A method and system for dynamic mapping and scheduling of network resources based on identity resolution

Abstract

This application discloses a method and system for dynamic mapping and scheduling of network resources based on identity resolution, relating to the field of computer network technology. The method includes: acquiring request information of the task to be scheduled, multi-source identity signals, resource node status, overall network load status, and service preference information; fusing and parsing the multi-source identity signals to construct a cross-domain unified identity identifier and generate an identity attribute vector; then, combining computing resource status, network link status, and service preferences, calculating the identity adaptability between the unified identity identifier and each resource node to generate a candidate resource node list; finally, making dynamic scheduling decisions based on overall network load and link status to determine the target resource node and its corresponding network path, and issuing network configuration rules and resource scheduling instructions. This achieves collaborative perception and dynamic matching of identity, network, and computing resources, improving the accuracy, real-time performance, and service adaptability of cross-domain resource scheduling.

Description

Technical Field

[0001] This application relates to the field of computer network technology, and in particular to a method and system for dynamic mapping and scheduling of network resources based on identity resolution. Background Technology

[0002] With the continuous development of cloud computing, edge computing, terminal intelligence, and computing networks, the types of services carried by network systems, access subjects, and resource requirements are showing a more complex and dynamic trend. Service requests in different application scenarios vary in terms of access location, service quality, resource consumption, and response time. Network resource management systems need to have stronger environmental awareness and resource coordination and scheduling capabilities to adapt to the ever-changing business operation status.

[0003] Currently, network resource management methods typically rely on fixed network identifiers, preset configuration rules, or relatively static resource allocation strategies for resource mapping and management. Such methods can meet basic needs in scenarios with stable network structures and relatively simple business types. However, in complex network environments where business requests change frequently and resource status fluctuates continuously, their ability to perceive the characteristics of request subjects, differences in business needs, and service quality preferences is relatively limited, which can easily lead to discrepancies between resource configuration and actual business needs.

[0004] To enhance network management capabilities, some solutions focus on leveraging identity information to strengthen network management or security control, while others focus on optimizing network or computing resource allocation based on changes in resource status. However, these solutions typically operate separately from the identity side or the resource side, lacking sufficient coordination between identity information and resource scheduling. This makes it difficult to simultaneously consider differences in requesting entities, business needs, and resource status changes during a unified scheduling process. Summary of the Invention

[0005] This application provides a method, system, storage medium, computer program product, and electronic device for dynamic mapping and scheduling of network resources based on identity resolution, in order to at least solve the problem that network resource scheduling in the prior art is difficult to adapt to complex dynamic business needs.

[0006] In a first aspect, embodiments of this application provide a method for dynamic mapping and scheduling of network resources based on identity identifier resolution. The method includes: acquiring task request information of a task to be scheduled, multi-source identity signals associated with the task to be scheduled, resource status data of multiple resource nodes, network-wide load status data, and service preference information corresponding to the task to be scheduled; wherein the task request information includes the request initiation location of the task to be scheduled, and the resource status data includes the computing resource status data of the resource nodes and the network link status data associated with the resource nodes; fusing and parsing the multi-source identity signals to construct a cross-domain unified identity identifier associated with the task to be scheduled, and generating an identity attribute vector corresponding to the unified identity identifier; and calculating the dynamic mapping and scheduling of network resources based on the identity attribute vector, the resource status data, and the service preference information. The system describes the identity compatibility between the unified identity identifier and each of the resource nodes, and generates a candidate resource node list based on the identity compatibility. Based on the candidate resource node list, the overall network load status data, and the network link status data, it performs dynamic scheduling decisions to determine a target resource node matching the task to be scheduled from the candidate resource node list, and to determine the network path from the request initiation location of the task to be scheduled to the target resource node. It generates network configuration rules corresponding to the network path and resource scheduling instructions corresponding to the target resource node, and sends the network configuration rules to the corresponding network devices for execution, and sends the resource scheduling instructions to the corresponding resource management platform for execution, so as to allocate the network path and the computing resources of the target resource node to the task to be scheduled associated with the unified identity identifier.

[0007] Secondly, embodiments of this application provide a network resource dynamic mapping and scheduling system based on identity identifier resolution. The system includes: a data sensing and acquisition unit, used to acquire task request information of a task to be scheduled, multi-source identity signals associated with the task to be scheduled, resource status data of multiple resource nodes, network-wide load status data, and service preference information corresponding to the task to be scheduled; wherein, the task request information includes the request initiation location of the task to be scheduled, and the resource status data includes computing resource status data of the resource nodes and network link status data associated with the resource nodes; an identity fusion and parsing unit, used to fuse and parse the multi-source identity signals to construct a cross-domain unified identity identifier associated with the task to be scheduled, and generate an identity attribute vector corresponding to the unified identity identifier; and an identity adaptation evaluation unit, used to evaluate the identity attribute vector, the resource status data, and the service preference information. The system includes: a unified identity identifier and a resource node; a joint scheduling decision unit, which performs dynamic scheduling decisions based on the candidate resource node list, the network load status data, and the network link status data to determine the target resource node matching the task to be scheduled from the candidate resource node list, and to determine the network path from the request initiation location of the task to be scheduled to the target resource node; and a policy orchestration and distribution unit, which generates network configuration rules corresponding to the network path and resource scheduling instructions corresponding to the target resource node, distributes the network configuration rules to the corresponding network devices for execution, and distributes the resource scheduling instructions to the corresponding resource management platform for execution, so as to allocate the network path and the computing resources of the target resource node to the task to be scheduled associated with the unified identity identifier.

[0008] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the network resource dynamic mapping and scheduling method based on identity resolution according to any embodiment of this application.

[0009] Fourthly, embodiments of this application provide a storage medium storing a computer program thereon, which, when executed by a processor, implements the steps of the network resource dynamic mapping and scheduling method based on identity resolution according to any embodiment of this application.

[0010] Fifthly, embodiments of this application provide a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the network resource dynamic mapping and scheduling method based on identity resolution according to any embodiment of this application.

[0011] The network resource dynamic mapping and scheduling method and system based on identity resolution provided in this application can achieve at least the following technical effects:

[0012] By fusing and parsing multi-source identity signals associated with tasks to be scheduled, a unified cross-domain identity identifier is constructed and a corresponding identity attribute vector is generated. This transforms identity-related information scattered across different sources, management domains, or business sides into a unified representation that can directly participate in resource scheduling. It is not only used for identity security confirmation, but also carries the request subject attributes, business association characteristics, and service preference basis, so that subsequent resource mapping can use a clearer identity dimension as the scheduling basis, thereby improving the accuracy of the correspondence between tasks to be scheduled and computing resource configuration objects.

[0013] Based on identity attribute vectors, resource status data, and service preference information, identity suitability is calculated to generate a candidate resource node list. This ensures that the initial selection process for resource nodes is no longer determined solely by resource availability or physical location, but rather reflects a comprehensive matching relationship between requester characteristics, business needs, and resource capacity. Furthermore, by combining the candidate resource node list, network-wide load status data, and network link status data to perform dynamic scheduling decisions, a linkage can be established between target resource node selection and network path determination. This ensures that the computational resource allocation results are consistent with the network transmission path configuration, thereby improving the adaptability of resource mapping results to real-time network operating conditions and task service requirements.

[0014] This technical solution incorporates the identity characteristics of the task requester into the dynamic mapping process of network resources through a cross-domain unified identity identifier. The identity adaptation result participates throughout the resource node selection, network path determination, and final scheduling and execution control, constructing a collaborative closed loop between identity resolution and resource scheduling. This improves the rationality of resource node selection, the accuracy of network path configuration, and the overall consistency of collaborative scheduling of computing resources and network link resources in complex network environments. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of this application 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 this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating an example of a network resource dynamic mapping and scheduling method based on identity resolution according to an embodiment of this application is shown;

[0017] Figure 2 This document illustrates an example of a process flow diagram for calculating the identity compatibility between a unified identity identifier and each resource node according to an embodiment of the present application.

[0018] Figure 3 This document illustrates an example of an operation flowchart for determining a target resource node and network path in a method according to an embodiment of this application.

[0019] Figure 4 This paper illustrates a schematic diagram of the system operation mechanism of an example of a network resource dynamic mapping and scheduling method based on identity resolution according to an embodiment of this application.

[0020] Figure 5 A schematic diagram of a comparative experiment simulation showing the request acceptance rate of different methods under dynamic load is shown;

[0021] Figure 6 This diagram illustrates a comparative simulation of the cumulative distribution of end-to-end task completion latency for latency-sensitive services under full load conditions using different methods.

[0022] Figure 7 A schematic diagram of the comparative experimental simulation results of different methods in terms of system control surface overhead and performance trade-offs is shown.

[0023] Figure 8 A structural block diagram of an example of a network resource dynamic mapping and scheduling system based on identity resolution, according to an embodiment of this application, is shown. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] It should be noted that in current related technologies, identity-aware network management has been applied to scenarios such as network segmentation, microservice communication, and service access control. For example, NSX micro-segmentation combined with NGFW (Next-Generation Firewall) solutions typically formulate access policies based on user, application, or workload identities; Consul-like service mesh architectures manage communication relationships between services through service identity authentication, authorization policies, and mTLS (mutual transport layer security). While these solutions can improve the granularity of network access control and communication security, their identity information is primarily used for authentication, authorization, and security isolation, and is generally not further used as a basis for scheduling compute node selection, network path determination, or link resource allocation. Therefore, in environments where business load, resource status, and service quality requirements are constantly changing, these solutions struggle to directly support the dynamic mapping and collaborative scheduling of compute network resources.

[0026] In the area of ​​computing network and computing power network scheduling, some IPCRSA (ICN-based On-Path Computing Resource Scheduling Architecture) solutions are beginning to incorporate business preferences or task requirements into the computing resource selection process. Other SD-CFN (Software-Defined Computing First Network) solutions focus on measuring computing resource status and network link performance to assist in computing routing or node selection. These solutions improve resource scheduling's adaptability to changes in business demands and network status to some extent, but their focus remains primarily on resource-side status, link performance, or path selection results. Even when identity information is introduced, it is often presented as an additional field or static label in business requests, and a cross-domain fusion and parsing process for multi-source identity signals has not yet been formed.

[0027] In the areas of virtual network embedding and dynamic resource mapping, some DVNE-DRL (Dynamic Virtual Network Embedding based on Deep Reinforcement Learning) solutions attempt to enhance the mapping capability between virtual network requests and physical network resources using learning models. These solutions can improve mapping flexibility under changing resource states, but their evaluation typically focuses on the resource matching relationships between virtual nodes, virtual links, and physical nodes and links, mainly considering resource-side factors such as node capacity, link status, topology connectivity, and resource utilization. For the identity attributes, business roles, service levels, and cross-domain identity associations of the requesting entity, these solutions often lack unified modeling and scheduling-side utilization.

[0028] Therefore, while current technologies have explored aspects such as identity-aware security control, service mesh access management, computing resource scheduling, virtual network mapping, and computing power network routing, certain disconnects remain between these approaches. Identity information is largely confined to authentication, authorization, and access boundary control, while resource scheduling relies more on node load, link status, topology, and physical resource utilization. For fragmented identity signals such as device identifiers, login accounts, application identifiers, network addresses, and user tags, current solutions typically lack a process to integrate them into a unified cross-domain identity view, and also lack a framework for using this identity view in conjunction with service preferences, resource status, and network link status for resource mapping. Consequently, in network environments with multiple stakeholders, multiple services, and continuously changing resource statuses, current solutions still struggle to generate more refined dynamic mapping and scheduling results for network resources based on the differences in requesting entities.

[0029] It should be understood that the above description of the relevant technologies is intended only to help the public better understand the inventive spirit and motivation of this application, and is not intended to limit this application. Furthermore, the technical solutions described in the above-mentioned relevant technologies are not prior art, and may also be undisclosed technical solutions, such as those under research or in the laboratory stage.

