Resource Scheduling Method, Apparatus and Device

Abstract

This application relates to the field of cloud computing technology, and discloses a resource scheduling method, device and equipment. The method includes: generating a multi-dimensional data topology graph of cloud environment monitoring data, decomposing the to-be-scheduled job set into a set of quantum computing units that can be processed in parallel, and adding priority annotations to the set of quantum computing units through a quantum heuristic algorithm. According to the priority annotations and the multi-dimensional data topology graph, the physical resources of the cloud environment are divided into non-uniform resource shards that match the task requirements. Based on the non-uniform resource shards, resource scheduling is performed through an improved hybrid particle swarm optimization algorithm, and weighted sum fitness evaluation and resource constraint correction are introduced. This application presents the resource status and associations comprehensively by generating a multi-dimensional data topology graph, then decomposes the to-be-scheduled job set into a set of quantum computing units and adds priority annotations, and then divides non-uniform resource shards for resource scheduling, improving the accuracy of resource adaptation, the scheduling efficiency and the resource utilization rate.

Description

Technical Field

[0001] This application relates to the field of cloud computing technology, and particularly to a resource scheduling method, apparatus, and device. Background Art

[0002] With the accelerating advancement of the construction of smart cities, the demand for real-time data processing in the field of intelligent transportation has shown an explosive growth. In the traffic management scenario of the cloud platform, due to the characteristics of high concurrency and low latency in services such as traffic flow monitoring, violation recognition, and path planning, and the significant fluctuations in service loads in different time periods / regions, a dynamic and accurate resource scheduling mechanism is urgently needed to ensure the stable operation of the system.

[0003] However, traditional scheduling algorithms have the problem of insufficient matching between resource allocation and task requirements. Especially in emerging scenarios such as new energy vehicle networking, due to the lack of coordinated optimization of task priorities and resource constraints, there are often contradictory phenomena where high-priority analysis tasks are delayed due to resource fragmentation, while low-priority batch processing tasks occupy excessive resources, resulting in low cluster resource utilization and resource scheduling efficiency, seriously restricting the real-time response ability of the intelligent transportation system. Summary of the Invention

[0004] The main purpose of this application is to provide a resource scheduling method, apparatus, and device, aiming to solve the technical problem in the prior art that due to inaccurate scheduling evaluation, resources cannot be accurately adapted to tasks, resulting in low resource scheduling efficiency.

[0005] To achieve the above object, this application proposes a resource scheduling method, and the method includes:

[0006] Generate a multi-dimensional data topology map of cloud environment monitoring data;

[0007] Decompose the job set to be scheduled into a set of quantum computing units that can be processed in parallel, and add priority labels to the set of quantum computing units through a quantum-inspired algorithm;

[0008] According to the priority label and the multi-dimensional data topology map, divide the physical resources of the cloud environment into non-uniform resource slices that match the task requirements;

[0009] Based on the non-uniform resource slices, perform resource scheduling through an improved hybrid particle swarm optimization algorithm, where the improved hybrid particle swarm optimization algorithm is an algorithm that introduces weighted sum fitness evaluation and resource constraint correction.

[0010] In one embodiment, the step of performing resource scheduling through an improved hybrid particle swarm optimization algorithm based on the non-uniform resource slices includes:

[0011] Map each resource shard of the non-uniform resource sharding into a particle in the particle swarm, and obtain an initial particle swarm by attaching a shard real-time status identifier to each particle;

[0012] Iteratively optimize the initial particle swarm through the improved hybrid particle swarm optimization algorithm to obtain an optimal particle set;

[0013] In the optimal particle set, determine the score of each particle corresponding to the resource shard according to the weighted sum of the preset multi-dimensional indexes, and select the resource shard with the highest score for resource scheduling.

[0014] In one embodiment, the step of iteratively optimizing the initial particle swarm through the improved hybrid particle swarm optimization algorithm to obtain an optimal particle set includes:

[0015] Based on the improved hybrid particle swarm optimization algorithm, set the dynamic parameter range for the initial particle swarm according to the cloud environment monitoring data to obtain a parameterized particle swarm;

[0016] Determine the weighted sum of the multi-dimensional monitoring data sets of each particle in the parameterized particle swarm to obtain a particle swarm with fitness scores;

[0017] Update the particle velocity and particle position based on the particle swarm with fitness scores to obtain an optimized particle swarm;

[0018] Extract multiple particles with optimal fitness from the optimized particle swarm to obtain an optimal particle set.

[0019] In one embodiment, the step of updating the particle velocity and particle position based on the particle swarm with fitness scores to obtain an optimized particle swarm includes:

[0020] Based on the real-time load data of the particle swarm with fitness scores, obtain a dynamic inertia weight parameter by determining the inertia weight of the current iteration period;

[0021] According to the priority annotation of the particle swarm with fitness scores, obtain optimized learning factor parameters by adjusting the individual learning factor and the social learning factor;

[0022] According to the current position information of the parameterized particle swarm and the resource constraint conditions, obtain a position update rule after constraint adjustment by correcting the particle position update range;

[0023] Based on the dynamic inertia weight parameter, the optimized learning factor parameters, and the position update rule after constraint adjustment, update the particle velocity and particle position to obtain an optimized particle swarm.

[0024] In one embodiment, the step of generating a multi-dimensional data topology map of the cloud environment monitoring data includes:

[0025] Preprocess the cloud environment monitoring data to obtain a standardized data matrix;

[0026] Based on the standardized data matrix, determine the resource correlation degree between nodes through an improved graph convolutional network algorithm, where the improved graph convolutional network algorithm is an algorithm with a dynamically introduced attention mechanism;

[0027] Generate a weighted resource association graph according to the resource correlation degree;

[0028] Fuse the weighted resource association graph with the physical network topology structure of the cloud environment to obtain a multi-dimensional data topology graph integrating resource status and network topology.

[0029] In one embodiment, the step of decomposing the to-be-scheduled job set into a set of quantum computing units that can be processed in parallel and adding priority annotations to the set of quantum computing units through a quantum heuristic algorithm includes:

[0030] Analyze the execution process of the to-be-scheduled job set based on the task dependency graph to obtain a task decomposition tree;

[0031] Determine the task parallelism threshold according to the resource distribution characteristics in the multi-dimensional data topology graph, and obtain the maximum number of parallelizable units based on the task parallelism threshold;

[0032] Divide the task decomposition tree into multiple independent subtask blocks according to the maximum number of parallelizable units to obtain a set of quantum computing units;

[0033] Build a quantum annealing cost function model containing the task characteristic parameters of the to-be-scheduled job set through a quantum heuristic algorithm based on the principle of quantum annealing;

[0034] Input the set of quantum computing units into the quantum annealing cost function model for optimization and solution to obtain the unit priority ranking result;

[0035] Add priority annotations to each quantum computing unit in the set of quantum computing units according to the unit priority ranking result.

