Multi-task joint scheduling method for edge-cloud system

Abstract

The present invention discloses a multi-task joint scheduling method for an end-edge cloud system. First, on the basis of a unified description of the nodes and tasks of the end-edge cloud system, a multi-task joint scheduling model for the end-edge cloud system is realized; secondly, maximizing the expected benefits of multi-task joint scheduling is used as the optimization goal, and optimization problem solving constraints are established to realize the construction of the optimization problem of multi-task joint scheduling of the end-edge cloud system; finally, a heuristic strategy is combined with a nonlinear optimization solution to realize step-by-step decision-making on task execution nodes and task allocation resource amounts, and complete the final generation of the multi-task joint scheduling scheme of the end-edge cloud system. The present invention can realize concurrent computing-intensive task joint scheduling that is highly adaptable to the end-edge cloud system from three aspects: multi-task joint scheduling model construction, multi-task joint scheduling optimization problem establishment, and multi-task joint scheduling scheme generation.

Description

Technical Field

[0001] The present invention belongs to the field of cloud computing and edge computing, and specifically is a multi-task joint scheduling method for an edge-cloud system. Background Art

[0002] With the continuous development of the Internet of Things, edge computing and cloud computing technologies, the advantages of the complementary integration of the three are increasingly evident. Widely deployed IoT devices will generate a large amount of perception data and computing tasks; edge computing reduces the delay and pressure of centralized transmission of data and tasks to the cloud by performing data analysis and task processing nearby; cloud computing provides a powerful data pool and resource pool, which can provide strong guarantees for the acceptance and processing of massive data and tasks. The end-edge-cloud integrated system architecture can combine the advantages of the Internet of Things, edge computing and cloud computing, which not only optimizes the overall resource utilization efficiency of the system, but also better responds to the dynamic changes in the network environment and task requirements, so that tasks can be scheduled to the appropriate location and allocated to the appropriate resources for processing, improving the real-time, flexibility and scalability of the system.

[0003] At the same time, compared with the traditional single or cluster architecture, the distributed architecture of the end-edge cloud system needs to face more adjustments in task scheduling and resource allocation. The first is the uneven distribution of resources: the resources of the three types of computing nodes, end, edge, and cloud, are usually heterogeneous and decentralized. Considering the instability of the network between nodes, there is a certain fluctuation in the available resources at the remote end; the second is the dynamic and uncertainty of tasks: the tasks in the end-edge cloud system are usually dynamic. On the one hand, there is a great deal of uncertainty in the arrival time, arrival quantity, and execution time of the tasks. On the other hand, different tasks may have very different QoS requirements for computing resources, completion time, etc.; in addition, the concurrency of tasks needs to be considered: in the burst load scenario, the system needs to complete the scheduling and calculation of a large number of tasks in a short time, which requires improving the utilization of system resources while ensuring the task completion rate to avoid system overload.

[0004] Traditional task scheduling methods can be divided into two categories: static scheduling algorithms and dynamic scheduling algorithms. Static scheduling algorithms usually make scheduling decisions before the arrival of tasks, and the execution order and resource allocation of tasks are fixed; they are mainly applicable to scenarios where the arrival time and demand of tasks are known and unchanged. In edge-cloud systems, tasks are usually highly dynamic, and static scheduling methods are difficult to meet the real-time changing task requirements. Dynamic scheduling algorithms make scheduling decisions based on real-time task information when tasks arrive, and have flexible flexibility and priority; common methods include priority-based scheduling, load balancing scheduling, resource scheduling, etc. These methods can usually make reasonable scheduling decisions based on information such as task priority, resource requirements, and node load. However, existing dynamic scheduling algorithms often fail to fully consider the collaborative relationship between tasks and the expected benefits in multi-task scheduling. In addition, current multi-task scheduling methods also face the problem of computational complexity, that is, as the number of tasks increases and the scale of the system expands, scheduling algorithms often find it difficult to generate scheduling plans within a reasonable time. Therefore, how to improve the efficiency and accuracy of scheduling through effective optimization methods has become the focus of research on multi-task scheduling in edge-cloud systems. Summary of the invention

[0005] The purpose of the present invention is to provide a multi-task joint scheduling method for an end-edge-cloud system, which realizes concurrent computing-intensive task joint scheduling that is highly adaptable to the end-edge-cloud system from three aspects: building a multi-task joint scheduling model, establishing a multi-task joint scheduling optimization problem, and generating a multi-task joint scheduling scheme.