[0030] The technical solutions in this application, including the collection, storage, use, processing, transmission, provision, and disclosure of users' personal information, comply with relevant laws and regulations and do not violate public order and good morals.

[0031] Figure 1 A flowchart illustrating an example of a network resource dynamic mapping and scheduling method based on identity resolution according to an embodiment of this application is shown.

[0032] Regarding the execution subject of the method in the embodiments of this application, it can be any controller or processor with computing or processing capabilities, such as a network resource management platform controller, a computing power network orchestration platform controller, an edge resource scheduling platform controller, a cloud-edge collaborative scheduling platform controller, an identity resolution service platform controller, or a software-defined network controller. The method in the embodiments of this application is implemented by running programs or instructions stored in a storage medium.

[0033] In some examples, it may be integrated into an electronic device or terminal through software, hardware, or a combination of both, and the type of terminal or electronic device may be diverse.

[0034] In one exemplary application scenario, this method can be applied to cross-regional smart manufacturing and industrial IoT factory cluster networks. In this scenario, the data plane can include edge computing nodes, core data centers, industrial gateways, access switching equipment, and heterogeneous network links connecting various computing facilities, distributed across different factory areas, workshops, or geographical regions. The terminal side can include automated production line control equipment, industrial robots, ultra-high-definition quality inspection cameras, mobile unmanned guided vehicles, sensor acquisition terminals, and factory business management systems. These terminal entities may continuously generate different types of tasks to be scheduled, such as industrial robot trajectory planning tasks, quality inspection image analysis tasks, equipment status monitoring tasks, production data archiving tasks, or background batch processing tasks. Different tasks have different requirements in terms of latency, bandwidth, computing power, reliability, and cost, and the device identities, application identities, or business entities behind different tasks may also be different.

[0035] The embodiments of this application are not limited to the above-mentioned industrial scenarios, but can also be applied to scenarios that require simultaneous scheduling of network resources and computing resources, such as smart parks, vehicle-road collaboration, cloud-edge collaborative video analysis, cross-domain office networks, and edge AI inference services. The following will use a task to be scheduled in an industrial Internet of Things scenario as an example for explanation.

[0036] like Figure 1 As shown, in step S110, the task request information of the task to be scheduled, the multi-source identity signal associated with the task to be scheduled, the resource status data of multiple resource nodes, the overall network load status data, and the service preference information corresponding to the task to be scheduled are obtained. The task request information includes the request initiation location of the task to be scheduled, and the resource status data includes the computing resource status data of the resource nodes and the network link status data associated with the resource nodes.

[0037] In practice, task request information can be reported by terminal devices, business systems, edge access gateways, or application service entry points, or it can be parsed by the control plane when it receives a business scheduling request. Task request information may include task type, request initiation time, request initiation location, business entry point, task size, task priority, resource requirement description, or task execution constraints. The request initiation location characterizes the source-side location of the task to be scheduled entering the network system, and can be determined by the access gateway identifier, edge node identifier, base station cell identifier, industrial park access switch identifier, business gateway address, or logical topology location. For example, when a mobile unmanned guided vehicle in a workshop initiates a path avoidance calculation task, the request initiation location can be determined based on the workshop edge gateway, wireless access point, or base station cell identifier it accesses; when an ultra-high-definition quality inspection camera initiates an image analysis task, the request initiation location can be determined based on the access switch or edge computing gateway of its production line.

[0038] Multi-source identity signals are used to describe the requesting entity associated with the task to be scheduled. This requesting entity can be a user, device, application, tenant, business system, or organizational entity. Multi-source identity signals can include device identifiers, hardware fingerprints, login accounts, business tokens, application identifiers, access addresses, session credentials, organizational tags, and historical access behavior records. For example, for a trajectory prediction task initiated by an industrial robot, multi-source identity signals can include the robot controller's device number, the access address assigned by the industrial gateway, the device registration account in the production control system, and the robot's control task records within a historical period; for an image recognition task initiated by a quality inspection camera, multi-source identity signals can include the camera device fingerprint, vision algorithm service call credentials, production line identifier, and the corresponding business system account.

[0039] Resource status data describes the current operational status of resource nodes and their associated network links that can participate in scheduling. Resource nodes can include cloud computing nodes, edge computing nodes, container instance nodes, storage nodes, GPU inference nodes, or other computing resource units capable of carrying tasks. Computing resource status data can include computing load, available computing capacity, storage margin, instance occupancy, task queuing status, accelerator availability, etc.; network link status data can include link latency, available bandwidth, link congestion level, link reachability, packet loss, and link stability, etc. Overall network load status data describes the overall load distribution of the network and resource system from a global perspective, such as the load levels of different factories, different resource pools, different edge node clusters, or different link segments. Service preference information characterizes the service quality requirements of the tasks to be scheduled, such as low latency, high bandwidth, low cost, high reliability, proximity processing, or security isolation, etc.

[0040] Therefore, the control side can form a multi-dimensional runtime context for the tasks to be scheduled, which can simultaneously cover the task source, identity subject, resource status, network status, and service preferences, enabling the control side to obtain a more complete scheduling basis. Compared with scheduling methods that rely solely on static network addresses, fixed configuration rules, or single resource status, this improves the completeness and real-time performance of scheduling input information and reduces resource allocation deviations caused by unclear request subjects, unclear resource status, or ambiguous service requirements.

[0041] In step S120, the multi-source identity signals are fused and parsed to construct a cross-domain unified identity identifier associated with the task to be scheduled, and an identity attribute vector corresponding to the unified identity identifier is generated.

[0042] In some implementations, multi-source identity signals can be standardized first, enabling them to enter a unified processing flow. For example, account identifiers, device identifiers, access addresses, application credentials, session information, and behavior records can undergo field standardization, deduplication, outlier removal, and time alignment. In industrial IoT scenarios, the same device may simultaneously possess a device asset number, control system account, network access address, application call credential, and operation log identifier; these information have different sources, field formats, and update cycles. Standardization can provide a unified parsing basis.

[0043] Subsequently, multi-source identity signals can be fused and analyzed to determine whether different identity signals correspond to the same requesting subject. For example, processing can be based on the degree of correlation between identity signals, consistency of source, business affiliation, access continuity, or historical record correlation. However, the embodiments of this application do not limit the specific identity fusion algorithm. For example, a quality inspection camera may experience a change in its access address at the network layer due to reconnection or address reassignment, but its hardware fingerprint, production line affiliation identifier, visual service call credential, and historical task behavior can still reflect the same subject attribute. The control side can merge multiple identity signals collected at different times or in different network domains into the same identity subject. As another example, when a mobile unmanned guided vehicle moves in different workshop areas, its wireless access point may change, but if the device registration information and business task type remain consistent, it can also be resolved to the same cross-domain identity subject.

[0044] After obtaining the unified identity identifier across domains, a corresponding identity attribute vector can be generated. This vector is used to describe the scheduling-related attributes of the requesting subject in a structured form, such as subject type, business category, service level, trust level, location attribute, historical resource usage characteristics, access behavior characteristics, or preference characteristics. Taking an industrial scenario as an example, the identity attribute vector corresponding to a robotic arm control device can include features such as control equipment, high-priority business, low-latency preference, and fixed workstation location; the identity attribute vector corresponding to a quality inspection camera can include features such as video acquisition device, high-bandwidth preference, and visual analysis business; and the identity attribute vector corresponding to a background data synchronization service can include features such as batch processing business, low-cost preference, and non-real-time task.

[0045] Thus, the dispersed and variable identity signals are transformed into stable cross-domain unified identity identifiers and computable identity attribute vectors, which can enhance the stability and consistency of the request subject's expression, reduce subject identification deviations caused by address changes, account dispersion, cross-domain device movement, or the lack of a single identity signal, and enable the resource scheduling process to carry out resource mapping based on a more stable identity view.

[0046] In step S130, based on the identity attribute vector, resource status data and service preference information, the identity adaptation degree between the unified identity identifier and each resource node is calculated, and a candidate resource node list is generated based on the identity adaptation degree.

[0047] In some implementations, identity attribute vectors can be used as features on the requesting subject side, resource status data as features on the resource node side, and service preference information as constraints on the task requirement side to evaluate the matching degree between the unified identity and each resource node. Identity attribute vectors can reflect the requesting subject's business level, trust status, location attributes, and historical behavioral characteristics; resource status data can reflect the resource node's available computing power, load status, and associated link service capabilities; service preference information can reflect the scheduling task's requirements for latency, bandwidth, cost, reliability, or security. By using the above information in combination, it is possible to evaluate whether a particular resource node is suitable for carrying the scheduling task associated with that unified identity.

[0048] When calculating identity suitability, multiple factors from the identity side, resource side, and preference side can be considered comprehensively, but the embodiments of this application do not limit the specific calculation method of suitability. For example, for the obstacle avoidance calculation task of a mobile unmanned guided vehicle, if the service preference information indicates that it has low latency and high reliability requirements, then edge computing nodes that are close to the workshop access side, have relatively stable link status, and sufficient resource margins are usually more suitable for carrying out the task; for the image analysis task of an ultra-high-definition quality inspection camera, if the service preference information indicates that it has high bandwidth and strong computing power requirements, then resource nodes with image processing capabilities and sufficient link bandwidth are more suitable for carrying out the task; for the production data archiving task, if its real-time requirements are low, then resource nodes with low cost and stable throughput are more suitable for carrying out the task. For tasks with regional compliance, security isolation, or dedicated business domain requirements, identity trust level, business affiliation area, or security policy can also be included in the suitability evaluation.

[0049] After obtaining the identity suitability of each resource node, the resource nodes participating in the evaluation can be sorted, filtered, or grouped to form a candidate resource node list. The resource nodes in the candidate resource node list are a group of alternative nodes that are most suitable for carrying the scheduled task under the current resource status and service preferences. The number of nodes or the filtering range of this list can be determined based on business requirements, resource scale, and scheduling strategies. For example, for low-latency control tasks, the candidate resource node list can mainly include edge computing nodes near the request initiation location that meet the resource status requirements; for high-bandwidth image analysis tasks, the candidate resource node list can mainly include resource nodes with image processing capabilities and network link status that meet the requirements.

[0050] Therefore, the control side can transform identity resolution results, resource status, and service preferences into adaptation evaluation results at the resource node level, which can improve the targeting of candidate resource screening, make the resource node list more in line with the request subject attributes and business service needs, and reduce the computational overhead caused by irrelevant or obviously unsuitable resource nodes participating in scheduling decisions.

[0051] In step S140, dynamic scheduling decisions are made based on the candidate resource node list, the overall network load status data, and the network link status data to determine the target resource node that matches the task to be scheduled from the candidate resource node list, and to determine the network path from the request initiation location of the task to be scheduled to the target resource node.

[0052] In some implementations, a candidate resource node list provides a range of alternative resource nodes. Dynamic scheduling decisions further combine network-wide load status data and network link status data to determine the final target resource node from this range. The control side can use network-wide load status data to determine whether there is excessive load, congestion, or task queuing in the region, resource pool, or link segment where each candidate resource node is located, and use network link status data to determine the reachability, latency, bandwidth, and stability from the request initiation location to each candidate resource node. This avoids selecting the target node based solely on a single adaptation evaluation result.

[0053] When determining target resource nodes, the compatibility of candidate resource nodes, remaining computing power, overall network load distribution, and the operational impact of nodes carrying scheduled tasks can be comprehensively considered. When determining network paths, the starting point can be the location where the scheduled task's request originates, and the target resource node can be the endpoint. The transmission path that meets the task's service requirements can be selected based on the network link status. For example, if a candidate resource node has high compatibility with the scheduled task but its access link is congested, the control side can choose other more stable candidate resource nodes. Similarly, if a cloud node has strong computing power but high cross-domain link latency, for real-time control tasks, the nearest edge resource node can be prioritized, while for non-real-time archiving tasks, a node with more suitable cost or load can be selected.

