A method for merging and scheduling tasks based on cloud-edge collaborative computing
By establishing a task scheduling model and a consolidation scheduling algorithm in the cloud-edge collaborative environment, combining the cloud-edge collaborative mechanism and preventive scheduling, the problem of difficult and time-consuming allocation timing of industrial computing tasks is solved, and fast and effective scheduling and resources are achieved Improved utilization.
Patent Information
- Application Number
- CN202110450162.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-04-25
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2041-04-25
AI Technical Summary
In the prior art, the timing of the allocation of industrial computing tasks in the cloud-edge collaborative environment is difficult to unify and take a long time, and it is difficult to meet the user's requirements for task response time.
A combined scheduling method based on cloud-edge collaborative computing tasks is proposed. By establishing a cloud-edge collaborative task scheduling model, combining cloud-assisted edge computing and edge-assisted cloud computer system, it adopts a combined scheduling algorithm and preventive scheduling judgment method, grouping and generating Sub-scheduling scheme, and uncertainty is evaluated through the interval number method, and the scheduling scheme is optimized to improve efficiency.
It realizes rapid and effective scheduling of large-scale industrial computing tasks in a cloud-edge collaborative environment, reduces energy consumption and task response time, and improves resource utilization.
Smart Images

Figure CN113553146B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a method for merging and scheduling cloud-edge collaborative computing tasks, and belongs to the field of industrial computing task algorithm optimization. Background Art
[0002] At the historic intersection of the rise of the fourth industrial revolution and the transformation and upgrading of my country's manufacturing industry, the Industrial Internet of Things (IIoT) was born with the deep integration of new-generation information technologies such as the Internet, big data, and artificial intelligence with the manufacturing industry. The Industrial Internet builds a new industrial production and service system that is fully connected with all factors, the entire industrial chain, and the entire value chain by realizing the comprehensive interconnection of people, machines, and things.
[0003] The terminal devices of the Industrial Internet are usually some sensors, smart devices and wearable devices, etc. Their computing resources and energy are limited, but the amount of data generated is huge. In order to meet the real-time requirements of industry and reduce the consumption of network and IT resources, cloud-edge collaboration is introduced into the Industrial Internet.
[0004] Task scheduling in the Industrial Internet refers to mapping industrial computing tasks to corresponding computing resources through appropriate configuration strategies to achieve full utilization of resources and efficient operation of tasks. It is a heterogeneous complex virtual task scheduling problem. The introduction of cloud-edge collaboration makes the structure of the Industrial Internet more complex and also brings greater challenges to task scheduling in the Industrial Internet. In the cloud-edge collaborative computing environment of the Industrial Internet, computing resources (including CPU, memory, I / O, etc.) are abstracted as virtual resources. These computing resources are often distributed in various heterogeneous computing devices such as local and remote locations, including servers, PCs, high-performance computers, etc. How to quickly configure virtual environments such as CPU, memory, and network to execute tasks according to task requirements, and how to improve the utilization of computing resources is not only an important part of the efficient operation of the platform, but also a difficulty in platform construction. Therefore, in the cloud-edge collaborative computing platform for the Industrial Internet, finding an efficient optimization algorithm to solve the scheduling problem of industrial computing tasks has important theoretical value and practical significance.
[0005] This type of scheduling problem has been proven to be an NP-Hard problem. For this type of scheduling problem, researchers usually use population-based metaheuristic algorithms to solve it. However, population-based metaheuristic algorithms themselves have high time complexity based on the characteristics of population evolution and iteration. At the same time, in the task scheduling problem of the industrial Internet, industrial computing tasks usually have more data and larger scale than mobile applications. It takes a long time to assign large-scale tasks simultaneously or in sequence, which makes it difficult to meet users' requirements for task response time. Summary of the invention
[0006] The technical problem solved by the present invention is: in view of the problem in the current existing technology that traditional industrial computing tasks have a large amount of computing data and a larger scale than mobile applications, and the allocation timing is difficult to unify and takes a long time, a method for merging and scheduling cloud-edge collaborative computing tasks is proposed.
[0007] The present invention solves the above technical problems by the following technical solutions:
[0008] A method for merging and scheduling cloud-edge collaborative computing tasks, the steps are as follows:
[0009] (1) External users send industrial computing tasks to the computing platform including the terminal device layer, cloud, and edge, and determine two cloud-edge collaboration mechanisms;
[0010] (2) Based on the cloud-edge collaboration mechanism determined in step (1), a cloud-edge collaboration task scheduling model is established;
[0011] (3) Run the merged scheduling algorithm, group and generate sub-scheduling plans, evaluate the sub-scheduling plans through the cloud-edge collaborative task scheduling model in step (2), and select the sub-scheduling plan with the best evaluation result for output within the specified number of iterations of the merged scheduling algorithm;
[0012] (4) Merge and fine-tune the sub-scheduling plans obtained in step (3) to obtain the final scheduling plan.
[0013] In step (1), the cloud-edge collaboration mechanism includes cloud-assisted edge computing and edge-assisted cloud computing, specifically:
[0014] When the edge server is executing a task, the cloud sends prior knowledge or historical data to the edge;
[0015] During the task execution process of the cloud server, the edge end collects and pre-processes the data of each manufacturing terminal at the terminal device layer and sends it to the cloud.
[0016] In step (2), the cloud-edge collaborative task scheduling model is specifically as follows:
[0017]
[0018] In the formula, the problem target It is mainly divided into two parts, It represents the time cost required to complete all tasks in a set of industrial computing task requests submitted. It represents the energy cost required to complete all tasks, T m and E mare two constants used to normalize the maximum task execution time and the total energy consumption of executing tasks. λ1,λ2 are objective weights, which are set according to the requirements of task execution time and energy consumption, satisfying λ1+λ2=1 and λ1,λ2∈[0,1]. In general, λ1=λ2=0.5 is set, indicating that the energy consumption target and the task execution time target are equally important.
