type Predefined Business Dictionary Video Conference, VoIP, Data Backup, Real-time Gaming SLA Indicator Adjective and Network Indicator Low Latency, High Bandwidth, Zero Packet Loss, Stable Network Object Network Component Term Path, Link, Tunnel, Connection Action Instruction Operation Verb Create, Delete, Modify, Guarantee (1) Exact matching process: traverse the basic tokenization and match with professional vocabulary in the network knowledge graph library. If the basic tokenization is exactly the same as the professional vocabulary, output the matching result. (2) Synonym matching process: for the basic tokenization that fails to match exactly, query the domain synonym dictionary to obtain all synonym variants of the word element. Match the synonym set with the network knowledge graph library. If any synonym matches the feature keyword, output the matching result with a medium confidence level. (3) Semantic matching process: calculate the cosine similarity between the word embedding vector and each feature description vector, sort them from high to low, and select the feature with the highest similarity that exceeds the threshold.

[0023] The specific process of step S220 includes: Step S221, establish the dependency relationship between the tokenized words, construct a tree structure with the action instruction as the root node based on the maximum spanning tree algorithm, find a tree that connects all the words, and make the sum of the weights of all edges in the tree maximum. Step S222, select the root node of the tree structure as the core action instruction in the input business policy text, and map the nodes in the tree to specific roles according to the pre-defined semantic role framework. Step S223, convert the identified discrete semantic roles into a structured representation.

[0024] The dependency relationship in step S221 includes the core dependency relationship and the Chinese-specific dependency relationship, as shown in the following table.

[0025] Table 2: Dependency Relationship List

[0026] The semantic role framework in step S222 is a triple structure, namely {role type, fill rule, output slot}. The role type defines a fixed set of semantic function tags commonly found in network business strategies; the fill rule specifies how to identify and extract the corresponding words or phrases from the dependency tree; and the output slot specifies which field in the structured representation the extracted information should ultimately be filled into.

[0027] Role types include core action instructions, target services / entities, subjects / constraints, attribute modifiers, degree / quantity, source address / endpoint, destination address / endpoint, time conditions, and methods / mechanisms. Descriptions and examples of each type are shown in the table below.

[0028] Table 3 Character Type Description Table Role Type Description Extraction Example from Business Policy Text Core Action Instruction The "verb" of a policy, indicating the core network operation to be performed Prioritize, Guarantee, Provide, Limit, Isolate, Re-route Target Service / Entity The object or service to which a policy is applied, usually guided by prepositions such as "for", "to", "towards", etc. Video Conference, VIP User, Financial System Traffic Object / Constraint Subject The network performance indicator or resource object directly acted upon or constrained by the action Latency, Bandwidth, Packet Loss Rate, Path Attribute Modification Further description or classification of the "Constraint Subject" End-to-End, Peak, Average, Interval Degree / Value Specific numerical requirements and comparison relationships for the "Constraint Subject" Less than 50 milliseconds, Greater than 100 Mbps, Minimum 1 Gbps Source Address / Endpoint The starting point of traffic to which the policy applies From Data Center A, Source IP 10.1.1.0 / 24 Destination Address / Endpoint The termination point of traffic to which the policy applies To the Internet, Destination IP 10.2.2.2 Time Condition The time range in which the policy takes effect From 9:00 to 18:00 on weekdays, Immediately Manner / Mechanism Optional ways to perform the action Through primary and backup paths, Using load balancing The fill rules are constructed based on dependency relationships, as shown in the table below.

[0029] Table 4. Dependency Relationships, Semantic Roles of Mappings, and Filling Rules Dependency Relationship Mapped Semantic Role Filler Rule Subject-Object Relationship Core Action Instruction ← Agent Determining the issuer of the action Object-Action Relationship Core Action Instruction → Object / Constraint Subject Determining the direct object of the action Attribute-Object Relationship Object → Attribute Modification / Degree / Value Extracting modification information of indicators Adverbial Relationship Core Action Instruction → Target Service / Entity / Manner / Mechanism Finding the adverb of the action Preposition Relationship Preposition → Destination Address / Source Address / Destination Extracting key entities after prepositions In step S223, the system predefines a structured template for business metrics, which is a JSON-like structure. Its fields correspond to various parameters required for network path calculation. The semantic roles identified in step S222 are then filled into the corresponding fields of the template according to the mapping rules. For example, if the input business strategy is "to ensure end-to-end latency of less than 50 milliseconds for video conferencing," the converted structured representation would be: json { "policy_intent": "guarantee_performance", "core_action": "protection", "target_service": { "type": "video conferencing" }, "constraints": { "metrics": [ { "type": "delay", "attribute": "end-to-end", "condition": "max", "value": 50, "unit": "milliseconds" } ]}, "source": null, / / Can be obtained from context or default policy "destination": null / / Can be obtained from context or default policy } Among them, "target_service" and "constraints" are slots.