[0036] In one embodiment, the step of dividing the physical resources of the cloud environment into non-uniform resource slices matching the task requirements according to the priority annotations and the multi-dimensional data topology graph includes:

[0037] Extract the resource demand characteristics of the set of quantum computing units based on the priority annotations to obtain a task resource demand vector;

[0038] Generate a node resource adaptation matrix according to the node resource status and network topology relationship of the multi-dimensional data topology graph;

[0039] According to the task resource requirement vector and the node resource adaptation matrix, a first resource shard is obtained by reserving exclusive resource blocks for the first-priority quantum computing units;

[0040] A second resource shard is obtained by allocating shared resource blocks to the second-priority quantum computing units;

[0041] Based on the network latency data in the topology graph, the communication path between shards is optimized to obtain the shard network topology relationship;

[0042] The first resource shard, the second resource shard, and the shard network topology relationship are combined and optimized to generate non-uniform resource shards matching the task requirements.

[0043] In one embodiment, after the step of performing resource scheduling through the improved hybrid particle swarm optimization algorithm based on the non-uniform resource shards, the following steps are further included:

[0044] Based on the comparison and analysis of the actual resource utilization data collected during the scheduling execution process with the expected scheduling target, a scheduling deviation index is obtained;

[0045] When the scheduling deviation index exceeds a preset threshold, the multi-dimensional data topology graph is updated according to the current actual resource state to obtain a corrected topology graph;

[0046] Based on the corrected topology graph, the priority labels of the quantum computing units with unfinished tasks are re-determined to obtain updated priority labels;

[0047] According to the updated priority labels, local adjustment is performed on the non-uniform resource shards to obtain an optimized resource shard configuration;

[0048] Re-scheduling is performed based on the optimized resource shard configuration.

[0049] In addition, to achieve the above object, the present application further provides a resource scheduling device, and the resource scheduling device includes:

[0050] A data integration module, configured to generate a multi-dimensional data topology graph of cloud environment monitoring data;

[0051] A unit decomposition module, configured to decompose a job set to be scheduled into a set of quantum computing units that can be processed in parallel, and add priority labels to the set of quantum computing units through a quantum heuristic algorithm;

[0052] A resource partitioning module, configured to partition the physical resources of the cloud environment into non-uniform resource shards matching the task requirements according to the priority labels and the multi-dimensional data topology graph;

[0053] A resource scheduling module, which is used to perform resource scheduling based on non-uniform resource sharding through an improved hybrid particle swarm optimization algorithm. Among them, the improved hybrid particle swarm optimization algorithm is an algorithm that introduces weighted sum fitness evaluation and resource constraint correction.

[0054] In addition, to achieve the above object, the present application also proposes a resource scheduling device, which includes: a memory, a processor, and a computer program stored on the memory and executable on the processor. The computer program is configured to implement the steps of the resource scheduling method as described above.

[0055] In addition, to achieve the above object, the present application also proposes a storage medium, which is a computer-readable storage medium. A computer program is stored on the storage medium, and when the computer program is executed by a processor, it implements the steps of the resource scheduling method as described above.

[0056] In addition, to achieve the above object, the present application also provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the steps of the resource scheduling method as described above.

[0057] The technical solution proposed by the present application generates a multi-dimensional data topology map of cloud environment monitoring data, decomposes the to-be-scheduled job set into a set of quantum computing units that can be processed in parallel, and adds priority markings to the set of quantum computing units through a quantum-inspired algorithm. According to the priority markings and the multi-dimensional data topology map, the physical resources of the cloud environment are divided into non-uniform resource shards that match the task requirements. Based on the non-uniform resource shards, resource scheduling is performed through an improved hybrid particle swarm optimization algorithm, and weighted sum fitness evaluation and resource constraint correction are introduced. The present application presents the resource status and associations comprehensively by generating a multi-dimensional data topology map, then decomposes the to-be-scheduled job set into a set of quantum computing units and adds priority markings, and then divides non-uniform resource shards for resource scheduling, improving the accuracy of resource adaptation and the scheduling efficiency and resource utilization rate. Description of the Drawings

[0058] The drawings here are incorporated into the specification and form a part of the specification, showing the embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application.

[0059] To more clearly illustrate the technical solutions in the embodiments of the present application or in the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, for those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0060] Figure 1It is a schematic flowchart provided for the first embodiment of the resource scheduling method of this application;

[0061] Figure 2 It is a schematic flowchart provided for the second embodiment of the resource scheduling method of this application;

[0062] Figure 3 It is a schematic flowchart provided for the third embodiment of the resource scheduling method of this application;

[0063] Figure 4 It is a schematic module structure diagram of the resource scheduling device in the embodiment of this application;

[0064] Figure 5 It is a schematic device structure diagram of the hardware operating environment involved in the resource scheduling method in the embodiment of this application.

[0065] The realization of the purpose, functional characteristics and advantages of this application will be further described with reference to the embodiments and the accompanying drawings. Specific embodiments

[0066] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of this application and are not used to limit this application.

[0067] In order to better understand the technical solutions of this application, the following will be described in detail in combination with the accompanying drawings of the specification and specific embodiments.

[0068] Traditional scheduling algorithms have the problem of insufficient matching between resource allocation and task requirements. Especially in emerging scenarios such as new energy vehicle networking, due to the lack of coordinated optimization of task priorities and resource constraints, there are often contradictions where high-priority analysis tasks are delayed due to resource fragmentation, while low-priority batch processing tasks occupy excessive resources, resulting in low cluster resource utilization and resource scheduling efficiency, seriously restricting the real-time response ability of the intelligent transportation system.

[0069] Therefore, in order to overcome the above defects, this application provides a solution. By generating a multi-dimensional data topology map to comprehensively present the resource status and associations, then decomposing the job set to be scheduled into a set of quantum computing units and adding priority annotations, and then dividing non-uniform resource slices for resource scheduling, the accuracy of resource adaptation is improved, and the scheduling efficiency and resource utilization are increased.

[0070] It should be noted that the execution subject of each embodiment of this application can be a computing service system with data processing, network communication, and program running functions, such as an electronic system, a resource scheduling system, etc. that can implement the above functions. Hereinafter, taking the resource scheduling system as an example (hereinafter referred to as "system"), the following embodiments will be described.

[0071] Based on this, the embodiments of this application provide a resource scheduling method, referring toFigure 1 , Figure 1 is a schematic flowchart of the first embodiment of the resource scheduling method of this application.

[0072] In this embodiment, the resource scheduling method includes steps S10 to S40:

[0073] Step S10, generate a multi-dimensional data topology map of cloud environment monitoring data.

[0074] It should be understood that the resource scheduling method proposed in this application first constructs a multi-dimensional data topology map by collecting the operation status data of various resources in the cloud computing environment in real time. This topology map is different from the traditional flat monitoring view. Instead, it presents physical resources such as computing nodes, storage devices, and network links in a three-dimensional manner. Among them, the size of the nodes represents the remaining computing power, the thickness of the connection lines reflects the actual available bandwidth, and the depth of the color indicates the current load level. This visualization method can intuitively display the association relationship and real-time status changes between resources. In particular, a dynamic weight adjustment mechanism is introduced in the multi-dimensional data topology map, so that the communication efficiency and dependence intensity between resources can be automatically updated according to the actual load situation, effectively solving the problem that the traditional static topology map cannot adapt to the dynamic changes of the cloud environment.

