An edge node scheduling method based on multi-dimensional evaluation and dynamic weight

Abstract

The application relates to an edge node scheduling method based on multi-dimensional evaluation and dynamic weight, which comprises the following steps: obtaining task resource demand and service quality requirement of a to-be-scheduled task, and screening to form a candidate node set; obtaining technical evaluation indexes and economic evaluation indexes of each candidate node, and obtaining technical scores and economic scores through normalization and weighting processing; determining task characteristic parameters and operation environment parameters according to task demand characteristics and resource operation states of the candidate nodes, and determining a task correction amount and an environment correction amount according to the task characteristic parameters and the operation environment parameters, so as to adjust a basic fusion weight to obtain a dynamic fusion weight; fusing the technical scores and the economic scores to obtain comprehensive scores, and determining a target node according to the comprehensive scores, and issuing the to-be-scheduled task after confirming that the target node is available. The application can improve the adaptability of edge node scheduling to task demand and operation environment.

Description

Technical Field

[0001] This application relates to the field of edge computing technology, and in particular to an edge node scheduling method based on multidimensional evaluation and dynamic weights. Background Technology

[0002] With the development of edge computing technology, more and more computing tasks are being migrated to edge nodes closer to the user. Since different edge nodes differ in terms of computing resources, network status, and service costs, it is necessary to select the appropriate edge node for task scheduling based on task requirements.

[0003] Existing edge node scheduling methods are difficult to adapt to different task types and dynamically changing network environments, which can easily lead to insufficient matching between node selection and task requirements, affecting task execution efficiency and service quality.

[0004] Therefore, how to dynamically adjust the node evaluation method according to the characteristics of task requirements and the operating environment of edge nodes, and improve the adaptability of edge node selection, has become an urgent technical problem to be solved. Summary of the Invention

[0005] To overcome, to some extent, the problem of insufficient matching between node selection and task requirements in edge node scheduling processes using fixed evaluation methods in related technologies, this application provides an edge node scheduling method based on multidimensional evaluation and dynamic weights.

[0006] The proposed solution is as follows:

[0007] An edge node scheduling method based on multidimensional evaluation and dynamic weights includes: Obtain the tasks to be scheduled and determine their resource requirements and quality of service requirements. From the edge nodes that are online, select edge nodes that meet the task resource requirements and service quality requirements to form a candidate node set; Based on the preset evaluation index system, obtain the technical evaluation index and economic evaluation index of each candidate node. The technical and economic evaluation indicators of each candidate node are normalized and weighted to obtain the technical and economic scores of each candidate node. The task characteristic parameters and operating environment parameters are determined based on the task requirement characteristics of the task to be scheduled and the resource operation status of the candidate node set, respectively. The task correction amount and the environment correction amount are determined based on the task characteristic parameters and the operating environment parameters, respectively. The pre-set basic fusion weights are adjusted based on the task correction amount and the environment correction amount to obtain dynamic fusion weights; The technical and economic scores of each candidate node are fused according to the dynamic fusion weight to obtain the comprehensive score of each candidate node; The target node is determined based on the comprehensive score of each candidate node, and the task to be scheduled is sent to the target node after confirming that the target node is in an available state.

[0008] Preferably, after obtaining the dynamic fusion weights, the method further includes: Obtain a pre-defined fusion weight range, and compare the dynamic fusion weight with the fusion weight range; When the dynamic fusion weight is lower than the lower limit of the fusion weight range, the dynamic fusion weight is adjusted to the lower limit of the fusion weight range; When the dynamic fusion weight is higher than the upper limit of the fusion weight range, the dynamic fusion weight is adjusted to the upper limit of the fusion weight range; When the dynamic fusion weight is within the range of the fusion weight, the dynamic fusion weight is directly output.

[0009] Preferably, after adjusting the range of the dynamic fusion weights, the method further includes: Obtain historical dynamic fusion weights; According to the preset smoothing weights, the dynamic fusion weights after range adjustment and the historical dynamic fusion weights are weighted and processed to output the smoothed dynamic fusion weights.

[0010] Preferably, the technical evaluation indicators and economic evaluation indicators of each candidate node are obtained, including: Send probe requests to each candidate node to obtain network round-trip latency, and combine the historical end-to-end latency of the corresponding candidate node to determine the estimated latency; Obtain the current available network bandwidth reported by each candidate node to get the available bandwidth; Based on the historical task completion records of each candidate node, determine the node reliability of the corresponding candidate node; The estimated latency, available bandwidth, and node reliability are used as the technical evaluation indicators. Obtain the service quotes submitted by each candidate node for the task to be scheduled; Query the blockchain records corresponding to each candidate node to obtain the credit score and the amount of collateral; The pledge gain is determined based on the pledged amount; The service price, credit score, and staking gain are used as the economic evaluation indicators.

[0011] Preferably, the technical evaluation indicators of each candidate node are normalized and weighted to obtain the technical score of each candidate node, including: The maximum and minimum values ​​corresponding to each technical evaluation indicator are determined according to the preset statistical range; For the estimated delay in the technical evaluation index, reverse normalization is performed according to the principle that the smaller the delay value corresponds to the higher the evaluation result, to obtain the delay evaluation value; For the available bandwidth in the aforementioned technical evaluation indicators, positive normalization is performed according to the principle that a larger bandwidth value corresponds to a higher evaluation result, to obtain the bandwidth evaluation value; The reliability evaluation value is determined based on the historical task completion status corresponding to the reliability of the node. Based on preset technical indicator weights, the latency evaluation value, bandwidth evaluation value, and reliability evaluation value are weighted to obtain the technical score of each candidate node.

[0012] Preferably, the economic evaluation indicators of each candidate node are normalized and weighted to obtain the economic score of each candidate node, including: The cost-benefit evaluation value is determined based on the service price of each candidate node, and the cost-benefit evaluation value is normalized according to the principle that the lower the service price, the higher the cost-benefit evaluation value. Obtain the credit score corresponding to each candidate node, and normalize the credit score according to the preset credit evaluation range to obtain the credit evaluation value; Obtain the staking amount corresponding to each candidate node, perform logarithmic mapping processing based on the staking amount to obtain the staking gain, and normalize the staking gain to obtain the staking evaluation value. Based on the preset economic indicator weights, the cost-benefit evaluation value, credit evaluation value, and pledge evaluation value are weighted to obtain the economic score of each candidate node.

[0013] Preferably, the task characteristic parameters and operating environment parameters are determined based on the task requirement characteristics of the task to be scheduled and the resource operating status of the candidate node set, including: Obtain the task type corresponding to the task to be scheduled; The latency sensitivity, bandwidth requirement, cost sensitivity, and reliability requirement of the task to be scheduled are determined based on the task type, task resource requirements, and service quality requirements. The latency sensitivity, bandwidth requirement, cost sensitivity, and reliability requirement are standardized to obtain the task characteristic parameters; Obtain the resource occupancy status and available resource status of each candidate node in the candidate node set; The global load pressure is determined based on the resource occupancy status of each candidate node in the candidate node set. The resource sufficiency is determined based on the available resource status of each candidate node in the candidate node set and the task resource requirements of the task to be scheduled. The operating environment parameters are determined based on the global load pressure and resource sufficiency.

