Dynamic resource isolation method and device in computing power network, equipment and medium

Abstract

The embodiment of the invention discloses a dynamic resource isolation method and device in a computing power network, equipment and a medium, and relates to the technical field of computing power networks. The method comprises the following steps: acquiring the priority of a target power service in a computing power network, and performing iterative calculation of an isolation index according to the priority, an initial resource demand vector of the target power service and an available resource vector; in the iterative calculation process of the isolation index, if the isolation index is greater than or equal to an isolation threshold value, issuing configuration to the resource management and control component through dynamic adaptive learning constraint based on the current resource demand vector, and if the isolation index is smaller than the isolation threshold value after iteration is finished, issuing configuration to the resource management and control component through dynamic adaptive learning constraint. If yes, preempting service resources according to the priority, the current resource demand vector, the current available resource vector and the current isolation index; and based on the preempted resource demand vector, configuration is issued to the resource management and control component through dynamic adaptive learning constraint, an end-to-end isolation path is established, and a flow table is updated.

Description

Technical Field

[0001] This invention relates to the field of computing network technology, and in particular to a method, apparatus, device and medium for dynamic resource isolation in computing networks. Background Technology

[0002] With the digital transformation of power systems, computing networks have become a key infrastructure supporting smart grids. This network achieves unified scheduling of computing, storage, and network resources through FlexE hard slicing, SRv6 deterministic forwarding, and Kubernetes container orchestration.

[0003] In existing technologies, 5G network slicing combined with APN6 service identification can initially achieve traffic classification and QoS guarantee; SRv6Policy and FlexE interface work together to provide path-level deterministic transmission; Kubernetes PriorityClass supports priority preemption of computing resources.

[0004] However, these solutions have significant drawbacks: their resource isolation schemes cannot meet actual business needs. Summary of the Invention

[0005] This invention provides a method, apparatus, device, and medium for dynamic resource isolation in computing networks, which can accurately isolate resources based on isolation indexes.

[0006] According to one aspect of the present invention, a dynamic resource isolation method in a computing power network is provided, the method comprising: In the computing power network, the priority of the target power service is obtained, and the isolation index is iteratively calculated based on the priority, the initial resource demand vector of the target power service, and the available resource vector; the isolation index reflects the degree to which the current available resources of the computing power network support the service demand. During the iterative calculation of the isolation index, if the isolation index is greater than or equal to the isolation threshold, then based on the current resource demand vector, a configuration is issued to the resource management component through dynamic adaptive learning constraints to establish an end-to-end isolation path and update the flow table, so as to process the target power business based on the isolation path; If the isolation index is less than the isolation threshold after the iteration, business resources are preempted based on the priority, the current resource demand vector, the current available resource vector, and the current isolation index. Based on the preempted resource demand vector, configuration is issued to the resource management component through dynamic adaptive learning constraints, an end-to-end isolation path is established, and the flow table is updated so as to process the target power business based on the isolation path.

[0007] According to another aspect of the present invention, a dynamic resource isolation device in a computing network is provided, comprising: The isolation index calculation module is used to obtain the priority of the target power service in the computing power network, and to perform iterative calculation of the isolation index based on the priority, the initial resource demand vector of the target power service, and the available resource vector; the isolation index reflects the degree to which the current available resources of the computing power network support the service demand. The path establishment module is used to, during the iterative calculation of the isolation index, if the isolation index is greater than or equal to the isolation threshold, issue a configuration to the resource management component based on the current resource demand vector through dynamic adaptive learning constraints, establish an end-to-end isolated path and update the flow table, so as to process the target power business based on the isolated path; The resource preemption module is used to preempt business resources based on the priority, current resource demand vector, current available resource vector, and current isolation index if the isolation index is less than the isolation threshold after the iteration ends. Based on the preempted resource demand vector, the module issues configuration to the resource management component through dynamic adaptive learning constraints, establishes an end-to-end isolation path, and updates the flow table to process the target power business based on the isolation path.

[0008] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the dynamic resource isolation method in the computing power network according to any embodiment of the present invention.

[0009] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions, the computer instructions being configured to cause a processor to execute and implement the dynamic resource isolation method in a computing network as described in any embodiment of the present invention.

[0010] The technical solution of this application embodiment includes: obtaining the priority of a target power service in a computing power network; iteratively calculating an isolation index based on the priority, the initial resource demand vector of the target power service, and the available resource vector; the isolation index reflects the degree of support of the current available resources of the computing power network for the service demand; during the iterative calculation of the isolation index, if the isolation index is greater than or equal to an isolation threshold, then based on the current resource demand vector, a configuration is issued to the resource management component through dynamic adaptive learning constraints to establish an end-to-end isolation path and update the flow table, so as to process the target power service based on the isolation path; if, after the iteration, the isolation index is less than the isolation threshold, then, based on the priority, the current resource demand vector, the current available resource vector, and the current isolation index, the service resource is preempted; based on the preempted resource demand vector, a configuration is issued to the resource management component through dynamic adaptive learning constraints to establish an end-to-end isolation path and update the flow table, so as to process the target power service based on the isolation path. This technical solution uses the isolation index to determine whether resource preemption should occur, achieving accurate judgment of whether preemption should occur and greatly improving the processing efficiency of various services in the entire computing network.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart of a dynamic resource isolation method in a computing network according to one embodiment of this application; Figure 2 This is a schematic diagram of the structure of a dynamic resource isolation device in a computing network according to one embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device that implements a dynamic resource isolation method in a computing network according to an embodiment of this application. Detailed Implementation

[0014] To enable those skilled in the art to better understand the present invention, 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 the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0015] It should be noted that the terms "first," "second," "target," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0016] Figure 1 This application provides a flowchart of a dynamic resource isolation method in a computing network, as one embodiment of which is applicable to resource isolation of service flows in a computing network. The method can be executed by a dynamic resource isolation device in the computing network, which can be implemented in hardware and / or software and can be configured in an electronic device with data processing capabilities. Figure 1 As shown, the method includes: S110, Obtain the priority of the target power service in the computing power network, and perform iterative calculation of the isolation index based on the priority, the initial resource demand vector of the target power service, and the available resource vector.