[0075] In the process of constructing the multi-dimensional data topology map, not only conventional monitoring metrics such as the Central Processing Unit (CPU) and memory are integrated, but also a resource affinity metric is creatively introduced to quantify the cooperation efficiency between different computing nodes. Through this integration method, the multi-dimensional data topology map can reflect both the physical resource status and the logical cooperation relationship. In addition, the system also designs an intelligent alarm function. When it detects that some key resources are about to reach the bottleneck, it will be highlighted in the multi-dimensional data topology map in advance. This prediction mechanism improves the foresight of resource scheduling.

[0076] Step S20, decompose the job set to be scheduled into a set of quantum computing units that can be processed in parallel, and add priority labels to the set of quantum computing units through a quantum-inspired algorithm.

[0077] It should be noted that the set of quantum computing units integrates multiple quantum computing units. A quantum computing unit refers to the smallest task module with a complete execution logic. Each unit encapsulates a specific computing task and resource requirements. Different from the traditional task decomposition method, modeling using the concept of quantum computing enables each task unit to have the superposition characteristics of quantum states, can consider multiple possible scheduling schemes at the same time, and can handle the dependence relationship and parallel possibility between tasks more flexibly.

[0078] In terms of task priority annotation, a heuristic algorithm based on the principle of quantum annealing is adopted. This algorithm comprehensively considers multiple dimensions such as task deadlines, resource demand intensity, and communication overhead between tasks, and calculates a dynamic priority score for each quantum computing unit. Notably, this algorithm introduces an environment perception mechanism that can automatically adjust the priority calculation strategy according to the overall load of the current cloud environment. For example, when the system load is high, the algorithm will appropriately increase the consideration weight of resource utilization; when the load is low, it will pay more attention to the task response speed to balance the system throughput and task response time.

[0079] Step S30: According to the priority annotation and the multi-dimensional data topology map, divide the physical resources of the cloud environment into non-uniform resource slices that match the task requirements.

[0080] Based on the priority annotation and the multi-dimensional data topology map obtained from the foregoing steps, non-uniform resource slicing is performed. Here, non-uniform resource slices refer to resource blocks that are dynamically divided according to the actual task requirements and have different sizes and configurations, which is in sharp contrast to the fixed-size resource allocation method in traditional cloud computing. Each resource slice is specifically optimized for the task requirements of a specific priority. High-priority tasks can obtain exclusive high-quality resources, while low-priority tasks share a common resource pool.

[0081] In the specific implementation, the system comprehensively analyzes the resource status information and task priority information shown in the multi-dimensional data topology map and dynamically determines the optimal slicing scheme. In particular, a slicing elastic adjustment mechanism is introduced in this process, which can dynamically optimize the resource slices according to the real-time feedback of the task execution progress. For example, when it is detected that a high-priority task is about to complete, the system will mark the resources it occupies as recyclable in advance to quickly respond to newly arrived high-priority tasks.

[0082] Step S40: Based on the non-uniform resource slices, perform resource scheduling through an improved hybrid particle swarm optimization algorithm, where the improved hybrid particle swarm optimization algorithm is an algorithm that introduces weighted sum fitness evaluation and resource constraint correction.

[0083] It should be noted that particle swarm optimization is an intelligent optimization algorithm that simulates the foraging behavior of bird flocks. In this application, each particle represents a possible resource allocation scheme. The improved hybrid particle swarm optimization algorithm has innovated mainly in two key aspects: one is to design a multi-dimensional fitness evaluation function that can optimize multiple objectives such as resource utilization rate and task completion time at the same time; the other is to add a resource constraint processing mechanism to ensure that the generated scheduling scheme always meets various resource limit conditions. These improvements enable the algorithm to better adapt to the complexity of the cloud computing environment.

[0084] The improved hybrid particle swarm optimization algorithm can automatically adjust the breadth and depth of the search according to the real-time feedback information during the scheduling process. At the initial stage of scheduling, a larger search step size is adopted to quickly locate the high-quality solution area, and at the later stage, the step size is reduced for fine optimization to ensure the scheduling efficiency and the quality of the solution. In addition, the historical high-quality scheduling solutions can be saved through a memory mechanism and directly called when encountering similar task combinations.

[0085] As an implementation manner, after the above step S40 in this embodiment, it may further include: comparing and analyzing the actual resource utilization data collected during the scheduling execution process with the expected scheduling target to obtain a scheduling deviation index; when the scheduling deviation index exceeds a preset threshold, updating the multi-dimensional data topology map according to the current actual resource status to obtain a corrected topology map; re-determining the priority annotation of the quantum computing units with unfinished tasks based on the corrected topology map to obtain an updated priority annotation; locally adjusting the non-uniform resource sharding according to the updated priority annotation to obtain an optimized resource sharding configuration; and performing re-scheduling based on the optimized resource sharding configuration.

[0086] It can be understood that after the initial resource scheduling is completed, the system will continuously monitor the system operation status, intelligently compare and analyze the actual utilization data of each resource shard collected in real time with the expected scheduling target, so as to calculate the execution deviation index of the current scheduling plan, including multiple dimensions such as resource utilization deviation and task delay deviation. Among them, the scheduling deviation index can quantify the gap between the current scheduling effect and the ideal state by weighted synthesis of multiple performance indicators. When the system detects that this index exceeds the preset safety threshold, it will automatically trigger the scheduling optimization process. Among them, the sliding window technology is adopted, which can distinguish instantaneous fluctuations and continuous deviations and avoid unnecessary scheduling adjustments.

[0087] After obtaining the scheduling deviation index, re-scan the current status of all physical resources and virtual resources in the cloud environment to obtain key parameters such as the latest resource load data and network status information. Then, combine these real-time data with the historical operation trend to generate a corrected topology map. In this process, an incremental update algorithm can be introduced, and only the changed part of the topology structure needs to be recalculated to reduce the update overhead. The updated topology map not only reflects the current resource status but also marks the resource bottleneck points and potential risk areas.

[0088] Based on the newly generated corrected topology graph, a priority re-evaluation is performed on the quantum computing units that have not yet completed execution. This step adopts a dynamic priority adjustment algorithm, comprehensively considering factors such as the remaining execution time of the task, the latest resource availability, and the business importance. Different from the initial priority annotation, the re-evaluation process particularly focuses on the deviation between the actual progress of the task and the expected plan, and will automatically increase the priority of tasks that lag behind the plan. The innovation in this link lies in the introduction of a priority decay factor. For tasks that have received multiple priority increases but still have not been completed, their priorities will be appropriately reduced to prevent individual tasks from occupying key resources for a long time, avoid the priority inversion problem, and ensure the fairness of system resources.

[0089] Furthermore, after obtaining the updated priority annotation, identify the shard area where the current resource allocation and task requirements are the least matched, and optimize it through defragmentation. Without affecting the running tasks, the dynamic reorganization of resource sharding can be completed to obtain an optimized resource sharding configuration. For example, for shards with too low load, the system will appropriately reduce their scale and allocate the released resources to shards with higher load; for communication-intensive task groups, they will be scheduled to adjacent positions on the network topology. Finally, based on the optimized resource sharding configuration, a rescheduling operation is performed, and the adjustment strategy is continuously optimized through machine learning to continuously adapt to the changes in the workload and maintain the best running state.