[0014] Preferably, the task correction amount and environment correction amount are determined according to the task characteristic parameters and the operating environment parameters, respectively, and the basic fusion weights are adjusted according to the task correction amount and environment correction amount to obtain dynamic fusion weights, including: The basic fusion weights are determined based on the correlation between latency sensitivity, bandwidth requirement, cost sensitivity, and reliability requirement in the task characteristic parameters. Based on the task features that meet the preset preference conditions in the task feature parameters, determine the task correction amount; when the latency sensitivity or reliability requirement of the task to be scheduled reaches the preset condition, increase the fusion ratio corresponding to the technical score; when the cost sensitivity of the task to be scheduled reaches the preset condition, increase the fusion ratio corresponding to the economic score. The environment correction amount is determined based on the global load pressure and resource adequacy in the operating environment parameters; when the global load pressure corresponding to the candidate node set increases or the resource adequacy decreases, the fusion ratio corresponding to the technical score is increased; The basic fusion weights are adjusted based on the task correction amount and the environment correction amount to obtain the dynamic fusion weights.

[0015] Preferably, the method further includes: The candidate node set is sorted according to the comprehensive score of each candidate node, the candidate node with the highest comprehensive score is determined as the target node, and the candidate nodes whose comprehensive score ranks within a preset ranking range are determined as backup nodes. Before sending the task to be scheduled to the target node, the node status information of the target node is reacquired, and the online status and resource availability status of the target node are reviewed. When the target node is online and meets the resource requirements of the task to be scheduled, the task to be scheduled is sent to the target node; When the target node does not meet the execution conditions of the task to be scheduled, the target node is re-determined based on the comprehensive score ranking of the backup nodes, and the task to be scheduled is sent to the re-determined target node.

[0016] Preferably, the method further includes: Obtain task characteristic parameters, runtime environment parameters, technical score, economic score, dynamic fusion weight, and corresponding task execution results during the historical scheduling process; The state information of the reinforcement learning model is constructed based on the task feature parameters and the operating environment parameters. The action information of the reinforcement learning model is determined based on the adjustment method of the technical score, economic score and dynamic fusion weight. The reward information of the reinforcement learning model is determined based on the task execution result. The reinforcement learning model is trained based on the state information, action information, and reward information to obtain a weight optimization model; The weight optimization model is used to optimize and adjust the dynamic fusion weight after inputting the task feature parameters, operating environment parameters and dynamic fusion weight corresponding to the current task to be scheduled, and output the optimized and adjusted dynamic fusion weight.

[0017] The technical solution provided in this application may include the following beneficial effects: This technical solution constructs a multi-dimensional evaluation system encompassing both technical and economic evaluation indicators to comprehensively assess the execution and service capabilities of candidate edge nodes. This avoids evaluation biases caused by relying solely on a single performance indicator, improving the comprehensiveness and accuracy of edge node selection. Simultaneously, task and environment correction amounts are determined based on the task characteristic parameters of the task to be scheduled and the resource operating status of the candidate node set. These correction amounts are then used to dynamically adjust the basic fusion weights, allowing the evaluation ratio between technical and economic scores to adaptively adjust with changes in task requirements and network environment. This improves the matching degree between the target node and the current task requirements and operating environment. Furthermore, this application uses dynamic fusion weights to comprehensively score candidate nodes and determines the target node based on the comprehensive score. This ensures that the edge node scheduling process considers task execution performance, resource utilization, and service costs, enhancing the flexibility and effectiveness of edge computing task scheduling.

[0018] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0020] Figure 1 This is a flowchart illustrating an edge node scheduling method based on multidimensional evaluation and dynamic weights provided in one embodiment of this application; Figure 2 This is a schematic diagram of a process for adjusting the dynamic fusion weights after obtaining them, according to one embodiment of this application. Figure 3This is a flowchart illustrating a process of normalizing and weighting the technical evaluation indicators of each candidate node to obtain the technical score of each candidate node, according to an embodiment of this application. Figure 4 This is a flowchart illustrating a process of normalizing and weighting the economic evaluation indicators of each candidate node to obtain the economic score of each candidate node, according to one embodiment of this application. Detailed Implementation

[0021] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0022] Example 1 Figure 1 This is a flowchart illustrating an edge node scheduling method based on multidimensional evaluation and dynamic weights, provided in one embodiment of this application. (Refer to...) Figure 1 An edge node scheduling method based on multidimensional evaluation and dynamic weights includes: S11. Obtain the tasks to be scheduled and determine the task resource requirements and service quality requirements of the tasks to be scheduled. S12. Select edge nodes that meet the task resource requirements and service quality requirements from the edge nodes that are online, and form a candidate node set; S13. Based on the preset evaluation index system, obtain the technical evaluation index and economic evaluation index of each candidate node; S14. Normalize and weight the technical and economic evaluation indicators of each candidate node to obtain the technical and economic scores of each candidate node. S15. Determine the task characteristic parameters and operating environment parameters based on the task requirement characteristics of the task to be scheduled and the resource operation status of the candidate node set, respectively. S16. Determine the task correction amount and environment correction amount based on the task characteristic parameters and the operating environment parameters, respectively; S17. Adjust the pre-set basic fusion weights according to the task correction amount and the environment correction amount to obtain dynamic fusion weights; S18. Based on the dynamic fusion weight, fuse the technical score and economic score of each candidate node to obtain the comprehensive score of each candidate node; S19. Determine the target node based on the comprehensive score of each candidate node, and after confirming that the target node is in an available state, send the task to be scheduled to the target node.

[0023] This embodiment provides an edge node scheduling method based on multidimensional evaluation and dynamic weights, which is applicable to distributed computing environments consisting of multiple edge nodes.

[0024] Specifically, the task to be scheduled is submitted to the scheduler by the user terminal or business system. The scheduler parses the task to be scheduled, determines the computing resources, storage resources and network transmission resources required during its execution, and determines the quality of service requirements based on the task execution requirements.

[0025] Quality of Service (QoS) requirements define the network transmission performance or task execution performance that must be met during task execution, such as maximum allowable latency, minimum available bandwidth, or task execution reliability requirements. The scheduler obtains the online status and resource operation status of the current edge nodes from the resource management module, retaining only edge nodes that are online and can meet the task resource requirements and QoS requirements, forming a candidate node set. This prevents edge nodes that clearly do not meet the task execution conditions from entering the subsequent evaluation process.

[0026] For the candidate node set, the scheduler obtains technical and economic evaluation indicators for each candidate node according to a preset evaluation indicator system. Technical evaluation indicators reflect the network transmission capacity and service stability of the candidate node when executing the scheduled task; in one implementation, these may include estimated latency, available bandwidth, and node reliability. Economic evaluation indicators reflect the cost status, trustworthiness, and fulfillment commitment of the candidate node when providing task services; in one implementation, these may include service price, credit score, and staking gain determined based on staking status. Technical evaluation indicators can be obtained based on the node's real-time network status, historical task execution records, and current probe results, while economic evaluation indicators can be obtained based on the service information submitted by the node and the corresponding blockchain records.