[0006] The technical solution to achieve the purpose of the present invention is: a multi-task joint scheduling method for an edge-cloud system, which specifically includes the following steps:

[0007] Step S1: construct a multi-task joint scheduling model for the end-edge-cloud system. The multi-task joint scheduling model for the end-edge-cloud system includes:

[0008] The multi-task joint scheduling matrix model is used to define the task-node scheduling matrix and the task-resource allocation matrix based on the unified description of the nodes and tasks of the end-edge-cloud system, and to formalize the expression of the multi-task scheduling strategy;

[0009] The expected task completion time model is used to determine the expected task completion time based on the scheduling, transmission, and execution time in the task life cycle;

[0010] The expected benefit model of the task is used to evaluate the benefits of task scheduling and execution from two aspects: expected reward and expected cost;

[0011] Step S2, based on the task expected benefit model, for the multi-task-multi-node scheduling scenario, construct a multi-task scheduling evaluation function; establish a multi-task scheduling optimization goal, and complete the construction of the multi-task scheduling optimization problem from the expected completion time, node resource constraints, the relationship between the task-node scheduling matrix and the task-resource allocation matrix;

[0012] Step S3, adopting a heuristic strategy, calculates the critical resource demand and pre-occupied resource amount of each task to be scheduled, and completes the execution node allocation and resource pre-occupation of each task in the order of the expected completion time of the task based on the calculation results; uses a nonlinear optimization method to optimize and solve the multi-task scheduling optimization problem to obtain the recommended resource allocation for each task; finally, based on resource constraints and associated constraints, verifies and corrects the task-node scheduling matrix and the task-resource allocation matrix to complete the final generation of the multi-task joint scheduling solution.

[0013] Compared with the prior art, the present invention has the following significant advantages:

[0014] (1) The multi-task joint scheduling framework of the end-edge-cloud system designed in the present invention is highly adaptable to multi-end concurrent request scenarios and end-edge-cloud heterogeneous computing systems. It breaks through the limitations of traditional single-task or single-system scheduling methods in ubiquitous interconnection and cloud-edge collaborative task scheduling, and has a high degree of originality.

[0015] (2) The multi-task joint scheduling and optimization model constructed by the present invention can accurately describe and formalize the multi-task joint scheduling problem from the aspects of system composition, task characteristics, expected benefits, scheduling objectives, constraints, etc., and provide model support for the generation and optimization of scheduling schemes, which is highly innovative.

[0016] (3) The two-step solution method combining heuristic strategy and nonlinear optimization adopted in the present invention can reduce the solution space and computational complexity of the optimization problem while realizing step-by-step decision-making of task execution nodes and task allocation resource amounts, significantly improving the speed of generating scheduling solutions, and has high practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 This is a flow chart of the multi-task joint scheduling method of the present invention. DETAILED DESCRIPTION

[0018] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0019] like Figure 1As shown, a multi-task joint scheduling method for an end-edge cloud system first realizes the formal expression of the task-node scheduling matrix, the task-resource allocation matrix, the expected completion time model, and the task expected benefit model on the basis of the unified description of the nodes and tasks of the end-edge cloud system, thereby constructing a multi-task joint scheduling model for the end-edge cloud system; secondly, taking the maximization of the expected benefit of the multi-task joint scheduling as the optimization goal, and based on the expected completion time, node resource constraints, the relationship between the task-node scheduling matrix and the task-resource allocation matrix, the optimization problem solving constraints are established to realize the construction of the optimization problem of the multi-task joint scheduling of the end-edge cloud system, and provide guidance for the subsequent scheduling scheme generation; finally, a heuristic strategy is combined with a nonlinear optimization solution to realize the step-by-step decision of the task execution node and the task allocation resource amount, and complete the final generation of the multi-task joint scheduling scheme of the end-edge cloud system. The present invention can realize the concurrent computing-intensive task joint scheduling that is highly adapted to the end-edge cloud system from three aspects: the construction of the multi-task joint scheduling model, the establishment of the multi-task joint scheduling optimization problem, and the generation of the multi-task joint scheduling scheme.