[0054] Here, the target resource node and network path can be determined collaboratively during the same scheduling process. This step reduces resource mismatch caused by the disconnect between computing resource selection and network path selection, such as when the computing node has sufficient capacity but the access path is congested, or when the network path is in good condition but the target node's carrying capacity is insufficient. As a result, the tasks to be scheduled can obtain more stable end-to-end resource carrying conditions, and improve the overall network resource utilization and load distribution.

[0055] In step S150, network configuration rules corresponding to network paths and resource scheduling instructions corresponding to target resource nodes are generated. The network configuration rules are then sent to the corresponding network devices for execution, and the resource scheduling instructions are sent to the corresponding resource management platform for execution, so as to allocate network paths and computing resources of target resource nodes to the scheduled tasks associated with the unified identity.

[0056] In some implementations, network configuration rules may include routing configurations, forwarding rules, flow table configurations, access control rules, service access policies, or other rules capable of controlling the transmission path of service traffic. Network devices may include edge switching devices, access gateways, core routing devices, software-defined network devices, or other devices capable of receiving control rules and executing traffic forwarding control. For example, for a quality inspection image analysis task initiated from a workshop edge gateway, network configuration rules can be used to direct service traffic along a defined link to the corresponding edge inference node or cloud processing node; for an obstacle avoidance task of a mobile unmanned guided vehicle, network configuration rules can be used to forward control traffic along a path that meets latency requirements to the target edge computing node.

[0057] Here, resource scheduling instructions are used to instruct the resource management platform to allocate, reserve, or start corresponding computing resources for the tasks to be scheduled on the target resource node. For example, the resource management platform can be a cloud resource management platform, an edge resource management platform, a container orchestration platform, a virtualization management platform, or other platforms capable of controlling resource nodes to perform task deployment and resource allocation. Resource scheduling instructions can include task deployment instructions, resource quota allocation instructions, service instance binding instructions, compute instance startup instructions, or task migration instructions, etc. For instance, when the target resource node is an edge computing node, the resource scheduling instructions can trigger that node to start the corresponding service instance or allocate computing resource quotas for the tasks to be scheduled; when the target resource node is a cloud resource pool, the resource scheduling instructions can trigger the resource management platform to create or bind compute instances in the corresponding resource pool.

[0058] In this embodiment, the scheduling decision formed by the control side is transformed into control actions that can be executed by network devices and resource management platforms, which enables the coordinated implementation of network path configuration and computing resource allocation, so that the task to be scheduled can obtain computing resources on the target resource node while obtaining the specified network transmission path, thereby improving resource delivery efficiency and service execution stability.

[0059] Regarding the implementation details of generating identity attribute vectors, in some examples of embodiments of this application, multi-source identity signals are mapped to a unified feature space with a preset feature dimension to obtain the feature vectors of each identity signal in the unified feature space.

[0060] In practical implementation, the collected multi-source identity signals can include categorical identity data and numerical identity data. Categorical identity data may include, for example, device type, business system identifier, application account type, access domain label, etc.; numerical identity data may include, for example, login frequency, request count, access time interval, historical resource call count, etc. Because different identity signals come from different sources, have different formats, and different dimensions, direct identity fusion is easily affected by differences in field types and numerical scales. Therefore, a pre-defined feature extractor, encoding rules, or embedding network can be used to uniformly encode different types of identity signals, mapping them to a feature space of the same dimension. After this mapping process, each identity signal can be represented as a feature vector with consistent dimensions and comparable scales, enabling similarity evaluation of identity signals from different sources within the same feature space.

[0061] Then, based on the matching degree and corresponding importance weight of each preset feature dimension, the feature matching similarity of any two identity signals in the unified feature space is calculated.

[0062] For example, any two identity signals in a multi-source identity signal and Feature matching similarity between It can be expressed by the following formula:

[0063] Equation (1)

[0064] In the formula, Indicates feature matching similarity. This represents the total number of feature dimensions. Indicates the first The importance weights of each feature dimension and , Indicates the first Identity signals in each feature dimension With identity signal degree of matching and This is a normalized value.

[0065] In equation (1), the matching relationship between two identity signals is comprehensively evaluated through dimensional weighted fusion. Specifically, different feature dimensions may have different degrees of importance for identity judgment. For example, device fingerprints, hardware identifiers, or business credentials usually have stronger identity stability than temporary network addresses, so they can be assigned higher importance weights; while features such as access addresses and temporary session information, which are prone to change with the network environment, can be assigned relatively lower importance weights. By weighted fusion of the matching degree of each dimension and normalization using the total weight, a feature matching similarity with a relatively stable value scale can be obtained, thereby reducing the impact of single feature fluctuations on identity association judgment.

[0066] Subsequently, using multi-source identity signals as graph vertices, identity signal pairs whose feature matching similarity meets a preset similarity threshold are constructed as associated edges, and the corresponding feature matching similarity is used as the edge weight of the associated edges to construct an identity relationship graph.

[0067] In some implementations, each vertex in the identity relationship graph corresponds to an identity signal, and associated edges represent a predefined association between two identity signals. By filtering identity signal pairs using a preset similarity threshold, the interference of low-confidence associations on the graph structure can be reduced. Edge weights reflect the similarity strength between two identity signals, enabling the identity relationship graph to express not only whether an association exists between identity signals but also the degree of that association. Thus, the originally scattered identity signals are organized into a topologically structured identity relationship graph, facilitating unified modeling of local adjacency relationships between identity signals.

[0068] Subsequently, the feature vectors of each identity signal in a unified feature space are used as the initial hidden layer representation vectors of the corresponding graph vertices, and a graph neural network model is used to perform feature aggregation on the identity relationship graph. The graph neural network model iteratively updates the hidden layer representation vectors of each graph vertex through the mean aggregation function, so as to reduce the impact of differences in neighborhood size and feature scale of different nodes on the identity aggregation results while utilizing the local structure of the adjacency topology.

[0069] To ensure the clarity of feature propagation relationships during the multi-layer iteration process of a graph neural network, graph vertices are defined. The initial hidden layer representation vector is ,and In the formula, This represents the feature vector of the corresponding identity signal in a unified feature space.

[0070] For example, the graph neural network model in the 1st... Layer-in-Diagram Vertex Hidden layer representation vector The feature aggregation result is updated using a standard mean aggregation function to reduce the impact of feature scale differences on the aggregation result while preserving the local structure of the adjacency topology. Thus, by means processing, the feature amplitude fluctuation caused by differences in the number of neighboring nodes can be reduced.

[0071] Equation (2)

[0072] In the formula, Represents a non-linear activation function. The graph neural network model represents the first... The learnable weight matrix of the layer, Representing the vertices of a graph The set of neighbor nodes in the identity relationship graph This represents the number of nodes in the neighbor node set. Representing the vertices of a graph In the Hidden layer representation vectors of the layer This indicates the layer number of the graph neural network model.

[0073] In the update process of Equation (2), the hidden layer representation vector of the graph vertex itself and the hidden layer representation vectors of its neighboring nodes participate in the aggregation, and the mean is applied by the number of neighboring nodes, so that the representation result of the graph vertex can simultaneously contain its own identity features and the local context information of the neighboring identity signals. It should be noted that in this example, the edge weights of the associated edges are mainly used for the construction of the identity relationship graph and the formation of adjacency relationships, while the mean aggregation function performs feature aggregation based on the formed adjacency topology. Through multi-level iterative updates, the final hidden layer representation vector of each graph vertex can express the identity signal's own attributes and its neighborhood association information.

[0074] Next, based on the final hidden layer representation vector obtained after multiple updates of the graph neural network model, a clustering algorithm is used to cluster the graph vertices to obtain identity signal clusters corresponding to the same identity entity.

[0075] In practical implementation, clustering algorithms can group graph vertices with similar representations into the same identity signal cluster based on distance, density, or similarity relationships between the final hidden layer representation vectors. Each identity signal cluster can correspond to an identity entity, which can be the same user, the same device, the same application instance, the same business system, or the same organizational entity. Through this clustering process, identity signals from different network layers or business systems can be merged into entity-level identity sets, thereby reducing the impact of missing, changed, or abnormal individual identity signals on the identity resolution results.

[0076] Furthermore, a unique cross-domain unified identity identifier is assigned to each identity signal cluster within the system, and cluster-level representation fusion processing is performed on the final hidden layer representation vectors of each signal vertex within the identity signal cluster to generate identity attribute vectors.

[0077] For example, the identity attribute vector corresponding to a cross-domain unified identity can be generated using the following cluster-level representation fusion processing formula:

[0078] Equation (3)

[0079] In the formula, Represents an identity attribute vector. Cluster-level fusion functions are represented. This represents the set of graph vertices contained in the identity signal cluster corresponding to the same identity entity. Representing the vertices of the graph within the identity signal cluster In the last layer of the graph neural network model, i.e. The final hidden layer representation vector output by the layer. This represents the total number of iterations in the graph neural network model.

[0080] In one example An average pooling function can be used to obtain the overall mean feature of the graph vertex representations within the identity signal cluster. In another example, Max pooling can also be used to preserve significant local features within identity signal clusters. Regardless of the specific fusion function used, cluster-level representation fusion processing compresses multiple graph vertex representations under the same identity entity into a unified entity-level vector representation. Thus, the system can generate structured identity attribute vectors for cross-domain unified identity identifiers, which can characterize the overall identity attributes and behavioral association features of the corresponding identity entity.

[0081] Through the embodiments of this application, multi-source identity signals can be transformed from their original discrete form into cross-domain unified identity identifiers and their corresponding identity attribute vectors. This improves the stability of identity signal fusion, reduces the impact of differences in format, scale, and source of heterogeneous identity data on the parsing results, and enhances the aggregation consistency of multiple types of identity signals under the same identity entity. For cloud-edge-device converged networks, industrial IoT, and multi-tenant computing network scheduling scenarios, this processing enables requesting entities to participate in the resource scheduling process in a structured vector form, thereby improving the computability and robustness of identity representation.

[0082] Figure 2 A flowchart illustrating an example of calculating the identity fit between a unified identity identifier and each resource node in a method according to an embodiment of this application is shown.

[0083] like Figure 2As shown, in step S210, resource status feature vectors for each resource node are extracted based on resource status data and preset resource cost rules. These resource status feature vectors include computational resource features obtained from the computational resource status data, link service features obtained from the network link status data, and cost features related to resource call costs.

[0084] It should be noted that different resource nodes typically differ in computing power, link conditions, and resource access costs. To enable different types of resources to participate in a unified evaluation, resource status data can first be converted into resource status feature vectors. For example, for the... resource nodes It is possible to extract its resource status feature vector. The resource status feature vector This can include computing resource characteristics, link service characteristics, and cost characteristics. Computing resource characteristics can be obtained from parameters such as processor utilization, available cores, available memory, accelerator status, and task queuing status of resource nodes. Link service characteristics can be obtained from parameters such as link latency, available bandwidth, packet loss, link reachability, or link stability associated with resource nodes. Cost characteristics can be obtained from resource node resource access prices, billing levels, energy costs, or preset resource cost rules.

[0085] Through the above processing, the computing power, network service capabilities, and call costs of resource nodes are organized into a unified data representation, reducing the format differences between different resource types and enabling resource nodes to have comparable and computable state representations, thereby improving the consistency of resource evaluation.

[0086] In step S220, the basic matching degree between the identity attribute vector and the resource status feature vector of the corresponding resource node is calculated using the dimension compatibility measurement function and the importance weight of each matching dimension on multiple preset matching dimensions.

[0087] For example, in the preset Calculate the identity attribute vector along each matching dimension. With the resource nodes Resource status feature vector Basic matching degree :

[0088] Equation (4)

[0089] In the formula, Indicates the basic matching degree. Indicates a unified identity identifier. This represents the total number of matching dimensions. Indicates the first The importance weights of each matching dimension, and satisfying and , This represents a dimensionality compatibility measure function that maps feature differences across different dimensions to a dimensionless space of [0,1]. and Representing identity attribute vectors respectively and resource state feature vector In the Feature values ​​on each matching dimension.