[0027] Because different evaluation indicators have different dimensions and value ranges, the scheduler normalizes each indicator before scoring, converting them to a unified evaluation range. For indicators where increased values ​​indicate improved node performance, forward normalization is used; for indicators where decreased values ​​indicate improved node performance, reverse normalization is used. After normalization, each technical evaluation indicator is weighted according to pre-configured technical indicator weights to obtain the technical score of the corresponding candidate node. Similarly, each economic evaluation indicator is weighted according to pre-configured economic indicator weights to obtain the economic score of the corresponding candidate node. For example, in a specific configuration, the technical indicator weights for latency, bandwidth, and reliability can be set to 0.4, 0.3, and 0.3, respectively, while the economic indicator weights for cost-effectiveness, credit, and staking gain can be set to 0.5, 0.3, and 0.2, respectively. The above weights are only used to illustrate one feasible configuration method and can be pre-configured according to the actual edge computing environment.

[0028] The scheduler further determines task characteristic parameters based on the task requirements of the tasks to be scheduled, and determines runtime environment parameters based on the resource operating status of the candidate node set. Task characteristic parameters are used to convert the different task requirements for node execution performance and service costs into parameters that can participate in weight adjustment. In one implementation, these parameters may include latency sensitivity, bandwidth requirement, cost sensitivity, and reliability requirement. Runtime environment parameters reflect the overall resource status of the current candidate node set. In one implementation, the global load pressure can be determined based on the resource occupancy status of each candidate node, and resource sufficiency can be determined based on the matching degree between the currently available resources of the candidate nodes and the resources required by the tasks to be scheduled. Thus, task characteristic parameters reflect the differences in the requirements of the tasks to be scheduled, while runtime environment parameters reflect the resource changes in the current edge computing environment. Both types of parameters serve as the basis for dynamically adjusting the fusion weights.

[0029] The scheduler can set basic fusion weight templates for common task types and load them according to the task type after receiving a task. This template is used to provide an initial decision-making starting point, and can be further modified later based on task characteristics and environmental conditions.

[0030] Example table of basic fusion weights:

[0031] The differences between the weights of technical indicators and economic indicators are shown in the table below:

[0032] After obtaining the corresponding basic fusion weight based on the current task type, the scheduler determines the task correction amount based on the task characteristic parameters and the environment correction amount based on the runtime environment parameters.

[0033] The task adjustment amount reflects the direction and degree of adjustment required for the current task to be scheduled relative to the basic fusion weight. For example, when a task has a high degree of latency sensitivity or reliability requirements, the task adjustment amount increases the impact of the technical score in the subsequent comprehensive evaluation; when a task has a high degree of cost sensitivity, the task adjustment amount increases the impact of the economic score. The environment adjustment amount reflects the impact of the overall resource status of candidate nodes on the node selection strategy. When the overall load of candidate nodes increases or the resource sufficiency decreases, the influence of technical performance on the node selection process is increased to reduce the possibility of selecting nodes with insufficient technical suitability under resource constraints. The scheduler uses the task adjustment amount and the environment adjustment amount to adjust the basic fusion weight to obtain a dynamic fusion weight corresponding to the current task to be scheduled and the current resource operating status.

[0034] It should be noted that after obtaining the dynamic fusion weights, the scheduler merges the technical and economic scores of each candidate node according to the dynamic fusion weights to obtain the comprehensive score of the corresponding candidate node. The dynamic fusion weights are used to determine the relative influence of the technical and economic scores in the comprehensive evaluation. Therefore, even if the technical and economic scores of the same candidate node remain unchanged under different scheduled tasks or different resource operating states, its comprehensive score can change with the current scheduling conditions. The scheduler compares the comprehensive scores of each candidate node and determines the candidate node whose comprehensive score meets the current node selection requirements as the target node. Before formally issuing the scheduled task, the online status of the target node and whether the resources required to execute the scheduled task are available are confirmed again. After confirming that the target node is available, the scheduled task and the corresponding task execution information are issued to the target node, which then executes the scheduled task.

[0035] For example, for scheduling tasks with high real-time requirements, their characteristic parameters can be characterized by high latency sensitivity and high reliability requirements. When the candidate node set is simultaneously under high load and available resources are scarce, both task adjustment and environment adjustment can adjust the dynamic fusion weights towards the technical evaluation side, allowing candidate nodes with low network latency, high matching degree between available resources and task requirements, and stable operation to obtain higher comprehensive evaluation results. For scheduling tasks with lower real-time requirements but higher cost sensitivity, when candidate resources are relatively abundant, the dynamic fusion weights can correspondingly increase the influence of the economic evaluation side, ensuring that the node scheduling results take into account service costs while meeting basic resource requirements and service quality requirements. Therefore, the same evaluation system can adapt to different types of scheduling tasks and different resource operating environments, without the need to reconstruct independent node scheduling rules for each task.

[0036] This technical solution constructs a multi-dimensional evaluation system encompassing both technical and economic evaluation indicators to comprehensively assess the execution and service capabilities of candidate edge nodes. This avoids evaluation biases caused by relying solely on a single performance indicator, improving the comprehensiveness and accuracy of edge node selection. Simultaneously, task and environment corrections are determined based on the task's characteristic parameters and the resource operating status of the candidate node set. These corrections are then used to dynamically adjust the basic fusion weights, allowing the evaluation ratio between technical and economic scores to adaptively adjust with changes in task requirements and network environment. This enhances the match between the target node and the current task requirements and operating environment. Furthermore, this application uses dynamic fusion weights to comprehensively score candidate nodes and determines the target node based on the comprehensive score. This ensures that the edge node scheduling process considers task execution performance, resource utilization, and service costs, improving the flexibility and effectiveness of edge computing task scheduling.

[0037] Example 2 Figure 2 This is a flowchart illustrating the process of adjusting the dynamic fusion weights after obtaining them, as provided in one embodiment of this application. Figure 2 The methods also include: S21. Obtain the preset fusion weight range and compare the dynamic fusion weight with the fusion weight range; S22. When the dynamic fusion weight is lower than the lower limit of the fusion weight range, adjust the dynamic fusion weight to the lower limit of the fusion weight range. S23. When the dynamic fusion weight is higher than the upper limit of the fusion weight range, adjust the dynamic fusion weight to the upper limit of the fusion weight range. S24. When the dynamic fusion weight is within the range of the fusion weight, the dynamic fusion weight is output directly. S25. Obtain historical dynamic fusion weights; S26. According to the preset smoothing weights, perform weighted processing on the dynamic fusion weights after range adjustment and the historical dynamic fusion weights, and output the smoothed dynamic fusion weights.

[0038] In this embodiment, after adjusting the basic fusion weights according to the task correction amount and the environment correction amount to obtain the dynamic fusion weights, the dynamic fusion weights are subject to range constraints in order to avoid the technical score or economic score from being overly biased towards a single evaluation dimension in the comprehensive score due to significant changes in task requirements or the resource operation status of the candidate node set.

[0039] In practice, a fusion weight range is pre-defined, which includes a lower limit and an upper limit. The fusion weight range can be pre-configured according to the operational requirements of the edge node scheduling system. In one implementation, the lower limit of the fusion weight is set to 0.2, and the upper limit of the fusion weight is set to 0.9.

[0040] Specifically, the dynamic fusion weights obtained based on task characteristic parameters and operating environment parameters are compared with the fusion weight range. When the dynamic fusion weight is lower than the lower limit, it is adjusted to the lower limit; when the dynamic fusion weight is higher than the upper limit, it is adjusted to the upper limit; when the dynamic fusion weight is between the lower and upper limits, no range adjustment is made, and the current dynamic fusion weight is retained. Therefore, even if the task to be scheduled has strong technical performance or cost preferences, the technical score and economic score can still participate in the comprehensive evaluation of candidate nodes, preventing one type of evaluation result from losing its practical value due to an excessively low fusion ratio.