[0020] The present invention specifically comprises the following steps:

[0021] Step S1: Building a multi-task joint scheduling model for the edge-cloud system:

[0022] In this step, we first build a multi-task joint scheduling matrix model. On the basis of a unified description of the nodes and tasks of the end-edge-cloud system, we define the task-node scheduling matrix and the task-resource allocation matrix to formalize the multi-task scheduling strategy. Secondly, we build a task expected completion time model that comprehensively considers the scheduling, transmission, and execution time in the task life cycle. Finally, we build a task expected benefit model to evaluate the benefits of task scheduling and execution from the two aspects of expected reward and expected cost, providing a basis for the construction of subsequent scheduling optimization problems.

[0023] Step S101, constructing a multi-task joint scheduling matrix model:

[0024] The end-edge cloud system involved in the present invention includes three types of nodes, namely cloud center nodes, edge server nodes and object-end device nodes. The number and functions of the three types of nodes in the system are as follows: the object-end device node (referred to as the end node) is the task initiation node, The edge server nodes (referred to as edge nodes) and one cloud center node (referred to as cloud node) are task computing nodes. Since computing-intensive tasks require a long period of exclusive use of computing resources and have a high tolerance for task completion time, in order to avoid resource waste caused by frequent scheduling, this method will wait for the number of tasks submitted by the end nodes to accumulate to a threshold. Then schedule multiple tasks. Set by the system user. During each scheduling process, a group of The task is recorded as ,in , according to the expected completion time of the task Incremental sorting, that is for The expected completion time and ; Available for dispatch The computing nodes are denoted as ,in for The edge node number, is the cloud node number. Then the task-node scheduling strategy is available matrix Indicates that the task-resource allocation strategy is available Matrix Indicates that

[0025] ,

[0026] in , Indicates the task Assigned to the node Execute, then there is Representation Node For the task The amount of computing resources allocated is ; Indicates the task Not assigned to a node Execute, then there is Since each task will be assigned to at most one node for computation, the matrix and Each row of has at most one non-zero element; if the matrix and No. If all rows are 0, it means the task Discarded and not calculated.

[0027] Step S102: constructing a model of the expected completion time of the task:

[0028] Computational tasks Scheduling to compute nodes Expected completion time of the execution Including: task scheduling and resource application time (approximately constant), transmission time (including network latency, calculation data upload, result download time), execution time ,therefore .

[0029] , where , They are Initiating Node and Computing Node The network latency and network bandwidth are determined by the selection of the node initiating node and the computing node; for The amount of data uploaded, for Download data volume, by It is determined by itself as a known quantity.

[0030] ,in for The computational complexity of is assigned by the system user based on historical experience, Self-determination is a known quantity; is the computing capacity-computing resource mapping function, which is obtained by fitting the system user based on historical experience and is a known function. The value is determined by the amount of computing node resources allocated.

[0031] therefore, .

[0032] Step S103: constructing the expected revenue model of the task:

[0033] Computational tasks Scheduling to compute nodes The expected reward for execution is , the expected cost is recorded as , the expected return is recorded as , the specific expression is as follows:

[0034] , Indicates when exist When executed, ,on the contrary ; Weight coefficient Set by the system user; constant Set by the system user for tasks The value of For the task The expected completion time indicates that when the expected completion time The reward for the task is positive when the task is completed earlier, and the reward is higher when the expected completion time is The reward value of the task is negative.

