A negotiation-driven method and system for multi-intent network telemetry configuration

Abstract

The application discloses a negotiation-driven method and system for multi-intent network telemetry configuration. First, a large language model is used to analyze and normalize natural language measurement intent, generating a standardized query pattern. Then, candidate sketches are filtered based on data flow sketch library coverage and pre-set device conditional error-resource profile library, generating measurement intent configuration (MIC). Subsequently, a topology capability graph is constructed based on network capability graph and current network state. Compatibility clustering is performed on multiple MICs, and consensus measurement envelope negotiation is performed on the topology capability graph. Then, ambiguity verification is performed through constraint subgraph matching from the measurement intent constraint graph to the topology capability graph. Finally, the verified consensus measurement envelope is locally solved and spliced into a global telemetry configuration. The application can improve the semantic consistency, structural compactness, solving efficiency and online maintainability of multi-intent network telemetry configuration, and reduce redundant deployment, configuration ambiguity and global recalculation overhead.

Description

Technical Field

[0001] This invention relates to the field of network telemetry and network management technology, and more specifically, to a negotiation-driven method and system for multi-intent network telemetry configuration, which can be used for multi-intent telemetry configuration generation, static bearer synthesis, envelope verification, local solution, and streaming update maintenance in programmable networks, software-defined networks, multi-tenant networks, and cloud data center environments. Background Technology

[0002] Network telemetry is used to collect and analyze network operational status, serving as a fundamental component for traffic anomaly detection, performance diagnosis, service assurance, and multi-tenant behavior analysis. With the development of programmable switches, removable network interface cards (NICs), software measurement modules, and data flow sketch (also known as sketch measurement structure) measurement methods, network telemetry has gradually shifted from passively collecting data on a small number of fixed indicators to proactively deploying and continuously observing data around specific business objectives. Among these, data flow sketches are a general summary structure in the network measurement field used to approximate the maintenance of statistics such as traffic frequency, cardinality, reflow, and variability under conditions of limited memory and high-speed data flow. The term "sketch" in the following text refers to this type of data flow sketch measurement structure or its specific algorithmic instances.

[0003] In real-world operations and maintenance scenarios, measurement requirements are typically not given as structured queries, but rather as high-level natural language intents. Examples include requests to continuously observe critical flows on shared core links, monitor sudden behaviors near hotspot edges, ensure accuracy boundaries on representative paths, or prioritize the timeliness and accuracy of certain tasks under resource constraints. These requirements often simultaneously impact the same network topology, the same set of paths, and the same limited number of device resources.

[0004] Existing telemetry configuration methods typically suffer from three main shortcomings. First, some methods require the input to be written as a structured query or operator graph, making it difficult to directly handle open, undefined, and potentially implicitly constrained multi-intent requests in natural language. Second, some methods focus on selecting data flow sketches, optimizing deployment locations, or allocating resources after the input has been given a structured input, lacking a comprehensive understanding of the structure shared by multiple intents before solving. Third, as new intents continue to arrive, existing methods often need to process them one by one or re-execute a large-scale configuration solution, which can easily lead to resource lock contention, redundant deployments, tail latency amplification, and historical configuration disturbances. Summary of the Invention

[0005] Through analysis of existing technologies, the inventors discovered that multi-intent network telemetry configuration requires the completion of natural language intent normalization, shared bearer structure synthesis, and streaming local updates before deployment and solution. Specifically, this includes: converting high-level natural language intents into controlled, discrete, and verifiable measurement intent configurations; synthesizing low-cost, low-ambiguity, and deployable shared bearer envelopes on the shared topology; and updating the affected envelopes and their local capability slices when streaming intents arrive.

[0006] Therefore, a negotiation-driven approach for multi-intent network telemetry configuration is needed, enabling the system to transform high-level natural language intents into deployable, verifiable, and locally maintainable global telemetry configurations according to the process of "intent parsing and configuration generation, topology capability abstraction, negotiation verification and solution".

[0007] Based on this, the present invention provides a negotiation-driven method and system for telemetry configuration in multi-intent networks, which solves the problems in existing telemetry configuration methods, such as the difficulty in converting natural language intents into executable telemetry configurations, the lack of modeling of multi-intent sharing relationships, the lack of static bearer synthesis and ambiguity verification in consensus measurement envelopes, and the easy triggering of global recalculation by streaming updates.

[0008] To achieve the above objectives, the present invention provides the following technical solution: The first aspect provides a negotiation-driven method for telemetry configuration in multi-intent networks, including: S1: Receives natural language measurement intent, processes it using a large language model, and generates a standardized query pattern; S2: For each normalized query pattern, combine the data flow sketch library coverage relationship with the pre-set device conditional error-resource profile library to generate a measurement intent configuration; where the data flow sketch library coverage relationship is used to characterize whether a specific sketch algorithm supports the corresponding query pattern, and the pre-set device conditional error-resource profile library is used to provide the basis for resource estimation of the sketch under different device types, node roles and error targets; S3: Construct a topology capability graph based on a pre-built network capability graph and the current network state; S4: Construct a measurement intent constraint graph for each measurement intent configuration, perform compatibility clustering on multiple measurement intent configurations, and negotiate a consensus measurement envelope on the topology capability graph to generate a consensus measurement envelope. The consensus measurement envelope represents a shared bearer object formed by a subset of compatible measurement intent configurations. S5: Perform constraint subgraph matching from the measurement intent constraint graph to the topology capability graph for each consensus measurement envelope, and perform ambiguity verification; S6: Perform envelope solving independently for each validated consensus measurement envelope to generate a local telemetry configuration, and then concatenate the local configurations into a global telemetry configuration.

[0009] In one implementation, S1 includes: S1.1: Receiving natural language measurement intent; S1.2: Utilize a large language model with knowledge enhancement and constraint decoding to perform multi-intent parsing and controlled normalization on the natural language measurement intent, extract the measurement intent label set, and generate a standardized query pattern.