[0041] Furthermore, to reduce the sudden changes in dynamic fusion weights caused by minor variations in task characteristic parameters or operating environment parameters during continuous scheduling, the dynamic fusion weights after range adjustment are smoothed.

[0042] The historical dynamic fusion weight is the dynamic fusion weight that has been dynamically adjusted in the previous scheduling cycle and actually used for the fusion of technical score and economic score. The scheduler saves the dynamic fusion weight after completing each scheduling cycle and reads it when performing smoothing in the next scheduling cycle.

[0043] In one implementation, a smoothing weight of 0.7 is assigned to the dynamic fusion weight of the current scheduling cycle after range adjustment, and a smoothing weight of 0.3 is assigned to the historical dynamic fusion weight. The two are then weighted, and the result is used as the smoothed dynamic fusion weight. The above smoothing weights can be pre-configured according to the rate of change of resource status in the edge computing environment, and their sum is kept to 1.

[0044] When the system performs edge node scheduling for the first time and there are no historical dynamic fusion weights, the dynamic fusion weights of the current scheduling period, adjusted for range, can be directly used as the dynamic fusion weights for this period. These weights are then saved after the current scheduling period is completed, so they can be used as historical dynamic fusion weights in the next scheduling period. Since the historical dynamic fusion weights are already constrained by the fusion weight range, when using the current dynamic fusion weights and historical dynamic fusion weights for weighted smoothing, the weight state of the previous scheduling period can be introduced while maintaining the influence of current task requirements and resource operating status, thus reducing the magnitude of weight changes between adjacent scheduling periods.

[0045] The technical solution in this embodiment limits the boundaries of dynamic fusion weights by limiting the range of fusion weights, which can prevent local anomalies in task characteristics or resource operation status from causing the comprehensive score to be overly biased towards technical or economic evaluation. On this basis, historical dynamic fusion weights are further introduced for cross-cycle smoothing, so that the dynamic fusion weights can respond to the current task requirements and resource status changes, and will not change frequently and significantly due to short-term state fluctuations. This improves the stability of the comprehensive score results during continuous edge node scheduling and reduces the possibility of target nodes frequently switching due to evaluation weight oscillations.

[0046] Example 3 In some embodiments, the technical evaluation indicators and economic evaluation indicators of each candidate node are obtained, including: Send probe requests to each candidate node to obtain network round-trip latency, and combine the historical end-to-end latency of the corresponding candidate node to determine the estimated latency; Obtain the current available network bandwidth reported by each candidate node to get the available bandwidth; Based on the historical task completion records of each candidate node, determine the node reliability of the corresponding candidate node; Estimated latency, available bandwidth, and node reliability are used as technical evaluation indicators. Obtain the service quotes submitted by each candidate node for the task to be scheduled; Query the blockchain records corresponding to each candidate node to obtain the credit score and the amount of collateral; The pledge gain is determined based on the amount pledged. Service pricing, credit score, and collateral gain are used as economic evaluation indicators.

[0047] In this embodiment, after the scheduler forms a set of candidate nodes, it obtains the technical evaluation indicators and economic evaluation indicators corresponding to each candidate node according to a preset evaluation index system.

[0048] Among them, the technical evaluation indicators are used to characterize the network transmission performance and historical operational stability of candidate nodes when executing scheduled tasks, including estimated latency, available bandwidth and node reliability; the economic evaluation indicators are used to characterize the cost level, historical reliability and performance commitment of candidate nodes when providing task services, including service price, credit score and staking gain.

[0049] Since the data sources, dimensions, and numerical ranges of each evaluation indicator are different, they are normalized after the indicators are collected to convert different evaluation indicators into evaluation values ​​that can be uniformly weighted and calculated.

[0050] Specifically, for estimated latency, the scheduler sends lightweight probe requests to each candidate node to obtain the current network round-trip time. Simultaneously, it reads the end-to-end latency records generated by the candidate node during historical task execution and combines the current probe results with historical latency status to determine the estimated latency. By utilizing both current network probe results and historical task execution data, the impact of a single network jitter on latency evaluation results can be reduced. For available bandwidth, candidate nodes obtain the network bandwidth available for task data transmission based on the current network connection status and report it to the scheduler. Nodes can report according to a preset period or re-report when the network status changes. For node reliability, it is determined based on the corresponding candidate node's task completion records within a preset historical task window. The task completion records must at least reflect the successful completion of tasks and can be combined with failure or timeout information to determine the stability of the candidate node's task execution during historical operation.

[0051] For economic evaluation indicators, service quotes are actively submitted by candidate nodes for tasks to be scheduled. The scheduler obtains the currently valid service quotes of candidate nodes when executing the current node scheduling. Credit scores are obtained based on the blockchain credit records of candidate nodes. These credit records are formed based on the historical performance of candidate nodes and reflect the credibility of candidate nodes in providing task services in the past. The staking amount is also obtained by querying the blockchain staking records of the corresponding candidate nodes, and the staking gain is further determined based on the staking amount, so that the performance commitments formed by the nodes through staking can participate in the economic evaluation. For blockchain records, while ensuring data validity, a short-term caching method can be used to reduce the communication overhead caused by repeated queries, but the corresponding record will be retrieved again after the cache expires or the corresponding on-chain state is updated.

[0052] Based on this, the technical evaluation indicators of each candidate node are normalized and weighted to obtain the technical score of each candidate node, which is then used as a reference. Figure 3 ,include: S1411. Determine the maximum and minimum values ​​corresponding to each technical evaluation indicator according to the preset statistical range; S1412. For the estimated delay in the technical evaluation indicators, reverse normalization is performed according to the principle that the smaller the delay value, the higher the evaluation result, to obtain the delay evaluation value. S1413. For the available bandwidth in the technical evaluation indicators, perform positive normalization according to the principle that the larger the bandwidth value, the higher the evaluation result, to obtain the bandwidth evaluation value. S1414. Determine the reliability evaluation value based on the historical task completion status corresponding to the node reliability. S1415. Based on the preset technical indicator weights, the latency evaluation value, bandwidth evaluation value, and reliability evaluation value are weighted to obtain the technical score of each candidate node.

[0053] When calculating the technical score, the maximum and minimum values ​​for technical evaluation indicators such as estimated latency and available bandwidth are determined based on a preset statistical range. The preset statistical range can be determined based on recent task samples or historical task samples of the same type as the currently scheduled task, and can be updated as the network status of edge nodes changes.

[0054] For the estimated latency, a reverse normalization method is used, that is, the lower the estimated latency of the candidate node, the higher the corresponding latency evaluation value.

[0055] For available bandwidth, a positive normalization method is used, meaning that the higher the available bandwidth, the higher the corresponding bandwidth evaluation value.

[0056] For node reliability, the reliability evaluation value can be determined based on the ratio between the number of successfully completed tasks within a preset historical task window and the total number of historical tasks. For example, the most recent 100 tasks can be used as the historical task window, but this number can be adjusted according to the actual node task execution frequency.

[0057] After normalizing each technical evaluation index, the delay evaluation value, bandwidth evaluation value, and reliability evaluation value are assigned corresponding weights according to the preset technical index weights, and the three are weighted to obtain the technical score of each candidate node.

