Task optimization scheduling method based on cloud edge-end cooperation mechanism

Abstract

The invention relates to a task optimization scheduling method based on a cloud edge-end cooperation mechanism. The method comprises the following steps: step 1, constructing the cloud edge-end cooperation mechanism; step 2, application optimization arrangement; step 3, service request intelligent assignment; and 4, dynamically optimizing and scheduling the tasks. Compared with an existing edge computing task scheduling method, the edge computing task scheduling method has the advantages that the throughput rate and the application stability of the system are maximized through application optimization arrangement based on the knowledge graph; and dynamic migration and resource scaling of the cloud side end task are realized based on dynamic optimization scheduling of the deep reinforcement learning task. The task optimization scheduling method designed according to the invention not only has high efficiency and strong performance, but also can effectively deal with the problems of low resource utilization rate, low service quality, unstable network and the like of the cloud side end, and provides a more efficient and reliable cloud side end task optimization scheduling method for industrial application.

Description

Technical Field

[0001] The present invention relates to a task optimization scheduling method, in particular to a task optimization scheduling method based on a cloud-edge-end collaboration mechanism. This method belongs to the field of artificial intelligence. Background Art

[0002] With the large-scale implementation of industries such as 5G, the Internet of Things, and industrial Internet, centralized cloud computing can no longer meet the requirements in terms of network latency, bandwidth cost, data security, service agility, etc. As the core capability base for the digital transformation of industries, edge computing has received extensive attention in the industry. Edge computing has entered the peak expectation period from the technology concept period and has become one of the important trends in future computing. At the same time, with the in-depth digital transformation, edge computing will be further strengthened in terms of the depth and breadth of industry applications and will be widely applied in many fields such as industry, healthcare, transportation, education, and energy, making full use of the innate advantages of being close to the user side to provide low-cost and high-quality services for industry users.

[0003] The integration of edge computing technologies is becoming the "booster" for the implementation of the edge computing industry. The digital transformation of industries requires that edge computing should have differentiated and customized capabilities for specific industries to meet the requirements of industry applications in terms of high-performance computing, massive access, intelligent analysis, security protection, etc. As the basic capability base, edge computing has a natural affinity with various emerging technologies such as artificial intelligence, big data, blockchain, and 5G. By "edge-computing-izing" various technologies and gradually forming a cloud-edge-end integrated architecture, it shields the underlying distributed heterogeneous resources and provides an application unified running environment upward to achieve unified application management, agile business deployment, reduction of latency and bandwidth costs, and secure data storage.

[0004] Due to the different resource scales and scattered locations of the cloud, edge, and end, and the different requirements of business applications in terms of latency, performance, quality of service, cost, etc., how to achieve unified cloud-edge-end computing resource scheduling, application optimization orchestration, and task dynamic scheduling, while maximizing resource utilization and system operation stability while ensuring the user experience, has become the main challenge. Although the industry is currently exploring cloud-edge-end resource optimization scheduling and task optimization scheduling from different dimensions such as resources, traffic, data, applications, and algorithms, in actual business scenarios, due to the diversity and complexity of business, there is still room for optimization and improvement in aspects such as access latency, traffic migration, and real-time monitoring. Summary of the Invention

[0005] The present invention proposes a task optimization scheduling method based on a cloud-edge-end collaborative mechanism, and its objectives are as follows: to build a cloud-edge-end integrated mechanism to achieve unified development, deployment, scheduling, and management of distributed applications, as well as unified global scheduling and management of resources, traffic, and services; further, to integrate technologies such as knowledge graphs and deep reinforcement learning to achieve cross-cloud, edge, and end application optimization orchestration, intelligent assignment of service requests, and dynamic optimization scheduling of tasks, maximizing application operation efficiency, cloud-edge node resource utilization rate, system operation stability, and enhancing the engineering practicability of the system in the cloud-edge-end environment.

[0006] The technical solution adopted by the present invention to achieve the above objectives is as follows:

[0007] A task optimization scheduling method based on a cloud-edge-end collaborative mechanism includes the following steps:

[0008] 1) Build a cloud-edge-end collaborative mechanism;

[0009] 2) Build a knowledge graph and optimize the orchestration of the application of the knowledge graph based on the feature information in the cloud-edge-end collaborative mechanism;

[0010] 3) Intelligently assign service requests;

[0011] 4) Perform dynamic optimization scheduling of tasks based on the deep reinforcement learning algorithm.

[0012] The said step 1) includes the following steps:

[0013] 1.1) Build a cloud-side cloud-native cluster based on kubernetes;

[0014] 1.2) On the cloud-edge link, build an edge-cloud collaboration engine based on the cloud-side controller and the edge-side controller to achieve unified development, deployment, scheduling, and management of distributed applications across the cloud and the edge, as well as unified global scheduling and management of resources, traffic, and services;

[0015] 1.3) Build an edge-side cloud-native cluster based on k3s;

[0016] 1.4) The end side uses the edge gateway and the edge all-in-one machine as carriers and is distributedly deployed near the device side of the industrial site for data collection, preprocessing, and control instruction execution.