[0019] In step (2), the uncertainty of the cloud-edge collaborative task scheduling model needs to be evaluated. The interval number method is used to evaluate the uncertainty of task I. i Calculation time Data transmission rate from terminal to edge Describe, if the task needs to be retransmitted, the amount of data from device l Upload time It is necessary to recalculate and calculate the evaluation results through the determined number of intervals and cloud-edge collaboration mechanism, specifically:
[0020] When the cloud-edge collaboration mechanism is cloud-assisted edge computing, the method is as follows:
[0021]
[0022] In the formula, Task I i The time it takes to upload data, download data from the cloud, and wait for the server to release available resources. is the task execution time, is the task termination time, Λi For Task I i The set of predecessor tasks, α1 and α2 are weight coefficients, satisfying the conditions of α1+α2=1 and α1,α2∈[0,1], R cpu,i , R mem,i Task I i CPU and memory requests, For equipment The transmission power C cpu,j , C mem,j Assign edge servers S edge,j CPU and memory capacity, and They are the static power and dynamic power of the server respectively;
[0023] When the cloud-edge collaboration mechanism is edge-assisted cloud computing, the method is as follows:
[0024]
[0025]
[0026] In the formula, β represents task Ii Whether preprocessing is required at the edge node. For tasks that do not require data preprocessing at the edge node, β=0; otherwise, β=1. Represents Task I i The time required to upload data to the cloud.
[0027] In step (3), when the merge scheduling algorithm starts, the grouping strategy is designed according to the cloud-edge collaboration mechanism, specifically:
[0028] (i) Sort the industrial computing tasks, calculate the delay time of each industrial computing task, and obtain the task attribute set C = {R cpu,i ,R mem,i ,time_delay i};
[0029] (ii) Arrange the three attributes in the task attribute set in descending order to obtain the comprehensive ranking result S i , and sort them in ascending order, specifically:
[0030]
[0031] Where N is the number of tasks, P c,i The position of the task in the independent sorting of c attributes;
[0032] (iii) Calculate the resource energy consumption cost of each server in the cloud and sort the resource costs of all servers in the cloud. The resource energy consumption cost calculation formula is as follows:
[0033]
[0034] In the formula, overhead E / C,j For cloud server S cloud,j Resource and energy consumption costs;
[0035] (iv) Group the industrial computing tasks evenly according to the ascending order results obtained in step (ii), and group the cloud servers according to the resource cost ranking. The ratio of the number of cloud servers allocated to the computing tasks in the first half of the process to the number of cloud servers allocated to the computing tasks in the second half of the process is 3:2.
[0036] In step (3), in the combined scheduling algorithm, the server availability is determined by the preventive scheduling determination method and the task waiting time t is calculated. wait,i , to obtain the sub-scheduling results, the specific steps of the preventive scheduling determination method are as follows:
[0037] (a) Calculate the delay time of current industrial computing tasks;
[0038] time_delayi =t max,i -t i
[0039] Where, t max,i and t i Mission I i The maximum execution time and expected computation time of
[0040] (b) within the delay time obtained in step (a), judging in the order of the available server list whether the current server meets the computing requirements of the current industrial computing task;
[0041] The specific determination method is as follows:
[0042]
[0043] In the formula, is the set of tasks being executed in the server, t rest,jk For task I in the task set jk The remaining execution time;
[0044] (c) Calculate the task I corresponding to the output of the merge scheduling algorithm i The estimated waiting time is:
[0045]
[0046] (d) According to the calculated task I i The estimated waiting time of industrial computing tasks is calculated based on the estimated waiting time, which is:
[0047]
[0048] In step (4), after obtaining the sub-scheduling schemes, the sub-scheduling schemes are merged according to the corresponding levels of each industrial computing task, the industrial computing task grouping results, and the cloud server grouping results. After obtaining the overall scheduling scheme, the Best Fit strategy is used to fine-tune the overall scheduling scheme using the maximum task completion time and total energy consumption cost to obtain the final scheduling scheme.
[0049] Among them, the calculation formula for the maximum task completion time and total energy consumption cost of the total scheduling plan is as follows:
[0050]
[0051] In the formula, M is the number of grouping processes, time max,i and E i are the maximum task completion time and energy consumption cost of the sub-scheduling solution obtained by process i, and Z final is the optimization target value of the overall scheduling solution.
[0052] The advantages of the present invention compared with the prior art are:
[0053] (1) The present invention provides a method for merging and scheduling cloud-edge collaborative computing tasks. Two cloud-edge collaborative modes are designed according to the different requirements of industrial computing tasks. The uncertainty scenarios in the cloud-edge collaborative process are combined to construct an uncertainty model for industrial computing task scheduling in the cloud-edge collaborative environment. For this purpose, an evaluation method based on interval numbers is designed, so that the algorithm can be applied to the uncertain cloud-edge environment. The proposed task scheduling model is more in line with engineering practice.
[0054] (2) The present invention adopts a preventive scheduling judgment method, which is not limited to the currently available resources. Instead, it predicts whether the server resources are available based on the resources that may be released in the server and the delay time of the task. At the same time, it proposes a grouping method for the purpose of reducing energy consumption for two cloud-edge collaboration modes, and applies the idea of group parallelization to improve the efficiency of algorithm solving. With this grouping method as the core, a merge scheduling algorithm is proposed. The basic search operator is extracted and improved from a variety of population-based metaheuristic algorithms, and a random step hyper-heuristic method is introduced to find the optimized search operator in the iterative process. According to the scheduling model, a scheduling scheme mapping mechanism is designed to decode the indirect code into a scheduling scheme. The overall scheduling scheme is obtained through the merging mechanism, and the refined adjustment algorithm is introduced to optimize the scheduling scheme. Based on the merge scheduling algorithm, fast and effective scheduling of large-scale tasks in cloud-edge collaboration scenarios can be achieved. BRIEF DESCRIPTION OF THE DRAWINGS
[0055] Figure 1 A flow chart of the balanced scheduling scheme provided for the invention;
[0056] Figure 2 A cloud-assisted edge computing model task and edge server failure decision flow chart provided for the invention;
[0057] Figure 3 A flow chart of the merge scheduling algorithm provided for the invention;
[0058] Figure 4 A schematic diagram of the operation of the case server tasks provided for the invention; DETAILED DESCRIPTION
[0059] A method for merging and scheduling computing tasks based on cloud-edge collaborative computing tasks is proposed based on the meta-heuristic algorithm. The algorithm explores the approximate optimal scheduling scheme for large-scale industrial computing tasks and improves the efficiency of problem solving. First, an evaluation method is proposed for the various uncertainties in the scheduling of computing tasks under cloud-edge collaboration, so that the algorithm can effectively solve the task scheduling problem in an uncertain environment. Then, a preventive scheduling judgment method is proposed to predict and judge the possible resources in the server and improve resource utilization. Finally, a grouping strategy is proposed for two cloud-edge collaboration modes. The rapid scheduling of large-scale tasks is achieved through three steps: dividing tasks, resources, and populations, solving sub-populations in parallel, and merging and fine-tuning scheduling schemes.
