A multi-agent global perception and multi-dimensional mapping method for cross-domain network resources

Abstract

This application relates to the field of computing power network resource management technology, specifically disclosing a global perception and multi-dimensional mapping method for multi-entity cross-domain computing network resources. It constructs a resource profile by extracting feature vectors from the indicators reported by each node, initiates active probing based on the profile boundary, and integrates historical task execution deviation records to obtain performance evidence. It then performs deep verification of the credit consistency between theoretical benchmark values ​​and observed comparison values ​​to obtain a consistency metric. Furthermore, it constructs a trust mapping model adjusted by credit penalties by combining topology connection records. Finally, it performs correlation analysis on the corrected node computing power distribution and cross-domain link bottlenecks to generate a global perception view. This solves the problem of distorted perception views caused by the lack of cross-verification of resource reporting in multi-entity environments. It also introduces structured deviation detection based on hardware coupling constraints between dimensions, overcoming the limitation of insensitivity to false reports of synchronization ratios in independent dimension-by-dimensional measurements, and achieving unified, reliable perception and accurate mapping of cross-domain heterogeneous resources.

Description

Technical Field

[0001] This application relates to the field of computing network resource management technology, and more specifically, to a method for global perception and multi-dimensional mapping of multi-entity cross-domain computing network resources. Background Technology

[0002] As computing networks rapidly evolve towards multi-entity, cross-regional, and heterogeneous integration, cross-domain computing networks composed of different cloud service providers, edge node providers, and data center owners have a wide variety of computing, storage, and network resources that lack a unified description standard. The dispersed resource status is difficult to fully perceive, and the intrinsic relationships between resources of different entities are even more difficult to effectively characterize. As a result, the upper-layer scheduling system cannot obtain an accurate global resource view, resource utilization remains low, and the reliability of task orchestration is difficult to guarantee.

[0003] In a multi-entity competitive operating environment, existing resource awareness solutions are generally based on the assumption of honest reporting, passively receiving resource status data generated by the entity's software layer. They lack effective means of cross-verification from underlying behavioral performance or historical task feedback, allowing entities to inflate metrics such as computing power, bandwidth, and storage throughput at extremely low cost to obtain excessive task allocation benefits, resulting in a severe deviation between the perceived view and physical reality. Furthermore, even with the introduction of a dimension-by-dimensional deviation detection mechanism to identify and penalize significant inflated values ​​in a single dimension, these methods still assume that each resource dimension is independent, measuring only the numerical differences of each dimension item by item. They fail to incorporate the objective hardware coupling constraints between dimensions such as processors, memory bandwidth, and network throughput into the deviation judgment framework. This leads to a lack of awareness of structured false reporting behavior that simultaneously inflates multiple metrics in similar proportions. Such nodes may appear complete in static profiling, but in large-scale model training, low-latency inference links, or data-intensive orchestration processes, they are highly susceptible to cascading performance collapses due to insufficient real coupling capabilities, ultimately causing scheduling systems to misjudge available resources, amplify cross-domain link congestion, and even breach service level agreement (SLA) regulations.

[0004] Therefore, an optimized scheme for global perception and multi-dimensional mapping of multi-agent cross-domain computing network resources is desired. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a method for global awareness and multi-dimensional mapping of multi-entity cross-domain computing network resources, comprising: Step 1: Extract the computational load and link connection metrics contained in the host reporting specifications and network interface status reported by each node in the multi-entity cross-domain computing network by feature vectorization to obtain a resource profile. Step 2: Based on the node mapping boundary determined by the resource profile, initiate active network probing for each node to obtain active probing data. At the same time, retrieve the execution deviation records of the corresponding nodes from the historical task logs, and then perform noise reduction and integration on the active probing data and execution deviation records to obtain performance evidence. Step 3: Perform a deep verification of resource credit consistency between the theoretical benchmark values ​​declared in the resource profile and the observed comparison values ​​recorded in the performance evidence to obtain a consistency measure. Step 4: Based on resource profiling, topology connection records, and consistency metrics, construct a trust mapping model; Step 5: Based on the trust mapping model, perform correlation analysis on the node computing power distribution after credit correction and the cross-domain link bottleneck to obtain a global perception view.

[0006] Compared with existing technologies, this application proposes a global perception and multi-dimensional mapping method for multi-agent cross-domain computing network resources. It constructs a resource profile by extracting feature vectors from the computational load and link connectivity metrics reported by each node. Based on the node boundaries determined by the profile, it initiates active network probing and integrates performance evidence by denoising execution deviation records from historical task logs. It then performs deep verification of resource credit consistency between the theoretical benchmark values ​​declared in the profile and the observed values ​​in the performance evidence to obtain a consistency metric. Finally, it constructs a trust mapping model adjusted by credit penalties based on the resource profile, topology connectivity records, and consistency metric. Finally, it performs correlation analysis between the corrected node computing power distribution and cross-domain link bottlenecks to generate a global perception view. This method solves the problem of severe deviation between the perceived view and physical reality caused by the lack of cross-verification of resource reporting in multi-agent environments. Furthermore, by introducing structured deviation detection with hardware coupling constraints between dimensions, it overcomes the limitation of insensitivity of independent dimension-by-dimensional metrics to structured false reporting behavior with proportional synchronous false reporting, achieving unified and reliable perception and accurate mapping of cross-domain heterogeneous resources. Attached Figure Description

[0007] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0008] Figure 1 This is a flowchart of a method for global perception and multi-dimensional mapping of multi-subject cross-domain computing network resources according to an embodiment of this application; Figure 2 This is a data flow diagram illustrating a global perception and multi-dimensional mapping method for multi-entity cross-domain computing network resources according to an embodiment of this application; Figure 3 This is a flowchart illustrating a method for global perception and multi-dimensional mapping of multi-subject cross-domain computing network resources according to an embodiment of this application. The flowchart describes how to initiate active network probing for each node to obtain active probing data, retrieve execution deviation records of the corresponding node from historical task logs, and then denoise and integrate the active probing data and execution deviation records to obtain performance evidence. Figure 4 This is a flowchart illustrating a method for global perception and multi-dimensional mapping of multi-subject cross-domain computing network resources according to an embodiment of this application. It involves performing a deep verification of resource credit consistency between the theoretical benchmark values ​​declared in the resource profile and the observed comparison values ​​recorded in the performance evidence to obtain a consistency measure. Figure 5 This is a flowchart illustrating the process of calculating the residual accumulation and summarization of the negative deviation between the theoretical baseline value and the observed comparison value of each dimension in the alignment matrix to obtain the resource deviation coefficient, according to a global perception and multi-dimensional mapping method for multi-subject cross-domain computing network resources in an embodiment of this application. Figure 6 This is a flowchart illustrating the construction of a trust mapping model based on resource profiling, topology connection records, and consistency measurement, according to an embodiment of this application, for a global perception and multi-dimensional mapping method for multi-subject cross-domain computing network resources. Detailed Implementation

[0009] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0010] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0011] While this application makes various references to certain modules of the systems according to embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The modules described are merely illustrative, and different aspects of the systems and methods may use different modules.