[0035] , Indicates when exist When executed, ,on the contrary ; Weight coefficient Set by the system user; For Node For the task The amount of computing resources allocated; for exist The expected completion time.

[0036] , indicating that when exist When executed, ;on the contrary .

[0037] Step S2: construct the multi-task joint scheduling optimization problem of the end-edge-cloud system:

[0038] In this step, firstly, based on the task expected benefit model in step S203, a multi-task scheduling evaluation function is constructed for the multi-task-multi-node scheduling scenario; then, a multi-task scheduling optimization goal is established, and the multi-task scheduling optimization problem is constructed from aspects such as expected completion time, node resource constraints, and the relationship between the task-node scheduling matrix and the task-resource allocation matrix, so as to provide guidance for the subsequent scheduling plan generation.

[0039] Step S201, constructing a multi-task scheduling evaluation function:

[0040] From S103, we can know that the calculation task Scheduling to compute nodes The expected return is recorded as , by step S101 task group The scheduling scheme is composed of the matrix , Therefore, the scheduling matrix is , In the case of Expected Return Then we can get the task group Expected returns:

[0041]

[0042] From step S102, it can be seen that ,remove All other constants are known, so we can write In addition, from step S101, hour , hour , so Therefore, the task force Expected Return

[0043] Step S202, multi-task scheduling optimization problem construction:

[0044] Note Node The upper limit of the remaining available resources that can be allocated is , and combined with the derivation process of steps S101 to S201, a group Mission Record and available for scheduling The computing nodes are denoted as The scheduling optimization problem between the two is constructed as follows:

[0045]

[0046] In the above optimization objectives and constraints, and is an unknown quantity, and the rest are known quantities. Among them, the optimization goal is to find the scheduling plan matrix , The best values ​​of all elements make the task group constructed in step S201 Expected Return Get the maximum value (i.e. The first constraint condition is derived in step S201. about The second constraint condition represents the node The task force The total amount of resources allocated shall not exceed The upper limit of remaining available resources ; The third constraint is the constraint in step S101 Definition of element; the fourth constraint indicates and Both are 0 or not 0 at the same time; the fifth constraint indicates that each task will be assigned to at most one node for calculation.

[0047] Step S3: Generate a multi-task joint scheduling solution for the end-edge-cloud system:

[0048] In this step, a heuristic strategy is first used to calculate the critical resource requirements and pre-occupied resource quantities of each task to be scheduled, and based on the calculation results, the execution node allocation and resource pre-occupancy of each task are completed in the order of the expected completion time of the tasks; then, a nonlinear optimization method is used to optimize the node task resource allocation scheme to obtain the recommended resource allocation for each task; finally, based on resource constraints and associated constraints, the task-node scheduling matrix and the task-resource allocation matrix are verified and corrected to complete the final generation of the multi-task joint scheduling scheme.

[0049] Step S301, perform node allocation and resource pre-occupancy:

[0050] For each task ,make , solve the task Critical resource requirements Considering the need to make the task The expected return of task The amount of pre-occupied resources should be greater than Therefore, the task in this step is The amount of pre-occupied resources set is ,in is a positive coefficient greater than 1, which is set by the system user. As can be seen from step S101, Sort by the expected completion time increment of the tasks. Since the expected completion time of the tasks in front is closer, this step is based on For each Select the execution node. The specific operations are as follows:

[0051] ①Get edge nodes The remaining allocatable resources , if exists , so that ,and ; then the task Assign to Node Execute and set the node The remaining allocatable resources are adjusted to ; This means that Assign to edge nodes Execute and set the pre-occupied resources to , so let

[0052] .

[0053] ②If no matching conditions are found in ① ,if , then the node The remaining allocatable resources are adjusted to ; This means that Distribute to cloud nodes Execute and set the pre-occupied resources to ; therefore,

[0054] .

[0055] if , then abandon the task , at this time .

[0056] ③ Repeat steps ① and ② until the task group is completed Execution node allocation and resource pre-occupancy.