[0060] There are few studies on large-scale task scheduling methods in existing cloud-edge environments. However, similarly, some researchers have proposed a method for large-scale task scheduling in cloud computing, namely the IPSO method, to solve this scheduling problem. The problem background considered in the study can be described as: the cloud computing system receives N independent tasks of different sizes and needs to schedule the tasks to M virtual machines. Each task can be assigned to any virtual machine, but each task can only be scheduled to one virtual machine. Under the above background, the large-scale task scheduling method is studied with the main goals of minimizing task completion time and maximizing resource utilization. The IPSO scheduling method proposed for this scheduling problem is mainly divided into three stages: batch collection of tasks, batch task scheduling, and balancing the final scheduling plan.
[0061] In the first stage, after a large-scale task sequence arrives, the tasks are divided into batches according to resources. In order to maximize resource utilization, the division index parameter μ is defined as the average value of available resources as follows, that is, the tasks are divided sequentially, and a new batch is started when the sum of the task lengths in the current batch reaches μ until the task division is completed.
[0062]
[0063] In the second stage, a scheduling scheme is generated for each batch of tasks based on the PSO algorithm, and all batches are scheduled in turn;
[0064] In the third stage, tasks are moved from the most heavily loaded VM with the largest completion time to the lightest loaded VM with the smallest completion time until the list of available VMs is cleared. The specific process is as follows: Figure 1 shown.
[0065] The above scheduling method for scheduling problems in cloud computing divides large-scale tasks into different batches and schedules them in sequence. According to experimental results, the scheduling efficiency is improved compared with all-at-once scheduling, but there are still problems and shortcomings. The main manifestations are that, first, the scheduling mechanism adopts a serial scheduling method, and there is still a problem of long scheduling time, that is, the scheduling efficiency needs to be further improved; second, the method uses traditional methods to determine whether the server has enough resources to provide computing services for the task, which is limited to the resources that the server can currently provide, and lacks the prediction of resources to be released within the waiting time.
[0066] Metaheuristic algorithms are widely used to solve scheduling problems due to their powerful global search capabilities. However, with more and more terminals in the industrial Internet, the number of requests to be processed is growing exponentially. When dealing with large-scale scheduling problems, the time complexity of population-based metaheuristic algorithms is relatively high. Based on the metaheuristic algorithm, this paper proposes a merged scheduling algorithm for cloud-edge collaborative tasks, explores the approximate optimal scheduling scheme for large-scale industrial computing tasks, and improves the efficiency of problem solving. The complete technical solution can be divided into three parts: "cloud-edge collaborative mechanism design", "multiple uncertainty analysis" and "merged scheduling algorithm", specifically:
[0067] (1) Cloud-edge collaboration mechanism
[0068] The task scheduling model in the cloud-edge collaborative environment of the industrial Internet consists of three levels: terminal device layer, edge end, and cloud end. The terminal device layer is located at the bottom of the entire model, including manufacturing equipment in the industrial Internet and users involved in manufacturing. The edge end is located in the middle of the model. In addition to providing computing resources close to the data source, it also plays a role in connecting the cloud and terminal device layers. The computing nodes on the edge end bring computing resources, storage, and services to the edge of the network, reducing the response delay of industrial computing tasks and the communication bandwidth required for data exchange. The edge node has both wireless data transmission and wired broadband connection functions, and can download trained models from the cloud or transmit data required for tasks to the cloud. The cloud mainly includes remote cloud data centers, which are responsible for providing large-scale computing resources and data storage space to achieve centralized management and resource sharing. The prior knowledge required in the calculation process of industrial computing tasks is stored on the computing nodes in the cloud. At the same time, tasks such as delay-insensitive computing tasks, data management, and service provision are usually performed in the cloud.
[0069] The model in this invention abstracts industrial computing tasks into two types of tasks. The first type of industrial computing task only needs to obtain real-time data or information transmitted from the bottom layer during the calculation process. In addition to obtaining real-time data from the terminal device layer during execution, the second type of task also needs to download prior knowledge or historical data from the cloud, and the two are combined for calculation. Therefore, assigning an industrial computing task to a cloud server or an edge server forms two cloud-edge collaboration mechanisms:
[0070] (a) Cloud-assisted edge computing: When a task is assigned to an edge server, in addition to collecting the required data from task-related industrial equipment, the cloud data center also collaboratively sends prior knowledge or historical data to the edge during task execution. Therefore, additional communication costs will be introduced from the cloud to the edge;
[0071] (b) Edge-assisted cloud computing: When a task is assigned to a cloud server, the edge server will collaborate with the cloud server to collect and pre-process data from multiple related manufacturing devices through wireless communication technology, and forward the data to the cloud server through wired optical fiber during task execution. Therefore, additional communication costs will be introduced from edge resources to cloud resources;
[0072] In the above two cloud-edge collaboration modes, there are multiple uncertainties from the network, nodes and tasks. Therefore, the present invention will analyze and model the uncertainties existing in cloud-edge collaboration based on the two collaboration modes.
[0073] (2) Multiple uncertainty analysis
[0074] The operation, management, and maintenance of a complex task processing platform consisting of "industrial terminal-edge-cloud" are very complex. In addition to the problems that may exist in each component itself, there are also uncertainties that are difficult to avoid or extremely costly to resolve in the communication and coordination process of each component. Therefore, the handling of uncertainty in the cloud-edge environment is crucial to maintaining the sustainable use of the platform. In actual application scenarios, the uncertainty of cloud-edge collaborative industrial computing task scheduling mainly comes from three parts: computing nodes, network links, and the task itself.
[0075] A set of industrial computing task requests I = {I1, I2, ..., I n} is submitted to the scheduling system. Tasks can be independent tasks or process-based tasks with predecessor and successor constraints. The completion of each task requires uploading status information and processing data from one or more industrial manufacturing equipment. Considering two cloud-edge collaboration modes, an industrial computing task can be described by a nine-tuple. Where R cpu,i ,R mem,i Represents Task I iThe number of CPU cores and memory size required for execution in the server; max,i Indicates the maximum task execution time acceptable to the user. Here, the task execution time is defined as the time from task submission to task completion calculation; t i It indicates the expected computing time of the task under the condition that the task's requirements for the number of CPU cores and memory size are met; m i Represents Task I i The corresponding number of industrial equipment, and the collection Contains all tasks related to I i Related industrial equipment; Describes the need to start from O i m i The amount of data uploaded by each device is i Keep the length consistent; It represents the amount of knowledge that needs to be downloaded from the cloud data center. The value can be 0, which means that the calculation of this task does not need to download prior knowledge from the cloud; Λ i Including Task I i All predecessor tasks, that is, task I i Must be in Λ i It can only start after all the predecessor tasks are completed.