[0090] In the calculation process of the above equation (4), different matching dimensions can correspond to different resource matching meanings, such as business level matching, regional location matching, trust level matching, computing power matching, network service capability matching, or cost constraint matching. For different types of features, the dimension compatibility measurement function can adopt different measurement methods. For example, for discrete features such as regional affiliation, security domain, and business category, consistency judgment or similar category mapping can be used; for continuous features such as computing capacity, expected processing capacity, and link service capability, interval mapping, distance decay, or normalized similarity mapping can be used. Through the dimension compatibility measurement function, features of different types and different dimensions can be converted into compatibility results under a unified value range.

[0091] Here, importance weight This is used to characterize the influence of different matching dimensions on the basic matching degree. For example, in tasks with high security isolation requirements, matching dimensions related to trust level or security domain can be assigned higher weights; in tasks with high requirements for geographical location or proximity processing, location-related matching dimensions can be assigned higher weights. By weighted fusion of the compatibility results of each dimension, the basic matching degree between the identity attribute vector and the resource status feature vector can be obtained, which reflects the degree of matching between the requesting subject and the resource node in terms of basic attributes, capability conditions, and constraints.

[0092] In step S230, demand preference vectors for latency, bandwidth, and cost are extracted from the service preference information. Based on the positive and negative correlation polarity of service quality for various environmental assessment indicators, the expected latency, available bandwidth, and call cost parameters of each resource node for the scheduled task are corrected and normalized to obtain positive utility indicators.

[0093] In practice, service preference information can be parsed into a demand preference vector for latency, bandwidth, and cost. .in, , and Let $\mathbf$ and $\mathbf$ represent the preference weights for latency, bandwidth, and cost, respectively, with each preference weight being greater than or equal to zero, and satisfying the following conditions: .

[0094] Because different environmental assessment indicators have different impacts on service quality, polarity correction is necessary. For example, higher available bandwidth usually indicates that resource nodes better meet bandwidth requirements, thus it is a positively correlated indicator; lower expected latency and lower call costs usually indicate that resource nodes better meet service requirements, thus it is a negatively correlated indicator. If weights are applied directly without polarity correction, the values ​​of different indicators will be inconsistent, which can easily affect the interpretability of the evaluation results. Therefore, different indicators can be uniformly converted into positive utility indicators, where higher values ​​indicate higher utility.

[0095] For example, the polarity correction and normalization of the expected latency and call cost, which exhibit negative correlation, can be performed using the following formula:

[0096] Equation (5)

[0097] in, , representing the delay dimension and the cost dimension, respectively.

[0098] For available bandwidth exhibiting positive correlation polarity, polarity correction and normalization can be performed using the following formula:

[0099] Equation (6)

[0100] In the formula, Represents resource nodes Positive utility metrics after polarity correction and normalization in the delay or cost dimension. Represents resource nodes Positive utility metrics after polarity correction and normalization in the bandwidth dimension; Represents resource nodes The original estimated latency or invocation cost parameters for the task to be scheduled. Represents resource nodes The raw available bandwidth parameters for the task to be scheduled; and These represent the maximum and minimum statistical values ​​of the resource nodes participating in the evaluation for the corresponding negative correlation indicators, respectively. and These represent the maximum and minimum statistical values ​​of the available bandwidth for the resource nodes participating in the evaluation, respectively.

[0101] In one example, when the maximum and minimum statistical values ​​corresponding to a certain indicator are equal, the positive utility indicator corresponding to that indicator can be set to a preset neutral value, or set to a fixed utility value according to preset rules, to avoid the division by zero problem in the normalization process. After the above processing, the delay utility indicator, bandwidth utility indicator, and cost utility indicator are all mapped to a unified dimensionless interval, and the larger the value, the more the corresponding resource node matches the service preferences of the task to be scheduled.

[0102] In step S240, the demand preference vector is weighted and combined with various positive utility indicators, and a balance control factor is introduced to fuse the weighted combination result with the basic matching degree to obtain the comprehensive fit degree, which serves as the identity fit degree between the unified identity identifier and each resource node.

[0103] For example, identity fit can be expressed as follows:

[0104] Equation (7)

[0105] In the formula, It indicates overall compatibility and serves as a unified identity identifier. With resource nodes Identity compatibility between them; Indicates the balance control factor, and ; Indicates the basic matching degree; , and These represent the preference weights for latency, bandwidth, and cost dimensions in the demand preference vector, respectively. , and Representing resource nodes respectively Positive utility metrics in terms of latency, bandwidth, and cost.

[0106] In equation (7), the basic matching degree reflects the degree of matching between the identity attribute vector and the resource state feature vector, while the preference weighting part reflects the degree to which resource nodes meet service preference information in terms of latency, bandwidth, and cost. Balanced control factor This is used to adjust the proportion of basic matching degree and preference utility results in the overall fit. For example, for tasks that emphasize identity attributes, regional constraints, or trust requirements, the impact of basic matching degree in the overall fit can be increased; for tasks that emphasize performance experience or cost control, the impact of service preference utility results in the overall fit can be increased. Through the above fusion processing, identity fit can simultaneously reflect the basic matching relationship between the requesting subject and the resource node, as well as the degree to which the resource node satisfies the task service preferences.

[0107] Through the embodiments of this application, the computing power, link service capabilities, resource call costs, and service preferences of the tasks to be scheduled are converted into adaptation evaluation results under a unified scale. This reduces the impact of differences in the dimensions and values ​​of different indicators on the evaluation results, giving the identity adaptation degree a more stable numerical meaning and a clearer technical interpretation. Therefore, for different types of tasks to be scheduled, this adaptation degree calculation method can generate differentiated evaluation results based on the relationship between identity attributes, resource status, and service preferences, thereby improving the accuracy and interpretability of resource node matching evaluation.

[0108] Regarding the implementation details of generating a candidate resource node list based on identity adaptability, in some examples of embodiments of this application, historical scheduling task records of all resource nodes in the entire network within a preset sliding time window are obtained, and the comprehensive resource consumption and task scheduling timestamp of the historical scheduling tasks allocated to each resource node are extracted. The comprehensive resource consumption is obtained by fusing computational resource consumption and network link occupancy according to a preset resource consumption weight.

[0109] It should be noted that when a resource node undertakes multiple scheduling tasks consecutively within a short period, even if its static resource status still appears available, it may experience residual load, link occupancy, or delayed resource recovery due to the accumulation of recent tasks. Therefore, the system can maintain a preset sliding time window on the control side, only counting historical scheduling task records that fall within this time window. The length of this sliding time window can be set according to the service scheduling frequency, average task execution time, resource release cycle, or network load fluctuation, so that the candidate resource selection process can reflect the recent usage of resource nodes.

[0110] In computing-network convergence scenarios, the pressure on resource nodes from historical scheduling tasks typically stems not only from computational resource consumption but also from network link occupancy. For example, video analytics tasks may simultaneously consume significant computational resources and uplink bandwidth; background synchronization tasks may have lower computational resource requirements but continuously occupy network links. Therefore, computational resource consumption and network link occupancy can be combined into the calculation of overall resource consumption to more comprehensively characterize the resource occupancy of a single historical scheduling task on resource nodes and their associated links.

[0111] For example, the first Comprehensive resource consumption of each historical scheduling task It can be calculated using the following fusion formula:

[0112] Equation (8)

[0113] In the formula, Indicates the first The total resource consumption of each historical scheduling task. Indicates the first The normalized value of the computational resource consumption of each historical scheduled task. Indicates the first The normalized value of network link occupancy for each historical scheduled task. This indicates the preset computational resource consumption weight. This represents the preset network link consumption weight, and satisfies... ,and , .

[0114] Through the fusion processing of the above equation (8), resource consumption data from different sources can be converted into comprehensive resource consumption under a unified scale, which can reflect the resource occupation of historical scheduling tasks on the computing side and network side, and avoid judging the recent pressure of nodes based on a single resource dimension.

[0115] Then, based on the time decay mechanism, the comprehensive resource consumption of each historical scheduling task is merged and aggregated with the task scheduling timestamp to calculate the dispersion penalty factor of each resource node; among which, the dispersion penalty factor is used to characterize the degree of historical scheduling concentration and residual load heat of the node after time decay.

[0116] For example, the dispersion penalty factor It can be expressed by the following formula:

[0117] Equation (9)

[0118] In the formula, This represents the dispersion penalty factor. Indicates the allocation to resource nodes within the sliding time window. The total number of historical scheduling tasks, Indicates the first The normalized value of the total resource consumption of each historical scheduling task. Indicates the current scheduling decision time. Indicates the first The task scheduling timestamps of each historical scheduled task. The time decay constant representing the control of the heat cooling rate and ,in, Time unit and time decay constant The units match.

[0119] In equation (9), the closer a historical scheduling task is to the current scheduling decision time, the larger its corresponding time decay term, and the more significant its impact on the dispersion penalty factor; the farther a historical scheduling task is from the current scheduling decision time, the smaller its corresponding time decay term, and the weaker its impact on the current node state. If a resource node is continuously assigned multiple tasks with high overall resource consumption within the sliding time window, the dispersion penalty factor of that resource node will increase accordingly; if a resource node has not been frequently scheduled recently, or if historical tasks have undergone a long period of decay, its dispersion penalty factor will remain at a low level. In one example, when there are no tasks assigned to a resource node within the sliding time window... When scheduling historical tasks, you can Set to zero.

[0120] Then, the identity fit of each resource node is adjusted exponentially by using a dispersion penalty factor to obtain the corrected identity fit after hotspot congestion adjustment.

[0121] For example, correcting identity adaptation It can be expressed by the following formula:

[0122] Equation (10)

[0123] In the formula, Indicates a unified identity identifier With resource nodes The corrected identity fit after hotspot congestion mitigation adjustments This indicates the overall fit (i.e., the identity fit before adjustment). This represents the adjustment constant that controls the intensity of the penalty, and .

[0124] In equation (10) above, the larger the dispersion penalty factor, the smaller the exponential decay term, and the more significant the reduction in the original identity fit; the smaller the dispersion penalty factor, the closer the exponential decay term is to one, and the weaker the impact on the original identity fit. Thus, even if a resource node has a high identity fit due to strong computing power or good link conditions, if it has recently undertaken a large number of tasks, the corrected identity fit will be appropriately reduced. Adjustment constant Used to control the sensitivity of this attenuation adjustment, The larger the value, the more significant the impact of recent scheduling concentration on correcting identity adaptation.

[0125] The above processing can reduce the probability of high-scoring resource nodes being repeatedly selected in a short period of time, allow the historical usage of resource nodes to participate in the candidate list generation process, alleviate the problem of excessive concentration of node selection, and improve the load distribution balance of the candidate resource node list.

[0126] Then, based on the corrected identity adaptation of each resource node, the resource nodes participating in the evaluation across the entire network are sorted in descending order, and a predetermined number of resource nodes at the top of the ranking are selected to form a candidate resource node list.

[0127] In some implementations, the resource nodes participating in the evaluation can be sorted according to the modified identity adaptation score, and the resource nodes entering the candidate resource node list can be determined according to a preset number, preset ratio, or preset adaptation score threshold. The preset number can be a fixed value or configured according to the type of the task to be scheduled, the size of the resource pool, or the overall network load status. By using the modified identity adaptation score for sorting, the candidate resource node list can not only reflect the identity adaptation relationship and resource service capabilities, but also reflect the recent scheduling concentration of resource nodes.

[0128] For example, this sorting and truncation process can reduce the number of resource nodes participating in subsequent scheduling selections, keeping the candidate resource node list within a suitable size for control-side processing. Simultaneously, resource nodes with concentrated recent task loads or high residual load activity are suppressed during sorting, helping to prevent excessive concentration of candidate nodes on a few high-performance nodes. Therefore, the candidate resource node list can achieve a more stable selection result between adaptability and load distribution.

[0129] Through this embodiment, the system can evaluate resource nodes based on identity adaptability while incorporating the impact of recent scheduling concentration and residual load heat of resource nodes. This can more accurately reflect the availability of resource nodes at the current scheduling moment, reduce the implicit impact of historical task accumulation on resource carrying capacity, and improve the load balance and screening stability of the candidate resource node list.