[0057] After the above steps ①②③, the matrix can be determined for the first time All elements in The value of , , is the task-node scheduling strategy matrix The current recommended value of The final value of is optimized and determined in subsequent steps.

[0058] Step S302: Optimizing the node task resource allocation scheme:

[0059]

[0060] Among them, the optimization objective is only is an unknown quantity. The first and second constraints are the same as step S202. The third constraint is expressed as for The actual amount of resources allocated does not exceed the amount pre-occupied in step S301 , the fourth constraint indicates is the matrix obtained in step S301 No. Line Column elements.

[0061] In order to solve the above constrained nonlinear optimization problem, we first transform it into an unconstrained optimization problem. The specific operations are as follows:

[0062] ①By bringing the first, second, and fourth constraints in the above optimization problem into the optimization objective, we can obtain:

[0063]

[0064] ② By constructing a penalty function (interior point method), the constrained problem is approximately transformed into an unconstrained problem, and the optimization iteration process is restricted to the feasible domain:

[0065]

[0066] in is the penalty item, is a small positive number set by the system user (for example ), the purpose is to Close to the feasible region boundary 0 and When , thus avoiding the search area from crossing the feasible domain; on the contrary, when In the feasible domain Internally, due to With the existence of , the penalty function is approximately equal to 0, which is approximately equal to the original optimization objective.

[0067] After completing the construction of the above unconstrained optimization problem, the nonlinear conjugate gradient method can be used to treat the optimization variables Perform iterative solution. During round iteration, The value of , optimize the objective function The gradient vector at this value point is ; then the iteration stop condition is the two norm of the gradient vector ,or .in The cut-off accuracy set by the system user, The number of rounds set by the system user. The number of rounds is recorded as If the iteration stop condition is met, then The solution result of the node task resource allocation optimization problem.

[0068] Step S303, multi-task node and resource scheduling scheme verification:

[0069] Note that the solution to the optimization problem in step S302 , at this time, we must The rationality of the value of is further verified. The specific operation is as follows: :

[0070] ①For or , then let , , .

[0071] ②For ,verify Is it greater than 0: If , then let , ;like , then let , , .

[0072] After steps ① and ②, the matrix can be finally determined and The values ​​of all elements in the task group To compute nodes The final scheduling solution is composed of the scheduling matrix and Absolutely. Specifically, for the task , if exists , so that , then Dispatching to Node Execute and assign The amount of resources; if for any , both have , then the task give up.

[0073] This architecture combines the powerful processing power of cloud computing with the low latency and high bandwidth advantages of edge computing. It can process and analyze data at the edge of the network, while using the computing resources in the cloud for large-scale data storage and in-depth analysis. The edge cloud system processes computing tasks at close range between the data source and the terminal device, improving the system's response speed and real-time performance. Through the collaborative work between the edge nodes and the cloud, the edge cloud architecture can provide more flexible, intelligent and efficient computing services, especially for applications such as the Internet of Things, smart manufacturing, and autonomous driving that require rapid processing of large amounts of data. This architecture not only optimizes resource utilization, but also better responds to dynamic changes in network environments and task requirements, and has huge development potential and application prospects.

[0074] The present invention proposes a multi-task joint scheduling method for the end-edge cloud system. By unified modeling of system nodes and tasks, a new systematic method is provided for the multi-task scheduling problem of the end-edge cloud system. The model can not only consider the resource requirements of each task, but also perform comprehensive scheduling according to the completion time and benefit of the task. By increasing the expected benefit of multi-task joint scheduling as the optimization target, the present invention fully considers the relationship collaboration and resource constraints between tasks, and provides clear guidance for the generation of scheduling schemes. By combining the adoption of strategy inspiration with the indirect optimization method, the present invention can gradually optimize the scheduling scheme of tasks under the premise of ensuring computing efficiency, and achieve better resource utilization and task scheduling efficiency. In summary, the present invention solves the problem of multi-task joint scheduling in the end-edge cloud system by establishing a multi-joint scheduling model, scheduling objectives, and combining heuristic strategies and nonlinear optimization methods. This method can greatly improve the resource utilization of the system, reduce the task execution delay, and is particularly suitable for the joint scheduling of computing-intensive tasks, with broad application prospects and practical value.