[0076] The edge server and the cloud server can be described by the same triplet, Among them C cpu,j ,C mem,j The number of CPU cores and memory that the server can provide are defined respectively; It represents the set of all tasks running in the server. Each task in progress can be described by three characteristics in Indicates that the server has completed the task The remaining time required can be estimated from the run log.
[0077] Based on the above background, in order to study the uncertainty in cloud-edge collaborative task scheduling, the present invention abstracts the uncertain scenarios mentioned above and summarizes them into the following four situations, and mainly uses interval numbers to describe the uncertainty;
[0078] (a) Wireless network transmission uncertainty
[0079] When mission data is transmitted through wireless communication technology such as 5G technology, it will be affected by communication bandwidth, interference from surrounding wireless networks, complex weather, etc., which will reduce the transmission rate to a certain extent on a theoretical basis. This model uses interval numbers to describe the uncertainty of the variable wireless transmission speed, and the value is uniformly distributed within the interval, as shown in the following formula, where r, Δr, r real They represent the static transmission rate, transmission rate uncertainty, dynamic transmission rate interval number and actual transmission rate respectively, where γ trans Represents transmission uncertainty, a positive number less than 1:
[0080]
[0081] (b) Task computation time uncertainty
[0082] Industrial manufacturing tasks processed in the Industrial Internet are basically real-time tasks. Although their execution time can be predicted, the accuracy of the prediction of the actual execution time cannot be guaranteed. Whether it is an edge computing node or a cloud computing node, it is difficult to ensure that its performance is relatively static and stable, especially in a cloud environment. In order to improve the utilization of server hardware resources, multiple virtual machines are configured in a physical machine to use resources synchronously. This resource sharing method is prone to resource interference between different virtual machines, which in turn leads to performance fluctuations of cloud computing nodes. This model task I i The number of intervals for calculating time The following formula shows that the actual value is evenly distributed within the interval:
[0083]
[0084] (c) Uncertainty in data transmission from industrial equipment
[0085] Industrial manufacturing tasks are usually completed by multiple terminal devices in collaboration, so the data required for task calculation also needs to be collected from multiple devices and transmitted to the edge server. Once a certain industrial device fails to transmit data, the device needs to retransmit the data. i For example, if the relevant equipment il If the data transmission fails, the device data upload time is as shown in the following formula. At this time, the time for task data to be transmitted from the terminal device layer to the edge end is as shown in the following two formulas, which may increase due to the secondary retransmission:
[0086]
[0087] (d) Edge server wired network interruption
[0088] The edge is usually connected to the remote cloud data center through a wired network. When the wired network fails due to reasons such as link aging, the edge server cannot transmit or interact with the cloud. edge,j There is an additional attribute Indicates S edge,j The probability of a wired network connection failure occurring.
[0089] When such uncertain events are expected to occur, in cloud-assisted edge computing, the faulty edge server cannot download the prior knowledge data from the cloud. In the edge-assisted cloud computing mode, the task cannot be transmitted to the cloud for execution through the selected edge server, so the selected server for the task needs to be adjusted. According to the different cloud-edge collaboration methods required by the task, the specific design is as follows.
[0090] For tasks that require cloud assistance when executed at the edge, when they are assigned to a faulty edge server, first consider whether the task can be transferred to other available edge servers, and complete the acquisition of cloud knowledge and task execution calculations through other edge servers that are working normally on the wired network. If all edge servers in the range are already fully loaded and cannot provide sufficient resources to complete the task, consider changing the execution layer of the task and assigning the task directly to the cloud computing node for execution. The decision-making process is as follows: Figure 2 For tasks that require edge data collection to be performed in the cloud, the tasks are transferred to other available edge servers and forwarded to the cloud.
[0091] (3) Merge Scheduling Algorithm
[0092] In the merge scheduling algorithm, the first step is to initialize the basic configuration of the algorithm, including population size, maximum number of iterations, number of groups, and complete the corresponding initialization requirements of various search operators. Considering that the algorithm is used to deal with large-scale task scheduling problems, the grouping strategy is first used to group tasks, cloud servers, and populations. The dimension reduction is achieved by decomposing the scheduling problem, and the sub-scheduling problems in each group are solved simultaneously by a multi-process approach. Then, five population-based metaheuristic algorithms are improved based on GA[7], PSO[8], and DE[9] to solve the sub-problems of cloud-edge collaborative scheduling, and the metaheuristic algorithm is selected by the random step-size hyper-heuristic method. When each sub-population meets the termination condition, the solutions given by each sub-population are merged through the merging scheme, and then the scheduling scheme is optimized by the fine-tuning method, and finally a scheduling scheme with superior performance in terms of time and energy consumption costs is given.
[0093] Before explaining in detail the grouping, merging and fine-tuning strategies in the merge algorithm, we first introduce the method of evaluating the scheme during the grouping optimization and fine-tuning process.
[0094] (a) Preventive Scheduling Decision Method
[0095] The present invention uses an indirect coding method to encode individuals. Each gene bit in the individual represents the priority of the task in resource allocation. Before performing individual evaluation, the mapping of the scheduling scheme must be completed first. In the process of scheduling scheme mapping, the determination of whether the server can meet the task requirements is a key link. In the study of scheduling problems in traditional cloud computing, edge computing and other scenarios, most of them use the following scheduling determination method: if the current server can provide the resources required for the task, the task can be assigned to the server; otherwise, it cannot. For the model considered in the present invention, this traditional scheduling determination condition can be summarized as the following formula:
[0096] R cpu,i ≤C cpu,j
[0097] R mem,i ≤C mem,j
[0098] Traditional methods of determining whether a server has sufficient resources to provide computing services for a task are limited to the resources that the server can provide in its current state. They lack predictive judgment on the release of resources and are unable to utilize fragmented resources in the server.