[0017] The said step 2) includes the following steps:

[0018] 2.1) Establish a knowledge graph for application orchestration and scheduling;

[0019] 2.2) Based on the feature information of the current cloud-edge-end collaborative mechanism, perform knowledge graph search and decision-making to achieve node scheduling and replica scaling for different load types;

[0020] 2.3) Automatically complete and correct the knowledge graph based on end-to-end entity recognition, relation extraction, and relation completion using machine learning.

[0021] Specifically, step 2.1) is as follows:

[0022] Construct a knowledge graph based on the multi-resource heterogeneous feature information of edge nodes and the feature information of published applications, with the optimization goals of maximizing system throughput and maximizing application stability, combined with the method of extracting artificial knowledge and experience.

[0023] Step 3) includes the following steps:

[0024] 3.1) Real-time calculate the network system status, load change characteristics, and resource usage of the cloud-edge-end collaboration mechanism;

[0025] 3.2) According to the requested resource type of the service request, preliminarily screen the service nodes from the aspects of resource constraints, application matching degree, and load status;

[0026] 3.3) With the goal of maximizing the requested service quality as the objective function, find the service node with the maximum objective function among the service nodes that have passed the preliminary screening, and assign the service request.

[0027] Step 4) includes the following steps:

[0028] 4.1) Real-time calculate and collect the network status and load status metrics of cloud-edge-end nodes, and use the ARIMA time series prediction model to predict the node metrics;

[0029] 4.2) Combine the real-time calculation and predicted metrics of the network status and load status of cloud-edge-end nodes, and use the deep reinforcement learning algorithm for task optimization scheduling analysis;

[0030] 4.3) Combine the results of task optimization scheduling analysis and execute dynamic scheduling decisions for application migration and resource scaling tasks.

[0031] Step 4.2) includes the following steps:

[0032] 4.2.1) Construct a sample pool:

[0033] Construct a sample pool with the node resource heterogeneous features, network status, task features, and application resource requirements as the state space, the application operation efficiency and node resource utilization rate as the target value, and the application migration and resource scaling as the action space;

[0034] 4.2.2) Construct a policy network and a value network:

[0035] Adopt a deep reinforcement learning algorithm based on the attention mechanism, take the state space of the sample pool as the input and the action space as the output to achieve dynamic optimization decisions for cloud-edge-end tasks, and continuously optimize the algorithm by combining historical decision results including node migration scheduling and resource scaling scheduling;

[0036] 4.2.3) Adaptive learning:

[0037] Perform adaptive training and learning of the deep reinforcement learning algorithm based on historical task optimization data.

[0038] A task optimization scheduling system based on the cloud-edge-end collaboration mechanism includes a memory and a processor; the memory is used to store computer programs; the processor is used to implement the task optimization scheduling method based on the cloud-edge-end collaboration mechanism when executing the computer programs.

[0039] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the task optimization scheduling method based on the cloud-edge-end collaboration mechanism is implemented.

[0040] The present invention has the following beneficial effects and advantages:

[0041] 1. The present invention solves the problem of application optimization and orchestration in cloud-edge-end collaboration, and improves the system throughput and application stability through node scheduling and replica scaling of application loads.

[0042] 2. The present invention combines factors such as network system status, load change characteristics, and requested resource types to perform intelligent assignment strategies for service requests and improve the quality of service requests.

[0043] 3. The present invention improves the application operation efficiency and node resource utilization rate through task dynamic optimization scheduling based on deep reinforcement learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 is a cloud-edge-end collaboration framework diagram;

[0045] Figure 2 is a schematic diagram of the task optimization scheduling method based on the cloud-edge-end collaboration mechanism provided by the embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0046] The following further describes the present invention in detail with reference to the drawings and embodiments.

[0047] As Figure 1 shown, the present invention includes the following steps:

[0048] Step 1: Construction of the cloud-edge-end collaboration mechanism, and the specific method is:

[0049] 1) Build a cloud-side cloud-native cluster based on Kubernetes to implement an AI model automated pipeline, big data processing, and application development platform;

[0050] 2) On the cloud-edge link, the cloud-side controller and the edge-side controller jointly form an edge-cloud collaboration engine to achieve unified development, deployment, scheduling, management of distributed applications across the cloud and the edge, as well as unified global scheduling and management of resources, traffic, and services;

[0051] 3) Build an edge-side cloud-native cluster based on k3s, which has capabilities such as application scheduling, data storage, and offline autonomy, achieving edge-cloud collaboration upwards and remote control and instruction communication with end nodes downwards;

[0052] 4) On the end side, with edge gateways and edge all-in-ones as carriers, they are distributed near the equipment side of the industrial site for data collection, preprocessing, control instruction execution, etc., and at the same time support the collaboration mechanism with the edge side.