[0017] The isolation index reflects the degree to which the currently available resources of the computing network support the service requirements. A higher isolation index indicates that the currently available computing resources in the computing network can complete the target power service; a lower isolation index indicates that the currently available computing resources in the computing network may not be able to complete the target power service. The priority of the target power service can be preset; for example, differential protection has a higher priority. The initial resource requirement vector of the target power service can be determined according to the user's SLA agreement (e.g., "differential protection delay cannot exceed 10ms"), and it includes three types of resources: FlexE time slots (network resources), SRv6 path bandwidth (transmission resources), and container computing resources (server resources). The available resource vector refers to the currently available resources in the computing network, which still includes the above three types of resources.

[0018] For example, the target power service is referred to as a service flow. Taking the service flow as an example, the APN6 header of the service flow is parsed at the computing power gateway, the service priority is determined according to the application identifier, and the resource feasibility is judged through isolation diffusion feedback iteration. The converged resource demand vector and isolation index are output to decide whether to directly allocate or trigger preemption.

[0019] In one specific implementation, the APN6 header of the service flow is parsed at the computing power gateway, and the service priority is determined based on the application identifier. Specifically, the APN6 header of the service flow entering the computing power gateway is parsed to extract the application identifier; the local policy library is queried based on the extracted application identifier to obtain the service priority P; an internal priority tag P is added to the service flow, and the service flow carrying the priority tag and the initial resource requirement vector generated according to the user service level agreement requirements are combined. The initial resource requirement vector is sent to the multi-dimensional resource isolation strategy engine. This includes the number of FlexE time slots, SRv6 path bandwidth, and container compute resource dimensions.

[0020] It should be noted that parsing the APN6 header of service flows at the computing power gateway and determining service priority based on the application identifier aims to establish a unified priority identifier for power service flows entering the computing power network. This ensures that all subsequent resource decisions (such as isolation assessment, preemption and reclamation, and path configuration) are guided by service criticality, avoiding critical service interruptions caused by indiscriminate competition. This step is deployed on the computing power gateway hardware (such as edge routers that support APN6 parsing), accelerating packet parsing through hardware to achieve microsecond-level processing, laying the foundation for system real-time performance.

[0021] Furthermore, the purpose of extracting the application identifier is to capture the original application characteristics of the business flow, serving as the sole input source for priority mapping. By parsing the APN6 extended fields using Deep Packet Inspection (DPI) technology, a standardized application identifier (App-ID, such as "differential protection" or "video surveillance") is extracted, avoiding reliance on fuzzy matching of IP / port.

[0022] Furthermore, the purpose of mapping business priorities is to convert application identifiers into internal system priority levels P (P1 for critical control, P2 for important monitoring, and P3 for general analysis), providing scaling factors and threshold benchmarks for subsequent isolation diffusion feedback iterations. The local policy library uses a hash table for storage (pre-configured in the gateway flash memory), with a query latency of <1ms.

[0023] Furthermore, by constructing standardized service request packets, priority information is ensured to accompany the flow throughout the process, while initial resource requirements are bundled to form a complete isolated request. This tag embeds flow metadata (internal VLAN or MPLS tag) and does not affect outer layer transmission; the initial resource requirement vector Generated by the SLA resolution module (automatically calculates FlexE time slots, SRv6 bandwidth, and container resources based on user contracts).

[0024] Optionally, perform an integrity check on the business request packet before sending it to the policy engine: verification Each dimension is non-negative and the sum does not exceed the system limit. If the verification fails, the business flow is rolled back to the gateway buffer queue and the alarm log is reported to ensure that invalid requests do not enter the decision chain.

[0025] In this embodiment of the application, optionally, the isolation index is calculated iteratively based on the priority, the initial resource demand vector of the target power service, and the available resource vector, including: performing isolation diffusion feedback iteration based on the priority, the initial resource demand vector of the target power service, and the available resource vector to obtain the isolation index for each iteration.

[0026] Specifically, based on the priority, the initial resource demand vector of the target power service, and the available resource vector, the first iteration of isolation diffusion feedback is performed. During the iteration, if the isolation index is greater than or equal to the isolation threshold, the iteration ends and operations such as resource allocation and path isolation are performed. Furthermore, during the iteration, the resource demand vector can be adjusted in each iteration until the isolation index is greater than or equal to the isolation threshold or the iteration ends. This configuration allows for gradual adjustment of the resource demand vector until the requirements are met.

[0027] In this embodiment of the application, optionally, isolation diffusion feedback iteration is performed based on the priority, the initial resource demand vector of the target power service, and the available resource vector to obtain the isolation index for each iteration, including: The isolation index is calculated using the following formula: ; in, For the first The isolation index for the next iteration, where t is the iteration number. Priority The Sigmoid normalization function, For the pre-trained resource coupling weight matrix, Let be the available resource vector for round t. Let be the intermediate resource vector after element-level diffusion scaling in the t-th round. Let be the resource demand vector at iteration t. This is the initial resource requirement vector. For resource diffusion sensitivity coefficient, This represents the historical feedback gain coefficient for isolation. For isolation degree gradient vector, This is the derivative of the Sigmoid function. For example, It can be element-wise division.

[0028] In one specific embodiment of the present invention, resource feasibility is determined through isolation degree diffusion feedback iteration, specifically as follows: Based on the service flow and initial resource requirement vector carrying priority tags Query the current resource pool to obtain the initial available resource vector. And calculate the isolation index based on the isolation diffusion feedback mechanism. ,like Then output , Used to establish end-to-end isolated paths; otherwise, update the resource requirement vector based on diffusion scaling. Continue until the convergence condition is met; if the iteration terminates... Then based on the converged , ,current And P performs preemption; among them, The isolation threshold corresponding to priority P. This is the final resource demand vector output after the isolation diffusion feedback iteration converges. This is the final isolation index after convergence.