[0012] Flowcharts are used in this application to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed in exact order. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0013] Current resource awareness schemes in multi-entity cross-domain computing networks are generally based on the assumption of honest reporting, passively receiving resource status data generated by the software layers of each entity without cross-verification methods. This allows entities to falsely report key indicators such as computing power and bandwidth at extremely low cost, causing a significant deviation between the perceived view and physical reality. Even with the introduction of a dimension-by-dimensional deviation detection mechanism, because each resource dimension is treated as independent, it is still impossible to effectively identify structured false reporting behavior that simultaneously falsely reports multiple indicators in similar proportions. Therefore, the technical solution of this application proposes a global awareness and multi-dimensional mapping method for resources in multi-entity cross-domain computing networks. This method first extracts feature vectors from the computing load and link connectivity metrics included in the host reporting specifications and network interface status reported by each node to construct a resource profile. Then, based on the node mapping boundaries determined by the resource profile, it initiates active network probing. Simultaneously, it retrieves the execution deviation records of the corresponding nodes from historical task logs, obtains performance evidence after noise reduction and integration, and performs a deep verification of resource credit consistency between the theoretical benchmark values ​​declared in the resource profile and the observed comparison values ​​recorded in the performance evidence to obtain a consistency measure. This verification process not only accumulates residuals for single-dimensional false alarms but also introduces inter-dimensional hardware coupling constraints for structured deviation detection to identify synchronization ratio false alarms. Then, based on the resource profile, topology connection records, and consistency measure, it constructs a trust mapping model adjusted by credit penalties, injects credit scores into the computing power attributes of graph vertices and the bandwidth attributes of graph edges for dynamic reduction, and finally performs correlation analysis on the corrected node computing power distribution and cross-domain link bottlenecks and renders and outputs a global perception view, thereby achieving pre-suppression of false resource reports and unified trust perception of cross-domain heterogeneous resources.

[0014] Figure 1 This is a flowchart of a method for global perception and multi-dimensional mapping of multi-subject cross-domain computing network resources according to an embodiment of this application. Figure 2 This is a data flow diagram illustrating a method for global perception and multi-dimensional mapping of multi-entity cross-domain computing network resources according to an embodiment of this application. Figure 1 and Figure 2As shown, a method for global perception and multi-dimensional mapping of multi-entity cross-domain computing network resources according to an embodiment of this application includes: S1, extracting the computational load indicators and link connection indicators contained in the host reporting specifications and network interface status reported by each node in the multi-entity cross-domain computing network by feature vectorization to obtain a resource profile; S2, based on the node mapping boundary determined by the resource profile, initiating active network probing for each node to obtain active probing data, and simultaneously retrieving the execution deviation records of the corresponding nodes from the historical task logs, and then integrating the active probing data and execution deviation records with noise reduction to obtain performance evidence; S3, performing deep verification of resource credit consistency between the theoretical benchmark values ​​declared in the resource profile and the observed comparison values ​​recorded in the performance evidence to obtain a consistency measure; S4, constructing a trust mapping model based on the resource profile, topology connection records, and consistency measure; S5, based on the trust mapping model, performing correlation analysis on the node computing power distribution and cross-domain link bottlenecks after credit correction to obtain a global perception view.

[0015] Specifically, in step S1, the computational load and link connectivity metrics included in the host reporting specifications and network interface status reported by each node in the multi-entity cross-domain computing network are extracted using feature vectorization to obtain a resource profile. It should be noted that due to the heterogeneous differences in hardware architecture and network configuration among nodes in the multi-entity cross-domain computing network, the original resource data reported by different entities are inconsistent in format and scale, and cannot be directly used for cross-domain comparison and correlation calculations. Based on this, the technical solution of this application first extracts the computational load and link connectivity metrics included in the host reporting specifications and network interface status reported by each node using feature vectorization to obtain a resource profile. Through the above processing, heterogeneous resource data can be transformed into a structured profile representation with unified scale, providing standardized input for subsequent credit verification and topology mapping.

[0016] More specifically, in a concrete example of this application, the raw data reported by each node is first subjected to structured splitting and cleaning. Each node reports its host reporting specifications and network interface status to the computing network control plane according to a predefined data interface protocol. The host reporting specifications encapsulate the node's computing power physical scalars in a serialized format, specifically including the number of CPU cores, GPU memory size, and memory capacity. The network interface status encapsulates link configuration parameters such as the network card bandwidth limit and maximum transmission unit. The data structure deserialization parsing engine performs field-level parsing and cleaning on the above two types of raw messages, removing redundant vendor identifiers and non-critical descriptive fields. The computing power-related physical scalars are aggregated into a computing load index set, and the link-related configuration parameters are aggregated into a link connection index set.

[0017] After splitting the indicator set, feature vectorization calculations were performed on the computational load indicator set and the link connectivity indicator set respectively. For the computational load indicator set, since the number of CPU cores is an integer from the units to the hundreds place while GPU memory is a floating-point number in the GB range, the dimensions of the various indicators differ significantly. Therefore, logarithmic scaling and max-min normalization need to be performed on each indicator to eliminate the influence of dimensions, resulting in a dimensionless comprehensive computing power index and generating a computational feature vector. Its calculation expression is as follows: in, This is the normalized comprehensive computing power index. The total number of resource dimensions involved in the calculation. For the first The original collected values ​​of each dimension, and The first The upper and lower limits of the preset physical configuration values ​​for each dimension. For the first The computational power contribution weight coefficients for each dimension are calculated, and the sum of the weights for each dimension is 1. For the link connectivity metric set, considering that the TCP and IP protocol stack headers occupy the effective payload space in network transmission, the effective throughput is calculated and a connection feature vector is generated based on the ratio between the network card bandwidth limit and the maximum transmission unit, after deducting the fixed protocol header overhead. The calculation expression is as follows: in, The effective throughput after deducting protocol overhead. This refers to the maximum physical bandwidth of the network card. The maximum transmission unit size is bytes. and These represent the fixed header byte lengths for the TCP and IP protocol layers, respectively. After generating both the computational feature vector and the connection feature vector, they undergo spatiotemporal alignment and structured encapsulation within a tolerance threshold. Due to slight discrepancies in the acquisition times of the computational load metric and the link connection metric, it is necessary to read the acquisition timestamp and node hardware identifier carried by each vector, and perform timestamp alignment and entity mapping alignment based on node identifiers within a preset tolerance threshold, such as 5 milliseconds. After alignment, the computational feature vector and the connection feature vector are concatenated into a high-dimensional tensor with added check bits, and encapsulated to output a resource profile with a unified representation format.

[0018] Specifically, in step S2, based on the node mapping boundaries determined by the resource profile, active network probing is initiated for each node to obtain active probing data. Simultaneously, execution deviation records for the corresponding nodes are retrieved from historical task logs. The active probing data and execution deviation records are then denoised and integrated to obtain performance evidence. It should be noted that, given that the resource profile only reflects the theoretical capabilities autonomously reported by each node's software layer and has not been cross-validated by actual physical behavior, and that nodes in a multi-entity competitive environment may have the motivation to overstate computing power and bandwidth to obtain excessive task distribution benefits, the true deliverable capability of a node cannot be determined solely based on declared data. Therefore, the technical solution of this application further initiates active network probing for each node based on the node mapping boundaries determined by the resource profile to obtain active probing data. Simultaneously, execution deviation records for the corresponding nodes are retrieved from historical task logs. The active probing data and execution deviation records are then denoised and integrated to obtain performance evidence. Through the above processing, real-time behavior observation data of nodes can be obtained from both real-time probing and historical feedback dimensions, providing quantifiable comparison basis for subsequent credit consistency verification.