[0053] Step 2: Application optimization and orchestration based on knowledge graph technology. The specific method is as follows:

[0054] 1) Build a knowledge graph for application orchestration and scheduling, mainly based on feature information such as multi-resource heterogeneity characteristics of edge nodes (system resources, communication capabilities, geographical characteristics, system architecture, etc.), published application characteristics (system resource requirements, data affinity, etc.), etc. With maximizing system throughput and maximizing application stability as optimization goals, build the knowledge graph by combining the method of extracting artificial knowledge and experience;

[0055] 2) Based on the current cloud-edge-end system feature information, perform knowledge graph search and decision-making to achieve node scheduling and replica scaling for load types such as AI applications, data collection applications, and business applications;

[0056] 3) Based on end-to-end entity recognition, relationship extraction, and relationship completion of machine learning, etc., achieve automatic completion and correction of the knowledge graph, and enhance the accuracy of knowledge graph decision-making.

[0057] Step 3: Intelligent assignment of service requests. The specific method is as follows:

[0058] 1) Real-time calculate the network system status, load change characteristics, and resource usage of the cloud-edge-end;

[0059] 2) Combine the requested resource type to initially screen service nodes from resource limitations, application matching degree, and load status;

[0060] 3) With maximizing the requested service quality as the objective function, find the service node with the maximum objective function among the service nodes that pass the initial screening and perform service request assignment.

[0061] Step 4: Task dynamic optimization scheduling based on deep reinforcement learning. The specific method is as follows:

[0062] (1) Construct a sample pool

[0063] Construct a sample pool with {node resource heterogeneous features, network status, task features, application resource requirements...} as the state space, application running efficiency and node resource utilization rate as the target value, and application migration and resource scaling as the action space.

[0064] (2) Construct a policy network and a value network

[0065] Adopt a deep reinforcement learning algorithm based on the attention mechanism to realize dynamic optimization decision-making of cloud-edge-end tasks, and continuously combine historical decision results, including node migration scheduling and resource scaling scheduling of applications, to ensure the operation stability of the cloud-edge-end collaborative cluster system.

[0066] (3) Adaptive learning

[0067] Based on historical task optimization data, conduct adaptive training and learning of the deep reinforcement learning algorithm to make task dynamic optimization more accurate and efficient.

[0068] Figure 2 The following is the process of the task optimization scheduling method based on the cloud-edge-end collaborative mechanism provided by the present invention. The specific design method mainly includes the following four steps:

[0069] 1) Construction of the cloud-edge-end collaborative mechanism

[0070] Build a cloud-side kubernetes cluster with 3 master nodes and 6 worker nodes in a single group based on rancherrke2; build an edge-side cluster with 1 master node and 3 worker nodes in 3 groups based on rancherk3s; deploy 5 groups of edge network gateway nodes based on the linux system. The cloud-to-edge and edge-to-end both have bidirectional reachable tcp networks.

[0071] Table 1 Cloud-edge-end resource configuration

[0072]

[0073]

[0074] 2) Application optimization and orchestration based on knowledge graph technology

[0075] Build a knowledge graph for application orchestration and scheduling, with a scale of more than 300 nodes, and realize node scheduling and replica scaling for load types such as AI applications, data collection applications, and business applications. The comparison with the cloud-edge-end system without application optimization and orchestration is shown in the following table.

[0076] Table 2 Comparison of Application Optimization Orchestration Effects Based on Knowledge Graph Technology

[0077] Index item With optimized layout algorithm Without optimized layout algorithm System throughput 95％ 50％ Application stability 90％ 70％ Application running stability Good Average

[0078] As can be seen from Table 2, the application optimization orchestration based on knowledge graph technology has obvious advantages in system throughput and application stability.

[0079] 3) Intelligent Assignment of Service Requests

[0080] Simulate 100 groups of random requests covering data types, business types, and intelligent service types, and send them to the cloud-edge-end system. Through the intelligent assignment mechanism of service requests, the service response time can be reduced by 20%, and the fault tolerance rate can be increased by 30%, achieving an overall improvement in service quality.

[0081] 4) Task Dynamic Optimization Scheduling Based on Deep Reinforcement Learning

[0082] Based on the completion of the construction of the cloud-edge-end system in steps 1 to 3, a task dynamic optimization scheduling mechanism based on deep reinforcement learning is constructed. The comparison with the system without the task dynamic optimization scheduling mechanism is shown in the following table.

