Time-sensitive task scheduling method for end-side cloud system

Abstract

The invention provides an end-side cloud system-oriented time-sensitive task scheduling method, which comprises the following steps of: firstly, carrying out unified modeling on nodes and networks, task characteristics, computing resources and task completion time in an end-side cloud system, and defining constraint conditions and target requirements; computing tasks are scheduled to a cloud center node to be executed, and edge computing resources are reserved as much as possible through preferential allocation of cloud computing resources; in the scheduling process, the cloud center node and the edge server node which are responsible for task execution try to insert the calculation task into the task queue of the started container in advance; and if the existing task queues do not meet the task deadline requirement, further trying to perform task scheduling in a mode of newly starting the execution container and the task queue. According to the time-sensitive task scheduling method for the end-side cloud system, time-sensitive task scheduling facing the end-side cloud system is realized from three aspects of system and task model construction, cloud priority time-sensitive task scheduling and edge-guaranteed time-sensitive task scheduling.

Description

Technical Field

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

[0002] In recent years, the application of IoT technology in time-sensitive scenarios such as Internet of Vehicles, industrial control, and smart grids has become increasingly widespread. In these applications, various IoT sensor devices will generate a large number of computing tasks in a short period of time, and these tasks are required to be processed quickly to ensure the stability and efficient operation of the system. However, the traditional centralized architecture based on cloud computing concentrates computing in the cloud, resulting in a long physical distance between the computing center and the terminal devices and users, which brings a lot of data transmission delays and makes it difficult to meet the low-latency service requirements. At the same time, with the surge in the number of devices, the continuous uploading of data to the ultra-large network puts huge pressure on the network bandwidth, further affecting the performance and stability of the system. To solve the above problems, the end-edge-cloud computing architecture has gradually become a research hotspot for time-sensitive IoT applications. By distributing computing resources between the end devices, edge server nodes, and cloud centers, it can achieve local computing and data processing and reduce data transmission delays. However, this hierarchical distributed architecture also brings new challenges. This is because the computing resources of a single edge server node are relatively limited, making it difficult to process massive data alone, and there is often a lack of efficient collaboration mechanisms between different nodes, making it difficult to achieve optimal task allocation and execution in a large-scale global environment. In this context, the task scheduling method of end-edge-cloud collaboration has attracted extensive attention. The task scheduling of end-edge-cloud collaboration is based on the resource status and task requirements of each layer of nodes to achieve dynamic task allocation and computing between the end, edge, and cloud. This scheduling mechanism flexibly allocates tasks between the object, edge, and cloud according to factors such as task deadline requirements, data transmission size, computing resource requirements, and system network status.

[0003] The current research methods in the field of end-cloud-edge collaborative task scheduling mainly include centralized scheduling, distributed scheduling, hierarchical scheduling, and intelligent scheduling based on machine learning. The main advantage of the centralized scheduling method is that it has a global resource perspective, which is easy to manage and maintain; but it needs to aggregate information from edge server nodes to the cloud for decision-making, and the data transmission delay is high. It is not suitable for delay-sensitive tasks that require fast response. In addition, a large amount of task and status information needs to be frequently uploaded to the cloud, which occupies bandwidth and may affect data transmission efficiency. The main advantage of the distributed scheduling method is that the data transmission delay is low and the system reliability is high; but each node only has a local perspective, and it is difficult to optimize the allocation of global resources of the system, which may lead to uneven resource utilization; and each node needs to communicate with each other and share resource information to achieve collaboration, which increases the complexity of management and coordination. The hierarchical scheduling method combines the characteristics of centralized and distributed scheduling, divides task scheduling into multiple levels, and has high flexibility. It can allocate tasks to the appropriate level for execution according to the urgency and computing requirements of the tasks, reduce latency and optimize resource usage; in addition, the hierarchical structure can make full use of the heterogeneity of end, edge, and cloud resources, improve overall performance through reasonable division of labor, and effectively utilize resources at each level; at the same time, hierarchical scheduling requires the design of task processing logic at different levels, which increases the complexity of system design. The scheduling method based on machine learning uses deep learning or reinforcement learning models to predict the resource requirements and latency requirements of tasks, dynamically adjusts the resource allocation between the end, edge, and cloud, and has the advantage of dynamically optimizing scheduling strategies; however, this method requires a large amount of data for training, and the model needs to be continuously updated during actual deployment, which increases the computing and storage burden of the system; and the prediction effect of the machine learning model depends on the data quality and model structure, which may be unstable in complex task scenarios. Summary of the invention

[0004] The purpose of the present invention is to provide a time-sensitive task scheduling method for edge-cloud systems, which realizes a hierarchical scheduling method that is highly adaptable to time-sensitive tasks from three aspects: system and task model construction, cloud-first time-sensitive task scheduling, and edge-guaranteed time-sensitive task scheduling.