[0010] In one implementation, S2 includes: S2.1: For each normalized query pattern, filter the candidate data flow sketch set that supports the normalized query pattern based on the data flow sketch library coverage relationship; S2.2: Based on the candidate data flow sketch set and the pre-set device conditional error-resource profile library, generate the intent instantiation error-resource profile of the current intent under different device and role combinations, and determine the path scope, structural support group, activation scope group and deployable device group. Use the candidate data flow sketch set, intent instantiation error-resource profile, path scope, structural support group, activation scope group and deployable device group as the measurement intent configuration.

[0011] In one implementation, the topology capability graph in S3 records node roles, device types, currently available resources, link capabilities, path attributes, and reachable path edges based on the physical topology. The reachable path edges are derived from routing tables, tunnel information, service chain paths, or path states maintained by the controller. They are used to indicate whether there is a valid observation path between two candidate devices or path anchors in the current network state, and record the upper bound of hop count, the lower bound of capacity, and path attribute labels.

[0012] In one implementation, S4 includes: S4.1: Configure and construct a measurement intention constraint graph for each measurement intention, wherein the nodes in the measurement intention constraint graph represent path endpoints, shared structure anchors, hotspot edge anchors, candidate measurement deployment points or intermediate carrier nodes, and the edges represent structural relationships that are located on the same path, adjacent to hotspot edges, must pass through a certain type of equipment or are limited by a certain capacity path. S4.2: Perform compatibility clustering on multiple measurement intent configurations based on one or more of the following: scope overlap, topological proximity, device candidate set overlap, accuracy tolerance interval, and resource attitude compatibility, to obtain a subset of compatible intents; S4.3: Construct a terminal group family for each compatible intent subset, obtain a low-cost connectivity shared core based on whether each terminal group in the restricted topology covers at least one candidate structural element, expand the activation range and candidate device set outside the shared core, and generate a consensus measurement envelope. The terminal group family is used to form group coverage constraints, and the restricted topology is obtained based on the compatible intent subset and the topology capability graph.

[0013] In one implementation, S5 includes: S5.1: For each consensus measurement envelope, check whether there is an embedding and check whether the embedding satisfies the constraints of node role, device type, edge capacity, path attribute, accuracy level and source tracing. Among them, the node role and device type constraints are used to limit the candidate deployment location, the edge capacity and path attribute constraints are used to limit the legal observation path, and the source tracing constraints are used to distinguish between user-given hard constraints and model completion soft constraints. S5.2: Ambiguity verification is performed from three dimensions: structural overlap, constraint relaxation, and mapping stability. If the verification result is a strict match or a controlled soft match, the subsequent local solution is entered. If the verification result is an ambiguous match or a hard constraint conflict, the relevant intent is split back into different envelopes or a local backoff correction is triggered. Structural overlap is used to compare the degree of overlap of supporting nodes, supporting edges, and deployment nodes of different intents in the consensus measurement envelope. Constraint relaxation is used to measure the proportion of wildcards or soft constraint backoffs used in the matching process. Mapping stability is used to evaluate whether there are too many alternative embeddings for the same intent within the envelope.

[0014] In one implementation, S6 includes: S6.1: Determine the data flow sketch instances and deployment locations required for each measurement intent configuration within the envelope; S6.2: Under the resource constraints of the candidate device set, allocate device and memory resources for each data flow sketch instance; S6.3: Based on the activation range group, configure the trigger range, acquisition strategy, and path boundary of the measurement task to obtain the local telemetry configuration; S6.4: Verify whether the local telemetry configuration meets the accuracy and resource constraints of each measurement intent configuration. If it does, then treat the local telemetry configuration as a component of the global telemetry configuration.

[0015] In one implementation, the method further includes processing continuously arriving new intents using dependency-aware micro-batch processing, specifically including: instantiating continuously arriving measurement intent configurations into scheduling primitives, each scheduling primitive recording a candidate envelope, a set of candidate devices, an envelope region to be written or modified, and a priority; constructing an update dependency graph based on the intersection of the candidate device sets and the intersection of the write regions between scheduling primitives; organizing scheduling primitives without write conflicts and device conflicts into the same micro-batch, and assigning scheduling primitives with dependencies to different micro-batches.

[0016] Based on the same inventive concept, a second aspect of this invention provides a negotiation-driven system for telemetry configuration in multi-intent networks, comprising: The multi-intent parsing and controlled normalization module is used to receive natural language measurement intents, process them using a large language model, and generate normalized query patterns. The measurement intent configuration generation module is used to generate a measurement intent configuration for each normalized query pattern by combining the data flow sketch library coverage relationship and the pre-set device conditional error-resource profile library. The data flow sketch library coverage relationship is used to characterize whether a specific sketch algorithm supports the corresponding query pattern, and the pre-set device conditional error-resource profile library is used to provide the basis for resource estimation of the sketch under different device types, node roles and error targets. The topology capability graph construction module is used to construct a topology capability graph based on a pre-built network capability graph and the current network state. The consensus measurement envelope negotiation module is used to construct a measurement intent constraint graph for each measurement intent configuration, perform compatibility clustering on multiple measurement intent configurations, and negotiate the consensus measurement envelope on the topology capability graph to generate a consensus measurement envelope. The consensus measurement envelope represents a shared bearer object formed by a subset of compatible measurement intent configurations. The ambiguity verification module is used to perform constraint subgraph matching from the measurement intent constraint graph to the topology capability graph for each consensus measurement envelope to perform ambiguity verification. The envelope solving and global stitching module is used to independently solve the envelope for each verified consensus measurement envelope, generate local telemetry configurations, and stitch the local configurations into a global telemetry configuration.

[0017] Based on the same inventive concept, a third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the negotiation-driven method for telemetry configuration of multi-intent network described in the first aspect.