[0019] Figure 3 This flowchart illustrates a method for global perception and multi-dimensional mapping of multi-entity cross-domain computing network resources according to an embodiment of this application. It describes how, based on resource profiling, node mapping boundaries are determined, proactive network probing is initiated for each node to obtain proactive probing data. Simultaneously, execution deviation records for the corresponding nodes are retrieved from historical task logs. Finally, the proactive probing data and execution deviation records are denoised and integrated to obtain performance evidence. Figure 3 As shown, step S2 includes: S21, based on the node communication address and service port declared in the resource profile, initiating asynchronous multipath active probing for each target node to obtain active probing data; S22, using the node identifier in the active probing data as an index key, calculating the time offset rate and time sequence correlation between the expected execution time and the actual completion time of the corresponding node in the historical task log to obtain the execution deviation record; S23, performing noise reduction and integration on the active probing data and the execution deviation record to obtain performance evidence.

[0020] In step S21, asynchronous multipath active probing is initiated against each target node based on the node communication addresses and service ports declared in the resource profile to obtain active probing data. It should be noted that since the computing power and bandwidth data recorded in the resource profile are all reported by the node's own software layer and have not undergone independent physical layer behavior verification, the actual response performance of the nodes needs to be obtained through actual network interaction. Based on this, the technical solution of this application further initiates asynchronous multipath active probing against each target node based on the node communication addresses and service ports declared in the resource profile to obtain active probing data. Through the above processing, the real-time physical response characteristics of each node can be obtained from the network transmission layer, providing independent observation samples for subsequent deviation comparison with theoretical benchmark values.

[0021] More specifically, in a concrete example of this application, the communication addresses and service ports declared by each target node are first extracted from the resource profile, serving as the addressing entry point for the probe targets. After determining the probe targets, the sending frequency and number of concurrent paths of probe packets are dynamically configured based on the network interface card bandwidth limit declared in the resource profile to avoid overloading the tested nodes with the probe traffic itself. Subsequently, probe requests using both ICMP and HTTP protocols are simultaneously initiated to multiple target nodes in an asynchronous, non-blocking manner. ICMP probes are used to collect network layer round-trip latency and packet loss rate, while HTTP probes are used to collect application layer service response time. After each probe request reaches the target node, the round-trip latency sample value and packet loss status mark of each probe path are recorded. The mean and variance of multiple round-trip latency samples for the same node are calculated. The mean reflects the average network distance of the node, while the variance reflects the jitter of the link. After completing all probe rounds, the average round-trip latency, packet loss rate, and latency variance of each node are weighted and aggregated according to a preset sensitivity coefficient, and the active probe data is encapsulated and output.

[0022] In step S22, the node identifiers in the active probe data are used as index keys to calculate the time offset rate and perform time-series correlation on the expected execution time and actual completion time of the corresponding node in the historical task log to obtain the execution deviation record. It should be noted that since the active probe data only reflects the instantaneous network response characteristics of the node at the probe moment, it cannot reflect the long-term performance stability of the node when carrying real computing tasks. Therefore, it is necessary to combine the historical task execution records to evaluate the node's continuous delivery capability. Based on this, the technical solution of this application further utilizes the node identifiers in the active probe data as index keys to calculate the time offset rate and perform time-series correlation on the expected execution time and actual completion time of the corresponding node in the historical task log to obtain the execution deviation record. Through the above processing, the real-time probe dimension and the historical execution dimension can be correlated at the node level, providing a performance deviation basis with temporal depth for subsequent noise reduction and integration.

[0023] More specifically, in a concrete example of this application, firstly, unique identifiers for each node are extracted from the active probing data. These identifiers are then used as index keys to perform exact matching searches on historical task logs, filtering out all job records carried by that node within the past observation period. For each job record, the expected execution time allocated during the scheduling phase and the actual completion time written back after the task is actually completed are extracted. The absolute deviation between these two values ​​is calculated for each record and divided by the expected execution time to obtain the time offset rate for a single job. After calculating each job record, the arithmetic mean of all time offset rates for that node within the observation period is taken to obtain the average time offset rate, which reflects the long-term execution reliability of that node. The calculation is expressed as follows: in, This represents the average time offset of the node. This represents the total number of historical tasks executed by this node within the observation period. For the first The actual completion time of each historical task For the first The expected execution time allocated to each historical task during the scheduling phase. After obtaining the average time offset rate, it is correlated with the detection score of the same node in the active detection data according to the timestamp, so that the real-time network performance and historical calculation deviation form a node-level correspondence and binding relationship, and the execution deviation record is encapsulated and output.

[0024] In step S23, the active detection data and execution deviation records are denoised and integrated to obtain performance evidence. It should be noted that since the active detection data and execution deviation records originate from two different acquisition channels—real-time network detection and historical task backtracking—they differ in time scale and data distribution, and each may contain noise interference introduced by instantaneous network jitter or occasional task anomalies. Direct splicing cannot form a reliable comprehensive performance criterion. Therefore, the technical solution of this application further denoises and integrates the active detection data and execution deviation records to obtain performance evidence. Through the above processing, information from both real-time observation and historical feedback can be fused into a single, credible performance credential, providing a stable comparison input for subsequent consistency verification with the theoretical benchmark value of the resource profile.

[0025] More specifically, in a specific example of this application, outlier removal is first performed on both the active probe data and the execution deviation record. For extreme latency sampling in the active probe data caused by instantaneous fluctuations in cross-domain routing, outlier samples exceeding the normal fluctuation range are identified and removed using the median absolute deviation method. For abnormal offset entries in the execution deviation record caused by temporary node maintenance or mid-task cancellation, they are filtered and removed based on the task status marker. After independent noise reduction for each channel, the cleaned probe comprehensive score in the active probe data and the average time offset rate in the execution deviation record are weighted and fused according to a preset historical evidence credibility weighting coefficient. The calculation is expressed as follows: in, The overall performance index after fusion. The overall score for active detection after noise reduction processing. The average time offset after filtering. To convert the offset rate into a reverse mapping value of the performance achievement rate, This is a weighting coefficient for the credibility of historical evidence, used to control the proportion of historical feedback in the fusion result. When this weighting coefficient is high, the fusion result focuses more on the long-term execution stability of nodes; when the value is low, it focuses more on real-time network response performance. The fused comprehensive performance index, along with the corresponding node identifier and timestamp, is encapsulated and output as performance evidence.