[0130] Figure 3 A flowchart illustrating an example of determining a target resource node and network path in a method according to an embodiment of this application is shown.

[0131] like Figure 3 As shown, in step S310, the entire network link delay matrix is ​​extracted based on the network link status data, and candidate network paths are planned from the request initiation position of the task to be scheduled to each candidate node in the candidate resource node list.

[0132] In some implementations, the control side can obtain the link delay between adjacent network nodes in the entire network topology based on network link state data, and form a network-wide link delay matrix accordingly, which is used to characterize the transmission cost of different links in the current network topology. For each candidate node in the candidate resource node list, the system uses the request initiation location of the task to be scheduled as the source end and the candidate node as the destination end, and plans the corresponding candidate network path in the topology space represented by the network-wide link delay matrix. The candidate network path can be determined based on factors such as link delay, link reachability, path hop count, and network policy constraints. For example, the shortest path algorithm, constrained path search algorithm, or multi-path routing strategy can be used to determine the candidate network path, and this application embodiment does not specifically limit this.

[0133] This allows each candidate resource node to be associated with one or more reachable network paths from the request initiation location to that candidate node. In subsequent calculations, a path that meets preset path selection criteria can be selected as the candidate network path corresponding to the candidate node. For example, a path with low latency, high reachability, or that conforms to network policy constraints can be selected. This ensures that resource node selection no longer depends solely on the node-side resource status, but also considers the network carrying capacity required for the task to reach the node.

[0134] In step S320, for each candidate node, the immediate scheduling benefit of scheduling the task to be scheduled to the corresponding candidate node is calculated based on the value assessment model. The immediate scheduling benefit is obtained by comprehensively evaluating the positive gain from correcting identity adaptation and the negative loss from the normalized physical link cost on the candidate network path. The negative loss from the normalized physical link cost is determined based on the path loss adjustment factor.

[0135] For example, the immediate benefit of scheduling can be calculated using the following formula:

[0136] Equation (11)

[0137] In the formula, This indicates that the task to be scheduled will be scheduled to a candidate node. The immediate benefits of scheduling Indicates a unified identity identifier With candidate nodes Correcting identity compatibility between them Indicates the positive gain adjustment factor and , Indicates the path loss adjustment factor and , This indicates the distance from the request initiation position of the task to be scheduled to the candidate node. Candidate network paths, Indicate candidate network paths A link between adjacent network nodes, Representing adjacent network nodes and The link delay cost after normalization of the link delay between them.

[0138] In equation (11) above, the modified identity fit degree is used to characterize the degree of fit between the candidate node and the scheduled task and its unified identity identifier, and the path link delay cost is used to characterize the network transmission cost required for the scheduled task to reach the candidate node. By introducing a positive gain adjustment factor and a path loss adjustment factor, the influence of node fit advantage and network path cost on the instantaneous benefit calculation can be adjusted. In order to enable the two types of quantities to participate in the same calculation process, the link delay can be converted into a dimensionless normalized link delay cost in advance. Thus, if a candidate node itself has a high modified identity fit degree, but its corresponding candidate network path has a high link delay, the instantaneous benefit of the node's scheduling will be suppressed accordingly; if a candidate node has both a high fit degree and a low path cost, its instantaneous benefit of scheduling will be correspondingly higher.

[0139] Therefore, node adaptability and network path cost can be evaluated simultaneously at the candidate node level, making the selection of target resource nodes more in line with end-to-end service requirements and reducing the risk of resource mismatch caused by selecting resource nodes solely based on node adaptability while ignoring path cost.

[0140] In step S330, the network load status data is parsed to obtain the normalized load occupancy rate of each candidate node, and the historical scheduling records and corresponding execution feedback records of each candidate node are retrieved to calculate the historical long-term service satisfaction score.

[0141] Here, the normalized load occupancy rate is used to characterize the resource consumption of candidate nodes in the current or near real-time state. It is determined comprehensively based on parameters such as the candidate node's processor utilization, memory utilization, queue length, instance load, and network access load, and mapped to a unified value range. The historical long-term service satisfaction score is determined based on the execution feedback records of candidate nodes in historical scheduled tasks. For example, it can be obtained by statistically analyzing indicators such as task completion rate, service quality compliance rate, failure rate, timeout count, and service stability. This score characterizes the service performance of candidate nodes during historical operation.

[0142] By incorporating normalized load occupancy and historical long-term service satisfaction scores into the evaluation, the system can simultaneously consider the current load status and historical service performance of candidate nodes. This avoids selecting nodes that seem suitable at present but have unstable long-term service performance based solely on a single immediate benefit, and also reduces the probability of nodes with excessively high current load being selected again.

[0143] In step S340, based on the normalized load occupancy rate and historical long-term service satisfaction score, the expected future revenue of each candidate node is constructed, and a future revenue weighting factor is introduced to integrate the scheduling immediate revenue and the expected future revenue to obtain the total expected value of each candidate node.

[0144] For example, candidate nodes Total value expected It can be expressed by the following formula:

[0145] Equation (12)

[0146] In the formula, Indicates a unified identity identifier The corresponding tasks to be scheduled are scheduled to candidate nodes. Total expected value at that time Indicates immediate benefits from scheduling. This represents the preset future return weighting factor and , Indicates candidate nodes Normalized load factor and , Indicates candidate nodes Historical long-term service satisfaction rating and The larger the value, the more likely it is to be a candidate node. The higher the historical service satisfaction, the better.

[0147] In the above equation (12), This can characterize the current relative idle level of candidate nodes. A higher value indicates a lower normalized load factor for a candidate node, while a lower value indicates a higher load factor for a candidate node. Combining the relative idle level with historical long-term service satisfaction scores yields the expected future revenue related to node load stability. A future revenue weighting factor is used to adjust the impact of this expected future revenue on the total expected value. By integrating the immediate scheduling revenue with the expected future revenue, the total expected value can simultaneously reflect the immediate adaptability of the current task scheduling, the cost of candidate paths, the current load status of candidate nodes, and the historical service performance of candidate nodes.

[0148] Therefore, the evaluation results of candidate nodes not only reflect the immediate benefits of the current scheduling, but also reflect the performance of candidate nodes in terms of load margin and historical stability. This can improve the stability of node value assessment and reduce the situation where nodes with excessive current load or poor historical service performance are selected due to short-term high adaptability.

[0149] In step S350, by introducing a soft maximum function of the temperature adjustment parameter, the total value expectation of each candidate node is mapped to the node selection probability, so that the candidate node with a higher total value expectation has a higher selection probability, and the degree of random exploration in the scheduling decision is controlled by the temperature adjustment parameter.

[0150] For example, based on each candidate node Total value expected Each candidate node is calculated using a soft maximum function. Node selection probability :

[0151] Equation (13)

[0152] In the formula, Indicates candidate nodes The probability of a node being selected as the target resource node. This represents the set consisting of each candidate node in the candidate resource node list. Represents a set Any candidate node in, Indicates candidate nodes Total value expected Indicates temperature control parameters and .

[0153] In equation (13) above, the soft maximum function transforms the total value expectation of each candidate node into a probability distribution, allowing each candidate node to obtain the corresponding node selection probability. Candidate nodes with higher total value expectations usually have a higher node selection probability, but candidate nodes with lower total value expectations can still retain a certain chance of being selected. The temperature adjustment parameter is used to control the smoothness of the probability distribution. When the temperature adjustment parameter is small, the probability distribution is more concentrated on candidate nodes with higher total value expectations; when the temperature adjustment parameter is large, the probability distribution is smoother, and the difference in selection probability among multiple candidate nodes decreases. Thus, an adjustment can be made between favoring high-value nodes and maintaining a certain degree of random exploration, so that node selection no longer adopts only the highest-scoring hard selection method, but forms a probabilistic selection result based on the total value expectation, which can reduce the possibility of multiple scheduled tasks continuously concentrating on the same high-scoring node, while retaining the selection tendency for high-value candidate nodes.

[0154] In step S360, probabilistic selection is performed according to the node selection probability to determine the target resource node from the list of candidate resource nodes, and the candidate network path corresponding to the target resource node is determined as the network path.

[0155] In some implementations, target resource nodes can be determined using random sampling based on probability distributions, or by selecting nodes according to probability intervals. Once a candidate node is selected as the target resource node, the candidate network path corresponding to that target resource node is determined as the network path for the task to be scheduled. This allows the expected total value of each candidate node to be transformed into the actual scheduling selection result, and the corresponding network path to be determined simultaneously. This ensures consistency between target resource node selection and network path selection, avoiding path mismatch problems caused by separate decision-making for resource nodes and network paths.

[0156] Through the embodiments of this application, the system can, based on a list of candidate resource nodes, jointly evaluate the corrected identity adaptation of candidate nodes, the cost of candidate network paths, current load status, and historical service performance, and determine the target resource node and network path using a probabilistic selection method. This implementation can improve the ability of scheduling decisions to perceive end-to-end resource conditions, reduce the risk of centralized node scheduling caused by selecting a single highest-scoring node, and improve the coordination between the target resource node and the network path.

[0157] In some examples of embodiments of this application, before calculating the identity adaptation degree between the unified identity identifier and each resource node, a forward-looking estimation of the future state of each resource node based on a time-series prediction model can be performed to dynamically adjust the available capacity.

[0158] More specifically, firstly, based on task request information and combined with service preference information, the business type to which the task to be scheduled belongs is identified, and the business prediction time window corresponding to the business type is determined.

[0159] In practical implementation, task request information can include task type, service entry point, task scale, request initiation location, application layer protocol identifier, or task execution constraints. Service preference information can include latency preferences, bandwidth preferences, cost preferences, and reliability preferences. The control side can identify the service type of the task to be scheduled based on the above information. For example, industrial control tasks typically have strong real-time requirements, and their resource usage duration may be short but fluctuates significantly; image analysis tasks may simultaneously consume computing resources and transmission bandwidth; batch archiving tasks typically have lower latency requirements but longer durations. Different service types correspond to different load change time scales, therefore, different service prediction time windows can be configured for different service types. .

[0160] Business forecast time window This time window is used to define the time range for estimating the future load status of resource nodes. This time window can be determined based on the service type configuration table, service level policy, or historical service statistics. For example, real-time control tasks may correspond to a shorter prediction time window, while data synchronization tasks may correspond to a longer prediction time window. By determining the prediction time window based on the service type, the time span of load estimation can be matched with the resource consumption characteristics of the tasks to be scheduled, reducing capacity estimation bias caused by using a fixed prediction span.

[0161] Then, based on historical load monitoring records and combined with the network-wide load status data, normalized actual load occupancy rate data of each resource node in the network is extracted from the historical time series before the current sampling time.

[0162] In practice, historical load monitoring records can be collected by resource monitoring components, node agents, edge gateways, resource management platforms, or network telemetry components. These records can include information such as the computing resource utilization rate, task queue utilization, instance load level, and network access load of resource nodes at multiple historical sampling times. Network-wide load status data can be used to supplement the overall load status of each resource node's region, resource pool, or link segment at the current moment.

[0163] Because different resource nodes may have different hardware specifications, resource capacities, and load metering methods, raw load data is not suitable for direct horizontal comparison. Therefore, the collected historical load data can be normalized to map the actual load occupancy of different nodes to a unified value range. After normalization, the resource nodes... At historical sampling time The actual load utilization rate can be expressed as .in, The index represents the historical time step. This normalized historical load sequence can express the load changes of different resource nodes over time in a relatively consistent manner.

[0164] Subsequently, the normalized actual load utilization rate data is input into the time series forecasting model, and a forward-looking time series extrapolation is performed using a dynamic autoregressive weighting mechanism that matches the business forecasting time window to obtain the forward-looking predicted load utilization rate of each resource node when it crosses the business forecasting time window in the future.