[0018] Compared with the prior art, the advantages and beneficial technical effects of the present invention are as follows: This invention discloses a negotiation-driven method for multi-intent network telemetry configuration, constructing a processing flow from intent parsing and configuration generation, topology capability abstraction, to negotiation verification and local solution. The method first utilizes a large language model with knowledge enhancement and constraint decoding to perform multi-intent parsing and controlled normalization of natural language measurement intents, generating a standardized query pattern. Then, it filters candidate sketches based on the coverage relationship of the data flow sketch library and a pre-set device conditional error-resource profile library, generating intent instantiation error-resource profile entries corresponding to the current intent, forming a MIC (Measurement Intent Configuration). Subsequently, it constructs a topology capability graph based on the network capability graph and the current network state. Next, it performs compatibility clustering on multiple MICs to construct terminal groups, and synthesizes low-cost connectivity sharing cores on the topology capability graph, expanding to form a consensus measurement envelope. Then, it performs ambiguity verification by matching constraint subgraphs from the measurement intent constraint graph to the topology capability graph. Finally, it performs local solution on the verified consensus measurement envelope and concatenates them into a global telemetry configuration. In streaming scenarios, it maintains the envelope forest through dependency-aware micro-batch processing and local actions. This invention can reduce redundant deployment, configuration ambiguity, and global recalculation overhead. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart of a negotiation-driven method for telemetry configuration of multi-intent networks in an embodiment of the present invention. Figure 2 This is a schematic diagram of the static bearing synthesis process in an embodiment of the present invention; Figure 3 This is a schematic diagram of dependency-aware micro-batch processing for streaming envelope maintenance in an embodiment of the present invention; Figure 4 This is a block diagram of a negotiation-driven system for multi-intent network telemetry configuration in an embodiment of the present invention. Detailed Implementation

[0021] This embodiment provides a negotiation-driven method for telemetry configuration in multi-intent networks, including: S1: Receives natural language measurement intent, processes it using a large language model, and generates a standardized query pattern; S2: For each normalized query pattern, combine the data flow sketch library coverage relationship with the pre-set device conditional error-resource profile library to generate a measurement intent configuration; where the data flow sketch library coverage relationship is used to characterize whether a specific sketch algorithm supports the corresponding query pattern, and the pre-set device conditional error-resource profile library is used to provide the basis for resource estimation of the sketch under different device types, node roles and error targets; S3: Construct a topology capability graph based on a pre-built network capability graph and the current network state; S4: Construct a measurement intent constraint graph for each measurement intent configuration, perform compatibility clustering on multiple measurement intent configurations, and negotiate a consensus measurement envelope on the topology capability graph to generate a consensus measurement envelope. The consensus measurement envelope represents a shared bearer object formed by a subset of compatible measurement intent configurations. S5: Perform constraint subgraph matching from the measurement intent constraint graph to the topology capability graph for each consensus measurement envelope, and perform ambiguity verification; S6: Perform envelope solving independently for each validated consensus measurement envelope to generate a local telemetry configuration, and then concatenate the local configurations into a global telemetry configuration.

[0022] In this invention, MIC refers to Measurement-Intent Configuration, which carries the query pattern, structural support group, activation scope group, and deployable device group obtained by natural language intent normalization; Consensus Measurement Envelope (CME) refers to the shared bearer object formed for a subset of compatible MICs, including a shared connectivity support structure, local execution scope, candidate device set, and the carried intent subset.

[0023] like Figure 1 As shown, the method of this invention includes three stages: intent parsing and configuration generation, topology capability abstraction, and negotiation verification and solution. Stage 1 executes S1 and S2 to convert natural language measurement intent into MIC; Stage 2 executes S3 to construct a topology capability graph based on the network capability graph and the current network state; Stage 3 executes S4 to S6 to generate consensus measurement envelopes, perform constraint subgraph matching, and perform local solution and global splicing after verification.

[0024] For new intentions arriving via streaming, the system first executes S1 and S2 to generate the corresponding MIC, and then locates the candidate envelope based on the topology capability graph of S3. When the new intention affects the existing envelope, the system only performs local negotiation, verification, solution and state submission of S4 to S6 on the relevant envelope.

[0025] This invention abstracts the underlying network into a network capability graph with capability annotations. Nodes in this graph represent candidate telemetry deployment locations and observable devices, while edges represent physical or logical links. Node capability information describes device type, remaining resource budget, role labels, and a set of deployable data flow sketches, while edge capability information describes capacity, path attributes, and observability labels. The system also maintains the current network state to characterize existing telemetry configurations, device utilization, and reusable objects.

[0026] The input can be a batch of intents or a stream of intents arriving in chronological order. For each intent, the system generates a Measurement Intent Configuration (MIC). This configuration includes a candidate data stream sketch set, an intent instantiation error-resource profile, a normalized query pattern, a path scope, a structural support group, an activation scope group, a deployable device group, and device preferences. The normalized query pattern describes the measurement metrics, the observed object, the path or scope of action, accuracy constraints, and deployment preferences. The structural support group describes the key structural elements that must be covered. The activation scope group describes the context accessible to the local solution. The deployable device group describes candidate devices with valid roles and types. The intent instantiation error-resource profile is a profile entry obtained from the instantiation of the current intent.

[0027] In one implementation, S1 specifically includes: S1.1: Receiving natural language measurement intent; S1.2: Utilize a large language model with knowledge enhancement and constraint decoding to perform multi-intent parsing and controlled normalization on the natural language measurement intent, extract the measurement intent label set, and generate a standardized query pattern.

[0028] The model is provided with a controlled measurement label space, sketch-query coverage knowledge, topological capability summaries, and output pattern constraints, enabling it to map measurement targets, flow objects, path ranges, accuracy levels, and deployment preferences from open text to a finite structured semantic domain. The sketch-query coverage knowledge corresponds to the data flow sketch library coverage relation Cover(s,q) in S2: in S1, it is used to limit the output of the large language model to the query range supported by the sketch library; in S2, the formal coverage relation Cover(s,q) performs candidate sketch filtering.