[0029] It should be noted that the resource feasibility assessment through isolation diffusion feedback iteration aims to quantify the current resource pool's support for business needs. By optimizing initial requirements through multiple rounds of cross-layer borrowing, it avoids blindly seizing low-priority business while providing a "soft landing" path for high-priority business. This step is deployed on a multi-dimensional resource isolation strategy engine (high-performance server or containerized microservice), and connects to the computing power gateway via an internal high-speed link, ensuring decision latency of <50ms, providing an elastic buffer for real-time assurance of critical power business.

[0030] Furthermore, an initial assessment framework for resource supply-demand matching is established, which obtains available resource vectors by querying the global resource pool in real time. This serves as a dynamic benchmark for iterative calculations. The query result... With the input from the previous step Together with the priority label P, the isolation degree diffusion feedback mechanism is input to initiate the first round of index calculation, ensuring that the evaluation is based on the actual state of the system.

[0031] S120, during the iterative calculation of the isolation index, if the isolation index is greater than or equal to the isolation threshold, then based on the current resource demand vector, a configuration is issued to the resource management component through dynamic adaptive learning constraints to establish an end-to-end isolation path and update the flow table, so as to process the target power business based on the isolation path.

[0032] The resource management components include: FlexE (Elastic Ethernet) controller, SRv6 (IPv6 segment routing) controller, and Kubernetes platform, which is a container orchestration framework that automates application deployment, scaling, and operation, and achieves resource scheduling and elastic scaling through microservices.

[0033] For example, if the isolation index is not lower than the isolation threshold corresponding to the business priority, then configuration instructions are directly issued to the FlexE controller, SRv6 controller and Kubernetes platform based on the converged resource demand vector through dynamic adaptive learning constraints to establish an end-to-end isolation path and update the flow table; Furthermore, to enable rapid path switching in resource-sufficient scenarios, when a threshold is met in a single or multiple rounds of computation, the current resource demand vector is directly locked. For the final and output the isolation index. This avoids unnecessary preemption costs. The decision is based on the threshold corresponding to priority P. (Based on engine configuration) Output results and As a direct input to the path configuration module (bypassing preemption), it ensures efficient system response when resources are plentiful, while providing baseline isolation strength for monitoring and compensation.

[0034] Furthermore, when isolation is insufficient, the demand vector is dynamically scaled. By activating a cross-layer resource borrowing mechanism, high-priority business processes are simulated to "borrow" from the relaxed dimension to the tense dimension, gradually approaching a feasible solution. This update uses an intermediate vector. Based on this, iteratively execute until convergence (maximum number of iterations or gradient convergence), and output the optimized result. This serves as the input for the next round of computation, forming an adaptive closed loop. This process reduces the actual resource gap, directly lowers the probability of subsequent preemption, and contributes to the final... Provide a more reasonable starting point.

[0035] S130, if the isolation index is less than the isolation threshold after the iteration, then the business resources are preempted according to the priority, the current resource demand vector, the current available resource vector and the current isolation index; based on the preempted resource demand vector, the configuration is issued to the resource management component through dynamic adaptive learning constraints, an end-to-end isolation path is established and the flow table is updated, so as to process the target power business based on the isolation path.

[0036] For example, if the isolation index is lower than the isolation threshold corresponding to the business priority, the debt diffusion wave is used to recursively select and reclaim low-priority business resources, update the available resource pool and debt matrix, until the isolation threshold is met. Then, based on the resource demand vector after preemption, configuration instructions are issued to the FlexE controller, SRv6 controller and Kubernetes platform through dynamic adaptive learning constraints to establish an end-to-end isolation path and update the flow table.

[0037] Furthermore, if the isolation index remains below the isolation threshold corresponding to the priority level after the iteration terminates, a mandatory guarantee mechanism after iteration failure is activated. When the diffusion borrowing is exhausted and the threshold still cannot be met, the debt diffusion wave preemption module is triggered to reclaim low-priority resources to fill the gap. This triggering is based on eventual convergence. (Minimum requirements have been optimized) (Current isolation level) (Latest resource pool) and P (priority context) are used as input parameters to ensure precise preemption targeting the remaining gap. The execution result (updated resource pool) is fed back to the path configuration, achieving a seamless transition from demand optimization to forced resource acquisition. Preferably, before preemption is triggered, a resource locking pre-operation is performed: [The text abruptly ends here, likely due to an incomplete sentence or missing information.] The corresponding dimension is marked as "pending" in the resource pool to prevent concurrent business decisions from causing issues. Race condition; if path establishment fails after preemption, the lock is rolled back and the diffusion iteration is retried once.

[0038] In this embodiment of the application, optionally, the preemption of service resources is carried out based on the priority, the current resource demand vector, the current available resource vector, and the current isolation index. This includes: identifying candidate services with a priority lower than that of the target power service among the services currently running in the computing power network; calculating the preemption utility of each candidate service in each iteration; preempting the candidate service with the lowest preemption utility in each iteration and updating the current available resource vector, and determining the isolation index based on the updated current available resource vector; if the isolation index is greater than the isolation threshold, then the iteration and the preemption of service resources are terminated.

[0039] Specifically, among the services currently running in the computing power network, candidate services with a priority lower than the target power service are identified. In each iteration, a candidate service with the lowest preemption utility can be preempted, as follows: the preemption utility of each candidate service is calculated; in each iteration, the candidate service with the lowest preemption utility is preempted, and the currently available resource vector is updated; the isolation index is determined based on the updated currently available resource vector; if the isolation index is greater than the isolation threshold, the iteration and the preemption of service resources are terminated.