[0165] For example, resource nodes Forward-looking forecast of load occupancy It can be expressed by the following formula:

[0166] Equation (14)

[0167] In the formula, Represents resource nodes For business forecasting time windows Forward-looking forecast of load occupancy, This represents the total number of historical time steps extracted. An index representing a historical time step. Indicates the current sampling time. Represents resource nodes At historical sampling time Normalized actual load factor Indicates the time window for business forecasting Matching the first The dynamic autoregressive weights for each historical time step, and , .

[0168] In equation (14) above, the dynamic autoregressive weights are used to characterize the contribution of different historical time steps to the prediction results. For shorter business prediction time windows, historical load data closer to the current sampling time can be assigned higher weights to enhance the response to short-term load changes; for longer business prediction time windows, the participation of earlier historical time steps can be appropriately increased to reflect the load change trend over a longer time scale. Since all dynamic autoregressive weights are non-negative and the sum of the weights is one, and To normalize the actual load utilization, therefore It can also be kept within a normalized scale consistent with historical load factors, making it convenient as a basis for capacity adjustment.

[0169] Next, the current available computing capacity of each resource node is obtained, and the current available computing capacity is redundantly reduced by using the corresponding forward-looking predicted load occupancy rate and preset capacity redundancy control factor to obtain the corrected available computing capacity.

[0170] For example, resource nodes Corrected available computing capacity It can be expressed by the following formula:

[0171] Equation (15)

[0172] In the formula, Represents resource nodes The corrected available computing capacity, Represents resource nodes The current available computing capacity, This represents the capacity redundancy control factor, and , Represents resource nodes For business forecasting time windows Forward-looking prediction of load occupancy.

[0173] In equation (15) above, a higher forward-looking load occupancy rate indicates that resource nodes may have a higher load level within the business forecast time window, resulting in a greater reduction in the current available computing capacity; a lower forward-looking load occupancy rate indicates that resource nodes experience less load pressure within the forecast time window, resulting in a smaller reduction in the current available computing capacity. Capacity redundancy control factor Used to control the reduction rate. When When the value is large, the system is more sensitive to predicting load changes; when When the value is small, the capacity correction is closer to the currently available computing capacity. Because... ,and Within the normalized load range, therefore the reduction factor It can be kept within a reasonable range to avoid unreasonable negative values ​​in the corrected available computing capacity.

[0174] Furthermore, the corrected available computing capacity is used as the available capacity feature of the corresponding resource node to update the resource status data, so as to inject load fluctuation redundancy in advance during the process of generating the candidate resource node list.

[0175] In practical implementation, the control side can write the corrected available computing capacity of each resource node into the resource status data to replace or supplement the original current available computing capacity. In this way, the available capacity feature in the resource status data no longer only represents the static idle resources at the current sampling time, but also includes load change estimates based on the business prediction time window. For resource nodes with high predicted load occupancy, their available capacity feature will be adjusted downwards accordingly; for resource nodes with low predicted load occupancy, the reduction in their available capacity feature will be smaller.

[0176] Through the embodiments of this application, the available computing capacity of resource nodes can be represented by combining the current state and predicted load changes, which can reduce the capacity estimation bias caused by relying solely on the current sampled values, and make the resource status data closer to the resource carrying capacity within the service prediction time window. For streaming services, bursty services, and services with long durations, this process can improve the stability of resource capacity evaluation and reduce the possibility of resource nodes being overestimated in terms of available capacity under short-term load changes.

[0177] In some examples of embodiments of this application, after the network configuration rules are sent to the corresponding network devices for execution and the resource scheduling instructions are sent to the corresponding resource management platform for execution, the actual running performance parameters of the task to be scheduled on the target resource node and network path can be continuously obtained, and the actual response latency and actual available bandwidth can be extracted.

[0178] In practice, actual operational performance parameters can be obtained from feedback by network telemetry components, resource node monitoring components, application layer probes, or resource management platforms. These parameters may include task processing time, end-to-end response latency, actual available bandwidth, link packet loss rate, task completion status, and resource usage. In this example, the actual response latency can be extracted from the actual operational performance parameters. and actual available bandwidth This is to evaluate whether the scheduling results that have been executed meet the performance requirements in the service preference information.

[0179] Because the underlying network and resource nodes may be affected by background traffic, resource contention, link fluctuations, or virtualization scheduling during operation, the evaluation results used in scheduling decisions may differ from the actual performance during task execution. Therefore, by continuously collecting actual response latency and actual available bandwidth, runtime data can be obtained to evaluate the scheduling execution results. This process enables the control side to grasp the true performance deviation during task execution, rather than relying solely on pre-scheduling estimates.

[0180] Subsequently, by combining the expected response latency, expected available bandwidth, and demand preference weights in the service preference information, the actual operating performance parameters are subjected to one-way truncation error accumulation only for performance indicators that fail to meet the standards, and the system execution error cost is calculated.

[0181] For example, the system execution error cost It can be expressed by the following formula:

[0182] Equation (16)

[0183] In the formula, This represents the cost of system execution errors. Indicates the actual response delay. Indicates the actual available bandwidth. Indicates the expected response delay and , Indicates the expected available bandwidth and , and These represent the preference weights for latency and bandwidth in the demand preference weights, respectively. This indicates a positive truncation function, used to accumulate errors only for performance indicators that fail to meet the target.

[0184] In equation (16) above, the error term in the latency dimension is zero when the actual response latency does not exceed the expected response latency; and the error term in the bandwidth dimension is zero when the actual available bandwidth is not lower than the expected available bandwidth. Therefore, the system execution error cost mainly reflects the portion of actual performance that does not meet service preference requirements, rather than using better-than-expected indicators as negative errors to offset other substandard indicators. By introducing latency preference weights and bandwidth preference weights, the error cost can be aligned with the task's focus on different service quality indicators. For example, for low-latency tasks, the impact of latency error in the system execution error cost can be higher; for bandwidth-sensitive tasks, the impact of bandwidth error can be higher.

[0185] Next, in response to the system execution error cost exceeding the preset tolerance threshold, a self-learning update closed loop is triggered for the importance weights of each matching dimension corresponding to the calculation of the basic matching degree, and the path loss adjustment factor corresponding to the calculation of the scheduling instant benefit.

[0186] In practice, the tolerance threshold is used to distinguish between acceptable operational fluctuations and execution deviations that require correction. Short-term jitter may occur during network transmission and resource scheduling. If parameters are updated for every tiny error, it can easily lead to frequent changes in model parameters. Therefore, the relevant parameters can be updated only when the system execution error cost exceeds the preset tolerance threshold. This reduces over-adjustment caused by occasional fluctuations and allows parameter updates to focus more on execution deviations that have a significant impact on service quality.

[0187] Then, for identity-resource mapping pairs that cause the system execution error cost to exceed the limit, based on the gradient direction of the system execution error cost relative to the weights of each matching dimension, a multiplicative weight update mechanism is adopted to adaptively and iteratively update the importance weights of each matching dimension when calculating the basic matching degree, so as to adjust the importance weight allocation of each matching dimension while ensuring that the sum of the importance weights after the update remains constant.

[0188] In one example, the system execution error cost relative to the first can be estimated based on historical execution samples, offline evaluation samples, or finite perturbation evaluation methods. The partial derivative gradient of the importance weights for each matching dimension. For example, this partial derivative gradient can be estimated using the following finite difference method:

[0189] Equation (17)

[0190] In the formula, Indicates the perturbation step size. Indicating in relation to the first After perturbing the importance weights of each matching dimension and renormalizing the importance weights of each matching dimension, the estimated system execution error cost is obtained. This represents the system execution error cost before the perturbation. This finite difference method is used to estimate the direction of the impact of changes in the importance weights of the corresponding matching dimension on the system execution error cost.

[0191] For example, the importance weights of each matching dimension can be updated using the following formula:

[0192] Equation (18)

[0193] In the formula, and These represent the numbers after the update and before the update, respectively. The importance weights of each matching dimension This represents the learning rate factor used to control the step size of weight adjustment. , and They represent the system execution error cost estimated based on historical execution samples or the finite difference method, respectively, relative to the first... The and the first The partial derivative gradient of the importance weights of each matching dimension This represents the matching dimension index for cumulative summation. This represents the total number of matching dimensions. In one example, the initial values ​​for the importance weights of each matching dimension can be set to positive values ​​to prevent a particular matching dimension from being permanently excluded in subsequent iterations.

[0194] In equation (18) above, the exponent term is used to adjust the importance weights of the corresponding matching dimensions according to the gradient direction. When increasing the importance weight of a certain matching dimension may lead to an increase in the system execution error cost, the corresponding gradient is positive, and the weight of that dimension will decrease accordingly after the update; when increasing the importance weight of a certain matching dimension may reduce the system execution error cost, the corresponding gradient is negative, and the weight of that dimension will increase accordingly after the update. The denominator is used to normalize the importance weights of each matching dimension, so that the updated importance weights maintain the same normalization constraint. Through this processing, the influence of each matching dimension in the basic matching degree calculation can be adjusted according to the execution error feedback.

[0195] Then, the historical congestion frequency and the total number of historical scheduling events on the network path are extracted within the historical time period. Based on the ratio of historical congestion frequency to the total number of historical scheduling events, the path loss adjustment factor corresponding to the calculation of the instantaneous scheduling benefit is dynamically increased.

[0196] For example, the path loss adjustment factor can be updated and adjusted using the following formula:

[0197] Equation (19)

[0198] In the formula, and These represent the path loss adjustment factors after the update and before the update, respectively. This represents the preset penalty compensation coefficient and , Indicates the frequency of historical congestion. Indicates the total number of historical scheduling events and .

[0199] Here, the path loss adjustment factor is used to control the impact of network path costs during the calculation of immediate scheduling benefits. If a network path or region experiences high congestion frequency during a historical period, it indicates that the actual performance of that path or region may be lower than the estimated results during the scheduling phase. By increasing the path loss adjustment factor according to the proportion of congestion frequency, the impact of network link costs in the calculation of immediate scheduling benefits can be increased, causing the same or similar paths to be subject to higher path cost constraints in the new scheduling calculation. If no congestion occurred during a historical period, or if the historical congestion frequency was low, the increment of the path loss adjustment factor is smaller.

[0200] In one example, to prevent the path loss adjustment factor from continuously increasing and exceeding the available adjustment range, a preset upper limit can be set for the path loss adjustment factor. When the calculated updated value exceeds the preset upper limit, the path loss adjustment factor can be limited to that preset upper limit, which can improve the stability of the parameter update process.

[0201] Furthermore, the updated importance weights are used to calculate the basic matching degree in subsequent calculations, and the updated path loss adjustment factors are used to calculate the real-time benefits of scheduling in subsequent calculations, so as to drive the long-term autonomous iteration and evolution of dynamic scheduling decisions.

[0202] In some implementations, the control side can write the updated importance weights and path loss adjustment factors into the parameter configuration data, allowing them to participate in the adaptation evaluation and scheduling calculation of new tasks to be scheduled. The updated importance weights can change the degree of influence of different matching dimensions in the basic matching degree; the updated path loss adjustment factor can change the degree of influence of network path cost in the scheduling immediate benefits. By continuously using execution performance parameters to correct the above parameters, the identity matching degree calculation and dynamic scheduling decision can be made closer to the actual operating state of the network and resource environment.

[0203] Through the embodiments of this application, the system can correct relevant parameters in identity adaptation calculation and path cost evaluation based on actual response latency, actual available bandwidth, and link congestion. This reduces the deviation between the estimation results and task execution results during the scheduling phase and improves the adaptability of parameter configuration to complex time-varying network environments. For scenarios with strong network state fluctuations, significant differences in resource node performance, or frequent changes in service load, this can improve the stability of scheduling parameters and the consistency of service quality of resource mapping results.

[0204] Figure 4 This paper illustrates a schematic diagram of the system operation mechanism of an example of a network resource dynamic mapping and scheduling method based on identity resolution according to an embodiment of this application.