[0029] Multi-intent parsing and controlled normalization include numerical semantic hierarchy, topological modularity abstraction, and source-based slot completion. Numerical semantic hierarchy is used to map continuous or unstable performance boundaries, sampling intensity, latency requirements, and resource budgets to finite levels. Topological modularity abstraction is used to normalize topological representations in natural language into controlled structural anchors. Source-based slot completion is used to perform restricted inferences on fields that are not explicitly given and to label the source type.

[0030] Specifically, the measurement intent label set includes one or more of the following: measurement metric type, traffic object type, path constraint type, accuracy level, and deployment preference type.

[0031] In one implementation, S2 specifically includes: S2.1: For each normalized query pattern, filter the candidate data flow sketch set that supports the normalized query pattern based on the data flow sketch library coverage relationship; S2.2: Based on the candidate data flow sketch set and the pre-set device conditional error-resource profile library, generate the intent instantiation error-resource profile of the current intent under different device and role combinations, and determine the path scope, structural support group, activation scope group and deployable device group. Use the candidate data flow sketch set, intent instantiation error-resource profile, path scope, structural support group, activation scope group and deployable device group as the measurement intent configuration.

[0032] In S2, the system maintains a data flow sketch library with a coverage relationship Cover(s,q) and a pre-defined device conditional error-resource profile library ProfileLib. Cover(s,q) is used to determine whether the sketch algorithm or sketch template s supports the normalized query pattern q. ProfileLib stores or provides the Profile(s,d,r,epsilon) function, which is used to estimate the memory, computation, and bandwidth resources required for sketch s under device type d, node role r, and target error epsilon. ProfileLib is the basis for generating MICs (Minimum Indicator Contexts). The error-resource profiles recorded in the MICs are instantiation results obtained by searching, interpolating, or calculating from ProfileLib based on the query pattern of the current intent, candidate sketches, candidate devices, node roles, and target errors. Based on the above information, the system filters a set of candidate data flow sketches for each intent and determines the path scope, structural support group, activation scope group, and deployable device group.

[0033] Through S1 and S2, the MIC output in Phase 1 simultaneously includes the normalized query from the semantic layer, the candidate support groups from the structural layer, and the device and resource preferences from the deployment layer. This MIC serves as the unified input for subsequent static bearer synthesis and envelope solving.

[0034] In one embodiment, the topology capability graph in S3 records node roles, device types, currently available resources, link capabilities, path attributes, and reachable path edges based on the physical topology. The reachable path edges are derived from routing tables, tunnel information, service chain paths, or path states maintained by the controller. They are used to indicate whether there is a valid observation path between two candidate devices or path anchors in the current network state, and record the upper bound of hop count, the lower bound of capacity, and path attribute labels.

[0035] S3 constructs a topology capability graph starting from the network capability graph G and the current network state C0. The topology capability graph includes physical devices, logical links, node roles, device types, available resources, link capabilities, and reachable path edges, providing a unified carrying space for subsequent negotiation, verification, and solution.

[0036] The network capability graph includes a set of nodes and a set of edges. Nodes represent candidate telemetry deployment locations, observable devices, or path anchors, while edges represent physical or logical links. Node capability information includes one or more of the following: device type, remaining resource budget, node role, set of deployable sketches, and current resource usage. Edge capability information includes one or more of the following: link capacity, latency, path attributes, observability label, and hotspot area label.

[0037] Reachable path edges are derived from routing tables, tunnel information, service chain paths, or path states maintained by the controller. They are used to indicate whether there is a valid observation path between two candidate devices or path anchors in the current network state, and record the upper bound of hop count, the lower bound of capacity, and path attribute labels.

[0038] In practical implementation, nodes in the topology capability graph represent observable devices, deployable devices, and path anchors. Node attributes include device type, whether it supports a programmable data plane, available memory, processing throughput, currently occupied resources, the measurement roles it can support, and supported data flow sketch types. Edges represent physical or logical connections; edge attributes include capacity, latency, packet loss risk, whether it is in a hotspot area, whether it belongs to a critical path, and whether it can be part of an observation path.

[0039] The method constructs reachable path edges based on routing tables, tunnel information, service chain paths, or path states maintained by the controller. Reachable path edges are used to indicate whether a valid observation path exists between two candidate devices or path anchors in the current network state, and record the upper bound of hop count, the lower bound of capacity, and path attribute labels.

[0040] In one implementation, S4 includes: S4.1: Configure and construct a measurement intention constraint graph for each measurement intention, wherein the nodes in the measurement intention constraint graph represent path endpoints, shared structure anchors, hotspot edge anchors, candidate measurement deployment points or intermediate carrier nodes, and the edges represent structural relationships that are located on the same path, adjacent to hotspot edges, must pass through a certain type of equipment or are limited by a certain capacity path. S4.2: Perform compatibility clustering on multiple measurement intent configurations based on one or more of the following: scope overlap, topological proximity, device candidate set overlap, accuracy tolerance interval, and resource attitude compatibility, to obtain a subset of compatible intents; S4.3: Construct a terminal group family for each compatible intent subset, obtain a low-cost connectivity shared core based on whether each terminal group in the restricted topology covers at least one candidate structural element, expand the activation range and candidate device set outside the shared core, and generate a consensus measurement envelope. The terminal group family is used to form group coverage constraints, and the restricted topology is obtained based on the compatible intent subset and the topology capability graph.

[0041] Specifically, S4 is for static bearer synthesis and consensus measurement envelope generation. After entering Phase 3, a measurement intent constraint graph is first constructed for each MIC. The nodes in the measurement intent constraint graph represent path endpoints, shared structure anchors, hotspot edge anchors, candidate measurement deployment points, and optional intermediate bearer nodes; the edges represent the structural relationships between these roles, such as being on the same path, adjacent to hotspot edges, requiring passage through a certain type of device, or being limited by a certain capacity path.