[0030] In step S300, a weighted comprehensive similarity calculation is performed to determine the similarity between the user-input business policy and the policies in the historical policy database, quickly identifying the most similar historical policy and its successful SID sequence as candidates. This involves business type similarity, SLA requirement similarity (comparing thresholds such as latency, bandwidth, and packet loss rate), and feature similarity (such as priority and security level).

[0031] In step S300, the specific process of calculating the business type similarity includes: Step S311: Construct the business feature vector of the business type; construct the business type ontology, which can be divided into three levels as shown in the table below; represent each business type as a multi-dimensional feature vector, one of which can be represented as {real-time requirements, bandwidth sensitivity, latency sensitivity, packet loss tolerance}, and use numerical representation to represent the intensity level of each dimension. Table 5 List of Business Type Levels Step S312: Calculate the business type ontology level similarity and feature vector similarity between the user-input business strategy and the historical strategy to obtain the business type similarity. Specifically, the business type ontology level similarity is obtained by calculating the distance between the two business types in the ontology tree; the feature vector similarity is obtained by calculating the cosine similarity between the feature vectors of the two business types.

[0032] In step S300, the SLA requires the following specific steps for similarity calculation: Step S321: Normalize the latency index using an inverse proportional function, normalize the bandwidth index using a proportional function, and normalize the packet loss rate using an exponential decay function. Step S322: Calculate the similarity of each index and then perform weighted summation.

[0033] In step S300, the features include priority, security level, etc. These features may be discrete; for example, the priority could be {high, medium, low}, and the security level could be {1, 2, 3, 4, 5}. The specific process of feature similarity calculation includes: Step S331, if the feature is ordered (such as priority, security level), it can be mapped to a numerical value, and then the similarity is calculated; Step S332, if the feature is an unordered category, whether it is equal or not can be used, and equal is 1 and unequal is 0.

[0034] In step S300, the obtained service type similarity, SLA requirement similarity and feature similarity are weighted and summed to obtain a comprehensive similarity. For each historical policy, the comprehensive similarity of the service policy input by the user and the new policy is calculated, and then the comprehensive similarity is sorted from high to low. The highest score that meets the threshold is selected as the most similar policy, and the SID sequence corresponding to the historical policy is obtained. If there is no historical policy that meets the threshold condition, the business indicators of the service policy input by the user are analyzed, and the business indicators are mapped to specific network path weights and constraint conditions. The specific process includes: Step S341, according to the pre-defined mapping rule, the business indicators are converted into constraint conditions and optimization objectives of network path calculation; Step S342, according to the mapping rule, specific constraint conditions are generated, such as maximum delay, minimum bandwidth, maximum packet loss rate, etc. Step S343, according to the business indicators, the optimization objectives of path calculation are determined, such as minimizing delay, minimizing link utilization, maximizing bandwidth, etc. Step S344, in the case of multiple optimization objectives, each objective needs to be assigned a weight, and a weighted sum method is used for multi-objective optimization.

[0035] In step S300, each node or link in the path is converted into a corresponding SRv6 SID according to the following table, and arranged into a SID sequence.

[0036] Table 6 SID sequence list corresponding to node or link The mapping rule in step S341 can be represented as follows: Rule 1, set the optimization objective to minimize delay, and set the jitter constraint; Rule 2, the delay requirement is directly mapped to the maximum delay constraint of path calculation; Rule 3, the bandwidth requirement is mapped to the minimum bandwidth constraint of path calculation; Rule 4, the packet loss rate requirement is mapped to the maximum packet loss rate constraint of path calculation; Rule 5, high priority is mapped to priority constraint in path calculation, that is, the path should use high priority queue, and in the case of resource conflict, it should be allocated preferentially.