[0205] like Figure 4 As shown, the system's operating mechanism mainly consists of four core processing modules and their interacting logical flows. First, the unified identity resolution engine, as the system's perception entry point, receives multi-source identity signals, parses and aggregates them into cross-domain unified identity identifiers and attribute vectors. Subsequently, the Weighted Identity-Resource Mapping Strategy (WIRMS) receives the aforementioned identity attribute vectors and, combined with network resource status and user service preferences, generates identity fit and candidate resource sets through multi-dimensional weighted matching calculations.

[0206] Next, Identity-Driven Resource Allocation (IDRA), based on the aforementioned candidate resource set, comprehensively considers network path constraints and immediate node benefits, outputting a probabilistic selection decision that satisfies the scheduling policy, thereby driving the underlying network devices to execute resource scheduling. Finally, the system monitoring and adaptive update module continuously collects actual operational performance parameters (such as actual response latency and actual available bandwidth) on the data plane, and calculates the system execution error cost, dynamically updating and iteratively evolving the matching weights and path loss adjustment factors according to the error gradient. Thus, through bidirectional data flow closed-loop collaboration, the entire dynamic scheduling process from identity awareness to resource mapping and then to policy self-learning is realized.

[0207] To evaluate the effectiveness of the proposed identity-based network resource dynamic mapping and scheduling mechanism (WIRMS+IDRA) in a near-realistic engineering environment, this application constructs a cloud-edge-device collaborative simulation environment integrating computing and network based on the OMNeT++ discrete event network simulation platform and combined with the Ryu SDN controller. In terms of physical topology and resource configuration, this simulation environment adopts the Fat-Tree topology commonly used in large-scale data centers, including 4 core switches, 8 aggregation switches, and 16 edge / access switches, and is configured with a total of 256 heterogeneous computing nodes; the link bandwidth between each network node is heterogeneously distributed between 1 Gbps and 10 Gbps.

[0208] Regarding traffic modeling and identity generation, the simulation environment simulated 5000 independent terminal entities, whose service request arrival patterns followed a Poisson distribution. To verify the application's ability to perceive and schedule multi-source identities and differentiated service preferences, the generated service requests were proportionally divided into three typical types: latency-sensitive (such as industrial real-time control services, accounting for 30%), bandwidth-intensive (such as ultra-high-definition video streaming services, accounting for 40%), and best-effort (such as background synchronization services, accounting for 30%).

[0209] To conduct objective performance comparison analysis, four baseline algorithms were set up for comparison with the method of this application. The specific baselines included: an IP-Base algorithm using traditional shortest path routing and static threshold allocation; an identity-aware segmentation algorithm (ID-Seg) that only uses identity for secure access control without interfering with resource scheduling; an in-Packet Preference-aware Composite Resource Scheduling Algorithm (IPRSA) that uses packet extension header preference fields combined with multiple weighted indicators; and a Dynamic Virtual Network Embedding based on Deep Reinforcement Learning (DVNE-DRL) algorithm that perceives network topology. The method proposed in this application serves as a comprehensive comparison group, fully integrating the entire process of Unified Identity Resolution (UI), Weighted Identity Resource Matching (WIRMS), and Dynamic Value Iterative Scheduling (IDRA).

[0210] Figure 5 This diagram illustrates a comparative simulation of the request acceptance rates of different methods under dynamic load.

[0211] like Figure 5 As shown, with the normalized network request arrival rate gradually increasing (from 0.2 to 1.0), due to the saturation of physical resources, the request acceptance rates of all compared algorithms show varying degrees of decline. The shaded line in the figure represents the mean acceptance rate and its variance under different random seeds in multiple simulation experiments. In the low-load stage (when the normalized network request arrival rate is less than 0.4), except for the traditional shortest path static allocation algorithm (IP-Base), all other algorithms can maintain a relatively high request acceptance rate of over 90%.

[0212] However, when the network enters a severely congested, high-load state (normalized network request arrival rate between 0.8 and 1.0), the performance of traditional algorithms degrades significantly. Specifically, the request acceptance rate of the identity-based access control algorithm (ID-Seg) and the traditional static allocation algorithm (IP-Base) drops sharply, falling to around 30% to 35%. The deep reinforcement learning-based embedding algorithm (DVNE-DRL) and the ICN-based heuristic scheduling algorithm (IPCRSA) perform slightly better, maintaining around 50% to 55%. In contrast, the dynamic mapping and scheduling method (IDRA) proposed in this application can still maintain the highest request acceptance rate of approximately 65% ​​when the network is fully loaded (arrival rate of 1.0), and its corresponding error shadow band is the most convergent, indicating that the method has extremely high stability in dynamically fluctuating network environments.

[0213] In this embodiment, by introducing a dispersion penalty factor based on a time-decrease mechanism during the resource matching phase, local "hotspot congestion" caused by excessive concentration of high-quality resources across the entire network is effectively avoided. Simultaneously, the dynamic scheduling decision mechanism can perform differentiated resource degradation and coordinated allocation based on multi-source identity preferences, thereby maximizing global network throughput and request response success rate under complex dynamic loads.

[0214] Figure 6 This diagram illustrates a comparative simulation of the cumulative distribution of end-to-end task completion delay for latency-sensitive services under full load conditions using different methods. The horizontal axis of the cumulative distribution function graph represents the end-to-end task completion delay (using logarithmic coordinates), and the vertical axis represents the cumulative probability.

[0215] like Figure 6As shown, traditional heuristic scheduling algorithms based on information-centric networks (IPCRSA) or virtual network embedding algorithms based on deep reinforcement learning (DVNE-DRL) often focus on the optimal mapping of the global physical topology, lacking deep awareness of the requesting entity's identity and specific business attributes. This leads to a severe "long tail effect" when processing high-priority tasks. It is clearly visible in the figure that these two traditional algorithms can only complete a response within 20ms for less than 60% of latency-sensitive tasks.

[0216] In contrast, the Dynamic Mapping and Scheduling (IDRA) method proposed in this application demonstrates better long-tail latency suppression and service quality differentiation. Its cumulative probability curve rises extremely rapidly, with approximately 85% of latency-sensitive tasks responding quickly within 20ms. This significant performance leap proves the effectiveness of this application's embodiments in addressing network jitter and queuing issues.

[0217] In this embodiment, by pre-constructing an identity attribute vector containing demand preference vectors (such as preference weights for latency), the underlying network resource scheduling is endowed with identity semantic awareness capabilities. When performing identity adaptation calculation and dynamic value optimization, the system can assign differentiated resource priority control and network path selection to "latency-sensitive" services, thereby providing stable and deterministic end-to-end latency guarantees for core services even in complex dynamic load environments.

[0218] Figure 7 The diagram shows a comparative simulation of different methods in terms of system control surface overhead and performance trade-offs. The diagram is a dual-axis stacked bar chart, where the horizontal axis represents the scale of concurrent requests (requests / second), the left main vertical axis (bar chart) represents the average decision delay per control surface (ms), and the right secondary vertical axis (line chart) represents the CPU resource utilization rate (%).

[0219] like Figure 7 As shown, as the number of concurrent requests increases from 1,000 to 50,000, the decision latency and computational resource utilization of all compared algorithms (including the traditional static allocation algorithm IP-Base, the deep reinforcement learning algorithm DVNE-DRL, and the proposed method) all show an upward trend. Specific data in the figure shows that in a high-concurrency scenario of 10,000 requests / second, the average latency for generating a single scheduling policy using the proposed method is approximately 4.2ms, which is about 45% and 15% higher than the IP-Base and DVNE-DRL schemes, respectively.

[0220] in accordance with Figure 7The stacked structure of the bar charts shown can be seen intuitively. The performance bottleneck of the additional computational burden on the control plane is mainly concentrated on "additional overhead (parsing + optimization)". This confirms the inevitable computational consumption brought about by the introduction of a cross-domain identity parsing engine (such as multi-source feature association graph calculation) and dynamic programming optimization mechanism in the decision-making stage of the embodiments of this application. It reveals the Pareto trade-off relationship in the system design, that is: the increase in millisecond-level latency of the control plane and the consumption of some CPU resources are reasonable costs in exchange for the global high throughput of the data plane under extremely high load and the suppression of long-tail latency of core business.

[0221] In response to the control plane overhead revealed above, in future engineering implementation and continuous system evolution, we can further reduce the computation time of cross-domain parsing and joint optimization by introducing dedicated hardware acceleration modules (such as FPGA parallel computing acceleration for graph neural networks) at the edge control nodes, or by establishing a flow table cache prefetching mechanism for low-frequency changing identity attribute features, thereby effectively making up for the lack of this control plane overhead.

[0222] In summary, this application proposes a dynamic mapping and scheduling method for network resources based on identity resolution. By constructing a cross-domain unified identity resolution mechanism and combining weighted identity resource matching and dynamic value scheduling decision algorithms, this scheme achieves accurate mapping from "request subject identity" to "underlying network resources." Multiple simulation results confirm that this scheme effectively alleviates the local hotspot problem that easily occurs in traditional scheduling, exhibits strong anti-congestion capabilities in complex, dynamic, and high-load scenarios, and improves global resource utilization and request acceptance rate. Simultaneously, through refined priority isolation of high-priority identities, this scheme successfully suppresses long-tail latency caused by network jitter, compensating for the shortcomings of traditional scheduling mechanisms in identity semantic awareness and differentiated service guarantees. Furthermore, this application achieves a reasonable trade-off between performance gains and computational overhead in its system architecture design, successfully exchanging millisecond-level latency costs on the control plane for extremely high mapping success rates and business determinism on the data plane, demonstrating significant engineering application value.

[0223] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of combined actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Secondly, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application. In the above embodiments, the descriptions of each embodiment have their own emphasis; for parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0224] Figure 8A structural block diagram of an example of a network resource dynamic mapping and scheduling system based on identity resolution, according to an embodiment of this application, is shown.

[0225] like Figure 8 As shown, the network resource dynamic mapping and scheduling system 800 based on identity resolution includes a data perception and acquisition unit 810, an identity fusion and parsing unit 820, an identity adaptation and evaluation unit 830, a joint scheduling decision unit 840, and a policy orchestration and distribution unit 850.

[0226] The data sensing and acquisition unit 810 is used to acquire task request information of the task to be scheduled, multi-source identity signals associated with the task to be scheduled, resource status data of multiple resource nodes, network load status data, and service preference information corresponding to the task to be scheduled; wherein, the task request information includes the request initiation location of the task to be scheduled, and the resource status data includes the computing resource status data of the resource nodes and the network link status data associated with the resource nodes.

[0227] The identity fusion and parsing unit 820 is used to fuse and parse the multi-source identity signals to construct a cross-domain unified identity identifier associated with the task to be scheduled, and generate an identity attribute vector corresponding to the unified identity identifier.

[0228] The identity adaptation evaluation unit 830 is used to calculate the identity adaptation degree between the unified identity identifier and each of the resource nodes based on the identity attribute vector, the resource status data and the service preference information, and generate a candidate resource node list based on the identity adaptation degree.

[0229] The joint scheduling decision unit 840 is used to perform dynamic scheduling decisions based on the candidate resource node list, the network load status data, and the network link status data, so as to determine the target resource node that matches the task to be scheduled from the candidate resource node list, and to determine the network path from the request initiation location of the task to be scheduled to the target resource node.

[0230] The strategy orchestration and distribution unit 850 is used to generate network configuration rules corresponding to the network path and resource scheduling instructions corresponding to the target resource node, and distribute the network configuration rules to the corresponding network devices for execution, and distribute the resource scheduling instructions to the corresponding resource management platform for execution, so as to allocate the network path and the target resource node's computing resources to the scheduled task associated with the unified identity.

[0231] In some embodiments, this application provides a non-volatile computer-readable storage medium storing one or more programs including execution instructions. The execution instructions can be read and executed by electronic devices (including but not limited to computers, servers, or network devices) to perform the steps of any of the above-described network resource dynamic mapping and scheduling methods based on identity resolution.

[0232] In some embodiments, this application also provides a computer program product, the computer program product including a computer program stored on a non-volatile computer-readable storage medium, the computer program including program instructions, which, when executed by a computer, cause the computer to perform the steps of any of the above-described network resource dynamic mapping and scheduling methods based on identity resolution.