Claims

1. A multi-task joint scheduling method for an edge-cloud system, characterized in that: The specific steps include: Step S1: construct a multi-task joint scheduling model for the end-edge-cloud system. The multi-task joint scheduling model for the end-edge-cloud system includes: The multi-task joint scheduling matrix model is used to define the task-node scheduling matrix and the task-resource allocation matrix based on the unified description of the nodes and tasks of the end-edge-cloud system, and to formalize the expression of the multi-task scheduling strategy; The expected task completion time model is used to determine the expected task completion time based on the scheduling, transmission, and execution time in the task life cycle; The task expected benefit model is used to evaluate the benefits of task scheduling and execution from two aspects: expected reward and expected cost. Specifically: ， , and when the task On the compute node When executed, ,otherwise ; For the task Scheduling to compute nodes Expected Rewards for Execution , specifically: ， In the formula, is the first weight coefficient; A constant value set by the user; For the task Expected completion time; For expected cost , specifically: , is the second weight coefficient; For Node For the task The amount of computing resources allocated; for exist Expected completion time on Step S2, based on the task expected benefit model, for the multi-task-multi-node scheduling scenario, construct a multi-task scheduling evaluation function; establish a multi-task scheduling optimization goal, and complete the construction of the multi-task scheduling optimization problem from the expected completion time, node resource constraints, the relationship between the task-node scheduling matrix and the task-resource allocation matrix; Step S3, using a heuristic strategy to calculate the critical resource requirements and pre-occupied resource amounts of each task to be scheduled, and based on the calculation results, complete the execution node allocation and resource pre-occupancy of each task in the order of the expected completion time of the tasks; The multi-task scheduling optimization problem is optimized and solved using nonlinear optimization methods to obtain the recommended resource allocation for each task. Finally, based on resource constraints and associated constraints, the task-node scheduling matrix and the task-resource allocation matrix are verified and corrected to complete the final generation of the multi-task joint scheduling solution.

2. The multi-task joint scheduling method for the edge-cloud system according to claim 1 is characterized in that: The specific method of constructing a multi-task joint scheduling matrix model is: Determine the physical device node as the task initiating node. The edge server nodes and one cloud center node are task computing nodes; The number of tasks waiting for the end node to submit accumulates to the threshold Then schedule multiple tasks; During each scheduling process, a group of The task is recorded as ,in , according to the expected completion time of the task Incremental sorting, that is for The expected completion time and ; Available for dispatch The computing nodes are denoted as ,in for The edge node number, is the cloud node number, then the task-node scheduling strategy is matrix Indicates that the task-resource allocation strategy is matrix Indicates that ， ， in , Indicates the task Assigned to the node Execute, then there is Representation Node For the task The amount of computing resources allocated is ; Indicates the task Not assigned to a node Execute, at this time ; The task-node scheduling strategy and the task-resource allocation strategy constitute a multi-task joint scheduling matrix model.

3. The multi-task joint scheduling method for the edge-cloud system according to claim 1 is characterized in that: The expected completion time model of the task is the task Scheduling to compute nodes Expected completion time of the execution , specifically: ， In the formula, Apply for time for task scheduling and resources, is the transmission time, is the execution time; Specifically: ， in, , Respectively for tasks The initiating node and computing node Network latency and bandwidth, The amount of data uploaded for the task, The amount of data downloaded for the task; Specifically: ， in, is the computational complexity of the task; It is the computing capability-computing resource mapping function.

4. The multi-task joint scheduling method for the edge-cloud system according to claim 1 is characterized in that: The constructed multi-task scheduling evaluation function is specifically: ， In the formula, is the first weight coefficient, is the second weight coefficient; For Node For the task The amount of computing resources allocated; , and when the task On the compute node When executed, ,otherwise ; For the task Expected completion time; , For the task Scheduling to compute nodes The expected completion time of the execution, Apply for time for task scheduling and resources, , Respectively for tasks The initiating node and computing node Network latency and bandwidth, The amount of data uploaded for the task, The amount of data downloaded for the task, is the computational complexity of the task; is the computing capability-computing resource mapping function, m is the number of tasks, is the number of computing nodes, A constant value set by the user.