[0040] In this embodiment of the application, optionally, the calculation formula for the preemption utility of the candidate service is as follows: ; in, The preemption utility of the k-th candidate service in the t-th iteration. The priority of candidate service k, Prioritizing the target power business Let K be the resource vector occupied by candidate business k in round t. This is the resource demand vector after the isolation index iteration. This is the debt penalty coefficient. ; Let k be the debt row vector of candidate business k in round t-1. In this invention, the debt matrix... This is a data structure used to quantify the resource debt accumulated by preempted services. Whenever a low-priority service is forcibly reclaimed (i.e., "preempted"), the system doesn't simply discard it. Instead, it records how many resources it lost (e.g., how many FlexE time slots, how much bandwidth, how much CPU); records the duration of the interruption; and accumulates this information along the resource dimension to form a vector-like "debt." This debt is subsequently used to: determine which services most need compensation (the greater the debt, the higher the priority); propagate the impact within the neighborhood (similar to a "ripple effect"); and control the fairness of preemption and system stability. Therefore, each service corresponds to a row of debt data, and each column corresponds to a resource type (e.g., time slots, bandwidth, CPU, etc.). Thus, the debt matrix is ​​a two-dimensional table: rows = services, columns = resource dimensions. Meaning: : The k-th business; t-1: The decision time of the previous round of preemption; ":": indicates "take all columns of this row", i.e., the entire row vector. It should be noted that if ":" is omitted, only the following is written: In technical terms, "total debt" is usually understood as a scalar value (a single number), such as "total debt amount." However, in this solution, debt is recorded separately according to resource dimensions for several reasons, including: different resources cannot be directly added together: 1 FlexE time slot ≠ 1Mbps bandwidth ≠ 1 CPU core; their units and scheduling mechanisms are different, so they must be recorded separately. Preemption utility calculation relies on vector operations: the formula uses... (That is, summing over all dimensions to obtain the "total debt"), which requires It must be a vector. If it's a scalar, it cannot be used to perform a dot product with a vector of all ones. Neighborhood sweep update is a vector-level operation: when the business... After being seized, the debt of its neighboring business m will also be inherited proportionally according to the resource dimension. This can only be achieved through vector operations. It is a column vector of all 1s with the same dimension as the debt row vector. The neighborhood sweep intensity coefficient, This is the distance attenuation coefficient. For based on The neighborhood is constructed based on resource vector similarity using Euclidean distance. It should be noted that... In reality, it is a set of other low-priority services that are highly similar to the resource vector of candidate service k, and m is a specific low-priority service in this set.

[0041] It should be noted that since this scheme uses the resource demand vector and isolation index both during and after iterations, and the isolation index changes continuously with each iteration, the usage... and The symbols represent the resource demand vector and isolation index after this iteration (in the above formula, since it is executed after the iteration, the value after this iteration is...). That is, the resource demand vector after the isolation index iteration.

[0042] The debt diffusion wave is used to recursively select and reclaim low-priority business resources. Specifically, based on , ,current and Query candidate service priority The currently running services constitute a candidate set, and the resource usage vector of each candidate service k is obtained. Priority and debt row vector ; based on Euclidean distance is used to construct neighborhoods based on resource vector similarity. ; The preemption utility of each candidate service is calculated using the following formula: ; Select the candidate business with the lowest preemptive utility. Synchronous recycling and updating of resources Measure the actual interruption duration And record debts ; Perform impact update on neighboring business ;in, The debt spillover attenuation coefficient in the neighboring region; Use the updated version Re-execute the single-round isolation calculation to obtain ;in, This represents the current isolation level index; Repeat the preemption loop until And based on the final , , , Establish an end-to-end isolated path. That is, the available resource vector after the preemption ends. It should be noted that the recursive selection and reclamation of low-priority business resources using debt diffusion waves aims to achieve precise and controllable preemption of low-priority businesses by minimizing quantified utility and achieving balanced propagation of debt within the neighborhood. This ensures that high-priority businesses receive absolute physical-level isolation while maintaining overall system fairness and preventing repeated interruptions of single businesses. This step is integrated into the preemption submodule of the multi-dimensional resource isolation strategy engine, interacting in real time with the isolation degree decision module. A distributed lock mechanism (such as ZooKeeper) coordinates the three controllers (FlexE, SRv6, and Kubernetes) to guarantee the atomic execution of preemption commands with a response time of <200ms. (The debt mentioned reflects the preemption status of low-priority businesses.) Furthermore, based on the converged resource demand vector, isolation index, currently available resource vector, and business priority, running businesses with priorities lower than the stated business priority are queried to form a candidate set, which is used to construct the preemption candidate pool and serve as the raw dataset for utility calculation. Low-priority running businesses are queried through real-time snapshots of the resource pool (filtering). ), obtain its resource usage vector (Three-dimensional real-time occupancy), priority and historical debt (Matrix rows).

[0043] Furthermore, a neighborhood based on resource vector similarity is constructed to provide a topological structure for debt spillover, preventing debt from concentrating on isolated businesses after preemption. A similarity neighborhood graph (an adjacency list in memory) is constructed by calculating the Euclidean distance between candidate resource vectors (with a dynamically adjusted threshold).

[0044] Furthermore, the preemption cost is quantified by calculating the preemption utility, taking into account priority ratios, resource scale, and historical / neighborhood debt. The candidate service with the lowest preemption utility is selected for a single round of preemption, and the system status is updated in real time. Interruption duration is measured in real time by issuing reclamation commands (FlexE release slots, SRv6 policy revocation, and K8s Pod eviction) in parallel to the three controllers. And accumulate debt based on resource-duration.

[0045] Preferably, a maximum preemption round limit (default 3 rounds) is added before the loop terminates. If the limit is exceeded, the system will continue to operate. Then the business priority P is downgraded and recorded in the system audit log, and subsequent businesses use this as input to avoid infinite loops.

[0046] Furthermore, the impact of updates on neighboring businesses is mitigated to achieve debt "contagion" equilibrium and prevent repeated occupation of the same cluster in the next round. This is achieved through exponential decay. Debt is distributed to its neighboring areas, and the D matrix is ​​updated accordingly. The isolation degree is then recalculated using the updated resource pool to verify the effectiveness of a single round of preemption, serving as a termination condition for the cycle. This spillover update prevents debt concentration and its impact on utility assessments in subsequent rounds. Furthermore, the updated available resource pool is used to re-execute the single-round isolation calculation to obtain the current isolation index. This preemption loop is repeated until the current isolation index is not lower than the isolation threshold corresponding to the business priority. An end-to-end isolated path is then established based on the final available resource pool, debt matrix, converged resource demand vector, and isolation index. The final available resource pool and debt matrix are directly passed to the path configuration step to ensure resource consistency. This step uses a closed-loop convergence preemption process to ultimately output path configuration parameters. The repetition is managed by a loop control module until the threshold is met. This final quadruple (…) , , , The encapsulation is passed to the path configuration to ensure a seamless transition from preemption and recovery to the establishment of hard isolation, while the debt status D_t supports subsequent compensation.