[0233] In some embodiments, this application also provides an electronic device, which includes: at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of a network resource dynamic mapping and scheduling method based on identity resolution.

[0234] The above-described product can perform the methods provided in the embodiments of this application, and has the corresponding functional modules and beneficial effects for performing the methods. Technical details not described in detail in this embodiment can be found in the methods provided in the embodiments of this application.

[0235] The electronic devices in this application can exist in various forms, including but not limited to: mobile communication devices, ultra-mobile personal computer devices, portable entertainment devices, or other airborne electronic devices with data interaction functions.

[0236] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0237] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented using software plus a general-purpose hardware platform, or of course, using hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0238] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for dynamic mapping and scheduling of network resources based on identity resolution, characterized in that, The method includes: The system acquires task request information of the task to be scheduled, multi-source identity signals associated with the task to be scheduled, resource status data of multiple resource nodes, network load status data, and service preference information corresponding to the task to be scheduled. The task request information includes the request initiation location of the task to be scheduled, and the resource status data includes the computing resource status data of the resource nodes and the network link status data associated with the resource nodes. The multi-source identity signals are fused and parsed to construct a cross-domain unified identity identifier associated with the task to be scheduled, and an identity attribute vector corresponding to the unified identity identifier is generated. Based on the identity attribute vector, the resource status data, and the service preference information, calculate the identity compatibility degree between the unified identity identifier and each of the resource nodes, and generate a candidate resource node list based on the identity compatibility degree; Dynamic scheduling decisions are made based on the candidate resource node list, the network load status data, and the network link status data to determine the target resource node that matches the task to be scheduled from the candidate resource node list, and to determine the network path from the request initiation location of the task to be scheduled to the target resource node. Generate network configuration rules corresponding to the network path and resource scheduling instructions corresponding to the target resource node, and send the network configuration rules to the corresponding network devices for execution, and send the resource scheduling instructions to the corresponding resource management platform for execution, so as to allocate the network path and the target resource node computing resources to the scheduled task associated with the unified identity; The step of fusing and parsing the multi-source identity signals to construct a cross-domain unified identity identifier associated with the task to be scheduled, and generating an identity attribute vector corresponding to the unified identity identifier, includes: Multi-source identity signals are mapped to a unified feature space with a preset feature dimension to obtain the feature vector of each identity signal in the unified feature space. Based on the matching degree and corresponding importance weight of each preset feature dimension, the feature matching similarity of any two identity signals in the unified feature space is calculated. Using the multi-source identity signals as graph vertices, identity signal pairs whose feature matching similarity satisfies a preset similarity threshold are constructed as associated edges, and the corresponding feature matching similarity is used as the edge weight of the associated edges to construct an identity relationship graph. The feature vectors of each identity signal in the unified feature space are used as the initial hidden layer representation vectors of the corresponding graph vertices, and a graph neural network model is used to perform feature aggregation on the identity relationship graph. The graph neural network model iteratively updates the hidden layer representation vectors of each graph vertex through the mean aggregation function, so as to reduce the impact of differences in neighborhood size and feature scale of different nodes on the identity aggregation results while utilizing the local structure of the adjacency topology. Based on the final hidden layer representation vector obtained after multiple updates of the graph neural network model, a clustering algorithm is used to cluster the graph vertices to obtain identity signal clusters corresponding to the same identity entity; A unique cross-domain unified identity identifier is assigned to the identity signal cluster within the system, and cluster-level representation fusion processing is performed on the final hidden layer representation vector of each signal vertex within the identity signal cluster to generate an identity attribute vector.

2. The method according to claim 1, characterized in that, The step of calculating the identity compatibility between the unified identity identifier and each of the resource nodes based on the identity attribute vector, the resource status data, and the service preference information includes: Based on resource status data and preset resource cost rules, resource status feature vectors are extracted for each resource node; wherein, the resource status feature vectors include computing resource features obtained from computing resource status data, link service features obtained from network link status data, and cost features related to resource call costs; On multiple preset matching dimensions, the basic matching degree between the identity attribute vector and the resource status feature vector of the corresponding resource node is calculated using the dimension compatibility measurement function and the importance weight of each matching dimension. Extract demand preference vectors for latency, bandwidth, and cost from service preference information, and perform polarity correction and normalization on the expected latency, available bandwidth, and call cost parameters of each resource node for the scheduled task based on the positive and negative correlation polarity of various environmental assessment indicators with service quality, to obtain positive utility indicators. The demand preference vector is weighted and combined with each of the positive utility indicators, and a balance control factor is introduced to fuse the weighted combination result with the basic matching degree to obtain a comprehensive fit degree, which serves as the identity fit degree between the unified identity identifier and each of the resource nodes.

3. The method according to claim 2, characterized in that, The generation of the candidate resource node list based on the identity adaptation degree includes: Obtain historical scheduling task records of all resource nodes in the entire network within a preset sliding time window, and extract the comprehensive resource consumption and task scheduling timestamp of the historical scheduling tasks allocated to each resource node; wherein, the comprehensive resource consumption is obtained by fusing the computational resource consumption and network link occupancy according to a preset resource consumption weight; Based on the time decay mechanism, the comprehensive resource consumption of each historical scheduling task is fused and aggregated with the task scheduling timestamp to calculate the dispersion penalty factor of each resource node; wherein, the dispersion penalty factor is used to characterize the degree of historical scheduling concentration and residual load heat of the node after time decay. The identity fit of each resource node is adjusted exponentially by using the aforementioned dispersion penalty factor to obtain the corrected identity fit after anti-hotspot congestion adjustment. Based on the corrected identity adaptation of each resource node, the resource nodes participating in the evaluation across the entire network are sorted in descending order, and a predetermined number of resource nodes at the top of the ranking are selected to form a candidate resource node list.

4. The method according to claim 3, characterized in that, The step of performing dynamic scheduling decisions based on the candidate resource node list, the overall network load status data, and the network link status data, to determine the target resource node matching the task to be scheduled from the candidate resource node list, and to determine the network path from the request initiation location of the task to be scheduled to the target resource node, includes: Based on network link status data, the entire network link delay matrix is ​​extracted, and candidate network paths are planned from the request initiation position of the task to be scheduled to each candidate node in the candidate resource node list. For each candidate node, the scheduling immediate benefit of scheduling the task to be scheduled to the corresponding candidate node is calculated based on the value assessment model; wherein, the scheduling immediate benefit is obtained by comprehensively evaluating the positive gain of correcting identity adaptation degree and the negative loss of normalized physical link cost on the candidate network path, and the negative loss of normalized physical link cost is determined based on the path loss adjustment factor. Analyze the network load status data to obtain the normalized load occupancy rate of each candidate node, and retrieve the historical scheduling records and corresponding execution feedback records of each candidate node to calculate the historical long-term service satisfaction score. Based on the normalized load occupancy rate and the historical long-term service satisfaction score, the expected future revenue of each candidate node is constructed, and a future revenue weighting factor is introduced to integrate the scheduling immediate revenue with the expected future revenue to obtain the total expected value of each candidate node. By introducing a soft maximum function of the temperature adjustment parameter, the total value expectation of each candidate node is mapped to the node selection probability, so that the candidate node with a higher total value expectation has a higher selection probability, and the degree of random exploration in the scheduling decision is controlled by the temperature adjustment parameter. Probabilistic selection is performed according to the node selection probability to determine the target resource node from the candidate resource node list, and the candidate network path corresponding to the target resource node is determined as the corresponding network path.

5. The method according to claim 1, characterized in that, Before calculating the identity compatibility between the unified identity identifier and each of the resource nodes, the method further includes making a forward-looking estimate of the future state of each of the resource nodes based on a time-series prediction model to dynamically adjust the available capacity, specifically including: Based on task request information and combined with service preference information, the business type to which the task to be scheduled belongs is identified, and a business prediction time window corresponding to the business type is determined. Based on historical load monitoring records and combined with network-wide load status data, normalized actual load occupancy rate data of each resource node in the network is extracted from the historical time series before the current sampling time. The normalized actual load occupancy rate data is input into the time series prediction model, and a forward-looking time series extrapolation is performed using a dynamic autoregressive weight mechanism that matches the business prediction time window to obtain the forward-looking predicted load occupancy rate of each resource node when it crosses the business prediction time window in the future. Obtain the current available computing capacity of each resource node, and use the corresponding forward-looking predicted load occupancy rate and preset capacity redundancy control factor to redundancy reduce the current available computing capacity to obtain the corrected available computing capacity. The corrected available computing capacity is used as the available capacity feature of the corresponding resource node to update the resource status data, so as to inject load fluctuation redundancy in advance during the generation of the candidate resource node list.

6. The method according to claim 4, characterized in that, After issuing the network configuration rules to the corresponding network devices for execution and issuing the resource scheduling instructions to the corresponding resource management platform for execution, the method further includes: Continuously acquire the actual runtime performance parameters of the tasks to be scheduled on the target resource nodes and network paths, and extract the actual response latency and actual available bandwidth; By combining the expected response latency, expected available bandwidth, and demand preference weights in the service preference information, the actual operating performance parameters are subjected to one-way truncation error accumulation only for the unmet performance indicators, and the system execution error cost is calculated. In response to the system execution error cost exceeding a preset tolerance threshold, a self-learning update closed loop is triggered for the importance weights of each matching dimension corresponding to the calculation of the basic matching degree, and the path loss adjustment factor corresponding to the calculation of the scheduling instant benefit. For identity-resource mapping pairs that cause the system execution error cost to exceed the standard, based on the gradient direction of the system execution error cost relative to the weights of each matching dimension, a multiplicative weight update mechanism is used to adaptively and iteratively update the importance weights of each matching dimension corresponding to the calculation of the basic matching degree, so as to adjust the importance weight allocation of each matching dimension while ensuring that the sum of the importance weights after the update remains constant. Extract the historical congestion frequency and the total number of historical scheduling events on the network path within a historical period, and dynamically increase the path loss adjustment factor corresponding to the calculation of real-time scheduling benefits based on the ratio of the historical congestion frequency to the total number of historical scheduling events. The updated importance weights are used to subsequently calculate the basic matching degree, and the updated path loss adjustment factor is used to subsequently calculate the scheduling instant benefit.

7. A network resource dynamic mapping and scheduling system based on identity resolution, used to implement the method as described in any one of claims 1-6; characterized in that, The system includes: The data sensing and acquisition unit is used to acquire task request information of the task to be scheduled, multi-source identity signals associated with the task to be scheduled, resource status data of multiple resource nodes, network load status data, and service preference information corresponding to the task to be scheduled; wherein, the task request information includes the request initiation location of the task to be scheduled, and the resource status data includes the computing resource status data of the resource nodes and the network link status data associated with the resource nodes. The identity fusion and parsing unit is used to fuse and parse the multi-source identity signals to construct a cross-domain unified identity identifier associated with the task to be scheduled, and to generate an identity attribute vector corresponding to the unified identity identifier. An identity adaptation evaluation unit is used to calculate the identity adaptation degree between the unified identity identifier and each of the resource nodes based on the identity attribute vector, the resource status data and the service preference information, and to generate a candidate resource node list based on the identity adaptation degree. The joint scheduling decision unit is used to perform dynamic scheduling decisions based on the candidate resource node list, the network-wide load status data, and the network link status data, so as to determine the target resource node that matches the task to be scheduled from the candidate resource node list, and to determine the network path from the request initiation location of the task to be scheduled to the target resource node. The strategy orchestration and distribution unit is used to generate network configuration rules corresponding to the network path and resource scheduling instructions corresponding to the target resource node, and to distribute the network configuration rules to the corresponding network devices for execution, and to distribute the resource scheduling instructions to the corresponding resource management platform for execution, so as to allocate the network path and the target resource node's computing resources to the scheduled task associated with the unified identity.

8. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor, characterized in that the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the method as described in any one of claims 1-6.

9. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-6.

Images