[0099] Since the scheduling model mentioned above takes into account the maximum tolerable completion time t of the task for the user max,i , the maximum delay time of a task can be defined as the following formula 1. This feature will be used in the model to conduct preventive analysis and judgment on the resources that the server can provide, so as to better utilize the server resources. When using the preventive judgment method, when task I i When the following equations 2 and 3 are satisfied, the server can be assigned to S edge,j superior:
[0100] time_delay i =t max,i -t i
[0101]
[0102] Based on the above content, a method for merging and scheduling cloud-edge collaborative computing tasks is proposed. The specific steps are as follows:
[0103] (1) External users send industrial computing tasks to the computing platform including the terminal device layer, cloud, and edge, and determine the cloud-edge collaboration mechanism corresponding to the industrial computing tasks;
[0104] Among them, the cloud-edge collaboration mechanism includes cloud-assisted edge computing and edge-assisted cloud computing, specifically:
[0105] When the edge server is executing a task, the cloud sends prior knowledge or historical data to the edge;
[0106] During the task execution process, the cloud server collects and pre-processes the data of each manufacturing terminal at the terminal device layer, and sends it to the cloud;
[0107] (2) Based on the cloud-edge collaboration mechanism determined in step (1), a cloud-edge collaboration task scheduling model is established;
[0108] The cloud-edge collaborative task scheduling model is as follows:
[0109]
[0110] In the formula, the optimization objective of the problem is It is mainly divided into two parts, It represents the time cost required to complete all tasks in a set of industrial computing task requests submitted. It represents the energy cost required to complete all tasks, T m and E m are two constants used to normalize the maximum task execution time and the total energy consumption of executing tasks. λ1,λ2 are objective weights, which are set according to the requirements of task execution time and energy consumption, satisfying λ1+λ2=1 and λ1,λ2∈[0,1]. In general, λ1=λ2=0.5 is set, indicating that the energy consumption target and the task execution time target are equally important.
[0111] It is necessary to evaluate the uncertainty of the cloud-edge collaborative task scheduling model. i Calculation time Data transmission rate from terminal to edge Describe, if the task needs to be retransmitted, the amount of data from device l Upload time It is necessary to recalculate and calculate the evaluation results through the determined number of intervals and cloud-edge collaboration mechanism, specifically:
[0112] When the cloud-edge collaboration mechanism is cloud-assisted edge computing, the method is as follows:
[0113]
[0114] In the formula, Task I i The time it takes to upload data, download data from the cloud, and wait for the server to release available resources. is the task execution time, is the task termination time, Λi For Task I iThe set of predecessor tasks, α1 and α2 are weight coefficients, satisfying the conditions of α1+α2=1 and α1,α2∈[0,1], R cpu,i , R mem,i Task I i CPU and memory requests, For equipment The transmission power C cpu,j , C mem,j Assign edge servers S edge,j CPU and memory capacity, and They are the static power and dynamic power of the server respectively;
[0115] When the cloud-edge collaboration mechanism is edge-assisted cloud computing, the method is as follows:
[0116]
[0117]
[0118] In the formula, β represents task I i Whether preprocessing is required at the edge node. For tasks that do not require data preprocessing at the edge node, β=0; otherwise, β=1. Represents Task I i The time required to upload data to the cloud;
[0119] (3) Run the merged scheduling algorithm, group and generate sub-scheduling plans, evaluate the sub-scheduling plans through the cloud-edge collaborative task scheduling model in step (2), and select the sub-scheduling plan with the best evaluation result for output within the specified number of iterations of the merged scheduling algorithm;
[0120] When the merge scheduling algorithm starts, a grouping strategy is designed based on the cloud-edge collaboration mechanism, specifically:
[0121] (i) Sort the industrial computing tasks, calculate the delay time of each industrial computing task, and obtain the task attribute set C = {R cpu,i ,R mem,i ,time_delay i};
[0122] (ii) Arrange the three attributes in the task attribute set in descending order to obtain the comprehensive ranking result S i , and sort them in ascending order, specifically:
[0123]
[0124] Where N is the number of tasks, P c,i The position of the task in the independent sorting of c attributes;
[0125] (iii) Calculate the resource energy consumption cost of each server in the cloud and sort the resource costs of all servers in the cloud. The resource energy consumption cost calculation formula is as follows:
[0126]
[0127] In the formula, overhead E / C,j For cloud server S cloud,j Resource and energy consumption costs;
[0128] (iv) Grouping the industrial computing tasks, uniformly grouping them in ascending order according to the results of step (ii), and grouping the cloud servers according to the resource cost ranking, with the ratio of the number of cloud servers allocated to the computing tasks in the first half of the process to the number of cloud servers allocated to the computing tasks in the second half of the process being 3:2;
[0129] Specifically, in the merge scheduling algorithm, the server resource availability is determined by the preventive scheduling determination method and the task waiting time t is calculated. wait,i , to obtain the sub-scheduling results, the specific steps of preventive scheduling determination are as follows:
[0130] (a) Calculate the delay time of current industrial computing tasks;
[0131] time_delay i =t max,i -t i
[0132] In the formula, t max,i and t i Mission I i The maximum execution time and expected computation time of
[0133] (b) within the delay time obtained in step (a), judging in the order of the available server list whether the current server meets the computing requirements of the current industrial computing task;
[0134] The specific determination method is as follows:
[0135]
[0136] In the formula, is the set of tasks being executed in the server, t rest,jk For task I in the task set jk The remaining execution time;
[0137] (c) Calculate the task I corresponding to the output of the merge scheduling algorithm i The estimated waiting time is:
[0138]
[0139] (d) According to the calculated task I i The estimated waiting time of industrial computing tasks is calculated based on the estimated waiting time, which is:
[0140]
[0141] (4) Merge and fine-tune the sub-scheduling plans obtained in step (3) to obtain the final scheduling plan.
[0142] Among them, after obtaining the sub-scheduling plan, the sub-scheduling plans are merged according to the corresponding levels of each industrial computing task, the grouping results of the industrial computing tasks, and the grouping results of the cloud servers. After obtaining the overall scheduling plan, the Best Fit strategy is used to fine-tune the maximum task completion time and total energy consumption cost of the overall scheduling plan to obtain the final scheduling plan;
[0143] Among them, the calculation formula for the maximum task completion time and total energy consumption cost of the total scheduling plan is as follows:
[0144]
[0145] In the formula, M is the number of grouping processes, time max,i and E i are the maximum task completion time and energy consumption cost of the sub-scheduling solution obtained by process i, and Z final is the optimization target value of the overall scheduling solution.