[0090] In this embodiment, a multi-dimensional data topology graph is generated to comprehensively present the resource status and associations, and then the set of jobs to be scheduled is decomposed into a set of quantum computing units and priority annotations are added. Next, non-uniform resource sharding is performed for resource scheduling, which improves the accuracy of resource adaptation, enhances the scheduling efficiency and resource utilization rate. Moreover, by adding a scheduling effect evaluation and adjustment mechanism, it can timely adapt to environmental changes and task adjustments, improving the scheduling effect.

[0091] Based on the first embodiment of the present application, in the second embodiment of the present application, the same or similar content as in the above-mentioned first embodiment can be referred to the above introduction and will not be elaborated hereinafter. On this basis, please refer to Figure 2 , the step S40 may include steps S401 to S403:

[0092] Step S401, map each resource shard of the non-uniform resource sharding to a particle in the particle swarm, and obtain an initial particle swarm by attaching a shard real-time status identifier to each particle.

[0093] It can be understood that when mapping the non-uniform resource sharding to the particles in the particle swarm optimization algorithm, each particle not only contains the basic configuration parameters of the resource shard, but also integrates real-time status identifiers, which dynamically reflect key running indicators such as the current load status, task processing progress, and resource availability of the shard.

[0094] Step S402: Iteratively optimize the initial particle swarm through the improved hybrid particle swarm optimization algorithm to obtain an optimal particle set.

[0095] In the particle swarm optimization stage, the improved hybrid optimization algorithm is used for iterative calculation to obtain an optimal particle set. This algorithm creatively integrates the crossover and mutation operations of the genetic algorithm and the temperature control mechanism of the simulated annealing algorithm, enabling the search process to maintain both global exploration ability and local refinement accuracy. During each iteration, the algorithm dynamically evaluates the diversity degree of the particle swarm. When a premature convergence trend is detected, a population recombination operation is automatically triggered.

[0096] As an implementation manner, in this embodiment, the above step S402 may include: Based on the improved hybrid particle swarm optimization algorithm, perform dynamic parameter range setting for the initial particle swarm according to the cloud environment monitoring data to obtain a parameterized particle swarm; determine the weighted sum of the multi-dimensional monitoring data sets of each particle in the parameterized particle swarm to obtain a particle swarm with fitness scores; update the particle velocity and particle position based on the particle swarm with fitness scores to obtain an optimized particle swarm; extract multiple particles with the best fitness from the optimized particle swarm to obtain an optimal particle set.

[0097] It should be understood that in the iterative optimization process of the improved hybrid particle swarm optimization algorithm adopted in this application, intelligent parameterization processing of the initial particle swarm is first performed based on real-time cloud environment monitoring data. This step can introduce a dynamic parameter range setting mechanism. The algorithm analyzes monitoring data such as the load characteristics and resource distribution status of the current cloud environment, and automatically adjusts the adjustable parameter range of each particle. For example, during high-load periods, the adjustment range of resource allocation parameters is narrowed to achieve fine scheduling, and during low-load periods, the search range is expanded to explore more possibilities.

[0098] After obtaining the parameterized particle swarm, the system performs multi-dimensional fitness evaluation, that is, comprehensively considering the weighted sum of multiple dimensions such as resource utilization rate, task response latency, energy consumption efficiency, and load balance to obtain a particle swarm with fitness scores. The innovation of this step lies in the proposal of a dynamic weight allocation mechanism. The weight coefficients are automatically adjusted according to real-time monitoring data. For example, when it is detected that the temperature of some computing nodes is too high, the weight of the energy consumption index is automatically increased; when there is a backlog of tasks during the business peak period, the evaluation of task response speed is emphasized. A fuzzy logic algorithm can also be introduced during the evaluation process to effectively handle the uncertainty and noise interference in the monitoring data.

[0099] In the particle update stage, different from the fixed update formula in the traditional particle swarm algorithm, the improved algorithm dynamically adjusts the learning factor of each particle according to the fitness score. For the elite particles with higher fitness, a conservative update strategy is adopted to maintain their excellent characteristics; for the particles with lower fitness, a more radical exploratory update is implemented.

[0100] In the selection of the optimal particle set, a strategy that emphasizes both elite retention and diversity preservation is adopted. Not only several particles with the highest fitness scores are selected, but also those characteristic particles that perform outstandingly in specific dimensions are especially retained to ensure that the finally obtained particle set contains both the global optimal solution and high-quality alternative solutions for different optimization focuses. In addition, the system establishes a detailed characteristic profile for each selected particle to record its performance characteristics during the iteration process. It is worth mentioning that even after obtaining the optimal particle set, the system still continuously monitors its actual execution effect and triggers the re-optimization process when necessary.

[0101] As an implementation manner, the step of obtaining the optimized particle swarm by updating the particle velocity and particle position based on the particle swarm with fitness scores in this embodiment includes: obtaining the dynamic inertia weight parameter by determining the inertia weight of the current iteration period based on the real-time load data of the particle swarm with fitness scores; obtaining the optimized learning factor parameter by adjusting the individual learning factor and the social learning factor according to the priority annotation of the particle swarm with fitness scores; obtaining the position update rule after constraint adjustment by correcting the particle position update range according to the current position information of the parameterized particle swarm and the resource constraint conditions; and obtaining the optimized particle swarm by updating the particle velocity and particle position based on the dynamic inertia weight parameter, the optimized learning factor parameter, and the position update rule after constraint adjustment.

[0102] It can be understood that based on the real-time load data of the particle swarm with fitness scores, the system designs an intelligent inertia weight adjustment mechanism, and this parameter no longer remains fixed but dynamically changes with the iteration process and the system state. A larger inertia weight is set in the initial stage of the search to enhance the global exploration ability, and the weight is gradually reduced as the iteration progresses to improve the local search accuracy. In particular, when it is detected that the diversity of the particle swarm decreases, the system will automatically increase the inertia weight to avoid premature convergence.

[0103] In terms of parameter optimization, the task priority annotation information is introduced into the learning factor adjustment process. For the particles corresponding to high-priority tasks (the first particles), the social learning factor is appropriately increased to promote the rapid spread of excellent experience; for the ordinary task particles (the second particles), a balanced learning factor configuration is maintained. The system automatically calculates the optimal learning factor ratio by analyzing the overall fitness distribution of the particle swarm, that is, the optimized learning factor parameter is obtained.

[0104] In view of the resource constraint characteristics of the cloud environment, a multi-dimensional resource constraint model is established to accurately quantify the limitation conditions of various resources such as CPU, memory, and bandwidth. When updating the particle position, through techniques such as projection transformation and boundary reflection, it is ensured that the new position is always within the feasible solution space. Among them, the algorithm dynamically adjusts the strictness of constraint processing according to the resource tension: allowing moderate over-allocation to explore better solutions when resources are abundant, and strictly enforcing the constraint conditions when resources are tense. A violation history record library can be maintained for adjusting the constraint processing strategy in subsequent iterations. The final velocity and position update process integrates the above dynamic inertia weight parameter, optimized learning factor parameter, and position update rule after constraint adjustment to obtain an optimized particle swarm. In addition, an anomaly detection mechanism is set during the update process, and when abnormal particle behavior is detected, the re-initialization process will be triggered.