[0047] In this embodiment of the application, optionally, configuration is issued to the resource management component through dynamic adaptive learning constraints, including: determining the current available resource vector, the current resource demand vector, the debt row vector, and the current coupling cost matrix corresponding to the isolation index, and calculating the initial coupling cost based on the current coupling cost matrix and the current resource demand vector; if the initial coupling cost is less than or equal to the dynamic cost upper limit, configuration instructions are issued to the FlexE controller, the SRv6 controller, and the Kubernetes platform to allocate resources of the corresponding dimensions respectively.

[0048] Optionally, in this embodiment of the application, the method further includes: if the initial coupling cost is greater than the dynamic cost upper limit, then the resource requirements that meet the high cost requirement conditions in the current resource requirement vector are locally downgraded to obtain an adjusted resource requirement vector; the adjusted cost is determined based on the adjusted resource requirement vector and the configuration success flag, until the adjusted cost is less than or equal to the dynamic cost upper limit, then configuration instructions are issued to the FlexE controller, SRv6 controller and Kubernetes platform to allocate resources of the corresponding dimension respectively.

[0049] In one specific embodiment of the present invention, the dynamic adaptive learning constraint-based issuance of FlexE time slots, SRv6Policy, and Kubernetes reserved configurations includes: Based on the above , , , Load the current coupling cost matrix Calculate the initial coupling cost ; like Then, configuration commands are issued to the FlexE controller, SRv6 controller, and Kubernetes platform to allocate resources for the corresponding dimensions; among them, This serves as the cost ceiling benchmark for priority P. The relaxation coefficient is used to compensate for the isolation degree. like Then, a partial downgrade adjustment will be implemented for the high-cost dimension. And based on the configuration success flag To perform online learning updates, use the following formula: ; in, For cost learning rate, This is a configuration success flag (it can be 1 if the configuration is successful, otherwise it can be 0). After successful configuration, the computing power gateway updates the flow table according to the allocated resource isolation combination, imports the service flow into the end-to-end hard isolation path, and... Feedback is fed back to the monitoring process for closed-loop optimization.

[0050] It should be noted that the configuration deployment via dynamic adaptive learning constraints aims to transform the resource requirements optimized / preempted in the previous steps into actual configuration instructions for the three-layer controller. Simultaneously, through cost threshold constraints and online learning mechanisms, it ensures controllable configuration linkage overhead, high synchronization success rate, and adaptability to system runtime variations. This step is deployed in the configuration submodule of the multi-dimensional resource isolation strategy engine, interfacing with the three controllers via gRPC / REST API, supporting parallel deployment and transaction rollback, guaranteeing configuration atomicity, and achieving an overall latency of <300ms.

[0051] Furthermore, based on the resource demand vector after direct convergence or preemption, the isolation index, and the final available resource pool, the current coupling cost matrix is ​​loaded, and the initial coupling cost is calculated, where the resource demand vector serves as the core vector for cost calculation. The potential linkage overhead of the three-layer resource allocation is evaluated as a preliminary check for configuration feasibility. This is achieved by loading persistent data. (From database or memory cache) Calculate the scalar cost for the core vector.

[0052] Furthermore, to achieve rapid path solidification in cost-compliant scenarios, when the initial cost meets a dynamic threshold, dedicated instructions (FlexE time slot slicing, SRv6 policy binding, and K8s Pod reservation) are issued in parallel. This threshold is based on... (Priority fixed benchmark) and (Isolation compensation) linkage, output configuration success flag. .

[0053] Furthermore, if the limits are exceeded, a partial downgrade adjustment will be implemented for the high-cost dimension. And based on the configuration success flag Perform online learning updates. This step aims to activate the self-healing mechanism for cost overruns, first by adjusting high-weight dimensions (reducing container resources). To reduce costs, and then based on actual (Post-configuration measurement) drives matrix updates. This degradation and learning are performed sequentially, outputting optimized results. and adjustment This process is triggered by an initial cost exceeding the limit, directly impacting retrying the delivery or the final outcome. This ensures that the configuration converges within a controllable cost range and accumulates experience for similar businesses in the future.

[0054] Furthermore, after successful configuration, the service flow is imported into the end-to-end hard-isolated path. This step aims to complete path activation and system-level feedback loop closure. The service flow is matched to the new path (FlexE channel + SRv6 SID + K8s Namespace) through the computing power gateway hardware flow table (TCAM acceleration), achieving physical-level isolation. This update is based on the resource combination confirmed by the three controllers, outputting an import completion signal, and simultaneously... Persistent feedback is sent to the system monitoring component. This feedback drives... The learning process iterates and provides a path benchmark for SLA verification, ensuring seamless integration and global optimization from configuration deployment to operational monitoring.

[0055] Optionally, in one embodiment of the present invention, a path connectivity pre-check is performed before the flow table is updated. A test packet is sent to FlexE / SRv6 to verify the availability of the slot / Policy. If the pre-check fails, a local degradation retry is triggered, and the weight of the failed dimension is doubled. Learning.

[0056] In one specific embodiment of the present invention, performance indicators are continuously collected in real time on the isolated path, including FlexE slot utilization, SRv6 path latency and jitter, and actual CPU / memory utilization of Kubernetes Pods. The collected metrics are compared with the requirements of the Service Level Agreement (SLA). If any metric is breached, it is recorded. And trigger emergency compensation: query the aforementioned debt matrix. Prioritize the total amount of debt The largest preempted business, and reuse based on The established end-to-end hard isolation path restores resources; Record during normal operation And update resource pool drift ;in, This refers to the amount of resources that are naturally released or newly added during the monitoring period.

[0057] It should be noted that continuously collecting performance metrics and performing compensation on isolated paths in real time aims to verify whether the actual performance of the path meets SLA requirements, promptly detect defaults, and quickly restore preempted services through a debt-driven path reuse mechanism. Simultaneously, it captures natural resource drift during normal operation to ensure the dynamic accuracy of the resource pool status. This step is deployed on a distributed monitoring agent (Prometheus + Exporter deployed on FlexE switches, SRv6 routers, and K8s nodes), interacting asynchronously with the policy engine via an event bus (Kafka), supporting millisecond-level alarms, with an overall compensation response time of <1 second.