[0058] In one implementation, the weights of the technical indicators corresponding to the latency evaluation value, bandwidth evaluation value, and reliability evaluation value can be set to 0.4, 0.3, and 0.3 respectively. The sum of the technical indicator weights remains at 1, ensuring that the technical score reflects the comprehensive technical performance of the candidate node under a unified evaluation scale. The above weights represent a specific configuration and can be pre-configured in practice according to the different latency, bandwidth, and reliability requirements of the task to be scheduled.

[0059] Based on this, the economic evaluation indicators of each candidate node are normalized and weighted to obtain the economic score of each candidate node, which is then used as a reference. Figure 4 ,include: S1421. Determine the cost-benefit evaluation value based on the service price of each candidate node, and normalize the cost-benefit evaluation value according to the principle that the lower the service price, the higher the cost-benefit evaluation value. S1422. Obtain the credit score corresponding to each candidate node, and normalize the credit score according to the preset credit evaluation range to obtain the credit evaluation value. S1423. Obtain the staking amount corresponding to each candidate node, perform logarithmic mapping based on the staking amount to obtain the staking gain, and normalize the staking gain to obtain the staking evaluation value. S1424. Based on the preset economic indicator weights, the cost-benefit evaluation value, credit evaluation value, and pledge evaluation value are weighted to obtain the economic score of each candidate node.

[0060] When calculating the economic score, the cost-benefit evaluation value is determined based on the service price of the candidate nodes, and the lower the service price, the higher the cost-benefit evaluation value.

[0061] Credit scores are converted into a unified credit rating value according to a preset credit rating range. For example, when blockchain credit records use a rating range of 0 to 100, they are converted into a unified normalized interval.

[0062] For staking gain, a logarithmic mapping is applied to the staking quantity of candidate nodes so that the staking gain increases as the staking quantity increases, while the gain generated by increasing the unit staking quantity gradually decreases. This avoids candidate nodes with excessively large staking quantities from having an excessive impact on the economic score solely based on the staking scale. Furthermore, the staking gain is normalized according to a preset maximum reference staking amount to obtain the staking evaluation value.

[0063] After completing the above processing, the cost-benefit evaluation value, credit evaluation value, and pledge evaluation value are weighted according to preset economic indicator weights to obtain the economic score of each candidate node. In one embodiment, the economic indicator weights corresponding to the cost-benefit evaluation value, credit evaluation value, and pledge evaluation value can be set to 0.5, 0.3, and 0.2 respectively, and the sum of the weights of each economic indicator remains at 1.

[0064] Therefore, the economic score is not determined solely by the service price of candidate nodes, but is evaluated by combining the service cost of candidate nodes, the reliability of historical performance, and the degree of performance commitment formed by staking, providing a unified quantitative basis for the dynamic integration of subsequent technical and economic scores.

[0065] The technical solution in this embodiment establishes separate data acquisition and normalization processing paths for technical and economic evaluation indicators. This allows node status data from different sources and with different dimensions to be converted into technical and economic scores with a unified evaluation scale. Specifically, using current network probing data and historical task data to jointly evaluate node technical performance reduces evaluation bias caused by instantaneous network fluctuations. Incorporating service pricing, credit status, and staking gains into the economic evaluation avoids the over-pricing of low-priced nodes due to relying solely on service prices. In particular, by applying a logarithmic mapping to the staking quantity, the staking gain exhibits diminishing marginal returns. This reflects the node's commitment to fulfillment while limiting the excessive influence of a single high staking amount on the node ranking results, providing a stable and discriminative scoring basis for subsequent comprehensive evaluation of candidate nodes based on dynamic fusion weights.

[0066] Example 4 In some embodiments, task characteristic parameters and runtime environment parameters are determined based on the task requirement characteristics of the task to be scheduled and the resource operation status of the candidate node set, including: Get the task type corresponding to the task to be scheduled; The latency sensitivity, bandwidth requirement, cost sensitivity, and reliability requirement of the task to be scheduled are determined based on the task type, task resource requirements, and service quality requirements. The latency sensitivity, bandwidth requirement, cost sensitivity, and reliability requirement are standardized to obtain task characteristic parameters; Obtain the resource occupancy status and available resource status of each candidate node in the candidate node set; The global load pressure is determined based on the resource occupancy status of each candidate node in the candidate node set; The resource sufficiency is determined based on the available resource status of each candidate node in the candidate node set and the task resource requirements of the task to be scheduled. The operating environment parameters are determined based on the overall load pressure and resource availability.

[0067] Furthermore, task correction and environment correction are determined based on task characteristic parameters and operating environment parameters, respectively. The basic fusion weights are then adjusted based on these corrections to obtain dynamic fusion weights, including: The basic fusion weights are determined based on the correlation between latency sensitivity, bandwidth requirement, cost sensitivity, and reliability requirement in the task characteristic parameters. Based on the task characteristics that meet the preset preference conditions in the task characteristic parameters, determine the task correction amount; when the latency sensitivity or reliability requirement of the task to be scheduled reaches the preset conditions, increase the fusion ratio corresponding to the technical score; when the cost sensitivity of the task to be scheduled reaches the preset conditions, increase the fusion ratio corresponding to the economic score. The environmental correction amount is determined based on the global load pressure and resource adequacy in the operating environment parameters; when the global load pressure corresponding to the candidate node set increases or the resource adequacy decreases, the fusion ratio corresponding to the technical score is increased; The basic fusion weights are adjusted based on the task correction amount and the environment correction amount to obtain the dynamic fusion weights.

[0068] This embodiment further explains the process of determining task characteristic parameters, operating environment parameters, and dynamic fusion weights. Based on the formation of the candidate node set and the completion of technical and economic evaluations of the candidate nodes, the fusion relationship between technical and economic scores is dynamically adjusted according to the differences in the needs of the tasks to be scheduled and the current resource operating status of the candidate node set. Specifically, task characteristic parameters characterize the scheduling task's demand tendencies in terms of latency, bandwidth, cost, and reliability; operating environment parameters characterize the current overall load level of the candidate node set and the sufficiency of resources available to execute the scheduling tasks. By establishing adjustment criteria from both the task side and the resource environment side, the fusion weights can change with actual scheduling conditions.

[0069] When determining the task characteristic parameters, first obtain the task type corresponding to the task to be scheduled, and then determine the latency sensitivity, bandwidth requirement, cost sensitivity and reliability requirement based on the task resource requirements and service quality requirements.

[0070] Delay sensitivity reflects the tolerance of a scheduled task for end-to-end response latency. It can be determined based on the maximum allowable latency in the service quality requirements. The smaller the maximum allowable latency, the higher the corresponding delay sensitivity. Its value ranges from 0 to 1, where 1 indicates a need for extremely low latency, such as less than 10ms; and 0 indicates a tolerance for higher latency, such as greater than 500ms. It can be calculated based on the maximum delay constraint in the task's SLA.

[0071] Bandwidth requirement reflects the data transmission capacity needs of the scheduled task and can be determined based on the amount of data to be transmitted, the task duration, and the minimum bandwidth requirement. Its value ranges from 0 to 1, where 1 indicates a requirement of Gbps bandwidth and 0 indicates a requirement of only Kbps bandwidth. It can be calculated based on the expected data volume and duration in the task description.