[0146] The following is further described in conjunction with specific embodiments:
[0147] In this embodiment, if Figure 4 As shown in the figure, the current status of a server with 16 cores is configured, where the x-axis represents the remaining time to complete the task and the y-axis represents the CPU resource usage of the server. There are four computing tasks in progress on the server. For task I i For example, the server can only provide 8 CPU cores in the current state. If the preventive judgment method in this model is adopted, within the maximum delay time of the task, that is, 0~time_delay i There are three tasks in total during the time period The calculation will be completed and then the computing resources will be released. It can be inferred that the actual number of CPU cores available on the server is 14 cores. The 8 cores marked as S2 are directly available resources, and the 6 cores marked as S1 shadow area are possible resources, which will be released by the other three tasks within the waiting time. Therefore, when task I i Can be assigned to server S k , its expected waiting time twait,i It can be calculated by the following formula:
[0148]
[0149] In modeling, various uncertainties need to be considered, and an evaluation method for uncertainty needs to be designed. In uncertainty analysis, the interval number method is used to describe the uncertainty of wireless network transmission rate and task calculation time. An interval number can be defined as shown in the following formula:
[0150]
[0151] where a - ,a + Represents the number of intervals Left and right limits over the real number field. - =a + = a, then is a real number.
[0152] Interval numbers can also be subjected to binary operations like general real numbers. The following equations 1, 2, and 3 give the extended interval number binary operations:
[0153]
[0154] The present invention uses the interval sequence method to realize the comparison of interval numbers. The midpoint of the interval and the half-width of the interval are defined as follows:
[0155]
[0156] Use the interval order relation ≤ mw To define the quality of the interval. That is, for the interval number and If and only if and When, the number of intervals Better than
[0157] At the same time, the above calculation operations and pros and cons comparison methods are combined with the two cloud-edge collaboration modes, according to the following formula:
[0158]
[0159] Calculation can obtain the objective function value of the scheduling solution in the form of interval numbers And use the above interval number comparison method in the population updating or selection process;
[0160] When Task I i Assigned to the edge computing node S edge,j , execution time and end time As shown in the above formulas 1 and 2, S edge,j The energy consumption required to complete the task is shown in the above formula 3, and the upload time of each task is shown in the above formula 4;
[0161] When Task I i Assigned to the cloud computing node S cloud,j The time when the task data is uploaded to the cloud data center and the time when the task is in S cloud,j The execution time and task termination time can be calculated by the above formulas 5, 6, and 7. cloud,j The energy consumption required to complete the task is shown in formula 8 above.
[0162] In summary, combined with the demand for improving the performance of the Industrial Internet in actual production, the present invention describes the goal of the cloud-edge collaborative task scheduling problem as follows:
[0163]
[0164] Among them, λ1,λ2 are objective weights, which are set according to the requirements of task execution time and energy consumption, satisfying λ1,λ2∈[0,1] and λ1+λ2=1. In general, λ1=λ2=0.5 is set, indicating that the energy consumption target and the task execution time target are equally important. This scheme uses the minimization formula as the optimization goal of the cloud-edge collaborative industrial computing task scheduling problem. The above-mentioned evaluation method for uncertainty will be used in the process of group optimization and fine-tuning of the combined scheduling algorithm, so that the algorithm can cope with various uncertainties and generate a scheduling plan.
[0165] In order to speed up the speed of solving large-scale task scheduling problems by the merged scheduling algorithm, this solution designs a grouping strategy to divide tasks, resources and populations, and performs multiple groups of parallel solutions, etc., and designs grouping strategies for the cloud-edge collaborative mode. Combining the two cloud-edge collaborative modes, based on the characteristics of fast response speed of the edge and powerful cloud storage and computing capabilities, the present invention aims to shorten the task response time and reduce energy consumption. The following two points are used as the design ideas of task sorting and grouping strategies: Tasks with larger resource (CPU, memory) requests and lower delay sensitivity tend to be assigned to cloud servers, and vice versa, they tend to be assigned to edge servers; tasks tend to be assigned to servers with lower energy consumption costs for execution.
[0166] Therefore, we first need to sort the tasks by multiple attributes (CPU request, memory request, delay sensitivity), and record this attribute set as C = {R cpu,i ,R mem,i ,time_delay i}. The present invention uses the formula:
[0167]
[0168] The maximum delay time calculated in represents the delay sensitivity of the task. After the three attributes are arranged in descending order independently, the task multi-attribute comprehensive ranking index is defined as the formula. Where N is the number of tasks, P is c,i The position of the task in the independent sorting of c attributes;
[0169] At the same time, the servers need to be sorted by resource cost. Since the edge has a strong regionality, the present invention only groups cloud servers. Cloud edge computing resources are abundant, and computing nodes have a certain degree of heterogeneity. It is possible to choose to schedule tasks to servers with lower resource energy consumption costs. On servers with lower resource energy consumption costs, tasks can occupy the same resources for calculation with lower energy consumption. The resource energy consumption cost is defined in the present invention as follows:
[0170]
[0171] The grouping strategy based on resource energy consumption cost firstly calculates the comprehensive ranking index S i The tasks are arranged in ascending order and grouped evenly in sequence. At this time, the tasks ranked higher have a larger delay time and more resource requests. In order to meet the requirements that such tasks are more inclined to be assigned to cloud servers and have lower energy consumption costs, the cloud servers are arranged in ascending order according to the resource energy consumption cost of the server, and the groups are not evenly grouped. The higher the grouping, the more cloud servers are required. Specifically, there are two types of cloud servers allocated to each group, and the ratio of the number is 3:2. The first half of the processes are allocated to the majority of servers, and the second half of the processes are allocated to the minority of servers. In this way, the problem is decomposed into multiple sub-scheduling problems. After the tasks, available cloud servers, and populations are grouped by the above grouping strategy, each process does not affect each other, and there is no additional communication overhead.
[0172] After solving the sub-scheduling problem, the scheduling solutions of each sub-problem need to be merged, so a scheduling solution merging mechanism is designed in the algorithm. First, the sub-scheduling solutions are mapped in each process, the level and specific resources assigned to each task are given, and the sequence numbers of the tasks and cloud nodes are adjusted according to the number of groups to obtain the overall scheduling solution. Then, according to the following formula:
[0173]
[0174] Dividing the scheduling problem into multiple sub-problems for solution reduces the dimension of the problem, but it will reduce the solution space, so some better solutions may be ignored. From another perspective, a task assigned to the cloud may not be grouped with the cloud server that is most economical in terms of time and energy consumption. Therefore, the algorithm needs to make further optimization adjustments on the merged scheduling scheme in order to obtain a smaller optimization target value, mainly optimizing energy consumption, and only adjusting the tasks assigned to the cloud, without changing the scheduling scheme for tasks assigned to the edge. This scheme uses Best Fit as the optimization strategy, traverses all servers in the cloud server list, records the server with the lowest cost of executing the task, and assigns the task to this server.