[0058] It should be noted that performance metrics are continuously collected in real time along the isolated path to establish a comprehensive observation basis for path operation, serving as the original data source for default judgment. Standardized metric streams are output through periodic collection (default 100ms) using dedicated probes (FlexE hardware counter, SRv6 segment statistics, K8scAdvisor).

[0059] Furthermore, to achieve immediate self-healing and fair compensation in default scenarios, when any indicator exceeds the SLA threshold, [the system will] set [a certain condition]. Trigger the compensation chain. (Through query) The business with the largest debt is selected as the recovery target. The hard-isolated path template fixed in the previous steps is reused (only the flow table import is adjusted), and the remaining redundancy is used for rapid recovery. This is triggered based on the default of the collected indicators, and outputs a compensation completion signal and an update. Direct feedback to configuration learning ( (Penalties) ensure that high-priority defaults do not sacrifice fairness for low-priority defaults.

[0060] Furthermore, to maintain the dynamic consistency of the system resource view, it is set to true when there is no breach. As positive learning samples, while accumulating natural drift (Release due to business completion, hardware failure recovery, or addition of a new node). This update follows the normal data collection cycle and outputs the latest data. This result is routed to the resource pool database. It runs in parallel with the default branch, directly serving the isolation decision for the next business cycle. (initial value), and for The cost matrix is ​​optimized to form an adaptive closed loop that monitors the global state.

[0061] Preferably, in one implementation scenario of the present invention, resource drift updates employ differential merging. First and current Perform vector clamping (not exceeding the physical limit), then atomically replace resource pool entries; when reusing default compensation paths, prioritize starting from... Allocation is done in positive increments to avoid triggering existing paths.

[0062] In one specific embodiment of the present invention, after receiving a normal service termination signal, the occupied resource vector is released. Element-level update resource pool And clear the record of this business in the debt matrix; Query the updated debt matrix Compensation requests are generated for all non-zero debt business units in descending order of total debt, and compensation paths are assigned until the debt is cleared; the compensation paths reuse the isolation degree decision and preemption mechanism. Set the compensation path configuration success flag. After final learning and after clearing Persist to the system database to form a self-healing feedback loop.

[0063] It should be noted that the resource release and compensation closed-loop execution after receiving a normal business termination signal aims to orderly reclaim high-priority business resources, prioritize compensation for historically preempted businesses, persist the learning state, ensure the debt matrix is ​​zeroed out, the cost matrix evolves, and provide a clean initial environment for the next business cycle. This step is integrated into the release submodule of the multi-dimensional resource isolation strategy engine, and interfaces with the monitoring module in an event-driven manner (the business termination signal is triggered via API or heartbeat loss), supports batch compensation queue processing, and the overall execution time is <500ms.

[0064] Furthermore, to promptly return path resources and clean up business footprints, as a prerequisite for compensation, release commands are issued in parallel to the three controllers (FlexE reclaims time slots, SRv6 revokes policies, and K8s deletes Pods), reducing element-level resource consumption. renew At the same time, set the corresponding row of the zero debt matrix.

[0065] Furthermore, to achieve debt priority compensation and form a systemic fairness closed loop, this is achieved through scanning. The total quantity is calculated in descending order. A request is generated for each non-zero business line, and a new path is allocated using the isolation diffusion feedback and preemption wave mechanism (no full preemption is required, only redundancy is released). This allocation is executed one by one until the matrix is ​​cleared, and the sequence of successful compensation is output.

[0066] Furthermore, to solidify the learning outcomes and clean state of the solidification cycle, it serves as a self-healing bridge across business processes. This is achieved through batch writing to a distributed database (etcd or MySQL) for storage. (Compensation learning signal) (Cumulative update matrix) and (All zeros). This persistence uses the compensation output of the previous step as the source to directly initialize W for the next cycle (if needed). (load )and (Zero matrix) ensures no state contamination and no inheritance of experience from the end to the newly identified.

[0067] Preferably, in one embodiment of the present invention, batch verification of compensation paths is performed before persistence. After all compensation requests are sent out in parallel, they are collected uniformly. The average success rate is calculated, and if it falls below a threshold, partial compensation is rolled back and high-failure dimensions are marked. This ensures that persistent data reflects the true learning outcomes.

[0068] The current QoS (Quality of Service) system only provides soft guarantees, which cannot meet the physical isolation requirements of critical power services (such as differential protection), leading to uncontrollable latency jitter under resource contention. Furthermore, the independent management of the network and computing layers lacks an end-to-end coordination mechanism, making configuration distribution prone to linked congestion. Additionally, the preemption strategy is limited to a single dimension, ignoring the fairness of low-priority services, easily resulting in debt hotspots and repeated outages. Simultaneously, existing technologies lack dynamic learning mechanisms, making it difficult for the system to adapt to resource drift and historical failures. These problems result in high SLA default risk for critical services, low resource utilization, and difficulty in meeting the "zero-interruption" requirement of the power system. This technical solution achieves collaborative hard isolation between FlexE, SRv6, and Kubernetes through a unified priority policy engine and cross-layer resource isolation, overcoming the limitations of soft QoS and ensuring physical-level guarantees for critical services. Debt propagation waves and dynamic learning mechanisms balance preemption fairness and adaptively configure overhead, avoiding debt concentration and linked congestion. Monitoring and compensation closed-loop reuse paths quickly restore resources, improving system robustness and utilization efficiency.