[0072] Cost sensitivity reflects the degree to which a scheduled task is constrained by execution costs, and can be determined based on the task budget and the cost requirements of the corresponding task type; its value ranges from 0 to 1, where 1 indicates a strictly limited budget and 0 indicates no price sensitivity. It can be set by the user or inferred from historical behavior.

[0073] The reliability requirement level reflects the degree to which a task depends on successful completion, and can be determined based on the number of failures allowed, retry requirements, or successful execution requirements. The requirement levels with different value ranges are standardized to a unified evaluation interval, resulting in task characteristic parameters; these parameters range from 0 to 1, where 1 indicates that successful completion is mandatory, and 0 indicates that retries or failures are allowed.

[0074] Let d, b, c, and r represent latency sensitivity, bandwidth requirement, cost sensitivity, and reliability requirement, respectively. The following table illustrates the values ​​of each task characteristic parameter for different task types:

[0075] For runtime environment parameters, the current resource occupancy and available resource status of each candidate node in the candidate node set are obtained. The resource occupancy status reflects the computing resources, network resources, and other task execution resources currently occupied by other tasks on the candidate node. The resource occupancy status of each candidate node is summarized to determine the global load pressure corresponding to the candidate node set. The global load pressure ranges from 0 to 1. The higher the global load pressure, the smaller the overall resource reserve of the candidate node set that can be used for new tasks.

[0076] The resource sufficiency is further determined based on the matching degree between the available resource status of each candidate node and the resource requirements of the task to be scheduled. The resource sufficiency value ranges from 0 to 1. When the number of available resources that can meet the resource requirements of the task is significantly higher than the resources required by the task to be scheduled, the resource sufficiency increases; when the number of candidate resources that can meet the resource requirements of the task decreases, the resource sufficiency decreases. The operating environment parameters are formed by combining the global load pressure and the resource sufficiency.

[0077] Furthermore, basic fusion weight configuration rules can be pre-established, so that different combinations of task characteristic parameters correspond to corresponding basic fusion weights. Based on the correlation between the latency sensitivity, bandwidth requirement, cost sensitivity, and reliability requirement of the current task to be scheduled, the basic fusion weight corresponding to the current task to be scheduled is determined from the pre-configured rules. When the overall latency sensitivity, bandwidth requirement, and reliability requirement are high, while the cost sensitivity is relatively low, the basic fusion weight is increased towards the technical score side; when the cost sensitivity is relatively high and the requirements for latency, bandwidth, and reliability are relatively low, the basic fusion weight is adjusted towards the economic score side. Thus, the basic fusion weight is used to reflect the overall requirement structure of the task to be scheduled, rather than being determined solely based on a single task characteristic.

[0078] Based on the basic fusion weights, the task correction amount is determined according to whether there are task features among the task feature parameters that meet preset preference conditions. Preset preference conditions are used to identify situations where the task to be scheduled has a significant preference for a certain evaluation direction. In one implementation, when the latency sensitivity or reliability requirement exceeds the corresponding preset threshold, a task correction amount is determined that increases the fusion ratio corresponding to the technical score; when the cost sensitivity exceeds the corresponding preset threshold, a task correction amount is determined that increases the fusion ratio corresponding to the economic score; when none of the task features meet the corresponding preset preference conditions, no additional task-side correction is required. Through this method, the basic fusion weights are used to reflect the overall task demand relationship, while the task correction amount is used to compensate for tasks with prominent demand characteristics; the roles of the two are distinct.

[0079] Furthermore, the environmental correction amount is determined based on the global load pressure and resource sufficiency parameters in the operating environment. When the global load pressure is low and the resource sufficiency is high, the candidate node set has a large selection margin, and the environmental correction amount can be kept at a low level. As the global load pressure increases or the resource sufficiency decreases, the environmental correction amount can be increased accordingly to improve the technical score fusion ratio, so that candidate nodes with higher network performance, service stability, and resource adaptability are given priority under resource constraints. In one specific implementation, the environmental correction amount can be determined jointly based on the degree of decrease in resource sufficiency and the degree of increase in global load pressure, so that the correction effect on the technical score side is more obvious when the global load pressure is higher and the resources are more scarce.

[0080] After determining the task correction and environment correction amounts, they are used to adjust the basic fusion weights corresponding to the currently scheduled tasks, resulting in dynamic fusion weights. Specifically, the correction amount that increases the fusion ratio of technical scores is used to enhance the influence of technical scores in the overall score of candidate nodes, while the correction amount that increases the fusion ratio of economic scores is used to correspondingly enhance the influence of economic scores. Therefore, when task requirements change or the overall resource status of candidate nodes changes, without altering the technical and economic evaluation indicators themselves, the same node evaluation framework can be adapted to different scheduling scenarios by adjusting the fusion relationship between the two types of scores.

[0081] The technical solution in this embodiment applies the overall demand relationship of the tasks to be scheduled, the prominent demand preferences, and the real-time resource status of the candidate node set to different adjustment stages of the fusion weight. This allows the dynamic fusion weight to reflect both the differences in task type and task demand, and to respond to changes in resource load in the edge computing environment. In particular, by determining the basic fusion weight, task correction amount, and environment correction amount in a hierarchical manner, the rigidity of the scheduling strategy caused by setting fixed weights solely based on task type can be avoided. Furthermore, the impact of technical performance on node selection can be appropriately increased when resources are scarce, thereby improving the ability of the comprehensive score to distinguish between current task demands and resource environment, and enhancing the adaptability of target node selection and scheduling stability.

[0082] Example 5 In some embodiments, the method further includes: The candidate node set is sorted according to the comprehensive score of each candidate node, the candidate node with the highest comprehensive score is determined as the target node, and the candidate nodes whose comprehensive score ranks within the preset ranking range are determined as backup nodes. Before sending the task to be scheduled to the target node, the node status information of the target node is retrieved again, and the online status and resource availability status of the target node are reviewed. When the target node is online and meets the resource requirements of the task to be scheduled, the task to be scheduled will be sent to the target node. When the target node does not meet the execution conditions of the task to be scheduled, the target node is re-determined based on the comprehensive score ranking of the backup nodes, and the task to be scheduled is sent to the re-determined target node.

[0083] This embodiment describes the availability verification after the target node is determined and the process of switching to a backup node.

[0084] After calculating the comprehensive score of each candidate node, the scheduler sorts the candidate node set according to the comprehensive score, determines the candidate node with the highest comprehensive score as the target node, and selects candidate nodes whose comprehensive score ranking falls within a preset ranking range from the remaining candidate nodes as backup nodes. The preset ranking range can be pre-set according to the number of candidate nodes, the importance of the task, and the system fault tolerance requirements. For example, when there are many candidate nodes, multiple backup nodes are reserved so that a quick switch can be completed when the status of the target node changes.

[0085] Specifically, since the comprehensive score of candidate nodes is formed at the node evaluation time, and the online status and resource usage of nodes may have changed by the time the task is officially issued, the scheduler re-acquires the node status information of the target node before issuing the task to be scheduled to the target node, and verifies the online status and resource availability status of the target node. The online status is used to confirm whether the target node can currently establish a valid communication connection with the scheduler; the resource availability status is used to confirm whether the remaining computing resources, storage resources, and network transmission resources of the target node can still meet the task resource requirements of the task to be scheduled. When the target node is online and its resource availability status meets the task resource requirements, the target node is confirmed to be in an available state, and the task to be scheduled is issued to that target node for execution.