[0083] Table 3 Comparison of Task Dynamic Optimization Scheduling Algorithm Effects Based on Deep Reinforcement Learning

[0084]

[0085]

[0086] As can be seen from Table 3, the task dynamic optimization scheduling based on deep reinforcement learning has obvious advantages in application operation efficiency and node resource utilization.

Claims

1. A task optimization scheduling method based on cloud-edge-device collaboration mechanism, characterized in that: The following steps are involved: 1) Build a cloud-edge-device collaboration mechanism; 2) Build a knowledge graph and optimize the application of the knowledge graph based on the feature information in the cloud-edge-end collaboration mechanism; 3) Intelligently assign service requests; 4) Dynamically optimize and schedule tasks based on deep reinforcement learning algorithms.

2. According to the method of claim 1, the task optimization scheduling method based on the cloud-edge-end collaboration mechanism is characterized in that: The step 1) comprises the following steps: 1.1) Build a cloud-native cluster on the cloud side based on Kubernetes; 1.2) On the cloud-edge link, an edge-cloud collaboration engine is built based on the cloud-side controller and the edge-side controller to achieve unified development, deployment, scheduling, and management of distributed applications across clouds and edges, as well as unified global scheduling management of resources, traffic, and services; 1.3) Build edge cloud native cluster based on k3s; 1.4) The end side uses edge gateways and edge all-in-one machines as carriers and is distributed and deployed near the equipment on the industrial site to perform data collection, preprocessing, and control instruction execution.

3. According to the task optimization scheduling method based on the cloud-edge-end collaboration mechanism of claim 1, it is characterized in that: The step 2) comprises the following steps: 2.1) Establish a knowledge graph for application orchestration and scheduling; 2.2) Based on the characteristic information of the current cloud-edge-device collaboration mechanism, a knowledge graph search decision is made to achieve node scheduling and replica scaling for different load types; 2.3) Automatically complete and correct the knowledge graph based on end-to-end entity recognition, relationship extraction and relationship completion based on machine learning.

4. According to the method of claim 3, the task optimization scheduling method based on the cloud-edge-end collaboration mechanism is characterized in that: The step 2.1) is specifically as follows: Based on the heterogeneous feature information of multiple resources of edge nodes and the feature information of published applications, the knowledge graph is constructed by combining artificial knowledge and experience extraction with the optimization goals of maximizing system throughput and maximizing application stability.

5. According to the method of claim 1, the task optimization scheduling method based on the cloud-edge-end collaboration mechanism is characterized in that: The step 3) comprises the following steps: 3.1) Real-time calculation of the network system status, load change characteristics and resource usage of the cloud-edge-end collaboration mechanism; 3.2) Based on the resource type requested by the service request, the service nodes are initially screened from the perspective of resource constraints, application matching, and load status; 3.3) Taking maximizing the request service quality as the objective function, find the service node with the largest objective function among the service nodes that pass the initial screening, and assign the service request.

6. According to the method of claim 1, the task optimization scheduling method based on the cloud-edge-end collaboration mechanism is characterized in that: The step 4) comprises the following steps: 4.1) Calculate and collect network status and load status indicators of cloud edge nodes in real time, and use the ARIMA time series prediction model to predict node indicators; 4.2) Combined with the real-time calculation and prediction indicators of the network status and load status of cloud edge nodes, a deep reinforcement learning algorithm is used to perform task optimization scheduling analysis; 4.3) Combined with the analysis results of task optimization scheduling, dynamic scheduling decisions are made for application migration and resource scaling tasks.

7. According to claim 6, a task optimization scheduling method based on cloud-edge-end collaboration mechanism is characterized in that: The step 4.2) comprises the following steps: 4.2.1) Constructing sample pool: The sample pool is constructed with node resource heterogeneity characteristics, network status, task characteristics, and application resource requirements as the state space, application operation efficiency and node resource utilization as the target value, and application migration and resource scaling as the action space; 4.2.2) Build strategy network and value network: Adopting a deep reinforcement learning algorithm based on the attention mechanism, taking the state space of the sample pool as input and the action space as output, it realizes dynamic optimization decision-making of cloud-edge tasks, and continuously optimizes the algorithm by combining historical decision results including node migration scheduling and resource scaling scheduling; 4.2.3) Adaptive Learning: Based on historical task optimization data, adaptive training and learning of deep reinforcement learning algorithms are performed.

8. A task optimization scheduling system based on cloud-edge-end collaboration mechanism, characterized in that: It includes a memory and a processor; the memory is used to store a computer program; the processor is used to implement a task optimization scheduling method based on a cloud-edge-end collaboration mechanism as described in any one of claims 1-7 when executing the computer program.

9. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, a task optimization scheduling method based on a cloud-edge-terminal collaboration mechanism as described in any one of claims 1 to 7 is implemented.

Images