[0042] The nodes and edges in the measurement intent constraint graph carry constraint attributes, including node role, device type preference, upper bound of path length, lower bound of link capacity, observability requirements, accuracy level, and soft constraint markers. For the same MIC, if a requirement can be carried by multiple equivalent candidate structures, these candidate structures form a terminal group; if a requirement must cover several key structural elements simultaneously, each key structural element forms a corresponding terminal group.

[0043] Subsequently, compatibility clustering was performed on multiple MICs. Compatibility clustering was based on one or more of the following: scope overlap, topological proximity, device candidate set overlap, accuracy tolerance interval, and resource attitude compatibility. Scope overlap was used to determine whether different intentions have a structural basis that can be shared; accuracy tolerance interval was used to avoid merging intentions with obvious conflicting observation granularities into the same compatible intention subset; device candidate set overlap was used to determine whether multiple intentions compete for the same batch of device resources.

[0044] In the specific implementation process, for each compatible intent subset The system according to The structural support groups, activation range groups, and deployable device groups of each MIC are cropped on the topology capability map obtained in S3 to generate a restricted map. Subsequently, the system operates within a restricted graph. The shared core of the overlay terminal group is synthesized, and a consensus measurement envelope is generated around the shared core.

[0045] like Figure 2 As shown, static bearer synthesis includes compatible MIC grouping, shared core synthesis, and envelope expansion. Compatible MIC grouping is used to form candidate member sets, terminal families, and restricted graphs. Shared core synthesis is used to find low-cost connected trees covering each terminal group on a constrained graph; envelope expansion is used to generate local execution subgraphs and candidate device views around the shared core.

[0046] Formally, for a set of candidate members, the system constructs a corresponding restricted topology and terminal group family. The goal of static bearer synthesis is to find a low-cost connected tree in the restricted topology, such that the connected tree can cover at least one candidate structural element in each terminal group. This problem can be formulated as a group Steiner tree problem, that is, finding a low-cost connected skeleton that jointly covers these groups, given that the input is a group of terminals rather than fixed terminals.

[0047] In one implementation, a local heuristic expansion algorithm is used to approximate the solution for the shared core. Specifically, local neighborhoods are extracted around the terminal group to obtain a local candidate graph; multi-source expansion is simultaneously initiated from the candidate terminals of each terminal group, and propagation is carried out along the direction with lower cumulative cost; when the expansion fronts of different terminal groups intersect in the local candidate graph, the corresponding low-cost paths are restored and candidate trees are generated; coverage completion and leaf node shrinking are performed on the candidate trees, and redundant leaf nodes that do not affect connectivity and group coverage are deleted to obtain the shared core.

[0048] After the shared core is generated, a local execution context is constructed around the shared core, forming a consensus measurement envelope. A consensus measurement envelope includes a subset of compatible intentions carried, a shared connectivity support structure, a local topological scope extended around the shared core, and a set of candidate devices that satisfy device legitimacy constraints. Figure 2 In the diagram, the red path represents the shared core, the yellow nodes represent deployable candidate nodes, the bar charts indicate deployment readiness assessments, and the pruning markers represent heuristic GSTP (Steiner Tree Problem in Graphs) pruning operations during shared core synthesis.

[0049] The generation of the consensus measurement envelope satisfies the following constraints: First, sufficient capacity, meaning that the nodes, edges, and candidate devices within the envelope must be sufficient to support the structural support group and deployed device group of all MICs within it. Second, structural compactness, meaning that the introduction of irrelevant local topology should be minimized while satisfying the capacity constraints. Third, semantic consistency, meaning that the proportion of soft constraints or wildcard fallbacks used does not exceed a preset threshold.

[0050] In one implementation, S5 includes: S5.1: For each consensus measurement envelope, check whether there is an embedding and check whether the embedding satisfies the constraints of node role, device type, edge capacity, path attribute, accuracy level and source tracing. Among them, the node role and device type constraints are used to limit the candidate deployment location, the edge capacity and path attribute constraints are used to limit the legal observation path, and the source tracing constraints are used to distinguish between user-given hard constraints and model completion soft constraints. S5.2: Ambiguity verification is performed from three dimensions: structural overlap, constraint relaxation, and mapping stability. If the verification result is a strict match or a controlled soft match, the subsequent local solution is entered. If the verification result is an ambiguous match or a hard constraint conflict, the relevant intent is split back into different envelopes or a local backoff correction is triggered. Structural overlap is used to compare the degree of overlap of supporting nodes, supporting edges, and deployment nodes of different intents in the consensus measurement envelope. Constraint relaxation is used to measure the proportion of wildcards or soft constraint backoffs used in the matching process. Mapping stability is used to evaluate whether there are too many alternative embeddings for the same intent within the envelope.

[0051] Specifically, S5 is constraint subgraph matching and ambiguity verification.

[0052] In multi-intent scenarios, even resource-feasible carrying areas can lead to the improper merging of different intents. To address this, this invention performs constraint subgraph matching from the measurement intent constraint graph to the topology capability graph for each consensus measurement envelope before entering the local solution.

[0053] The constraint subgraph matching check examines the embedded relationships and their constraint satisfaction, including node roles, device types, edge capacities, path attributes, accuracy levels, and source tracing constraints. Among these, node role and device type constraints are used to limit candidate deployment locations, edge capacity and path attribute constraints are used to limit legal observation paths, and source tracing constraints are used to distinguish between user-given hard constraints and model completion soft constraints.

[0054] The system distinguishes three states based on the matching results. Strict matching indicates that the measurement intent constraint graph can be uniquely or stably embedded in the envelope without relaxing constraints. Soft matching indicates that the measurement intent constraint graph can be embedded after enabling controlled wildcards or soft constraint fallback, but this embedding needs to be labeled and monitored in subsequent solutions. Ambiguous matching indicates that multiple semantically different MICs have highly overlapping embedding sets in the same envelope, or that a certain MIC has a large number of semantically equivalent but structurally different embeddings, making it difficult to distinguish the final deployment relationships.