[0037] In steps S342 and S343, according to the mapping rules described above, the constraint conditions and optimization targets are set as follows for the example of video conference: (1) Constraint conditions Maximum delay: 50 ms; Minimum bandwidth: 10 Mbps; Maximum packet loss rate: 0.001; Maximum jitter: 10 ms; (2) Optimization targets Primary optimization target: Minimize delay (since video conference is low delay sensitive); Secondary optimization target: Maximize bandwidth (although video conference has bandwidth requirements, delay is more important, so bandwidth is a constraint, and bandwidth may not be primarily optimized in the optimization target unless there are multiple paths that meet the constraints).

[0038] In step S300, the specific process of calculating the physical path that meets the service indicators according to the network path weight and the constraint conditions includes Step S351: Abstract the network as a graph, where nodes represent network devices and edges represent links. Each link has multiple attributes, such as delay, bandwidth, utilization, cost, etc., according to service indicators and network performance. Step S352: Convert the constraint conditions obtained by mapping into constraints on the path. Step S353: Calculate a comprehensive weight for each link according to the optimization target. Step S354: Use Dijkstra's algorithm to find the shortest path from the source to the destination. Step S355: Check if the path meets all the constraint conditions. If not, adjust the weights and re-execute step S354. If it does, return the path as an SRv6-aware path. Step S356: Convert each node or link in the path to the corresponding SRv6 SID and arrange it into a SID sequence.

[0039] The specific process of step S356 includes: Step S3561: Collect the topology information of the network through the network controller, and understand the SRv6 functions supported by each node, such as whether it supports End, End.X, End.DT4 / End.DT6, End.B6, etc. Step S3562, according to the service index, determine the SRv6 function that needs to be used, for example, if the service needs to pass through a certain intermediate node, End.X SID may be used; if the service needs VPN function, End.DT4 / End.DT6 SID may be used; if the service needs traffic engineering, End.B6 SID may be used to guide the traffic to pass through the specified path; Step S3563, filter the original topology according to the SRv6 constraint, and retain the nodes and links that support the required SRv6 function, for example, if the service needs End.X SID, only the nodes that support End.X SID can be selected as intermediate nodes; Step S3564, use the constrained shortest path algorithm to calculate the path from the source node to the destination node on the SRv6 capability constraint graph, and consider both traditional constraints (such as bandwidth, delay, etc.) and SRv6 specific constraints (such as SID depth limit, nodes that must be passed through, etc.) when calculating; Step S3565, according to the calculated physical path, assign appropriate SIDs to each node in the path to form a SID sequence, for example, for each node on the path, select End SID, End.X SID, etc. according to its position and function; if the path needs to insert service functions (such as firewall, load balancing, etc.), corresponding SIDs also need to be inserted in the SID sequence; Step S3566, verify whether the generated SID sequence meets the service index, including the delay and bandwidth of the path, and the SRv6 specific requirements; Step S3567, issue the generated SID sequence to the source node, and the source node will use the SID sequence to encapsulate the data packet.

[0040] Step S3564 specifically includes: Based on the original topology and the SRv6 capability constraint graph of the nodes, wherein the nodes have SRv6 capability attributes, and the edges have traditional metrics (bandwidth, delay, etc.); Map the service index to traditional constraints and SRv6 specific constraints; Use the constrained shortest path algorithm (CSPF) to calculate the shortest path from the source node to the destination node.

[0041] When the SLA index exceeds the constraint condition, for example, the maximum delay exceeds 50ms, diagnose, re-optimize and update the affected service path, trigger the recalculation process of step S400, the specific process includes: Step S401, when the SLA index exceeds the threshold, generate an alarm information; Step S402, the network management system parses the alarm information and extracts key identifiers; the identifiers include source IP address, destination IP address, service flow ID, SID sequence, and application type, protocol, port, etc. Step S403, locate the physical path and the carried service policy according to the key identifiers; Step S404, the network management system detects the real-time state or historical performance data of each network element (node or link) along the located physical path, and identifies the faulty element; Step S405, set the inheriting constraint condition and the excluding constraint condition; Step S406, recalculate the physical path, convert each node or link in the path into the corresponding SRv6 SID, and arrange it into a SID sequence.