[0026] Specifically, in step S3, a deep verification of resource credit consistency is performed between the theoretical benchmark values ​​declared in the resource profile and the observed comparison values ​​recorded in the performance evidence to obtain a consistency measure. It should be noted that, given that the resource profile reflects the theoretical capabilities declared by the node itself, while the performance evidence reflects the actual behavioral performance after the fusion of real-time detection and historical feedback, the degree of deviation between the two directly determines the credibility level of the node's reported data. Furthermore, in a multi-entity competitive environment, nodes may employ various false reporting strategies, such as single-dimensional false reporting or proportional synchronous false reporting. Therefore, a deep, structured comparison of theoretical and observed values ​​is necessary to effectively identify different types of distorted behavior. Based on this, the technical solution of this application further performs a deep verification of resource credit consistency between the theoretical benchmark values ​​declared in the resource profile and the observed comparison values ​​recorded in the performance evidence to obtain a consistency measure. Through the above processing, the degree of deviation between the node's reported data and the actual observed data can be quantified into a credit score within a standard probability space, providing a node-level credibility weight basis for the subsequent construction of the trust mapping model.

[0027] Figure 4This is a flowchart illustrating a method for global perception and multi-dimensional mapping of multi-agent cross-domain computing network resources according to an embodiment of this application. It involves performing a deep verification of resource credit consistency between the theoretical benchmark values ​​declared in the resource profile and the observed comparison values ​​recorded in the performance evidence to obtain a consistency measure. (See flowchart for example.) Figure 4 As shown, step S3 includes: S31, performing dimensionality reduction and alignment on the theoretical computing power and bandwidth benchmark values ​​declared in the resource profile and the observed performance indices recorded in the performance evidence to obtain an alignment comparison matrix; S32, performing residual accumulation and summarization calculation on the negative deviations between the theoretical benchmark values ​​and observed comparison values ​​of each dimension in the alignment comparison matrix to obtain the resource deviation coefficient; S33, performing quantization mapping on the resource deviation coefficient in the standard probability space to obtain a consistency measure.

[0028] In step S31, the theoretical computing power and bandwidth benchmark values ​​declared in the resource profile and the observed performance indices recorded in the performance evidence are dimensionality-reduced and aligned to obtain an alignment comparison matrix. It should be noted that since the theoretical benchmark values ​​in the resource profile represent the on-paper capabilities of independent dimensions such as computing power and bandwidth in the form of multi-dimensional feature vectors, while the observed performance indices in the performance evidence are comprehensive scalars after multi-channel fusion, the two are not equal in terms of the number of dimensions and data representation space, making direct item-by-item deviation comparison impossible. Based on this, the technical solution of this application further dimensionality-reduced and aligned the theoretical computing power and bandwidth benchmark values ​​declared in the resource profile and the observed performance indices recorded in the performance evidence to obtain an alignment comparison matrix. Through the above processing, the theoretical declared values ​​and actual observed values ​​can be uniformly mapped to the same dimension space and dimensional structure, providing a consistent comparison input for subsequent residual calculations in each dimension.

[0029] More specifically, in a concrete example of this application, the theoretical baseline values ​​declared by each node in the computational and network dimensions are first extracted from the resource profile according to the node identifier. The computational dimension corresponds to the normalized comprehensive computing power index, and the network dimension corresponds to the effective throughput after deducting protocol overhead. Subsequently, the comprehensive performance index corresponding to the same node is extracted from the performance evidence. This index is a weighted fusion result of real-time detection score and historical performance achievement rate, and its value has been normalized to the range of 0 to 1. Since the theoretical baseline value contains two independent dimensions, computation and network, while the observed performance index is a single fused scalar, the observed performance index needs to be decomposed and restored according to the same dimensional division method as the theoretical baseline value. Specifically, the comprehensive performance index is inversely weighted using the weight coefficients of the fusion stage, decomposing it into observed component values ​​corresponding to the computational and network dimensions respectively. After completing the dimension alignment, the theoretical baseline value and the corresponding observed component value of each node in each resource dimension are arranged by row and each resource dimension is arranged by column. This constructs an alignment comparison matrix with the node as the row index and the resource dimension as the column index. Each position in the matrix stores a pair of theoretical and observed data for direct reading in subsequent residual calculations.

[0030] In step S32, residual accumulation and summarization are performed on the negative deviations between the theoretical baseline values ​​and observed comparison values ​​of each dimension in the alignment comparison matrix to obtain the resource deviation coefficient. It should be noted that, given that the alignment comparison matrix simultaneously contains the theoretical declared values ​​and observed measured values ​​of each node across multiple resource dimensions, and that the typical manifestation of a node falsely reporting resources in a multi-entity competitive environment is a negative deviation where the theoretical value is higher than the observed value, it is necessary to perform directional residual extraction and cross-dimensional summarization to aggregate the false reporting signals scattered across various dimensions into a unified deviation quantification index. Based on this, the technical solution of this application further performs residual accumulation and summarization on the negative deviations between the theoretical baseline values ​​and observed comparison values ​​of each dimension in the alignment comparison matrix to obtain the resource deviation coefficient. Through the above processing, the scattered false reporting deviations across various dimensions can be condensed into a node-level comprehensive deviation scalar, providing continuous and comparable numerical input for subsequent nonlinear credit decay mapping.

[0031] Figure 5 This document describes a flowchart illustrating the process of calculating the resource deviation coefficient by accumulating and summarizing the negative deviations between the theoretical baseline values ​​and observed comparison values ​​of each dimension in the alignment matrix of a global perception and multi-dimensional mapping method for multi-subject cross-domain computing network resources, according to an embodiment of this application. Figure 5As shown, step S32 includes: S321, extracting inter-dimensional coupling prior information from the alignment comparison matrix to obtain a coupling prior matrix and a dimensional residual vector; S322, measuring the coupling anomaly deviation from the coupling prior matrix and the dimensional residual vector to obtain a coupling anomaly score; S323, adaptively fusing the dimensional residual vector and the coupling anomaly score with dual-channel deviation coefficients to obtain a resource deviation coefficient.

[0032] In step S321, inter-dimensional coupling prior information is extracted from the alignment comparison matrix to obtain a coupling prior matrix and a dimensional residual vector. It should be noted that, given that the original dimensional deviation detection assumes each resource dimension is independent, it only measures the numerical difference of each dimension item by item, failing to incorporate the objectively existing hardware coupling constraints between dimensions such as computing, storage, and network into the deviation judgment framework. This results in a lack of ability to identify structured false alarms that simultaneously falsely report multiple indicators in similar proportions. Based on this, the technical solution of this application further extracts inter-dimensional coupling prior information from the alignment comparison matrix to obtain a coupling prior matrix and a dimensional residual vector. Through the above processing, while retaining sensitivity to single-dimensional false alarms, the hardware coupling rules between multi-dimensional resources can be explicitly encoded into the subsequent computation link, providing structured prior constraints and directional residual input for measuring coupling anomaly deviation.