[0055] To quantify the degree of ambiguity, the system evaluates it from three dimensions: structural overlap, constraint relaxation, and mapping stability. Structural overlap is used to compare the degree of overlap of supporting nodes, supporting edges, and deployment nodes in the envelope for different intentions; constraint relaxation is used to measure the proportion of wildcards or soft constraint backoffs used during the matching process; and mapping stability is used to assess whether there are too many substitute embeddings for the same intention within the envelope. If the degree of ambiguity exceeds a threshold, the system splits the relevant intentions back into different envelopes or performs local backoff corrections on the MIC field and envelope structure.

[0056] In one implementation, S6 includes: S6.1: Determine the data flow sketch instances and deployment locations required for each measurement intent configuration within the envelope; S6.2: Under the resource constraints of the candidate device set, allocate device and memory resources for each data flow sketch instance; S6.3: Based on the activation range group, configure the trigger range, acquisition strategy, and path boundary of the measurement task to obtain the local telemetry configuration; S6.4: Verify whether the local telemetry configuration meets the accuracy and resource constraints of each measurement intent configuration. If it does, then treat the local telemetry configuration as a component of the global telemetry configuration.

[0057] Specifically, S6 is consensus envelope solving and global splicing.

[0058] For consensus measurement envelope verified by S5 Deployment solving is performed on envelope-induced local capability slices. Consensus measurement envelope. Depend on , , and Composition; among which, local capability slicing refers to slicing from the topological capability map of S3 according to... The range of nodes and edges The candidate device set is cropped into a subgraph with resource attributes, and then... The load-bearing frame must be retained. The shared connectivity support structure representing the consensus measurement envelope. This represents the activation range or local execution subgraph obtained by extending around the shared core. This represents the set of candidate devices within the local execution subgraph that satisfy the role, type, and resource constraints. This indicates the compatible MIC subset carried by the envelope.

[0059] After the envelope is solved according to steps S6.1 to S6.4, the local solution results of multiple envelopes are concatenated into a global telemetry configuration after a consistency check. The consistency check includes at least device resource conflict checks, path coverage omission checks, duplicate data acquisition checks, and global policy consistency checks. If an envelope solution fails, the system only performs renegotiation, constraint relaxation, or splitting and reconstruction on that envelope, without affecting other verified and successfully solved envelopes.

[0060] In one implementation, the method further includes processing continuously arriving new intents using dependency-aware micro-batch processing, specifically including: instantiating continuously arriving measurement intent configurations into scheduling primitives, each scheduling primitive recording a candidate envelope, a set of candidate devices, an envelope region to be written or modified, and a priority; constructing an update dependency graph based on the intersection of the candidate device sets and the intersection of the write regions between scheduling primitives; organizing scheduling primitives without write conflicts and device conflicts into the same micro-batch, and assigning scheduling primitives with dependencies to different micro-batches.

[0061] Specifically, this step corresponds to Figure 1 The closed-loop update part of S6, and by Figure 3 To elaborate further, in streaming scenarios, the system maintains a continuously updated runtime state, which includes the current consensus measurement envelope forest, local deployment results, global stitching state, and a set of feasible embeddings from intent to envelope.

[0062] like Figure 3 As shown, consecutively arriving MICs are converted into scheduling primitives and input into the dependency-aware micro-batch processor. Each scheduling primitive can be represented as... ,in As candidate envelope, For structure type or write region type, For the expected write area, For the set of candidate devices, Priority is given to this. Figure 3 In For simplification, where This indicates the target envelope and its core write region. This represents the set of candidate devices.

[0063] An update dependency graph is constructed based on the candidate device set and the expected write region. If the candidate device sets or write regions of two scheduling primitives intersect, they are assigned to different batches; otherwise, they can enter the same micro-batch for parallel preprocessing. Each micro-batch is submitted sequentially, and updates that modify the global state are executed serially according to priority.

[0064] After obtaining ordered micro-batches, the system processes batches A, Batch B, Batch C, etc., sequentially. For each batch, the system performs feasibility pre-screening, candidate pruning, and action evaluation in parallel within the batch; this modifies the global state. Updates are submitted serially according to batch order and priority. By batching based on dependencies, the system performs parallel preprocessing on scheduling primitives without dependencies and submits scheduling primitives with dependencies in batches.

[0065] For each newly arrived MIC, the system selects a local maintenance action based on the candidate envelope and the current resource state. These actions include: Attach (attaching the new intent to an existing envelope); Merge (merging two compatible and highly shared envelopes); Split (removing ambiguous or conflicting intents from the current envelope); and Rebuild-affected-subset (reconstructing only the affected subset of local envelopes). Action selection follows a constrained perturbation principle: prioritizing the stability of existing configurations, then limiting the scope of impact, and finally comparing the cost of local reconstruction.

[0066] Compared with the prior art, the present invention has the following beneficial effects: First, by converting natural language measurement intent into a controlled MIC, the risk of configuration inconsistencies or unexecutability caused by unstructured input is reduced.

[0067] Second, by generating candidate sketches and device preferences through sketch-query coverage relationships and error-resource profiles, the MIC can be directly called by subsequent bearer synthesis and local solution modules.

[0068] Third, by synthesizing the consensus measurement envelope before deployment and solving, the shared structure with compatible intentions is concentrated on the low-cost connectivity support structure, reducing the duplication of the carrying structure.

[0069] Fourth, by performing constraint subgraph matching at the envelope layer, hard constraint conflicts, soft constraint relaxations, and structural mapping ambiguities are identified before resource solving.

[0070] Fifth, by performing local solutions and global stitching on an envelope-by-envelope basis, the solution scope is limited to relevant local capability slices.