[0005] The technical solution to achieve the purpose of the present invention is: a time-sensitive task scheduling method for an edge-cloud system, which specifically includes the following steps:

[0006] Step S1: Use the cloud center node as the task computing node to calculate the sum of the task scheduling time, data upload time, and result return time. ;

[0007] like , then the expected completion time from task initiation to result reception is Exceeding the task deadline , proceed to step S3, otherwise proceed to step S2;

[0008] Step S2, use the cloud center node task queue insertion strategy to schedule the task, if successful, end the scheduling, otherwise, use the cloud center node task queue new creation strategy to schedule the task, if successful, end the scheduling, if failed, proceed to step S3;

[0009] Step S3: Use the edge server node as the task computing node to recalculate the sum of the task scheduling time, data upload time, and result return time. :

[0010] If for all edge server nodes, there are , then the expected completion time from task initiation to result reception for all edge server nodes as task computing nodes exceeds the task deadline , that is, the expected completion time of all edge server nodes exceeds the task deadline, so the task is abandoned And end the scheduling;

[0011] If there is an edge server node, so that the expected completion time from task initiation to result reception does not exceed the task deadline when the task is calculated, the edge server nodes that meet the conditions are scheduled in turn using the edge server node task queue insertion strategy. If successful, the scheduling is terminated. If all edge server nodes that meet the conditions fail, the edge server node task queue creation strategy is used to schedule the task. If successful, the scheduling is terminated. If failed, the task is abandoned. And end the scheduling.

[0012] Compared with the prior art, the present invention has the following significant advantages:

[0013] (1) The task scheduling framework designed by the present invention combines cloud priority with edge guarantee, which breaks through the limitation of edge-first and cloud-later scheduling in the traditional hierarchical scheduling mode. It can effectively avoid the situation where the cloud task scheduling delay cannot meet the sudden task with extremely short deadline after the edge resources are exhausted first, and has a high degree of originality.

[0014] (2) The cloud-edge two-level task scheduling method proposed in the present invention can flexibly select two strategies, task queue insertion and task queue new start, based on the original computing container and task queue, taking into account the latency requirements of the current task and the original task, and is highly innovative.

[0015] (3) The edge-cloud system and task model adopted by the present invention can uniformly model the nodes and networks, task characteristics, computing resources, and task completion time in the edge-cloud system, and cover the entire life cycle of task initiation, transmission, scheduling, waiting, execution, and return, and has high practicality. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of the time-sensitive task scheduling method of the present invention. DETAILED DESCRIPTION

[0017] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0018] Combination Figure 1 A time-sensitive task scheduling method for end-edge-cloud systems is proposed. Based on the hierarchical scheduling design idea, a unified model of nodes and networks, task characteristics, computing resources, and task completion time in the end-edge-cloud system is established. The constraints and target requirements of the entire life cycle of task initiation, transmission, scheduling, waiting, execution, and return are clarified. By giving priority to the allocation of computing resources on the cloud in exchange for the maximum retention of edge computing resources, it is used to cope with sudden tasks with extremely short deadlines. From three aspects, system and task model construction, cloud-prioritized time-sensitive task scheduling, and edge-guaranteed time-sensitive task scheduling, time-sensitive task scheduling for end-edge-cloud systems is realized.

[0019] First, determine the system and task model under the edge-cloud architecture:

[0020] A unified description is given to the composition, numbering, communication links, and core parameters of the three types of nodes: end, edge, and cloud. Secondly, the tasks initiated by the device nodes on the physical end are modeled, and the main characteristics of the task model are determined, including computing resource requirements, deadlines, task data volume, and task computing complexity description functions. After that, the cloud and edge task execution nodes are modeled, and the node computing resource model, node container, and its task queue model are determined. Finally, the completion time of the computing task is modeled, including executing tasks in existing containers and executing tasks in newly created containers.