[0175] In combination with the above content, in this embodiment, if Figure 3 As shown in the figure, the scheduling method based on cloud-edge collaborative computing tasks is as follows:
[0176] (1) External users send industrial computing tasks to the computing platform including the terminal device layer, cloud, and edge, and determine the cloud-edge collaboration mechanism corresponding to the industrial computing tasks;
[0177] (2) Based on the cloud-edge collaboration mechanism determined in step (1), a cloud-edge collaboration task scheduling model is established;
[0178] (3) Run the merged scheduling algorithm, group and generate sub-scheduling plans, evaluate the sub-scheduling plans through the cloud-edge collaborative task scheduling model in step (2), and select the sub-scheduling plan with the best evaluation result for output within the specified number of iterations of the merged scheduling algorithm;
[0179] (4) Merge and fine-tune the sub-scheduling plans obtained in step (3) to obtain the final scheduling plan.
[0180] Among them, the cloud-edge collaboration mechanism includes cloud-assisted edge computing and edge-assisted cloud computing, specifically:
[0181] When the edge server is executing a task, the cloud sends prior knowledge or historical data to the edge;
[0182] During the task execution process, the cloud server collects and pre-processes the data of each manufacturing terminal at the terminal device layer, and sends it to the cloud;
[0183] In step (2), the cloud-edge collaborative task scheduling model is specifically as follows:
[0184]
[0185] In the formula, the optimization objective of the problem is It is mainly divided into two parts, It represents the time cost required to complete all tasks in a set of industrial computing task requests submitted. It represents the energy cost required to complete all tasks, T m and E m are two constants used to normalize the maximum task execution time and the total energy consumption of executing tasks. λ1,λ2 are objective weights, which are set according to the requirements of task execution time and energy consumption, satisfying λ1+λ2=1 and λ1,λ2∈[0,1]. In general, λ1=λ2=0.5 is set, indicating that the energy consumption target and the task execution time target are equally important.
[0186] In step (2), the uncertainty of the cloud-edge collaborative task scheduling model needs to be evaluated. The interval number method is used to evaluate the uncertainty of task I. i Calculation time Data transmission rate from terminal to edge Describe, if the task needs to be retransmitted, the amount of data from device l Upload time It is necessary to recalculate and calculate the evaluation results through the determined number of intervals and cloud-edge collaboration mechanism, specifically:
[0187] When the cloud-edge collaboration mechanism is cloud-assisted edge computing, the method is as follows:
[0188]
[0189] In the formula, Task I i The time it takes to upload data, download data from the cloud, and wait for the server to release available resources. is the task execution time, is the task termination time, Λi For Task I i The set of predecessor tasks, α1 and α2 are weight coefficients, satisfying the conditions of α1+α2=1 and α1,α2∈[0,1], R cpu,i , R mem,i Task I i CPU and memory requests, For equipment The transmission power C cpu,j , C mem,j Assign edge servers S edge,j CPU and memory capacity, and They are the static power and dynamic power of the server respectively;
[0190] When the cloud-edge collaboration mechanism is edge-assisted cloud computing, the method is as follows:
[0191]
[0192]
[0193] In the formula, β represents task I i Whether preprocessing is required at the edge node. For tasks that do not require data preprocessing at the edge node, β=0; otherwise, β=1. Represents Task I i The time required to upload data to the cloud.
[0194] In step (3), when the merge scheduling algorithm starts, the grouping strategy is designed according to the cloud-edge collaboration mechanism, specifically:
[0195] (i) Sort the industrial computing tasks, calculate the delay time of each industrial computing task, and obtain the task attribute set C = {R cpu,i ,R mem,i ,time_delay i};
[0196] (ii) Arrange the three attributes in the task attribute set in descending order to obtain the comprehensive ranking result S i , and sort them in ascending order, specifically:
[0197]
[0198] Where N is the number of tasks, P c,i The position of the task in the independent sorting of c attributes;
[0199] (iii) Calculate the resource energy consumption cost of each server in the cloud and sort the resource costs of all servers in the cloud. The resource energy consumption cost calculation formula is as follows:
[0200]
[0201] In the formula, overhead E / C,j For cloud server S cloud,j Resource and energy consumption costs;
[0202] (iv) Group the industrial computing tasks evenly according to the ascending order results obtained in step (ii), and group the cloud servers according to the resource cost ranking. The ratio of the number of cloud servers allocated to the computing tasks in the first half of the process to the number of cloud servers allocated to the computing tasks in the second half of the process is 3:2.
[0203] In step (3), in the combined scheduling algorithm, the server resource availability is determined by the preventive scheduling determination method and the task waiting time t is calculated. wait,i,The specific steps of preventive scheduling determination are as follows:
[0204] (a) Calculate the delay time of current industrial computing tasks;
[0205] time_delay i =t max,i -t i
[0206] Where, t max,i and t i Mission I i The maximum execution time and expected computation time of
[0207] (b) within the delay time obtained in step (a), judging in the order of the available server list whether the current server meets the computing requirements of the current industrial computing task;
[0208] The specific determination method is as follows:
[0209]
[0210] In the formula, is the set of tasks being executed in the server, t rest,jk For task I in the task set jk The remaining execution time;
[0211] (c) Calculate the task I corresponding to the output of the merge scheduling algorithm i The estimated waiting time is:
[0212]
[0213] (d) According to the calculated task I i The estimated waiting time of industrial computing tasks is calculated based on the estimated waiting time, which is:
[0214]
[0215] In step (4), after obtaining the sub-scheduling schemes, the sub-scheduling schemes are merged according to the corresponding levels of each industrial computing task, the industrial computing task grouping results, and the cloud server grouping results. After obtaining the overall scheduling scheme, the Best Fit strategy is used to fine-tune the overall scheduling scheme using the maximum task completion time and total energy consumption cost to obtain the final scheduling scheme.