5. The multi-task joint scheduling method for the edge-cloud system according to claim 4 is characterized in that: The multi-task scheduling optimization problem constructed is specifically: ， In the formula, For Node The upper limit of the remaining available resources that can be allocated, m is the number of tasks, is the number of computing nodes, A constant value set by the user; is the computational complexity of the task; is the computing capability-computing resource mapping function, For Node For the task The amount of computing resources allocated.

6. The multi-task joint scheduling method for the edge-cloud system according to claim 5 is characterized in that: The heuristic strategy is adopted to calculate the critical resource demand and pre-occupied resource quantity of each task to be scheduled, and based on the calculation results, the execution node allocation and resource pre-occupancy of each task are completed in the order of the expected completion time of the task. The specific process is as follows: For each task ,make , solve the task Critical resource requirements , for exist = The function value when ; For the task The amount of pre-occupied resources set is ,in is a positive coefficient greater than 1; according to For each Select the execution node. The specific operations are as follows: ①Get the node The remaining allocatable resources , if exists , so that ,and ; then the task Assign to Node Execute and set the node The remaining allocatable resources are adjusted to ; This means that Assign to edge nodes Execute and set the pre-occupied resources to , so let ； ②If no matching conditions are found in ① ,if , then let the task Assign to Node Execute and set the node The remaining allocatable resources are adjusted to ; This means that Distribute to cloud nodes Execute and set the pre-occupied resources to ; therefore, ； if , then abandon the task , at this time ; ③ Repeat steps ① and ② until the task group is completed The execution node allocation and resource pre-occupancy of the matrix All elements in The value of , , is the task-node scheduling strategy matrix The current recommended value of The final value of is optimized and determined in subsequent steps.

7. The multi-task joint scheduling method for the edge-cloud system according to claim 6 is characterized in that: Using nonlinear optimization methods, the multi-task scheduling optimization problem is optimized and solved, and the specific method for obtaining the recommended resource allocation for each task is as follows: The multi-task scheduling optimization problem is transformed into: ， Continue to transform the multi-task scheduling optimization problem into an unconstrained optimization problem. The specific operations are as follows: ① Substitute the first, second, and fourth constraints in the multi-task scheduling optimization problem into the optimization objective and obtain: ， in: ； ② By constructing a penalty function, the constrained problem is approximately transformed into an unconstrained problem, and the optimization iteration process is limited to the feasible domain: ， in is the penalty item, is a positive number set by the system; The nonlinear conjugate gradient method is used to treat the optimization variables Perform iterative solution, ; Record During round iteration, The value of , optimize the objective function The gradient vector at this value point is ; The iteration stop condition is the two norm of the gradient vector ,or ,in The cut-off accuracy set by the system user, The cutoff number of rounds set by the system user is recorded in the number of iteration rounds. When the iteration stop condition is met, ,but The solution result of the node task resource allocation optimization problem.

8. The multi-task joint scheduling method for the edge-cloud system according to claim 7 is characterized in that: Based on resource constraints and associated constraints, the task-node scheduling matrix and the task-resource allocation matrix are verified and corrected to complete the final generation of the multi-task joint scheduling solution. The specific method is as follows: Remember ,right The rationality of the value is verified, and the specific operation method is: For each : ①For or , then let , , ; ②For ,verify Is it greater than 0: If , then let , ;like , then let , , ; After steps ① and ②, the matrix is ​​finally determined and The values ​​of all elements in the task group To compute nodes The final scheduling solution is composed of the scheduling matrix and Completely sure, that is, for the task , If exists , so that , then Dispatching to Node Execute and assign The amount of resources; if for any , both , then the task give up.

Images