[0105] Step S403, in the optimal particle set, determine the score of the resource shard corresponding to each particle according to the weighted sum of the preset multi-dimensional indicators, and select the resource shard with the highest score for resource scheduling.

[0106] It should be understood that each candidate particle (i.e., the particle in the optimized particle swarm) needs to pass the assessment of multiple indicators, which include both traditional quantitative parameters such as resource utilization rate and task completion rate, and qualitative evaluation dimensions such as service quality satisfaction and scheduling scheme stability. The evaluation process adopts a dynamic weighting method, and the weight allocation will be automatically adjusted according to the characteristics of the business period: during the peak daytime business period, the task response speed is emphasized, and during the night batch processing period, the energy efficiency is given priority. The system will finally select the solution with the highest score as the execution basis, and through the solution elastic adaptation mechanism, it is also allowed to make fine-tuning according to the actual operation situation during the implementation process.

[0107] In this embodiment, by mapping non-uniform resource shards to particles and attaching real-time status identifiers, the resource changes are dynamically reflected, the iterative optimization of the hybrid particle swarm optimization algorithm is improved, the score is determined according to the weighted sum of the preset multi-dimensional indicators, and the optimal resource shard is selected, which improves the scheduling accuracy and adaptability, meets the complex task requirements, and avoids resource conflicts and waste.

[0108] Based on the first embodiment of the present application, in the third embodiment of the present application, the same or similar content as the above-mentioned embodiment 1 can be referred to the above introduction and will not be repeated hereinafter. On this basis, please refer to Figure 3 , the step S10 may include steps S101~S104:

[0109] Step S101, preprocess the cloud environment monitoring data to obtain a standardized data matrix.

[0110] It is understandable that when generating a multi-dimensional data topology graph of cloud environment monitoring data, the original monitoring data is first subjected to intelligent preprocessing. This preprocessing process can include conventional data cleaning and missing value handling, and can also adopt suitable normalization methods for different types of metrics such as CPU utilization rate, memory occupancy rate, network throughput, etc. through multi-scale normalization techniques. For example, through sliding time window analysis, the system automatically identifies and eliminates instantaneous outliers while retaining the true load fluctuation characteristics. The standardized data matrix generated after preprocessing retains the spatio-temporal correlation characteristics of the original monitoring data. In particular, this preprocessing stage can also automatically increase the data collection frequency through an adaptive sampling mechanism when the system load fluctuates violently to ensure that key state changes can be accurately captured.

[0111] Step S102: Based on the standardized data matrix, determine the resource correlation degree between nodes through an improved graph convolutional network algorithm, where the improved graph convolutional network algorithm is an algorithm with a dynamically introduced attention mechanism.

[0112] In the resource correlation degree analysis stage, an improved graph convolutional network algorithm is adopted. Its core innovation is the introduction of a dynamic attention mechanism, which enables the network to automatically adjust the importance weights of connections between nodes according to the real-time load situation. For example, when the CPU load of a certain node exceeds the threshold, the correlation degree weight between it and the storage node will be dynamically increased. The improved graph convolutional network algorithm captures the complex non-linear relationships between resource nodes through multi-layer feature propagation and aggregation. Compared with the traditional graph convolutional network, this improvement increases the response speed of the model to the dynamic changes in the cloud environment. During the training process, a transfer learning algorithm can also be adopted to enable the model to quickly adapt to the deployment requirements of cloud platforms of different scales.

[0113] Step S103: Generate a weighted resource correlation graph according to the resource correlation degree.

[0114] Based on the calculated resource correlation degree, the system constructs a weighted resource correlation graph. This graph adopts a multi-dimensional edge weight representation method, and each edge contains weight information in multiple dimensions such as performance correlation degree, communication frequency, and dependence strength, and will regularly recalculate the edge weight values to ensure that they reflect the latest resource interaction status. For large-scale cloud environments, the present application can also adopt a hierarchical graph structure construction algorithm to first construct subgraphs in each local area and then perform global connection through key hub nodes to reduce the graph construction time complexity.

[0115] Step S104: Perform fusion processing on the weighted resource correlation graph and the physical network topology structure of the cloud environment to obtain a multi-dimensional data topology graph that combines resource status and network topology.

[0116] It should be understood that an accurate mapping relationship between virtual resource nodes and physical devices can be established through a graph alignment algorithm. The generated multi-dimensional data topology graph adopts a visualization encoding scheme: the size of the node represents the remaining computing power, the depth of the node color reflects the current load rate, the width of the edge represents the communication bandwidth, the edge color indicates the network latency, and the resource bottleneck area and critical communication path are highlighted through the dynamic focusing function.

[0117] As an implementation manner, the step S20 may include: analyzing the execution process of the job set to be scheduled based on the task dependency graph to obtain a task decomposition tree; determining a task parallelism threshold according to the resource distribution characteristics in the multi-dimensional data topology graph, and obtaining the maximum number of parallelizable units based on the task parallelism threshold; dividing the task decomposition tree into multiple independent sub-task blocks according to the maximum number of parallelizable units to obtain a set of quantum computing units; constructing a quantum annealing cost function model including the task characteristic parameters of the job set to be scheduled through a quantum heuristic algorithm based on the principle of quantum annealing; inputting the set of quantum computing units into the quantum annealing cost function model for optimization and solution to obtain a unit priority sorting result; and adding a priority annotation to each quantum computing unit in the set of quantum computing units according to the unit priority sorting result.

[0118] It should be understood that in the stage of decomposing the job set to be scheduled, first, a task decomposition tree is constructed by deeply analyzing the task dependency graph. This process gradually splits complex jobs into sub-task modules with clear input-output relationships, and the task granularity evaluation algorithm is used to identify the most suitable task splitting points for parallelization. In particular, the system can guide the decomposition process of the current task by maintaining a task pattern library and comparing the optimal decomposition schemes of historical similar tasks.

[0119] When determining the task parallelism threshold, calculate the maximum parallel processing capabilities of each computing node in the current environment. Based on these data, determine the optimal task parallelism threshold through a sliding window. This threshold will be automatically adjusted according to the resource load status, and the maximum number of task units that can be executed simultaneously without causing resource competition is calculated based on this threshold. Based on the above analysis results, the system divides the task decomposition tree into a set of quantum computing units, and each quantum computing unit in it encapsulates a complete execution context and resource requirement description. The division process intelligently adjusts the size and composition of the unit according to the computing characteristics and data dependency relationships of the sub-tasks, and through the unit integrity verification mechanism, ensures that each unit can be independently scheduled and executed without generating external dependencies.