[0216] Among them, the calculation formula for the maximum task completion time and total energy consumption cost of the total scheduling plan is as follows:
[0217]
[0218] In the formula, M is the number of grouping processes, timemax,i and E i are the maximum task completion time and energy consumption cost of the sub-scheduling solution obtained by process i, and Z final is the optimization target value of the overall scheduling solution.
[0219] Although the present invention has been disclosed as above in the form of a preferred embodiment, it is not intended to limit the present invention. Any person skilled in the art may make possible changes and modifications to the technical solution of the present invention by using the methods and technical contents disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope of the technical solution of the present invention.
[0220] The contents not described in detail in the specification of the present invention belong to the common knowledge of those skilled in the art.
Claims
1. A method for merging and scheduling cloud-edge collaborative computing tasks, characterized in that Here are the steps: (1) External users send industrial computing tasks to the computing platform including the terminal device layer, cloud, and edge, and determine two cloud-edge collaboration mechanisms; (2) Based on the cloud-edge collaboration mechanism determined in step (1), a cloud-edge collaboration task scheduling model is established; (3) Run the merged scheduling algorithm, group and generate sub-scheduling plans, evaluate the sub-scheduling plans through the cloud-edge collaborative task scheduling model in step (2), and select the sub-scheduling plan with the best evaluation result for output within the specified number of iterations of the merged scheduling algorithm; (4) merging and fine-tuning the sub-scheduling plans obtained in step (3) to obtain a final scheduling plan; In step (2), the cloud-edge collaborative task scheduling model is specifically: In the formula, the problem target It is mainly divided into two parts, It represents the time cost required to complete all tasks in a set of industrial computing task requests submitted. It represents the energy cost required to complete all tasks, T m and E m are two constants used to normalize the maximum task execution time and the total energy consumption of executing tasks; λ1,λ2 are objective weights, which are set according to the requirements of task execution time and energy consumption, satisfying λ1+λ2=1 and λ1,λ2∈[0,1]; In step (2), the uncertainty of the cloud-edge collaborative task scheduling model needs to be evaluated, and the interval number method is used to evaluate the uncertainty of task I. i Calculation time Data transmission rate from terminal to edge Describe, if the task needs to be retransmitted, the amount of data from device l Upload time It is necessary to recalculate and calculate the evaluation results through the determined number of intervals and cloud-edge collaboration mechanism, specifically: When the cloud-edge collaboration mechanism is cloud-assisted edge computing, the method is as follows: In the formula, t wait,i Task I i The time it takes to upload data, download data from the cloud, and wait for the server to release available resources. is the task execution time, is the task termination time, Λi For Task I i The set of predecessor tasks, α1 and α2 are weight coefficients, satisfying the conditions of α1+α2=1 and α1,α2∈[0,1], R cpu,i , R mem,i Task I i CPU and memory requests, For equipment The transmission power C cpu,j , C mem,j Assign edge servers S edge,j CPU and memory capacity, and They are the static power and dynamic power of the server respectively; When the cloud-edge collaboration mechanism is edge-assisted cloud computing, the method is as follows: In the formula, β represents task I i Whether preprocessing is required at the edge node. For tasks that do not require data preprocessing at the edge node, β = 0; On the contrary, β = 1; Represents Task I i The time required to upload data to the cloud; In step (3), when the merge scheduling algorithm starts, a grouping strategy is designed according to the cloud-edge collaboration mechanism, specifically: (i) Sort the industrial computing tasks, calculate the delay time of each industrial computing task, and obtain the task attribute set C = {R cpu,i ,R mem,i ,time_delay i }; (ii) Arrange the three attributes in the task attribute set in descending order to obtain the comprehensive ranking result S i , and sort them in ascending order, specifically: Where N is the number of tasks, P c,i The position of the task in the independent sorting of c attributes; (iii) Calculate the resource energy consumption cost of each server in the cloud and sort the resource costs of all servers in the cloud. The resource energy consumption cost calculation formula is as follows: In the formula, overhead E / C,j For cloud server S cloud,j Resource and energy consumption costs; (iv) Group the industrial computing tasks evenly according to the ascending order results obtained in step (ii), and group the cloud servers according to the resource cost ranking. The ratio of the number of cloud servers allocated to the computing tasks in the first half of the process to the number of cloud servers allocated to the computing tasks in the second half of the process is 3:
2.
2. According to claim 1, a method for merging and scheduling cloud-edge collaborative computing tasks, characterized in that: In step (1), the cloud-edge collaboration mechanism includes cloud-assisted edge computing and edge-assisted cloud computing, specifically: When the edge server is executing a task, the cloud sends prior knowledge or historical data to the edge; During the task execution process of the cloud server, the edge end collects and pre-processes the data of each manufacturing terminal at the terminal device layer and sends it to the cloud.
3. According to claim 2, a method for merging and scheduling cloud-edge collaborative computing tasks, characterized in that: In step (3), in the combined scheduling algorithm, the server availability is determined by the preventive scheduling determination method and the task waiting time t is calculated. wait,i , to obtain the sub-scheduling results, the specific steps of the preventive scheduling determination method are as follows: (a) Calculate the delay time of current industrial computing tasks; time_delay i =t max,i -t i Where, t max,i and t i Mission I i The maximum execution time and expected computation time; (b) within the delay time obtained in step (a), judging in the order of the available server list whether the current server meets the computing requirements of the current industrial computing task; The specific determination method is as follows: In the formula, is the set of tasks being executed in the server, t rest,jk For task I in the task set jk The remaining execution time; (c) Calculate the task I corresponding to the output of the merge scheduling algorithm i The estimated waiting time is: (d) According to the calculated task I i The estimated waiting time of industrial computing tasks is calculated based on the estimated waiting time, which is:
4. According to claim 1, a method for merging and scheduling cloud-edge collaborative computing tasks, characterized in that: In the step (4), after obtaining the sub-scheduling schemes, the sub-scheduling schemes are merged according to the corresponding levels of each industrial computing task, the industrial computing task grouping results, and the cloud server grouping results. After obtaining the overall scheduling scheme, the Best Fit strategy is used to fine-tune the overall scheduling scheme using the maximum task completion time and total energy consumption cost to obtain the final scheduling scheme. Among them, the calculation formula for the maximum task completion time and total energy consumption cost of the total scheduling plan is as follows: In the formula, M is the number of grouping processes, time max,i and E i are the maximum task completion time and energy consumption cost of the sub-scheduling solution obtained by process i, and Z final is the optimization target value of the overall scheduling solution.