[0042] The service flow ID in step S402 is a service flow ID dynamically created by the system when the first matching service policy data packet appears.

[0043] In step S403, according to the identifier type, the location is divided into the following three cases: (1) The alarm information contains the SID sequence Step S4031, find the corresponding path through the SID sequence; Step S4032, find the carried service policy through the corresponding relationship between the path and the service index; (2) The alarm information contains the source IP and the destination IP Step S4033, find the matching service policy using the service index where the source IP and the destination IP are located; Step S4034, find the corresponding physical path and SID sequence through the service policy; (3) The alarm information contains the service flow ID Step S4035, find the corresponding service policy according to the service flow ID; Step S4036, find the corresponding physical path and SID sequence through the service policy.

[0044] The inheriting constraint condition in step S405 is the original constraint condition of the service policy, and the excluding constraint condition is to exclude the identified faulty element from the newly obtained physical path.

Claims

1. A knowledge graph based SRv6 path selection method, characterized in that, include: Step S100: Construct a network knowledge graph library and a policy and SID mapping library; Step S200: Perform natural language parsing on the business strategy input by the user to obtain business metrics; Step S300: Arrange the SID sequence corresponding to the business strategy according to the business indicators. If the business strategy input by the user is similar to the historical strategy that meets the conditions, query the strategy and SID mapping library and output the SID sequence corresponding to the similar historical strategy. If there is no similar superior historical strategy, map the business indicators to specific network path weights and constraints, calculate the physical path that meets the business indicators, convert each node or link in the path into the corresponding SRv6 SID, arrange it into a SID sequence, and store the SID sequence and the corresponding strategy in the strategy and SID mapping library.

2. The method of claim 1, wherein, It also includes step S400, which specifically includes: Step S401: When the SLA metric in the input business strategy exceeds the threshold, generate an alarm message; Step S402: The network management system parses the alarm information and extracts key identifiers, including source IP address, destination IP address, service flow ID, SID sequence, application type, protocol, port, etc. Step S403: Locate the physical path and the service policy carried by the key identifier; Step S404: The network management system detects the real-time status or historical performance data of each network element along the located physical path and identifies the faulty element. Step S405: Set inheritance constraints and exclusion constraints. Inheritance constraints are the original constraints of the business strategy, and exclusion constraints exclude the identified faulty elements. Step S406: Recalculate the physical path, convert each node or link in the path into a corresponding SRv6 SID, and arrange them into a SID sequence.

3. The method of claim 2, wherein, In step S403, the location process is divided into the following three cases according to the identifier type: (1) The alarm information contains a SID sequence. Step S4031: Find the corresponding path using the SID sequence; Step S4032: Find the business strategy carried by the path through the correspondence between the path and the business indicators; (2) The alarm information includes the source IP and destination IP. Step S4033: Use the service metrics of the source IP and destination IP to find the matching service strategy; Step S4034: Find the corresponding physical path and SID sequence through business strategy; (3) The alarm information includes the service flow ID. Step S4035: Find the corresponding business strategy based on the business flow ID; Step S4036: Find the corresponding physical path and SID sequence through the business strategy.

4. The method according to claim 1 or 2, characterized in that, Step S200 specifically includes: Step S210: Perform word segmentation on the input business strategy text; Step S220: Analyze the sentence structure and grammatical relationships in the input business strategy text based on the segmented words, establish a structured representation, and obtain business metrics; If the business strategy is described in Chinese, step S210 specifically includes: Step S211: Clean the input business strategy text and perform basic word segmentation using a basic word segmenter; Step S212: Adjust the sliding window to scan the basic word segmentation results, and perform multi-level matching of the scanned words or adjacent word combinations with professional vocabulary in the network knowledge graph database; Step S213: Perform part-of-speech tagging and named entity recognition on the matched words, where named entities include business type, SLA indicator, network object and action instruction.