[0120] In the priority annotation stage, a quantum-inspired algorithm based on the principle of quantum annealing is adopted. The quantum annealing cost function model constructed by this algorithm innovatively integrates multiple optimization dimensions such as task deadline, resource demand intensity, and data dependence depth. During the training process of this model, transfer learning is introduced to quickly adapt to different types of job characteristics. When the set of quantum computing units is input into the quantum annealing cost function model for optimization and solution, an optimization strategy that simulates the quantum tunneling effect is adopted, enabling the algorithm to have the ability to jump out of the local optimal solution.

[0121] Based on the obtained unit priority sorting results, the finally generated priority annotation not only includes a simple rank sorting, but also innovatively introduces dynamic adjustment coefficients. These coefficients will be automatically adjusted according to the system load, ensuring that high-priority tasks can receive more attention when resources are scarce, and appropriately relaxing the restrictions when resources are abundant to improve the overall throughput. The system can also enable related task units to share some priority characteristics through the priority propagation mechanism.

[0122] As an implementation manner, the step S30 may include: extracting the resource demand characteristics of the set of quantum computing units based on the priority annotation to obtain a task resource demand vector; generating a node resource adaptation matrix according to the node resource status and network topology relationship of the multi-dimensional data topology graph; based on the task resource demand vector and the node resource adaptation matrix, obtaining a first resource slice by reserving exclusive resource blocks for the first-priority quantum computing units; obtaining a second resource slice by allocating shared resource blocks for the second-priority quantum computing units; optimizing the communication path between slices based on the network delay data in the topology graph to obtain a slice network topology relationship; combining and optimizing the first resource slice, the second resource slice, and the slice network topology relationship to generate a non-uniform resource slice that matches the task requirements.

[0123] It can be understood that in the resource slice division stage, first, the resource demands of the set of quantum computing units are analyzed through feature extraction. Based on the priority annotation information, the system constructs a multi-dimensional task resource demand vector, which includes conventional computing resource demands such as CPU and memory, as well as feature dimensions such as communication bandwidth demand and data locality preference. The extraction process uses an adaptive feature weighting algorithm to automatically adjust the weight ratio of each dimension according to the task type. For example, the CPU weight is increased for compute-intensive tasks, and the storage performance is emphasized for data-intensive tasks.

[0124] In terms of resource adaptation analysis, a node resource adaptation matrix is generated based on a multi-dimensional data topology graph. This matrix is represented in a tensor structure, encoding both the static configuration information and the dynamic load status of computing nodes. During the matrix construction process, incremental update techniques are used to ensure data timeliness. In particular, the system can efficiently encode the topological relationships of a super-large-scale cloud environment through a topology-aware matrix compression algorithm without losing key information.

[0125] For tasks with different priorities, exclusive resource blocks reserved for first-priority tasks use hard isolation technology to ensure stable performance for critical tasks, and the scale of these resource blocks can be dynamically adjusted according to the actual requirements of the tasks; the shared resource blocks allocated to second-priority tasks use a soft isolation mechanism and implement over-allocation and recycling of resources. In addition, the system allows low-priority shards to be temporarily lent to high-priority shards for use when resources are idle to improve resource utilization.

[0126] In terms of network optimization, communication paths between shards are optimized based on the latency data of the topology graph, and the network connection strategy between shards is dynamically adjusted. Among them, optimized paths can be pre-established for shard pairs with high-frequency communication to reduce communication latency. The system builds a shard network topology graph through multi-layer abstract representation, which contains physical link information and encodes the deployment of virtual network functions.

[0127] The final non-uniform resource shard generation process uses combinatorial optimization to model the resource allocation problem as a multi-objective optimization problem and finds the optimal trade-off solution through Pareto front analysis. In addition, the system also monitors the running status of each shard in real time and dynamically adjusts the shard configuration accordingly.

[0128] For ease of understanding, an example is given below, but it does not limit the resource scheduling method of this application. In the intelligent transportation management cloud platform of a provincial capital city, it is necessary to simultaneously process real-time monitoring video streams from more than 5,000 intersections and perform analysis tasks such as vehicle recognition, violation detection, and traffic flow statistics. Under the traditional scheduling method, task backlogs are serious during peak hours, the average delay of critical violation recognition tasks is relatively long, and the utilization rate of the GPU cluster is low.

[0129] After applying the resource scheduling method of this application to the cloud platform, the system first constructs a multi-dimensional topology map containing 200 GPU servers, dynamically tracks the computing power, video memory status of each server, and the network latency between nodes. When the morning and evening rush hours arrive, the platform decomposes the sudden video analysis tasks into quantum computing units, and automatically identifies high-priority tasks such as illegal detection through the quantum-inspired algorithm, and marks their emergency levels. Based on the priority marking, the system divides the GPU cluster into non-uniform slices: reserves dedicated slices for the illegal detection tasks with high real-time requirements (such as each slice is equipped with 2 GPUs and exclusive 100 Gbps network bandwidth), and allocates shared slices for the batch processing traffic statistics tasks (multiple tasks share GPU resources). Dynamically optimize resource allocation through the improved hybrid particle swarm algorithm. When a traffic accident occurs in a certain area, automatically elevate the priority of the analysis tasks at the relevant intersections and trigger resource rescheduling.

[0130] By applying the resource scheduling method of this application, the platform can reduce the latency of the illegal recognition task, improve the GPU utilization rate, and at the same time shorten the completion time of the batch processing task. In high traffic, through dynamic topology map update and incremental rescheduling, it can successfully handle traffic peaks without discarding key tasks throughout the process.

[0131] In this embodiment, the standardized data matrix is obtained by preprocessing the monitoring data, and the improved graph convolutional network algorithm introducing the dynamic attention mechanism is used to determine the resource correlation degree, generate a weighted resource correlation graph, and fuse it with the physical network topology structure to obtain a multi-dimensional data topology map, which more comprehensively and accurately reflects the resource status and network topology relationship in the cloud environment. And, comprehensively considering task dependencies and resource distributions, improve the accuracy of decomposition and annotation, extract resource requirement features based on priority annotation, generate a node resource adaptation matrix, reserve exclusive resource blocks or allocate shared resource blocks for different priority quantum computing units, optimize the communication path between slices, and combine and optimize to generate non-uniform resource slices, which better meet the task requirements and improve resource utilization.

[0132] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the resource scheduling method of this application. Based on this technical concept, more forms of simple transformations are within the protection scope of this application.

[0133] This application also provides a resource scheduling device. Please refer to Figure 4 The resource scheduling device includes:

[0134] A data integration module 10 for generating a multi-dimensional data topology map of cloud environment monitoring data;

[0135] A unit decomposition module 20 for decomposing the job set to be scheduled into a set of quantum computing units that can be processed in parallel, and adding priority annotations to the set of quantum computing units through a quantum heuristic algorithm;

[0136] A resource partitioning module 30, configured to partition physical resources of a cloud environment into non-uniform resource shards that match task requirements according to the priority annotation and the multi-dimensional data topology graph.

[0137] A resource scheduling module 40, configured to perform resource scheduling based on the non-uniform resource shards through an improved hybrid particle swarm optimization algorithm, where the improved hybrid particle swarm optimization algorithm is an algorithm that introduces weighted sum fitness evaluation and resource constraint correction.