[0086] If the verification result indicates that the target node is offline, or that the target node's remaining resources are insufficient to meet the task resource requirements of the scheduled task, then the target node is determined not to meet the execution conditions of the scheduled task. The scheduler does not re-perform the complete scoring process for all candidate nodes. Instead, based on the existing comprehensive score ranking of the backup nodes, it re-determines the target node from the backup nodes and performs the same online status and resource availability status verification on the re-determined target node. If the re-determined target node meets the execution conditions, the scheduled task is assigned to that target node; if it still does not meet the execution conditions, the next backup node is selected according to its comprehensive score ranking, until a target node that meets the execution conditions is determined.

[0087] In one specific implementation, the scheduler can save the candidate node ranking results after the candidate node scoring is completed and set a corresponding backup node list. During the task distribution phase, only the status of the currently selected nodes is reviewed, thereby reducing the processing overhead caused by re-executing all technical evaluations, economic evaluations, and dynamic fusion calculations due to short-term node failures. When there are no backup nodes in the candidate node set that meet the execution conditions, the current distribution process can be terminated, and the candidate node screening and scoring process can be re-triggered to avoid assigning tasks to edge nodes that cannot execute normally.

[0088] The technical solution in this embodiment further incorporates availability verification and backup node switching mechanisms after the target node is determined through comprehensive scoring. This compensates for scheduling deviations caused by changes in node status between the node scoring time and the task issuance time. By utilizing existing comprehensive score ranking results for backup node switching, it eliminates the need to immediately re-execute the entire scoring process when the target node fails. This reduces redundant calculation overhead, shortens node switching time in abnormal situations, and improves the actual task issuance success rate and the continuity of edge node scheduling.

[0089] Example 6 In some embodiments, the method further includes: Obtain task characteristic parameters, runtime environment parameters, technical score, economic score, dynamic fusion weight, and corresponding task execution results during the historical scheduling process; The state information of the reinforcement learning model is constructed based on the task characteristic parameters and the operating environment parameters. The action information of the reinforcement learning model is determined based on the adjustment method of technical score, economic score and dynamic fusion weight. The reward information of the reinforcement learning model is determined based on the task execution result. The reinforcement learning model is trained based on state information, action information, and reward information to obtain a weight optimization model; The weight optimization model is used to optimize and adjust the dynamic fusion weights after taking the task feature parameters, runtime environment parameters and dynamic fusion weights corresponding to the current task to be scheduled as input, and outputs the optimized and adjusted dynamic fusion weights.

[0090] This embodiment describes the process of long-term optimization of dynamic fusion weights based on historical scheduling results. During the continuous operation of the edge node scheduling system, each completed scheduling process is recorded, forming historical scheduling data. This historical scheduling data includes task characteristic parameters for the corresponding task, operating environment parameters corresponding to the candidate node set, technical and economic scores of the candidate nodes, the actual dynamic fusion weights used, and the task execution results. The task execution results reflect the actual execution status after this node scheduling, and may include information such as whether the task was successfully completed, whether service quality requirements were met, and task execution costs, thereby establishing a correspondence between scheduling conditions, weight configuration, and actual execution results.

[0091] Training samples for the reinforcement learning model are constructed based on historical scheduling data. Task characteristic parameters and operating environment parameters from the historical scheduling process are used as state information to describe the task requirements and edge node resource environment when executing corresponding scheduling decisions. Action information is determined based on the adjustments to technical scores, economic scores, and dynamic fusion weights during the corresponding scheduling process, representing the adjustments made to the relative impact on the technical and economic evaluation sides in the current state. Reward information is determined based on the task execution results. Higher rewards are given when the task is successfully completed, the service quality meets preset requirements, and the task execution cost is within a reasonable range; rewards are reduced when the task fails, the service quality does not meet preset requirements, or the task execution cost is too high, so that the reinforcement learning process can distinguish different dynamic fusion weight adjustment methods based on the actual scheduling effect.

[0092] Specifically, historical scheduling data generated within a preset training period can be aggregated, and the reinforcement learning model can be trained using state information, action information, and reward information. In one implementation, historical scheduling data generated in the last 24 hours can be used for batch training; the training period can also be set according to the task scheduling frequency and the rate of historical sample accumulation. During model training, based on the reward results corresponding to each dynamic fusion weight adjustment method under different task feature parameters and operating environment parameters, the correlation between task requirements, resource operating status, and appropriate weight adjustment strategies is gradually established. After training is completed, the training results are verified, and the model that meets the preset verification requirements is used as the weight optimization model for dynamic fusion weight optimization in subsequent scheduling processes.

[0093] When executing a new task to be scheduled, the dynamic fusion weight is first determined based on task characteristic parameters and operating environment parameters according to the aforementioned implementation method. Then, the task characteristic parameters, operating environment parameters, and dynamic fusion weight corresponding to the current task to be scheduled are input into the weight optimization model. The weight optimization model judges the degree of adaptation between the current dynamic fusion weight and the current task and resource environment based on the learned historical scheduling patterns, and optimizes the dynamic fusion weight when adjustments are needed, outputting the optimized and adjusted dynamic fusion weight. The optimized and adjusted dynamic fusion weight is used for the subsequent fusion calculation of technical score and economic score, thereby retaining the basic scheduling mechanism of real-time weight calculation based on task characteristics and operating environment, while further correcting it using actual scheduling results accumulated during long-term operation.

[0094] To avoid instability in weight optimization results due to insufficient training samples, when historical scheduling data has not met the preset training conditions, node scheduling can continue to use dynamically fused weights determined based on task feature parameters and runtime environment parameters. Once the historical scheduling data meets the training conditions and model validation is completed, the weight optimization model can be activated. Subsequently, as new task execution results continue to be generated, the training samples can be updated according to the preset training cycle, and the weight optimization model can be retrained or calibrated to adapt to long-term changes in task type distribution, node resource status, and network environment.

[0095] The technical solution in this embodiment, based on the real-time determination of dynamic fusion weights according to task characteristic parameters and operating environment parameters, further introduces actual task execution results to form a closed-loop feedback. This allows the adjustment of dynamic fusion weights not only to be based on the current task and resource status, but also to be corrected using the execution effects accumulated during long-term scheduling. Therefore, it can reduce the adaptation deviation of preset weight adjustment rules after long-term changes in the operating environment, enabling the fusion relationship between technical scores and economic scores to gradually adapt to the actual task distribution and resource operation characteristics, thereby improving the long-term adaptability of the edge node scheduling strategy.

[0096] It is understood that the same or similar parts in the above embodiments can be referred to each other, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0097] It should be noted that in the description of this application, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this application, unless otherwise stated, "a plurality of" means at least two.

[0098] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the function involved, as will be understood by those skilled in the art to which embodiments of this application pertain.

[0099] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0100] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0101] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0102] The storage media mentioned above can be read-only memory, disk, or optical disk, etc.