[0021] The edge-to-edge cloud system involved in the present invention includes three types of nodes, namely cloud center nodes, edge server nodes and physical device nodes. The number and functions of the three types of nodes in the system are as follows: The physical device node is the task initiating node. The edge server nodes and one cloud center node are task computing nodes. The best communication link from the task initiation node to the task computing node (i.e., the link with the fastest data transmission among the accessible communication links) is recorded as ,link The propagation delay and bandwidth are denoted as and ,in It is the task initiating node number. All task initiating nodes are physical device nodes. is the task computing node, when When the task computing node is the edge server node, The task computing node is the cloud center node.

[0022] Computational tasks The task is initiated by the device node at the physical end, and after the decision of the task scheduler, it is transmitted to the designated task queue of the designated computing node to wait. After the waiting is completed, the task is actually executed and the result of the calculation is returned to the task initiating node.

[0023] Computing resource requirements Representation Task Minimum requirements for computing resources, including but not limited to CPU, GPU, memory, etc.; deadline Representation Task The longest tolerable waiting time from initiation to receiving the returned result; the amount of task data Representation Task The amount of data that needs to be transmitted; the function describing the computational complexity of the task (Subscript represents the meaning of calculation), then the expected computing time of the task is The time consumed is determined by the computational complexity of the task itself and the amount of task data. , and the actual allocated computing resources The three are jointly determined and obtained by fitting the historical task execution time data.

[0024] The cloud edge server node itself has certain computing resources such as CPU, GPU, memory, etc., and is deployed with running containers that can handle different types of tasks. edge server nodes and 1 cloud center node, currently started and able to handle computing tasks The set of all running containers is recorded as ,in is the total number of containers in the collection. For each container , including the following features:

[0025] a) Representation container The actual allocated computing resources include but are not limited to CPU, GPU, memory, etc.

[0026] b) Representation container The current task queue, and make the following agreements for the task queue: ① is the queue length, that is, the queue has A task, when When indicating container There is no computing task at present; ②When hour, Indicates the first tasks, including , Indicates the task currently being calculated; ③ Indicates the first The remaining deadline margin of the task, that is, The difference between the deadline and expected completion time of a task; ④ Indicates the first The expected completion time of the task is Time Task Expected to be able to be container Execution completed.

[0027] c) Representation container Currently calculating the task The expected remaining computing time, the amount of task data currently being calculated , the task has been calculated for ,but

[0028]

[0029] When the task The expected completion time from task initiation to result reception when the corresponding execution container has been started Includes: Task scheduling time , Data upload time , Task queue waiting time , Task execution time , Result return time .Right now

[0030] (Formula 1)

[0031] When the task When the corresponding execution container needs to be created, the expected completion time from task initiation to result reception Includes: Task scheduling time , Data upload time , Execute container startup time , Task execution time , Result return time .

[0032] (Formula 2)

[0033] In formula 1 and formula 2: ① Can be approximated as a constant, determined by the mean of the system's historical monitoring data; ② It is determined by the task data size, communication link bandwidth, and communication link delay, that is, ,in Respectively for tasks ③ Since the amount of data returned is usually small, the data transmission time can be approximately ignored. ④If the task Actually allocated to the container Execute, task execution time ,in For container The allocated computing resources.

[0034] In formula 1: ,in For container The task queue The amount of task data for each task, For container The task queue The expected computation time of a task, For container The computation time that has elapsed for the first task in the task queue.

[0035] In formula 2: It can be approximated as a constant and is determined by the mean of the system's historical monitoring data.

[0036] A time-sensitive task scheduling method for edge-cloud systems, the specific steps are:

[0037] Step S1: Use the cloud center node as the task computing node to calculate the sum of the task scheduling time, data upload time, and result return time. ;

[0038] like , then the expected completion time from task initiation to result reception is Exceeding the task deadline , proceed to step S3, otherwise proceed to step S2. The specific process is:

[0039] Remember the task The initiating node number is , the task execution node number is ,remember Calculate the node number for the task. When the task computing node is the edge server node, The task computing node is the cloud center node.

[0040] From formula 1, we can know that (Formula 3). is the expected completion time in (1) and (2) The public fixed overhead portion of is a known constant related to tasks and nodes. If the fixed time overhead is greater than or equal to the task deadline, the task deadline requirement cannot be met no matter how it is scheduled.