5. The method of claim 4, wherein, The multi-level matching described in step S212, in order of confidence level, consists of exact matching, synonym matching, and semantic matching. (1) The precise matching process includes traversing the basic word segmentation and matching it with the professional vocabulary in the network knowledge graph database. If the basic word segmentation and the professional vocabulary are completely consistent, the matching result is output. (2) The synonym matching process includes querying the domain thesaurus, obtaining all synonym variants of the word element, matching the thesaurus with the network knowledge graph database, and outputting the matching result if any synonym matches the feature keyword. (3) The semantic matching process includes calculating the cosine similarity between the word embedding vector and each feature description vector, sorting them from high to low similarity, and selecting the feature with the largest similarity exceeding the threshold.

6. The method of claim 3, wherein, Step S220 specifically includes: Step S221: Establish the dependency relationship between words after word segmentation, construct a tree structure with the action instruction entity as the root node based on the maximum spanning tree algorithm, and find a tree that connects all words so that the sum of the weights of all edges in the tree is maximized. Step S222: Take the root node of the selected tree structure as the core action instruction in the input business strategy text, and map the nodes in the tree to specific roles according to the predefined semantic role framework; Step S223: The identified discrete semantic roles are converted into structured representations.

7. The method according to claim 1 or 2, characterized in that, Step S300 determines the similarity between the user-input business strategy and the strategies in the historical strategy library by calculating a weighted comprehensive similarity. The comprehensive similarity is obtained by weighted summation of business type similarity, SLA requirement similarity, and feature similarity. The method for obtaining business type similarity includes: Step S311: Construct the business feature vector of the three-level business type ontology; Step S312: Calculate the business type ontology level similarity and feature vector similarity between the user-input business strategy and the historical strategy, and sum them by weight to obtain the business type similarity; wherein the business type ontology level similarity is obtained by calculating the distance between the two business types in the ontology tree; the feature vector similarity is obtained by calculating the cosine similarity between the feature vectors of the two business types. SLA requires similarity acquisition methods to include: Step S321: Normalize the various indicators in the SLA requirements; Step S322: Weighted sum of the similarities of each indicator to obtain the SLA required similarity; Methods for obtaining feature similarity include: Step S331: For ordered features, map them to numerical values ​​and then calculate the similarity. Step S332: For unordered features, directly determine whether they are equal. If they are equal, the value is 1; if they are not equal, the value is 0.

8. The method of claim 7, wherein, In step S300, if there are no historical policies that meet the threshold conditions, the business metrics of the user-input business policy are parsed out, and the business metrics are mapped to specific network path weights and constraints, including: Step S341: Based on predefined mapping rules, convert business metrics into constraints and optimization objectives for network path calculation; Step S342: Generate specific constraints according to the mapping rules; Step S343: Determine several optimization objectives for path calculation based on business metrics; Step S344: Assign weights to each objective and perform multi-objective optimization using a weighted sum method.

9. The method according to claim 8, characterized in that, The specific process of calculating the physical path in step S300 includes: Step S351: Abstract the network into a graph, where nodes represent network devices and edges represent links, and each link has multiple attributes. Step S352: Transform the constraints obtained from the mapping into constraints on the path; Step S353: Calculate a comprehensive weight for each link based on the optimization objective; Step S354: Use Dijkstra's algorithm to obtain the shortest path from the source to the destination; Step S355: Check if the path meets all constraints. If not, adjust the overall weight and re-execute step S353. If it meets the constraints, return the path as an SRv6 perception path. Step S356: Convert each node or link in the path into a corresponding SRv6 SID and arrange them into a SID sequence.

10. The method according to claim 9, characterized in that, The specific process of converting each node or link in the path to the corresponding SRv6 SID in step S356 includes: Step S3561: Collect network topology information through the network controller and obtain the SRv6 functions supported by each node; Step S3562: Determine the SRv6 functions to be used based on business metrics; Step S3563: Filter the original topology according to SRv6 constraints, retain the nodes and links that support the required SRv6 functions, and establish an SRv6 capability constraint graph. Step S3564: Using the constrained shortest path algorithm, calculate the path from the source node to the destination node on the SRv6 capability constraint graph; Step S3565: Based on the calculated physical path, assign a suitable SID to each node in the path to form a SID sequence; Step S3566: Verify whether the generated SID sequence meets the business indicators; Step S3567: The generated SID sequence is sent to the source node, which will use the SID sequence to encapsulate data packets.

Images