[0071] Sixth, by using dependency-aware micro-batch processing, updates without write conflicts and candidate device conflicts are preprocessed in parallel, and global configuration state consistency is maintained through ordered commits.

[0072] This invention also provides a negotiation-driven system for telemetry configuration in multi-intent networks; please refer to [link to relevant documentation]. Figure 4 ,include: The multi-intent parsing and controlled normalization module 101 is used to receive natural language measurement intents, process them using a large language model, and generate a normalized query pattern. The measurement intent configuration generation module 102 is used to generate a measurement intent configuration for each normalized query pattern by combining the data flow sketch library coverage relationship and the preset device conditional error-resource profile library. The data flow sketch library coverage relationship is used to characterize whether a specific sketch algorithm supports the corresponding query pattern, and the preset device conditional error-resource profile library is used to provide the basis for resource estimation of the sketch under different device types, node roles and error targets. The topology capability graph construction module 103 is used to construct a topology capability graph based on a pre-built network capability graph and the current network state. The consensus measurement envelope negotiation module 104 is used to construct a measurement intention constraint graph for each measurement intention configuration, perform compatibility clustering on multiple measurement intention configurations, and negotiate the consensus measurement envelope on the topology capability graph to generate a consensus measurement envelope. The consensus measurement envelope represents a shared bearer object formed oriented towards a subset of compatible measurement intention configurations. The ambiguity verification module 105 is used to perform constraint subgraph matching from the measurement intent constraint graph to the topology capability graph for each consensus measurement envelope to perform ambiguity verification. The envelope solving and global stitching module 106 is used to independently perform envelope solving for each verified consensus measurement envelope, generate local telemetry configurations, and stitch the local configurations into a global telemetry configuration.

[0073] Specifically, the consensus measurement envelope negotiation module includes a compatibility clustering unit, a terminal group construction unit, a shared core synthesis unit, and an envelope expansion unit; wherein, the shared core synthesis unit is used to solve or approximately solve for a low-cost connectivity skeleton covering the terminal group on a constrained topology. The streaming maintenance module includes a scheduling primitive instantiation unit, an update dependency graph construction unit, a micro-batch generation unit, and an ordered commit unit; wherein, the update dependency graph construction unit is used to determine the dependencies between different update requests based on the intersection of candidate device sets and the intersection of write regions.

[0074] As a preferred option, the system also includes: The streaming maintenance module is used to execute S1~S3 to normalize the new intents and locate candidate envelopes when new intents continue to arrive. Then, it instantiates scheduling primitives, constructs and updates the dependency graph, generates ordered micro-batches, and maintains the affected envelope forest and global configuration state through local actions such as Attach, Merge, Split, and Rebuild-affected-subset.

[0075] The closed-loop verification and update module is used to perform phased verification and online correction of the MIC field, envelope structure, candidate device view, and local solution strategy based on solution failures, resource violations, ambiguous matching, and deployment feedback.

[0076] This invention can be applied to scenarios such as carrier backbone networks, data center networks, enterprise networks, campus networks, industrial internet, and multi-tenant cloud networks. The method can be used in conjunction with programmable switches, P4 switches, SmartNICs, software telemetry agents, and various data flow sketch measurement modules. It can also be integrated with software-defined network controllers, network automation platforms, and intent-driven network management systems.

[0077] This invention can perform controlled parsing, measurement intent configuration generation, topology capability abstraction, consensus measurement envelope synthesis, ambiguity verification, local solution, global splicing, and streaming maintenance based on multiple high-level natural language measurement intents, reducing redundant configuration and global recalculation processes, and is suitable for network management scenarios that require continuous telemetry configuration and maintenance.

[0078] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the negotiation-driven method for telemetry configuration of multi-intent networks described above.

[0079] In specific implementation, the computer device may include a processor, a communications interface, memory, and a communication bus. The processor, communications interface, and memory communicate with each other via the communication bus. The processor can call logical instructions from memory to execute a negotiation-driven method for multi-intent network telemetry configuration, mainly including the software processing portion described above.

[0080] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0081] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0082] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various modifications and variations to the embodiments of the invention without departing from the spirit and scope of the invention. Thus, if these modifications and variations of the embodiments of the invention fall within the scope of the claims of the invention and their equivalents, the invention also intends to include these modifications and variations.

Claims

1. A negotiation-driven method for telemetry configuration in multi-intent networks, characterized in that, include: S1: Receives natural language measurement intent, processes it using a large language model, and generates a standardized query pattern; S2: For each normalized query pattern, combine the data flow sketch library coverage relationship with the pre-set device conditional error-resource profile library to generate a measurement intent configuration; where the data flow sketch library coverage relationship is used to characterize whether a specific sketch algorithm supports the corresponding query pattern, and the pre-set device conditional error-resource profile library is used to provide the basis for resource estimation of the sketch under different device types, node roles and error targets; S3: Construct a topology capability graph based on a pre-built network capability graph and the current network state; S4: Construct a measurement intent constraint graph for each measurement intent configuration, perform compatibility clustering on multiple measurement intent configurations, and negotiate a consensus measurement envelope on the topology capability graph to generate a consensus measurement envelope. The consensus measurement envelope represents a shared bearer object formed by a subset of compatible measurement intent configurations. S5: Perform constraint subgraph matching from the measurement intent constraint graph to the topology capability graph for each consensus measurement envelope, and perform ambiguity verification; S6: Independently solve the envelope for each verified consensus measurement envelope to generate a local telemetry configuration, and then concatenate the local configurations into a global telemetry configuration; S4 includes: S4.1: Configure and construct a measurement intention constraint graph for each measurement intention, wherein the nodes in the measurement intention constraint graph represent path endpoints, shared structure anchors, hotspot edge anchors, candidate measurement deployment points or intermediate carrier nodes, and the edges represent structural relationships that are located on the same path, adjacent to hotspot edges, must pass through a certain type of equipment or are limited by a certain capacity path. S4.2: Perform compatibility clustering on multiple measurement intent configurations based on one or more of the following: scope overlap, topological proximity, device candidate set overlap, accuracy tolerance interval, and resource attitude compatibility, to obtain a subset of compatible intents; S4.3: Construct a terminal group family for each compatible intent subset, obtain a low-cost connectivity shared core based on whether each terminal group in the restricted topology covers at least one candidate structural element, expand the activation range and candidate device set outside the shared core, and generate a consensus measurement envelope. The terminal group family is used to form group coverage constraints, and the restricted topology is obtained based on the compatible intent subset and the topology capability graph. S5 includes: S5.1: For each consensus measurement envelope, check whether there is an embedding and check whether the embedding satisfies the constraints of node role, device type, edge capacity, path attribute, accuracy level and source tracing. Among them, the node role and device type constraints are used to limit the candidate deployment location, the edge capacity and path attribute constraints are used to limit the legal observation path, and the source tracing constraints are used to distinguish between user-given hard constraints and model completion soft constraints. S5.2: Ambiguity verification is performed from three dimensions: structural overlap, constraint relaxation, and mapping stability. If the verification result is a strict match or a controlled soft match, the subsequent local solution is entered. If the verification result is an ambiguous match or a hard constraint conflict, the relevant intent is split back into different envelopes or a local backoff correction is triggered. Structural overlap is used to compare the degree of overlap of supporting nodes, supporting edges, and deployment nodes of different intents in the consensus measurement envelope. Constraint relaxation is used to measure the proportion of wildcards or soft constraint backoffs used in the matching process. Mapping stability is used to evaluate whether there are too many alternative embeddings for the same intent within the envelope.