[0033] More specifically, in a concrete example of this application, the theoretical baseline values ​​for all participating nodes in each resource dimension are first extracted from the alignment comparison matrix. The coupling prior matrix is ​​constructed using the correlation coefficients between the theoretical baseline values. For any two resource dimensions, using the theoretical baseline values ​​of all nodes in these two dimensions as statistical samples, the Pearson correlation coefficient is calculated to quantify the statistical coupling strength between the two dimensions under the condition of honest reporting. Its calculation is expressed as follows: in, The first in the coupling prior matrix Wei and Di Coupling coefficients between dimensions and Indexed for two resource dimensions. For node indexing, This represents the total number of nodes participating in the statistics. For the first The node at the th The theoretical benchmark value in dimension For the first The mean of the theoretical benchmark value, For the first The node at the th The theoretical benchmark value in dimension For the first The mean of the theoretical baseline values. This matrix describes not a single-point deviation, but rather the achievable proportional structure between multi-dimensional resources under honest reporting conditions. It is suitable for characterizing the physical rationality of different subject resource templates in hybrid scenarios such as multi-cloud, edge, and data center. In the same step, the dimensional residual vector is rectified dimension-by-dimensionally based on the theoretical baseline value and observed value of the current node to be verified, specifically retaining the deviation that is higher on paper than in reality, so as to focus the verification focus on the risk of false reporting rather than accidental extraordinary performance. Its calculation expression is: in, For the dimension of the residual vector at the th dimension The residual components on the dimension, For the current node at the th The theoretical benchmark value in dimension For the current node at the th Observations on the dimension, This is used to filter out cases where observed values ​​are higher than theoretical values. The resulting dimensional residual vector can both support the original bias detection logic and provide a unified input for the next step of constructing the coupled anomaly deviation.

[0034] In practical multi-entity cross-domain computing networks, computing power, video memory, memory bandwidth, network bandwidth, and storage throughput are not isolated metrics that can be freely combined. Instead, they are interconnected parameters constrained by processor sockets, memory channels, bus channel count, device architecture generations, and board interconnection methods. Specifically, for an edge computing node equipped with a 64-core high-performance processor, its memory bandwidth configuration must match the number of memory channels of the processor. Higher core count CPUs are accompanied by higher memory bandwidth configurations, and high-performance graphics processors correspond to higher video memory bandwidth. Furthermore, the sustained throughput of 10 Gigabit or higher speed network cards is also limited by the combined effects of storage I / O capabilities and bus bottlenecks. If a node is actually equipped with only a 32-core processor, 256GB of memory, and a 25Gbps network card, but reports all indicators in a similar proportion to 64 cores, 512GB, and 100Gbps, then the dimensional deviation detection may fail to trigger sufficient penalties due to the uniformity of deviations across dimensions. However, the coupling prior matrix can capture that the dimension ratio combination declared by the node has deviated from the statistical coupling relationship between dimensions under normal hardware architecture, and the dimension residual vector simultaneously records the directional deviation magnitude of each dimension. Together, they provide a structured basis for subsequent calculation of coupling anomaly deviation.

[0035] In step S322, the coupling prior matrix and dimensional residual vector are subjected to a coupling anomaly deviation measurement to obtain a coupling anomaly score. It should be noted that, given that the dimensional residual vector only records the single-dimensional deviation magnitude between the theoretical and observed values ​​in each dimension, and has not yet structurally determined whether the directional combinations of deviations violate hardware coupling rules, in the adversarial reporting environment of a multi-agent computing network, many false reports do not simply exaggerate a single indicator, but attempt to maintain a seemingly reasonable high-specification configuration. Only by examining these deviations in the coupling space can their structural irrationality be identified. Based on this, the technical solution of this application further measures the coupling anomaly deviation of the coupling prior matrix and dimensional residual vector to obtain a coupling anomaly score. Through the above processing, the deviation judgment can be extended from the level of single-dimensional numerical comparison to the level of multi-dimensional structural rationality, enabling the effective identification of nodes where single-dimensional deviations are not prominent but the deviation directions violate coupling constraints.

[0036] More specifically, in a concrete example of this application, the coupling prior matrix is ​​transformed into a covariance metric space, and then the Mahalanobis distance is calculated on the dimensional residual vectors to obtain the coupling anomaly score. First, a covariance transformation is performed on the coupling prior matrix, converting it from a correlation coefficient matrix into a covariance inverse matrix. This inverse matrix serves to measure and encode the strength of coupling constraints between resource dimensions; the more tightly coupled a dimension is, the higher its corresponding weight in the inverse matrix, and the greater the anomaly contribution of the deviation in that direction. Then, the dimensional residual vectors and their transposes are multiplied with the covariance inverse matrix, and the square root is taken. Instead of treating the residuals of each dimension as mutually independent Cartesian coordinates, the physical coupling relationship between resource dimensions is incorporated into the distance definition itself, resulting in the calculation of the coupling anomaly score, expressed as follows: in, For coupling anomaly score, Let the residual vector be composed of the residual components of each dimension. This is the transpose of the vector. The covariance inverse matrix, derived from the coupling prior matrix, is used to characterize the strength of coupling constraints between resources in each dimension. If the deviation of a node falls within the physically acceptable range of cooperative fluctuations, the distance value is relatively limited; if the deviation direction breaks through the normal coupling boundary, even if the absolute value of the deviation in each dimension is not prominent, a higher anomaly score will be obtained. This effectively addresses the problem of false alarms in synchronization ratios.

[0037] Continuing with the aforementioned edge computing node scenario, suppose this node falsely reports a 32-core processor as 64 cores, 256GB of memory as 512GB, and a 25Gbps network card as 100Gbps. The false reporting ratios for each dimension are close to 2x, and the residual components obtained from the dimension-by-dimensional deviation detection are relatively uniform, failing to trigger a sufficiently strong penalty. However, in the coupling metric space, there is a definite proportional constraint between the memory bandwidth configuration corresponding to the 64-core processor and the number of channels for the 512GB of memory. The continuous throughput of the 100Gbps network card requires matching storage input / output capabilities and bus bandwidth far exceeding the actual carrying capacity limit of the hardware platform where the 25Gbps network card resides. These dimensional combinations are low-probability configurations under the statistical regularities of normal hardware architecture encoded by the coupling prior matrix. After Mahalanobis distance calculation, the node's coupling anomaly score will deviate from the distribution range of normal nodes, thus having a stronger corrective effect on the resource deviation coefficient in the subsequent fusion stage.

[0038] In step S323, the dimensional residual vector and the coupling anomaly score are adaptively fused using dual-channel deviation coefficients to obtain the resource deviation coefficient. It should be noted that, given that independent dimensional deviation detection and coupling anomaly deviation measurement target two different distortion modes—one-dimensional false alarms and structured false alarms—simple superposition with fixed weights would introduce new rigidity problems when the two types of deviation intensities are asymmetrical, failing to automatically adjust the contribution ratio of the two detection channels based on the actual false alarm characteristics of the node. Therefore, the technical solution of this application further performs adaptive fusion of the dimensional residual vector and the coupling anomaly score using dual-channel deviation coefficients to obtain the resource deviation coefficient. Through the above processing, one-dimensional distortion identification and structured anomaly identification can be uniformly incorporated into the same credit verification caliber, allowing the resource deviation coefficient to remain stable under normal fluctuations while adaptively increasing towards the higher penalty range when anomalies occur.