[0138] The resource scheduling device provided in this application adopts the resource scheduling method in the above embodiment, and can solve the technical problem in the prior art that due to inaccurate scheduling evaluation, resources cannot be accurately matched with tasks, resulting in low resource scheduling efficiency. Compared with the prior art, the beneficial effects of the resource scheduling device provided in this application are the same as those of the resource scheduling method provided in the above embodiment, and other technical features in the resource scheduling device are the same as those disclosed in the method of the above embodiment, and will not be elaborated here.

[0139] This application provides a resource scheduling device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the resource scheduling method in the first embodiment above.

[0140] Next, refer to Figure 5 , which shows a schematic structural diagram of a resource scheduling device suitable for implementing the embodiments of this application. The resource scheduling device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Desctions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 5 The resource scheduling device shown is only an example and should not impose any limitations on the functions and usage scopes of the embodiments of this application.

[0141] As Figure 5As shown, the resource scheduling device may include a processing device 1001 (such as a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to the program stored in the read-only memory 1002 or the program loaded from the storage device 1003 into the random access memory 1004. In the random access memory 1004, various programs and data required for the operation of the resource scheduling device are also stored. The processing device 1001, the read-only memory 1002, and the random access memory 1004 are connected to each other through a bus 1005. The input / output interface 1006 is also connected to the bus. Generally, the following systems may be connected to the input / output interface 1006: an input device 1007 including, for example, a touch screen, a touchpad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the resource scheduling device to communicate with other devices wirelessly or wiredly to exchange data. Although the figure shows a resource scheduling device with various systems, it should be understood that it is not required to implement or have all the shown systems. More or fewer systems can be implemented or had alternatively.

[0142] Specifically, according to the embodiments disclosed in the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, the embodiments disclosed in the present application include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program codes for performing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device, or installed from the storage device 1003, or installed from the read-only memory 1002. When the computer program is executed by the processing device 1001, the above functions defined in the methods of the embodiments disclosed in the present application are executed.

[0143] The resource scheduling device provided by the present application adopts the resource scheduling method in the above embodiments, and can solve the technical problem in the prior art that due to inaccurate scheduling evaluation, resources cannot be accurately matched with tasks, resulting in low resource scheduling efficiency. Compared with the prior art, the beneficial effects of the resource scheduling device provided by the present application are the same as those of the resource scheduling method provided by the above embodiments, and other technical features in the resource scheduling device are the same as those disclosed in the method of the previous embodiment, and will not be elaborated here.

[0144] It should be understood that each part disclosed in this application can be implemented by hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in a suitable manner in any one or more embodiments or examples.

[0145] As described above, this is only the specific implementation manner of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in this application, and all should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

[0146] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., computer programs) stored thereon, and the computer-readable program instructions are used to execute the resource scheduling method in the above embodiments.

[0147] The computer-readable storage medium provided by this application can be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination of the above. More specific examples of computer-readable storage media can include, but are not limited to: electrical connections with one or more wires, portable computer disks, hard disks, random access memory (RAM: Random Access Memory), read-only memory (ROM: Read Only Memory), erasable programmable read-only memory (EPROM: Erasable Programmable Read Only Memory or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM: CD-Read Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the above. In this embodiment, the computer-readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system or device. The program code contained on the computer-readable storage medium can be transmitted by any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination of the above.

[0148] The above computer-readable storage medium can be included in the resource scheduling device; or it can exist separately without being assembled into the resource scheduling device.

[0149] The above computer-readable storage medium carries one or more programs which, when executed by a resource scheduling device, cause the resource scheduling device to: generate a multi-dimensional data topology map of cloud environment monitoring data, decompose a set of jobs to be scheduled into a set of quantum computing units that can be processed in parallel, and add priority annotations to the set of quantum computing units through a quantum-inspired algorithm, divide the physical resources of the cloud environment into non-uniform resource slices that match the task requirements according to the priority annotations and the multi-dimensional data topology map, and perform resource scheduling based on the non-uniform resource slices through an improved hybrid particle swarm optimization algorithm, where the improved hybrid particle swarm optimization algorithm is an algorithm that introduces weighted sum fitness evaluation and resource constraint correction.

[0150] Computer program code for performing the operations of the present application may be written in one or more programming languages or combinations thereof. The programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code may execute entirely on the user's computer, partially on the user's computer, execute as a stand-alone software package, execute partially on the user's computer and partially on a remote computer, execute entirely on a remote computer or server, or execute on an ARM (Advanced RISC Machines) development board. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (for example, by using an Internet service provider to connect through the Internet).

[0151] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present application. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code that contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combinations of blocks in the block diagram and / or flowchart, may be implemented by a dedicated hardware-based system for performing the specified functions or operations, or may be implemented by a combination of dedicated hardware and computer instructions.

[0152] The modules involved in the embodiments of the present application can be implemented in software or in hardware. In some cases, the name of the module does not constitute a limitation on the unit itself.

[0153] The readable storage medium provided by the present application is a computer-readable storage medium, and the computer-readable storage medium stores computer-readable program instructions (i.e., computer programs) for executing the above resource scheduling method, which can solve the technical problem in the prior art that due to inaccurate scheduling evaluation, resources cannot be accurately matched with tasks, resulting in low resource scheduling efficiency. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the present application are the same as those of the resource scheduling method provided by the above embodiments, and will not be elaborated here.

[0154] The present application also provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the steps of the resource scheduling method as described above.

[0155] The computer program product provided by the present application can solve the technical problem in the prior art that due to inaccurate scheduling evaluation, resources cannot be accurately matched with tasks, resulting in low resource scheduling efficiency. Compared with the prior art, the beneficial effects of the computer program product provided by the present application are the same as those of the resource scheduling method provided by the above embodiments, and will not be elaborated here.

[0156] The above are only some embodiments of the present application, and do not limit the patent scope of the present application. Any equivalent structural transformation made by using the description and drawings of the present application under the technical concept of the present application, or any direct / indirect application in other related technical fields, is included in the patent protection scope of the present application.