2. The negotiation-driven method for telemetry configuration in multi-intent networks as described in claim 1, characterized in that, S1 includes: S1.1: Receiving natural language measurement intent; S1.2: Utilize a large language model with knowledge enhancement and constraint decoding to perform multi-intent parsing and controlled normalization on the natural language measurement intent, extract the measurement intent label set, and generate a standardized query pattern.

3. The negotiation-driven method for telemetry configuration in multi-intent networks as described in claim 1, characterized in that, S2 include: S2.1: For each normalized query pattern, filter the candidate data flow sketch set that supports the normalized query pattern based on the data flow sketch library coverage relationship; S2.2: Based on the candidate data flow sketch set and the pre-set device conditional error-resource profile library, generate the intent instantiation error-resource profile of the current intent under different device and role combinations, and determine the path scope, structural support group, activation scope group and deployable device group. Use the candidate data flow sketch set, intent instantiation error-resource profile, path scope, structural support group, activation scope group and deployable device group as the measurement intent configuration.

4. The negotiation-driven method for telemetry configuration in multi-intent networks as described in claim 1, characterized in that, The topology capability graph in S3 records node roles, device types, currently available resources, link capabilities, path attributes, and reachable path edges based on the physical topology. Reachable path edges are derived from routing tables, tunnel information, service chain paths, or path states maintained by the controller. They are used to indicate whether there is a valid observation path between two candidate devices or path anchors in the current network state, and record the upper bound of hop count, the lower bound of capacity, and path attribute labels.

5. The negotiation-driven method for telemetry configuration in multi-intent networks as described in claim 1, characterized in that, S6 include: S6.1: Determine the data flow sketch instances and deployment locations required for each measurement intent configuration within the envelope; S6.2: Under the resource constraints of the candidate device set, allocate device and memory resources for each data flow sketch instance; S6.3: Based on the activation range group, configure the trigger range, acquisition strategy, and path boundary of the measurement task to obtain the local telemetry configuration; S6.4: Verify whether the local telemetry configuration meets the accuracy and resource constraints of each measurement intent configuration. If it does, then treat the local telemetry configuration as a component of the global telemetry configuration.

6. The negotiation-driven method for telemetry configuration in multi-intent networks as described in claim 1, characterized in that, The method further includes processing continuously arriving new intents using dependency-aware micro-batch processing, specifically including: instantiating continuously arriving measurement intent configurations into scheduling primitives, each scheduling primitive recording a candidate envelope, a set of candidate devices, an envelope region to be written or modified, and a priority; constructing an update dependency graph based on the intersection of the candidate device sets and the intersection of the write regions between scheduling primitives; organizing scheduling primitives without write conflicts and device conflicts into the same micro-batch, and assigning scheduling primitives with dependencies to different micro-batches.

7. A negotiation-driven system for telemetry configuration in multi-intent networks, characterized in that, Based on the method described in claim 1, it includes: The multi-intent parsing and controlled normalization module is used to receive natural language measurement intents, process them using a large language model, and generate normalized query patterns. The measurement intent configuration generation module is used to generate a measurement intent configuration for each normalized query pattern by combining the data flow sketch library coverage relationship and the pre-set device conditional error-resource profile library. The data flow sketch library coverage relationship is used to characterize whether a specific sketch algorithm supports the corresponding query pattern, and the pre-set device conditional error-resource profile library is used to provide the basis for resource estimation of the sketch under different device types, node roles and error targets. The topology capability graph construction module is used to construct a topology capability graph based on a pre-built network capability graph and the current network state. The consensus measurement envelope negotiation module is used to construct a measurement intent constraint graph for each measurement intent configuration, perform compatibility clustering on multiple measurement intent configurations, and negotiate the consensus measurement envelope on the topology capability graph to generate a consensus measurement envelope. The consensus measurement envelope represents a shared bearer object formed by a subset of compatible measurement intent configurations. The ambiguity verification module is used to perform constraint subgraph matching from the measurement intent constraint graph to the topology capability graph for each consensus measurement envelope to perform ambiguity verification. The envelope solving and global stitching module is used to independently solve the envelope for each verified consensus measurement envelope, generate local telemetry configurations, and stitch the local configurations into a global telemetry configuration.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the negotiation-driven method for telemetry configuration of multi-intent networks as described in any one of claims 1 to 6.

Images