[0039] More specifically, in a concrete example of this application, the dimensional residual vector and coupling anomaly score are incorporated into a dual-channel fusion framework. First, the independent dimensional deviation is calculated based on the dimensional residual vector, maintaining the original weighted Euclidean summation approach. Then, the residual components of each dimension are weighted and summed squared according to preset deviation weights, and the square root is taken. The calculation is expressed as follows: in, For independent dimension deviation, The total number of resource dimensions involved in the calculation. For the first Dimensional bias weights, For the first The residual component of each resource dimension measures the deviation between its reported value and the observed value, making it suitable for identifying single-point distortions such as high computing power, high bandwidth, and high storage throughput. Subsequently, adaptive fusion weights are constructed. To avoid the rigidity problems introduced by fixed-weight fusion, the fusion weights are jointly determined by the coupling anomaly score and the deviation of independent dimensions. When the coupling anomaly is stronger, the contribution ratio of structural anomaly channels is automatically increased, making the model more adaptively sensitive to hidden false alarms. The calculation of the adaptive fusion weights is expressed as follows: in, For adaptive fusion weights, For coupling anomaly score, For independent dimension deviation, To prevent extremely small positive numbers with a denominator of zero, after calculating the independent dimension deviation and adaptive fusion weights, the outputs of the two detection channels are weighted and synthesized. The calculation is expressed as follows: in, This is the resource deviation coefficient in the final output. For adaptive fusion weights, For independent dimension deviation, This is the score for coupling anomaly. Because... As the intensity of the two types of deviations changes dynamically, the resource deviation coefficient remains stable under normal fluctuations, but rapidly increases to the high penalty range when there are structural anomalies, thus forming a closer linkage with the subsequent nonlinear credit decay mapping.

[0040] Continuing with the aforementioned edge computing node scenario, for nodes that falsely report 32 cores as 64 cores only in the computing power dimension while reporting the truth for other dimensions, their independent dimension deviation is at a high level due to the large residual component in the computing power dimension. However, their coupling anomaly score is at a low level because the combination of other dimensions still conforms to normal hardware coupling rules. In this case, adaptive fusion weights are appropriate. The resource deviation coefficient is relatively low, primarily driven by the deviation of independent dimensions. For nodes that synchronously inflate computing power, memory, and bandwidth in roughly equal proportions, the deviation of independent dimensions is at a moderate level due to the uniform distribution of residuals across dimensions. However, the coupling anomaly score is at a high level because the dimension combination deviates from the statistical coupling boundary of the normal hardware architecture. The contribution of coupled abnormal channels in the resource deviation coefficient is automatically increased, so nodes with obvious single-dimensional fraud will not be missed, and nodes with structured fraud will not escape punishment due to uniform deviations in all dimensions. Both types of deviations are uniformly included in the same credit verification caliber.

[0041] In step S33, the resource deviation coefficient is quantized and mapped within a standard probability space to obtain a consistency measure. It should be noted that since the resource deviation coefficient is an unbounded continuous value, its range is not constrained by an upper limit as it varies with the magnitude of the deviation and the degree of structural anomaly. Therefore, it cannot be directly used as the penalty weight for subsequent graph vertex and edge attributes and needs to be mapped to a bounded interval with clear physical meaning. Based on this, the technical solution of this application further quantizes and maps the resource deviation coefficient within a standard probability space to obtain a consistency measure. Through the above processing, the degree of deviation can be converted into a credit score between 0 and 1, providing directly multiplicative weighting factors for computational power reduction and bandwidth pressure reduction in the subsequent trust mapping model.

[0042] More specifically, in a concrete example of this application, the resource deviation coefficient is input into a preset nonlinear credit decay function to perform quantization mapping. This decay function uses a dynamically shifted Sigmoid variant, whose output characteristics are: when the resource deviation coefficient is close to 0 (i.e., the node reports truthfully), the credit score approaches 1; when the resource deviation coefficient gradually increases but remains within the range allowed by normal hardware fluctuations, the credit score decreases slowly; and when the resource deviation coefficient exceeds a preset deviation tolerance threshold, the credit score exhibits a steep decay trend. Its calculation is expressed as follows: in, This refers to the credit score in the consistency measure, and its value is an open interval ranging from 0 to 1. This is the resource deviation coefficient input from the upstream. The decay slope control factor is used to control the severity of credit score decay after deviation exceeds the tolerance level. The deviation tolerance threshold represents the median of the maximum legal deviation allowed within the jitter range of normal network operation. After completing the mapping calculation, the credit score, along with the corresponding node identifier, is encapsulated and output as a consistency metric for direct use in subsequent trust mapping model construction.

[0043] Specifically, in step S4, a trust mapping model is constructed based on resource profiles, topology connection records, and consistency metrics. It should be noted that while the consistency metric quantifies the reliability between the data reported by each node and actual observations, this credit score is not yet integrated with the physical topology of the computing network and cannot directly reflect the true availability of cross-domain communication links between nodes. Therefore, node-level credit penalties need to be injected into the topology graph structure containing resource entities and connection relationships to form a global trustworthy resource mapping. Based on this, the technical solution of this application further constructs a trust mapping model based on resource profiles, topology connection records, and consistency metrics. Through the above processing, the computing power attributes of each node and the link bandwidth attributes between nodes can be penalized and reduced according to the credit score, making the resource capacity represented by the mapping model closer to the true deliverable capacity of the nodes, providing a credit-corrected topology basis for the subsequent generation of a global perception view.

[0044] Figure 6 This is a flowchart illustrating the construction of a trust mapping model based on resource profiling, topology connection records, and consistency metrics, according to an embodiment of this application for a global perception and multi-dimensional mapping method for multi-entity cross-domain computing network resources. Figure 6 As shown, step S4 includes: S41, instantiating graph vertices based on the resource entities described in the resource profile, and using the credit score of the corresponding node in the consistency metric as the attenuation multiplier to linearly reduce the penalty injection on the theoretical computing power attribute of each graph vertex to obtain a modified graph vertex set; S42, based on the data dependency relationship and network connectivity status in the topology connection record, establishing associated graph edges between each vertex in the modified graph vertex set and assigning theoretical physical bandwidth weights to obtain a basic mapping graph model; S43, taking the minimum value of the credit score of the two vertices at both ends of each graph edge in the basic mapping graph model as the link penalty factor, and performing joint attenuation voltage drop constraints and solidification encapsulation on the theoretical bandwidth of the graph edge to obtain a trust mapping model.

[0045] In step S41, graph vertices are instantiated based on the resource entities described in the resource profile. The credit score of the corresponding node in the consistency metric is used as a decay multiplier to linearly reduce the theoretical computing power attribute of each graph vertex, resulting in a corrected graph vertex set. It should be noted that since the theoretical computing power attribute of each node recorded in the resource profile is still an uncorrected book value, directly constructing the mapping graph structure using it would result in the computing power of falsely reported nodes being included in the global resource graph, causing subsequent scheduling decisions to allocate tasks based on distorted capacity information. Therefore, the technical solution of this application further instantiates graph vertices based on the resource entities described in the resource profile and uses the credit score of the corresponding node in the consistency metric as a decay multiplier to linearly reduce the theoretical computing power attribute of each graph vertex, resulting in a corrected graph vertex set. Through the above processing, the computing power attribute value carried by each graph vertex can be matched with the trustworthiness of that node, thereby achieving pre-suppression of falsely reported computing power at the graph structure level.