[0069] The technical solution of this application embodiment includes: obtaining the priority of a target power service in a computing power network; iteratively calculating an isolation index based on the priority, the initial resource demand vector of the target power service, and the available resource vector; the isolation index reflects the degree of support of the current available resources of the computing power network for the service demand; during the iterative calculation of the isolation index, if the isolation index is greater than or equal to an isolation threshold, then based on the current resource demand vector, a configuration is issued to the resource management component through dynamic adaptive learning constraints to establish an end-to-end isolation path and update the flow table, so as to process the target power service based on the isolation path; if, after the iteration, the isolation index is less than the isolation threshold, then, based on the priority, the current resource demand vector, the current available resource vector, and the current isolation index, the service resource is preempted; based on the preempted resource demand vector, a configuration is issued to the resource management component through dynamic adaptive learning constraints to establish an end-to-end isolation path and update the flow table, so as to process the target power service based on the isolation path. This technical solution uses the isolation index to determine whether resource preemption should occur, achieving accurate judgment of whether preemption should occur and greatly improving the processing efficiency of various services in the entire computing network.

[0070] Figure 2 This is a schematic diagram of a dynamic resource isolation device in a computing network, provided as an optional embodiment of this application. This device can execute the dynamic resource isolation method in a computing network provided in any embodiment of this invention, and possesses the corresponding functional modules and beneficial effects for executing the method. For example... Figure 2 As shown, the device includes: The isolation index calculation module 210 is used to obtain the priority of the target power service in the computing power network, and to perform iterative calculation of the isolation index based on the priority, the initial resource demand vector of the target power service, and the available resource vector; the isolation index reflects the degree of support of the current available resources of the computing power network for the service demand; The path establishment module 220 is used to, during the iterative calculation of the isolation index, if the isolation index is greater than or equal to the isolation threshold, issue a configuration to the resource management component based on the current resource demand vector through dynamic adaptive learning constraints, establish an end-to-end isolation path and update the flow table, so as to process the target power business based on the isolation path; The resource preemption module 230 is used to preempt business resources based on the priority, current resource demand vector, current available resource vector, and current isolation index if the isolation index is less than the isolation threshold after the iteration ends; and to issue configuration to the resource management component based on the preempted resource demand vector through dynamic adaptive learning constraints, establish an end-to-end isolation path and update the flow table, so as to process the target power business based on the isolation path.

[0071] The technical solution of this application embodiment includes: an isolation index calculation module 210, used to obtain the priority of the target power service in the computing power network, and perform iterative calculation of the isolation index based on the priority, the initial resource demand vector of the target power service, and the available resource vector; the isolation index reflects the degree of support of the current available resources of the computing power network for the service demand; a path establishment module 220, used to, during the iterative calculation of the isolation index, if the isolation index is greater than or equal to the isolation threshold, issue a configuration to the resource management component based on the current resource demand vector through dynamic adaptive learning constraints, establish an end-to-end isolated path and update the flow table, so as to process the target power service based on the isolated path; a resource preemption module 230, used to, if the isolation index is less than the isolation threshold after the iteration, preempt the service resources based on the priority, the current resource demand vector, the current available resource vector, and the current isolation index; issue a configuration to the resource management component based on the preempted resource demand vector through dynamic adaptive learning constraints, establish an end-to-end isolated path and update the flow table, so as to process the target power service based on the isolated path. This technical solution uses the isolation index to determine whether resource preemption should occur, achieving accurate judgment of whether preemption should occur and greatly improving the processing efficiency of various services in the entire computing network.

[0072] Optionally, in this embodiment of the application, the isolation index calculation module 210 includes: The isolation index calculation unit is used to perform isolation diffusion feedback iteration based on the priority, the initial resource demand vector of the target power service, and the available resource vector to obtain the isolation index for each iteration.

[0073] Optionally, in this embodiment of the application, the isolation index calculation unit is specifically used for: The isolation index is calculated using the following formula: ; in, For the first The isolation index for the next iteration, where t is the iteration number. Priority The Sigmoid normalization function, For the pre-trained resource coupling weight matrix, Let be the available resource vector for round t. Let be the intermediate resource vector after element-level diffusion scaling in the t-th round. Let be the resource demand vector at iteration t. This is the initial resource requirement vector. For resource diffusion sensitivity coefficient, This represents the historical feedback gain coefficient for isolation. For isolation degree gradient vector, This is the derivative of the Sigmoid function.

[0074] In this embodiment of the application, optionally, the resource preemption module 230 includes: The candidate service determination unit is used to determine candidate services with a lower priority than the target power service among the services currently running in the computing power network. The preemption utility calculation unit is used to calculate the preemption utility of each candidate service in each iteration. The available resource vector update unit is used to preempt the candidate service with the lowest preemption utility in each iteration, update the current available resource vector, and determine the isolation index based on the updated current available resource vector. The preemption termination unit is used to terminate the iteration and preemption of business resources if the isolation index is greater than the isolation threshold.

[0075] In this embodiment of the application, optionally, the calculation formula for the preemption utility of the candidate service is as follows: ; in, The preemption utility of the k-th candidate service in the t-th iteration. The priority of candidate service k, Prioritizing the target power business Let K be the resource vector occupied by candidate business k in round t. This is the resource demand vector after the isolation index iteration. This is the debt penalty coefficient. ; Let k be the debt row vector of candidate business k in round t-1. It is a column vector of all 1s with the same dimension as the debt row vector. The neighborhood sweep intensity coefficient, This is the distance attenuation coefficient. For based on The neighborhood is constructed based on the similarity of resource vectors using Euclidean distance.

[0076] Optionally, in this embodiment of the application, configuration is distributed to the resource management component through dynamic adaptive learning constraints, including: Determine the current coupling cost matrix corresponding to the current available resource vector, current resource demand vector, debt row vector, and isolation index, and calculate the initial coupling cost based on the current coupling cost matrix and the current resource demand vector; If the initial coupling cost is less than or equal to the dynamic cost upper limit, configuration instructions are issued to the FlexE controller, SRv6 controller, and Kubernetes platform to allocate resources in the corresponding dimensions.

[0077] Optionally, in this embodiment of the application, the device further includes: The local degradation module is used to locally downgrade the resource requirements that meet the high-cost requirement conditions in the current resource requirement vector if the initial coupling cost is greater than the dynamic cost upper limit, so as to obtain an adjusted resource requirement vector. The configuration instruction issuing unit is used to determine the adjusted cost based on the adjusted resource demand vector and the configuration success flag. When the adjusted cost is less than or equal to the dynamic cost upper limit, configuration instructions are issued to the FlexE controller, SRv6 controller and Kubernetes platform to allocate resources in the corresponding dimensions.