[0103] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0104] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. An edge node scheduling method based on multidimensional evaluation and dynamic weights, characterized in that, include: Obtain the tasks to be scheduled and determine their resource requirements and quality of service requirements. From the edge nodes that are online, select edge nodes that meet the task resource requirements and service quality requirements to form a candidate node set; Based on the preset evaluation index system, obtain the technical evaluation index and economic evaluation index of each candidate node. The technical and economic evaluation indicators of each candidate node are normalized and weighted to obtain the technical and economic scores of each candidate node. The task characteristic parameters and operating environment parameters are determined based on the task requirement characteristics of the task to be scheduled and the resource operation status of the candidate node set, respectively. The task correction amount and the environment correction amount are determined based on the task characteristic parameters and the operating environment parameters, respectively. The pre-set basic fusion weights are adjusted based on the task correction amount and the environment correction amount to obtain dynamic fusion weights; The technical and economic scores of each candidate node are fused according to the dynamic fusion weight to obtain the comprehensive score of each candidate node; The target node is determined based on the comprehensive score of each candidate node, and the task to be scheduled is sent to the target node after confirming that the target node is in an available state.

2. The method according to claim 1, characterized in that, After obtaining the dynamic fusion weights, the method further includes: Obtain a pre-defined fusion weight range, and compare the dynamic fusion weight with the fusion weight range; When the dynamic fusion weight is lower than the lower limit of the fusion weight range, the dynamic fusion weight is adjusted to the lower limit of the fusion weight range; When the dynamic fusion weight is higher than the upper limit of the fusion weight range, the dynamic fusion weight is adjusted to the upper limit of the fusion weight range; When the dynamic fusion weight is within the range of the fusion weight, the dynamic fusion weight is directly output.

3. The method according to claim 2, characterized in that, After adjusting the range of the dynamic fusion weights, the method further includes: Obtain historical dynamic fusion weights; According to the preset smoothing weights, the dynamic fusion weights after range adjustment and the historical dynamic fusion weights are weighted and processed to output the smoothed dynamic fusion weights.

4. The method according to claim 1, characterized in that, Obtain the technical and economic evaluation indicators for each candidate node, including: Send probe requests to each candidate node to obtain network round-trip latency, and combine the historical end-to-end latency of the corresponding candidate node to determine the estimated latency; Obtain the current available network bandwidth reported by each candidate node to get the available bandwidth; Based on the historical task completion records of each candidate node, determine the node reliability of the corresponding candidate node; The estimated latency, available bandwidth, and node reliability are used as the technical evaluation indicators. Obtain the service quotes submitted by each candidate node for the task to be scheduled; Query the blockchain records corresponding to each candidate node to obtain the credit score and the amount of collateral; The pledge gain is determined based on the pledged amount; The service price, credit score, and staking gain are used as the economic evaluation indicators.

5. The method according to claim 4, characterized in that, The technical evaluation metrics for each candidate node are normalized and weighted to obtain the technical score for each candidate node, including: The maximum and minimum values ​​corresponding to each technical evaluation indicator are determined according to the preset statistical range; For the estimated delay in the technical evaluation index, reverse normalization is performed according to the principle that the smaller the delay value corresponds to the higher the evaluation result, to obtain the delay evaluation value; For the available bandwidth in the aforementioned technical evaluation indicators, positive normalization is performed according to the principle that a larger bandwidth value corresponds to a higher evaluation result, to obtain the bandwidth evaluation value; The reliability evaluation value is determined based on the historical task completion status corresponding to the reliability of the node. Based on preset technical indicator weights, the latency evaluation value, bandwidth evaluation value, and reliability evaluation value are weighted to obtain the technical score of each candidate node.

6. The method according to claim 4, characterized in that, The economic evaluation indicators of each candidate node are normalized and weighted to obtain the economic score of each candidate node, including: The cost-benefit evaluation value is determined based on the service price of each candidate node, and the cost-benefit evaluation value is normalized according to the principle that the lower the service price, the higher the cost-benefit evaluation value. Obtain the credit score corresponding to each candidate node, and normalize the credit score according to the preset credit evaluation range to obtain the credit evaluation value; Obtain the staking amount corresponding to each candidate node, perform logarithmic mapping processing based on the staking amount to obtain the staking gain, and normalize the staking gain to obtain the staking evaluation value. Based on the preset economic indicator weights, the cost-benefit evaluation value, credit evaluation value, and pledge evaluation value are weighted to obtain the economic score of each candidate node.

7. The method according to claim 1, characterized in that, Based on the task requirement characteristics of the task to be scheduled and the resource operation status of the candidate node set, the task characteristic parameters and operation environment parameters are determined respectively, including: Obtain the task type corresponding to the task to be scheduled; The latency sensitivity, bandwidth requirement, cost sensitivity, and reliability requirement of the task to be scheduled are determined based on the task type, task resource requirements, and service quality requirements. The latency sensitivity, bandwidth requirement, cost sensitivity, and reliability requirement are standardized to obtain the task characteristic parameters; Obtain the resource occupancy status and available resource status of each candidate node in the candidate node set; The global load pressure is determined based on the resource occupancy status of each candidate node in the candidate node set; The resource sufficiency is determined based on the available resource status of each candidate node in the candidate node set and the task resource requirements of the task to be scheduled. The operating environment parameters are determined based on the global load pressure and resource sufficiency.

8. The method according to claim 7, characterized in that, Based on the task characteristic parameters and operating environment parameters, the task correction amount and environment correction amount are determined respectively. The basic fusion weights are then adjusted based on the task correction amount and environment correction amount to obtain dynamic fusion weights, including: The basic fusion weights are determined based on the correlation between latency sensitivity, bandwidth requirement, cost sensitivity, and reliability requirement in the task characteristic parameters. Based on the task features that meet the preset preference conditions in the task feature parameters, determine the task correction amount; when the latency sensitivity or reliability requirement of the task to be scheduled reaches the preset condition, increase the fusion ratio corresponding to the technical score; when the cost sensitivity of the task to be scheduled reaches the preset condition, increase the fusion ratio corresponding to the economic score. The environment correction amount is determined based on the global load pressure and resource adequacy in the operating environment parameters; when the global load pressure corresponding to the candidate node set increases or the resource adequacy decreases, the fusion ratio corresponding to the technical score is increased; The basic fusion weights are adjusted based on the task correction amount and the environment correction amount to obtain the dynamic fusion weights.

9. The method according to claim 1, characterized in that, The method further includes: The candidate node set is sorted according to the comprehensive score of each candidate node, the candidate node with the highest comprehensive score is determined as the target node, and the candidate nodes whose comprehensive score ranks within a preset ranking range are determined as backup nodes. Before sending the task to be scheduled to the target node, the node status information of the target node is reacquired, and the online status and resource availability status of the target node are reviewed. When the target node is online and meets the resource requirements of the task to be scheduled, the task to be scheduled is sent to the target node; When the target node does not meet the execution conditions of the task to be scheduled, the target node is re-determined based on the comprehensive score ranking of the backup nodes, and the task to be scheduled is sent to the re-determined target node.

10. The method according to claim 1, characterized in that, The method further includes: Obtain task characteristic parameters, runtime environment parameters, technical score, economic score, dynamic fusion weight, and corresponding task execution results from the historical scheduling process; The state information of the reinforcement learning model is constructed based on the task feature parameters and the operating environment parameters. The action information of the reinforcement learning model is determined based on the adjustment method of the technical score, economic score and dynamic fusion weight. The reward information of the reinforcement learning model is determined based on the task execution result. The reinforcement learning model is trained based on the state information, action information, and reward information to obtain a weight optimization model; The weight optimization model is used to optimize and adjust the dynamic fusion weight after inputting the task feature parameters, operating environment parameters and dynamic fusion weight corresponding to the current task to be scheduled, and output the optimized and adjusted dynamic fusion weight.

Images