[0046] More specifically, in a concrete example of this application, firstly, all resource entities described in the resource profile are traversed, a graph vertex object is created for each independent computing network node, and the theoretical computing power value declared by the node in the resource profile is written into the computing power attribute field of the graph vertex. After all graph vertices are instantiated, the credit score of the corresponding node is matched one by one from the consistency metric based on the node identifier carried by each graph vertex. The credit score is used as a decay multiplier and multiplied point-by-point with the theoretical computing power attribute of the graph vertex, and its calculation is expressed as follows: ,in, For the first The actual usable computing power of each graph vertex after penalty. For the first The theoretical computing power book value declared by each node in the resource profile. For the first Each node corresponds to a credit score in the consistency metric. When a node's credit score is 0.6, the available computing power of its graph vertices will be reduced to 60% of the theoretical value. As the credit score approaches 1, the computing power remains essentially unchanged. After injecting computing power penalties into all graph vertices, all corrected graph vertices are encapsulated and output as a corrected graph vertex set.

[0047] In step S42, based on the data dependencies and network connectivity status in the topology connection record, associated graph edges are established between the vertices of the modified graph vertex set and assigned theoretical physical bandwidth weights to obtain the basic mapping graph model. It should be noted that since the modified graph vertex set only contains the computing power attributes of each node after credit penalty, and network connections and data dependencies between nodes have not yet been established, isolated vertices cannot represent the communication paths and collaborative structures between resources in a cross-domain computing network. Based on this, the technical solution of this application further establishes associated graph edges between the vertices of the modified graph vertex set and assigns theoretical physical bandwidth weights based on the data dependencies and network connectivity status in the topology connection record to obtain the basic mapping graph model. Through the above processing, discrete modified graph vertices can be organized into a graph structure with topological connections, providing a complete topological carrier for subsequent credit bandwidth reduction at the graph edge level.

[0048] More specifically, in a concrete example of this application, the topology connection record is first obtained from the network control plane. This record contains the physical network connectivity status and cross-subject data dependencies between nodes. Each connection entry in the topology connection record is parsed, extracting the source and destination node identifiers for each connection. The corresponding two graph vertices are located in the modified graph vertex set, and a graph edge object is created between them. For node pairs with bidirectional communication requirements, two directed graph edges, one forward and one reverse, are established. After the graph edge is created, the physical link bandwidth capacity corresponding to the connection is extracted from the topology connection record and assigned as a theoretical physical bandwidth weight to the bandwidth attribute field of the graph edge. This weight reflects the upper limit of the original transmission capacity of the link before credit correction. After all connection entries have been traversed, the modified graph vertex set and all graph edges are assembled to form the basic mapping graph model.

[0049] In step S43, the minimum credit score of the two vertices at both ends of each graph edge in the basic mapping graph model is taken as the link penalty factor. The theoretical bandwidth of the graph edge is then subject to joint attenuation and voltage drop constraints and solidified encapsulation to obtain the trust mapping model. It should be noted that since the bandwidth weights of each graph edge in the basic mapping graph model are still theoretical values ​​without credit correction, and the actual data transmission capacity of a cross-domain communication link is limited by the actual carrying capacity of the less trustworthy node at either end, any false reporting behavior by any node will cause the actual usable bandwidth of the link to be lower than the theoretical nominal value. Based on this, the technical solution of this application further takes the minimum credit score of the two vertices at both ends of each graph edge in the basic mapping graph model as the link penalty factor, and performs joint attenuation and voltage drop constraints and solidified encapsulation on the theoretical bandwidth of the graph edge to obtain the trust mapping model. Through the above processing, the bandwidth attribute of the graph edge can reflect the constraint effect of the weakest link in terms of trustworthiness at both ends, thereby completing the voltage drop correction of false reporting bandwidth at the topology level.

[0050] More specifically, in a concrete example of this application, all edges in the underlying mapping graph model are traversed one by one. For each edge, the credit scores carried by the two endpoints are extracted, and the smaller of the two scores is taken as the penalty factor for that link. This strategy follows the "weakest link principle," meaning that the reliable transmission capability of a cross-domain link is determined by the node with the lower credit score. After determining the link penalty factor, it is multiplied by the theoretical physical bandwidth weight currently carried by the edge. The calculation is expressed as follows: in, For connecting nodes With nodes The actual available bandwidth of the graph edges after credit penalty mapping. This represents the original theoretical physical bandwidth capacity of the graph edges as marked in the topology connection record. and These are the source nodes at both ends of the graph edge. With the target node The credit scores of the two endpoints are calculated. When the credit scores of the two endpoints are 0.9 and 0.5 respectively, the available bandwidth of the link will be reduced to 50% of the theoretical value. After the bandwidth reduction is completed for all graph edges, the set of corrected graph vertices and all graph edges after bandwidth attenuation are solidified and encapsulated to output the trust mapping model.

[0051] Specifically, in step S5, based on the trust mapping model, the node computing power distribution after credit correction and cross-domain link bottlenecks are correlated and analyzed to obtain a global awareness view. It should be noted that since the trust mapping model stores all vertex computing power attributes and edge bandwidth attributes after credit correction in the form of a graph structure, this graph structure has not yet been transformed into a resource view that can be directly consumed by upper-layer scheduling decisions. Therefore, it is necessary to extract and integrate the node computing power distribution and link bottleneck features in the graph. Based on this, the technical solution of this application further uses the trust mapping model to correlate and analyze the node computing power distribution after credit correction and cross-domain link bottlenecks to obtain a global awareness view. Through the above processing, the trusted resource information in the graph structure can be transformed into a unified view that takes into account both computing power distribution and network bottlenecks, providing a globally resource map with credit verification for upper-layer scheduling decisions.

[0052] More specifically, in a concrete example of this application, the trust mapping model is first separated into node and edge sets, extracting the graph vertex and edge sets as two independent data subsets. For the graph vertex set, nodes are clustered according to their primary domain affiliation in the computing network, grouping nodes belonging to the same cloud service provider or edge computing region into the same cluster. The modified values ​​of all nodes within each cluster are then summed and aggregated using available computing power, expressed as follows: in, For the first The region of a cluster can be aggregated using computing power. To belong to the A set of nodes for a cluster. For nodes The actual usable computing power value after credit penalties For nodes The online activity status indicator coefficient is set to 1 when a node is online and schedulable, and 0 when it is offline or under maintenance. The aggregated available computing power values ​​of all clusters are arranged by cluster index and encapsulated into a global availability matrix, while the extracted graph edge set is retained as a link topology subgraph.

[0053] After the global availability matrix and link topology subgraph are generated, congestion bottlenecks are identified for each edge in the link topology subgraph. Each edge in the link topology subgraph is traversed, and the congestion index of that edge is calculated by combining the historical baseline normal traffic of the link preset by the computing network with the instantaneous burst traffic increment captured in the current detection period. The calculation is expressed as follows: in, For the graph edge The congestion bottleneck index. The historical baseline normal flow is located on the edge of this graph. For sudden surges in traffic, This represents the actual available bandwidth of the graph edge after credit-based attenuation. A higher congestion index indicates that the transmission load of the link is closer to or exceeds its trusted capacity limit. After calculating the congestion index of all graph edges, a preset health threshold is used for filtering. Graph edges with congestion indices exceeding the health threshold are marked as high-risk links. These high-risk links and their connected endpoint nodes are extracted, reconstructed, and the output network bottleneck topology is encapsulated.