Claims

1. A resource scheduling method, characterized in that, The method includes the following steps: Generate a multi-dimensional data topology graph of cloud environment monitoring data; Decompose the job set to be scheduled into a set of quantum computing units that can be processed in parallel, and add priority labels to the set of quantum computing units through a quantum-inspired algorithm; According to the priority labels and the multi-dimensional data topology graph, divide the physical resources of the cloud environment into non-uniform resource slices that match the task requirements; Based on the non-uniform resource slices, perform resource scheduling through an improved hybrid particle swarm optimization algorithm, where the improved hybrid particle swarm optimization algorithm is an algorithm that introduces weighted sum fitness evaluation and resource constraint correction; The step of decomposing the job set to be scheduled into a set of quantum computing units that can be processed in parallel, and adding priority labels to the set of quantum computing units through a quantum-inspired algorithm includes: Analyze the execution process of the job set to be scheduled based on the task dependency graph to obtain a task decomposition tree; Determine the task parallelism threshold according to the resource distribution characteristics in the multi-dimensional data topology graph, and obtain the maximum number of parallelizable units based on the task parallelism threshold; Divide the task decomposition tree into multiple independent sub-task blocks according to the maximum number of parallelizable units to obtain a set of quantum computing units; Through a quantum-inspired algorithm based on the principle of quantum annealing, construct a quantum annealing cost function model including the task characteristic parameters of the job set to be scheduled; Input the set of quantum computing units into the quantum annealing cost function model for optimization and solution to obtain the unit priority sorting result; According to the unit priority sorting result, add priority labels to each quantum computing unit in the set of quantum computing units; The step of dividing the physical resources of the cloud environment into non-uniform resource slices that match the task requirements according to the priority labels and the multi-dimensional data topology graph includes: Extract the resource requirement characteristics of the set of quantum computing units based on the priority labels to obtain a task resource requirement vector; Generate a node resource adaptation matrix according to the node resource status and network topology relationship in the multi-dimensional data topology graph; According to the task resource requirement vector and the node resource adaptation matrix, obtain the first resource slice by reserving a unique resource block for the first-priority quantum computing unit; Obtain the second resource slice by allocating shared resource blocks to the second-priority quantum computing units; Optimize the communication path between slices based on the network delay data in the topology graph to obtain the slice network topology relationship; Combine and optimize the first resource slice, the second resource slice, and the slice network topology relationship to generate non-uniform resource slices that match the task requirements.

2. The resource scheduling method according to claim 1, wherein The step of performing resource scheduling through an improved hybrid particle swarm optimization algorithm based on the non-uniform resource slices includes: Map each resource slice of the non-uniform resource slices to a particle in the particle swarm, and obtain an initial particle swarm by attaching a slice real-time status identifier to each particle; Iteratively optimize the initial particle swarm through the improved hybrid particle swarm optimization algorithm to obtain an optimal particle set; In the optimal particle set, the score of the resource shard corresponding to each particle is determined according to the weighted sum of the preset multi-dimensional indexes, and the resource shard with the highest score is selected for resource scheduling.

3. The resource scheduling method according to claim 2, wherein The step of iteratively optimizing the initial particle swarm through the improved hybrid particle swarm optimization algorithm to obtain the optimal particle set includes: Based on the improved hybrid particle swarm optimization algorithm, the dynamic parameter range is set for the initial particle swarm according to the cloud environment monitoring data to obtain a parameterized particle swarm; Determine the weighted sum of the multi-dimensional monitoring data sets of each particle in the parameterized particle swarm to obtain a particle swarm with fitness scores; Update the particle velocity and particle position based on the particle swarm with fitness scores to obtain an optimized particle swarm; Extract multiple fitness-optimal particles from the optimized particle swarm to obtain the optimal particle set.

4. The resource scheduling method according to claim 3, wherein The step of updating the particle velocity and particle position based on the particle swarm with fitness scores to obtain the optimized particle swarm includes: Based on the real-time load data of the particle swarm with fitness scores, obtain the dynamic inertia weight parameter by determining the inertia weight of the current iteration cycle; According to the priority annotation of the particle swarm with fitness scores, obtain the optimized learning factor parameter by adjusting the individual learning factor and the social learning factor; According to the current position information of the parameterized particle swarm and the resource constraint conditions, obtain the position update rule after constraint adjustment by correcting the particle position update range; Based on the dynamic inertia weight parameter, the optimized learning factor parameter, and the position update rule after constraint adjustment, update the particle velocity and particle position to obtain the optimized particle swarm.

5. The resource scheduling method according to any one of claims 1 to 4, characterized in that The step of generating the multi-dimensional data topology map of the cloud environment monitoring data includes: Preprocess the cloud environment monitoring data to obtain a standardized data matrix; Based on the standardized data matrix, determine the resource correlation degree between nodes through the improved graph convolutional network algorithm, where the improved graph convolutional network algorithm is an algorithm with an additional dynamic attention mechanism introduced; Generate a weighted resource correlation graph according to the resource correlation degree; Fuse the weighted resource correlation graph with the physical network topology structure of the cloud environment to obtain a multi-dimensional data topology map integrating resource status and network topology.

6. The resource scheduling method according to any one of claims 1 to 4, characterized in that After the step of resource scheduling through the improved hybrid particle swarm optimization algorithm based on non-uniform resource shards, it further includes: Compare and analyze the actual resource utilization data collected during the scheduling execution process with the expected scheduling target to obtain a scheduling deviation index; When the scheduling deviation index exceeds the preset threshold, update the multi-dimensional data topology map according to the current actual resource status to obtain a corrected topology map; Re-determine the priority annotation for the quantum computing units with unfinished tasks based on the corrected topology map to obtain an updated priority annotation; Perform local adjustment on the non-uniform resource shards according to the updated priority annotation to obtain an optimized resource shard configuration; Perform re-scheduling based on the optimized resource shard configuration.

7. A resource scheduling device, characterized in that, The resource scheduling device includes: A data integration module for generating a multi-dimensional data topology map of the cloud environment monitoring data; The unit division module is used to decompose the job set to be scheduled into a set of quantum computing units that can be processed in parallel, and add priority annotations to the set of quantum computing units through a quantum heuristic algorithm; The resource division module is used to divide the physical resources of the cloud environment into non-uniform resource slices that match the task requirements according to the priority annotations and the multi-dimensional data topology graph; The resource scheduling module is used to perform resource scheduling based on the non-uniform resource slices through an improved hybrid particle swarm optimization algorithm, where the improved hybrid particle swarm optimization algorithm is an algorithm that introduces weighted sum fitness evaluation and resource constraint correction; The unit division module is further used to analyze the execution process of the job set to be scheduled based on the task dependency graph to obtain a task decomposition tree; determine the task parallelism threshold according to the resource distribution characteristics in the multi-dimensional data topology graph, and obtain the maximum number of parallelizable units based on the task parallelism threshold; divide the task decomposition tree into multiple independent subtask blocks according to the maximum number of parallelizable units to obtain a set of quantum computing units; construct a quantum annealing cost function model including the task characteristic parameters of the job set to be scheduled through a quantum heuristic algorithm based on the principle of quantum annealing; input the set of quantum computing units into the quantum annealing cost function model for optimization and solution to obtain the unit priority sorting result; add priority annotations to each quantum computing unit of the set of quantum computing units according to the unit priority sorting result; The resource division module is further used to extract the resource demand characteristics of the set of quantum computing units based on the priority annotations to obtain a task resource demand vector; generate a node resource adaptation matrix according to the node resource status and network topology relationship in the multi-dimensional data topology graph; obtain the first resource slice by reserving a unique resource block for the first-priority quantum computing unit according to the task resource demand vector and the node resource adaptation matrix; obtain the second resource slice by allocating a shared resource block for the second-priority quantum computing unit; optimize the communication path between slices based on the network delay data in the topology graph to obtain the slice network topology relationship; combine and optimize the first resource slice, the second resource slice, and the slice network topology relationship to generate non-uniform resource slices that match the task requirements.

8. A resource scheduling device, characterized in that, The resource scheduling device includes: a memory, a processor, and a resource scheduling program stored on the memory and executable on the processor. When the resource scheduling program is executed by the processor, it implements the resource scheduling method according to any one of claims 1 to 6.

Images