[0078] The dynamic resource isolation device in a computing network provided in this application embodiment can execute the dynamic resource isolation method in a computing network provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of executing the method.

[0079] Figure 3 A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0080] like Figure 3As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0081] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0082] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as dynamic resource isolation methods in computing networks.

[0083] In some embodiments, the dynamic resource isolation method in a computing network can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the dynamic resource isolation method in a computing network described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the dynamic resource isolation method in a computing network by any other suitable means (e.g., by means of firmware).

[0084] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0085] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0086] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0087] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0088] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0089] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS services, such as high management difficulty and weak business scalability.

[0090] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0091] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A dynamic resource isolation method in a computing power network, characterized in that, include: In the computing power network, the priority of the target power service is obtained, and the isolation index is iteratively calculated based on the priority, the initial resource demand vector of the target power service, and the available resource vector. The isolation index reflects the degree to which the currently available resources of the computing network support business needs; During the iterative calculation of the isolation index, if the isolation index is greater than or equal to the isolation threshold, then based on the current resource demand vector, a configuration is issued to the resource management component through dynamic adaptive learning constraints to establish an end-to-end isolation path and update the flow table, so as to process the target power business based on the isolation path; If the isolation index is less than the isolation threshold after the iteration, business resources are preempted based on the priority, the current resource demand vector, the current available resource vector, and the current isolation index. Based on the preempted resource demand vector, configuration is issued to the resource management component through dynamic adaptive learning constraints, an end-to-end isolation path is established, and the flow table is updated so as to process the target power business based on the isolation path.

2. The method according to claim 1, characterized in that, The isolation index is iteratively calculated based on the priority, the initial resource demand vector of the target power service, and the available resource vector, including: Based on the priority, the initial resource demand vector of the target power service, and the available resource vector, an isolation degree diffusion feedback iteration is performed to obtain the isolation degree index for each iteration.

3. The method according to claim 2, characterized in that, Based on the priority, the initial resource demand vector of the target power service, and the available resource vector, an isolation degree diffusion feedback iteration is performed to obtain the isolation degree index for each iteration, including: The isolation index is calculated using the following formula: ； in, For the first The isolation index for the next iteration, where t is the iteration number. Priority For the Sigmoid normalization function, For the pre-trained resource coupling weight matrix, Let be the available resource vector for round t. Let be the intermediate resource vector after element-level diffusion scaling in the t-th round. Let be the resource demand vector at iteration t. This is the initial resource requirement vector. For resource diffusion sensitivity coefficient, This represents the historical feedback gain coefficient for isolation. For isolation degree gradient vector, This is the derivative of the Sigmoid function.

4. The method according to claim 1, characterized in that, Based on the aforementioned priority, current resource demand vector, current available resource vector, and current isolation index, business resources are preempted, including: In the computing power network, identify candidate services with a priority lower than that of the target power service among the services currently in operation; In each iteration, the preemption utility of each candidate service is calculated; In each iteration, the candidate service with the lowest preemption utility is preempted, and the current available resource vector is updated. The isolation index is determined based on the updated current available resource vector. If the isolation index is greater than the isolation threshold, the iteration and the preemption of business resources will end.

5. The method according to claim 4, characterized in that, The formula for calculating the preemption utility of candidate services is as follows: ； in, The preemption utility of the k-th candidate service in the t-th iteration. The priority of candidate service k, Prioritizing the target power business Let K be the resource vector occupied by candidate business k in round t. This is the resource demand vector after the isolation index iteration. This is the debt penalty coefficient. ; Let k be the debt row vector of candidate business k in round t-1. It is a column vector of all 1s with the same dimension as the debt row vector. The neighborhood sweep intensity coefficient, This is the distance attenuation coefficient. Based on The neighborhood is constructed based on the similarity of resource vectors using Euclidean distance.

6. The method according to claim 1, characterized in that, Configurations are distributed to the resource management component through dynamic adaptive learning constraints, including: Determine the current coupling cost matrix corresponding to the current available resource vector, current resource demand vector, debt row vector, and isolation index, and calculate the initial coupling cost based on the current coupling cost matrix and the current resource demand vector; If the initial coupling cost is less than or equal to the dynamic cost upper limit, configuration instructions are issued to the FlexE controller, SRv6 controller, and Kubernetes platform to allocate resources in the corresponding dimensions.

7. The method according to claim 6, characterized in that, The method further includes: If the initial coupling cost is greater than the dynamic cost upper limit, then the resource requirements in the current resource requirement vector that meet the high cost requirement condition will be locally downgraded to obtain the adjusted resource requirement vector. Based on the adjusted resource demand vector and the configuration success flag, the adjusted cost is determined until the adjusted cost is less than or equal to the dynamic cost upper limit. Then, configuration instructions are issued to the FlexE controller, SRv6 controller, and Kubernetes platform to allocate resources in the corresponding dimensions.

8. A dynamic resource isolation device in a computing power network, characterized in that, include: The isolation index calculation module is used to obtain the priority of the target power service in the computing power network, and to perform iterative calculation of the isolation index based on the priority, the initial resource demand vector of the target power service, and the available resource vector. The isolation index reflects the degree to which the currently available resources of the computing network support business needs; The path establishment module is used to, during the iterative calculation of the isolation index, if the isolation index is greater than or equal to the isolation threshold, issue a configuration to the resource management component based on the current resource demand vector through dynamic adaptive learning constraints, establish an end-to-end isolated path and update the flow table, so as to process the target power business based on the isolated path; The resource preemption module is used to preempt business resources based on the priority, current resource demand vector, current available resource vector, and current isolation index if the isolation index is less than the isolation threshold after the iteration ends. Based on the preempted resource demand vector, the module issues configuration to the resource management component through dynamic adaptive learning constraints, establishes an end-to-end isolation path, and updates the flow table to process the target power business based on the isolation path.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the dynamic resource isolation method in the computing power network according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the dynamic resource isolation method in the computing network according to any one of claims 1-7.

Images