[0054] After the global availability matrix and network bottleneck topology are ready, spatial alignment and cross-merging are performed on both. The regional computing power distribution of each cluster in the global availability matrix is ​​used as the lower-level capability layer, and the high-risk links and their endpoint distribution marked in the network bottleneck topology are used as the upper-level constraint layer. Spatial alignment between the layers is performed according to the geographical location of the nodes and their main domain identifiers. After alignment, the data from the two layers are cross-merged, linking the available computing power information of each region with the bottleneck link information involved in that region. The merged associated data is converted into a standardized weighted label graph format, and view rendering is performed to output a globally perceptual view, which can be directly invoked by upper-level intelligent scheduling decisions.

[0055] As described above, the global perception and multi-dimensional mapping method for multi-entity cross-domain computing network resources according to the embodiments of this application can be implemented in various cross-domain computing network resource management and collaborative scheduling scenarios, such as multi-cloud computing power transaction matching scenarios, cloud-edge-device collaborative task orchestration scenarios, or joint resource scheduling scenarios among national computing power hub nodes. In one possible implementation, this method can be integrated as a trusted resource perception and mapping engine based on credit consistency verification into a computing power network resource management platform, a multi-entity computing network collaborative scheduling platform, or a computing power transaction operation and control platform. For example, this method can be an independent resource awareness application running on the computing network control plane server, or it can be a resource credibility assessment and topology mapping processing component in an existing computing network orchestration and scheduling device, or it can be an intelligent perception and analysis middleware service deployed on the computing power hub side and receiving data reported by each subject node through cross-domain communication links. Of course, the resource profile extraction, active detection and historical feedback fusion, credit consistency deep verification and trust mapping model construction processing in this method can also run on edge computing nodes, regional computing power scheduling centers or cross-domain gateway devices, as the underlying real-time analysis and execution foundation of the multi-subject cross-domain computing network resource credibility perception and global mapping system.

[0056] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for global perception and multi-dimensional mapping of multi-entity cross-domain computing network resources, characterized in that, include: Step 1: Extract the computational load and link connection metrics contained in the host reporting specifications and network interface status reported by each node in the multi-entity cross-domain computing network by feature vectorization to obtain a resource profile. Step 2: Based on the node mapping boundary determined by the resource profile, initiate active network probing for each node to obtain active probing data. At the same time, retrieve the execution deviation records of the corresponding nodes from the historical task logs, and then perform noise reduction and integration on the active probing data and execution deviation records to obtain performance evidence. Step 3: Perform a deep verification of resource credit consistency between the theoretical benchmark values ​​declared in the resource profile and the observed comparison values ​​recorded in the performance evidence to obtain a consistency measure. Step 4: Based on resource profiling, topology connection records, and consistency metrics, construct a trust mapping model; Step 5: Based on the trust mapping model, perform correlation analysis on the node computing power distribution after credit correction and the cross-domain link bottleneck to obtain a global perception view.

2. The method for global perception and multi-dimensional mapping of multi-entity cross-domain computing network resources according to claim 1, characterized in that, Step one includes: Step 1.1: Using a data structure deserialization parsing engine, the computing power physical scalar in the host report specifications and the link configuration parameters in the network interface status are split, cleaned and extracted to obtain the computing load index set and the link connection index set. Step 1.2: Perform feature vectorization calculation on the computational load index set to obtain computational feature vectors, and simultaneously perform effective throughput calculation on the link connection index set to obtain connection feature vectors. Step 1.3: Perform spatiotemporal alignment and structured high-dimensional tensor concatenation on the calculated feature vector and the connected feature vector within the tolerance threshold to obtain the resource profile.

3. The method for global perception and multi-dimensional mapping of multi-entity cross-domain computing network resources according to claim 1, characterized in that, Step two includes: Step 2.1: Based on the node communication addresses and service ports declared in the resource profile, initiate asynchronous multipath active probing for each target node to obtain active probing data; Step 2.2: Using the node identifier in the active probe data as the index key, calculate the time offset rate and perform time-series correlation on the expected execution time and actual completion time of the corresponding node in the historical task log to obtain the execution deviation record; Step 2.3 involves denoising and integrating the active detection data and execution deviation records to obtain performance evidence.

4. The method for global perception and multi-dimensional mapping of multi-entity cross-domain computing network resources according to claim 1, characterized in that, Step three includes: Step 3.1: Perform dimensionality reduction and alignment on the theoretical computing power and bandwidth benchmarks declared in the resource profile and the observed performance indices recorded in the performance evidence to obtain the alignment comparison matrix; Step 3.2: Perform residual accumulation and summarization calculations on the negative deviations between the theoretical baseline values ​​and observed comparison values ​​of each dimension in the alignment comparison matrix to obtain the resource deviation coefficient; Step 3.3: Perform a quantization mapping of the resource deviation coefficient within the standard probability space to obtain a consistency measure.

5. The method for global perception and multi-dimensional mapping of multi-entity cross-domain computing network resources according to claim 4, characterized in that, Step 3.3 includes: quantizing the resource deviation coefficient within the standard probability space using the following formula: in, This is the resource deviation coefficient. It is the attenuation slope control factor, and For deviations from the tolerance threshold, It is the credit score in the consistency measure, and its value is an open interval ranging from 0 to 1.

6. The method for global perception and multi-dimensional mapping of multi-entity cross-domain computing network resources according to claim 1, characterized in that, Step four includes: Step 4.1: Instantiate and construct graph vertices based on the resource entities described in the resource profile, and use the credit score of the corresponding node in the consistency metric as a decay multiplier to linearly reduce the theoretical computing power attribute of each graph vertex to obtain the corrected graph vertex set. Step 4.2: Based on the data dependencies and network connectivity status in the topology connection record, establish the associated graph edges between the vertices of the modified graph vertex set and assign theoretical physical bandwidth weights to obtain the basic mapping graph model. Step 4.3: Take the minimum credit score of the two vertices at both ends of each graph edge in the basic mapping graph model as the link penalty factor, and perform joint attenuation voltage drop constraint and solidification encapsulation on the theoretical bandwidth of the graph edge to obtain the trust mapping model.

7. The method for global perception and multi-dimensional mapping of multi-entity cross-domain computing network resources according to claim 1, characterized in that, Step five includes: Step 5.1: Separate the node set and edge set of the trust mapping model, and perform clustering and computing power aggregation calculation on the node set according to the main membership domain to obtain the global availability matrix and link topology subgraph; Step 5.2: Based on the background baseline traffic and instantaneous burst traffic increments preset by the computing network, the congestion index is calculated and the health threshold is filtered for each edge in the link topology subgraph, and the endpoint reconstruction is performed to extract the high-risk links that exceed the threshold to obtain the network bottleneck topology. Step 5.3: Spatial alignment and cross-merging of the global availability matrix and network bottleneck topology, and conversion of the merged result into a standardized graph format for view rendering output to obtain a globally perceived view.

8. The method for global perception and multi-dimensional mapping of multi-entity cross-domain computing network resources according to claim 4, characterized in that, Step 3.2 includes: Interdimensional coupling prior information is extracted from the alignment ratio matrix to obtain the coupling prior matrix and the dimension residual vector; The coupling anomaly deviation is measured on the coupling prior matrix and the dimension residual vector to obtain a coupling anomaly score. The resource deviation coefficient is obtained by adaptively fusing the dimensional residual vector and the coupling anomaly score using a dual-channel deviation coefficient.

Images