[0041] First try to assign the task Dispatching to the cloud center node, the specific steps are as follows:

[0042] For the cloud center node, , and calculated according to (3) :

[0043] like , then it means , that is, the expected completion time of the cloud center node exceeds the task deadline, so step S3 is performed.

[0044] like , then proceed to step S2.

[0045] Step S2, use the cloud center node task queue insertion strategy to schedule the task, if successful, end the scheduling, otherwise, use the cloud center node task queue new creation strategy to schedule the task, if successful, end the scheduling, if failed, proceed to step S3;

[0046] Schedule time-sensitive tasks to cloud center nodes, save edge server node computing resources as much as possible, and use them to process tasks with low communication delay requirements. The specific strategies for scheduling time-sensitive tasks to cloud center nodes include cloud center node task queue insertion strategy and cloud center node task queue creation strategy. The former aims to insert computing tasks into the existing task queue of the cloud center node and ensure that tasks at and after the insertion position will not time out; the latter, when there is no suitable insertion position, creates a new container and a new task queue in the cloud center node to complete the computing task before the deadline.

[0047] Specifically, the goal of using the cloud center node task queue insertion strategy for task scheduling is to sequentially Schedule to the existing task queue of the cloud center node until a task with the expected completion time is found Less than the deadline The target container on the cloud center node , and the target insertion position of the container task queue When the containers and insertion positions on the above cloud center nodes can be found, this scheduling plan is the task The actual scheduling plan.

[0048] Since the scheduling target is , so it is necessary to satisfy For cloud center nodes ( ) can handle computing tasks The set of all containers is recorded as ,in is the total number of containers in the collection. For each container Task queue Try inserting a task (Note that for each given container , are known constants related to tasks and containers). The specific steps are as follows:

[0049] ① Separate order ,like , so that , explain the task Even if inserted at the end of the task queue, it can still meet the deadline requirements and will not affect the expected completion time of the original tasks in the queue. Insert cloud center node The task queue is at the end of the task, and the task scheduling ends. Then the That is , End of the queue That is ;like If all attempts to insert fail, proceed to step ②.

[0050] ② Separate orders , , compare in turn and Size until you find The task queue ( ) bit task, so that , and for , both , explain the task Insert into Task queue No. The position after will not cause The following task has timed out. Therefore, the task Queued to cloud center node The task queue of Then the cloud center node That is , The queue of The position is At this time, if the task Dispatching to cloud center nodes Task queue No. The task is one digit after the task. The deadline can be met, but the task queue originally queued The expected completion time of the task All increased , that is, the remaining deadline margin All reduced , so we need to judge Are they all greater than 0?

[0051] On the contrary, if , then it means to The task queue attempt to insert failed; this is because if This means that the task The first task queued or earlier, will result in the existence of The task timed out; on the other hand, have , indicating that the task The first task queued The later position will lead to the task It has timed out. Therefore, in this case, the task Insert into the cloud center node Any position in the task queue will cause the task in the queue to time out. ,as well as , no results were found that met the above conditions. and , it means that the attempt has failed, and the new strategy of the cloud center node task queue is used to try task scheduling.

[0052] Specifically, the goal of using the new strategy of the cloud center node task queue for task scheduling is to try to place tasks in the cloud center nodes (node ​​number ) Create a new container To perform tasks If you create a new container on a cloud center node , and the container The resources are , which enables the task exist Expected completion time of the execution Less than the deadline . Then at this time when the above container , allocate resources That is the task The actual scheduling plan.

[0053] The condition for using the new strategy of the cloud center node task queue for task scheduling is that there is no suitable position to insert the task in the task queue of the existing container in the cloud center node Therefore, you need to try to execute the task by creating a new container in the cloud center node. . Record the cloud center node (node ​​number ) The current container number is Then, the newly created container is recorded as At this time, according to formula 2, the task The expected completion time is:

[0054]

[0055] can be approximated as a constant, is a known constant related to tasks and nodes, so let Also a constant. The fixed time overhead of the new strategy for the task queue is adopted. If the fixed time overhead is greater than or equal to the task deadline, the task deadline requirement cannot be met no matter how it is scheduled.

[0056] if , it means that even if the task execution time is not considered, the task cannot be completed before the deadline, so directly go to step S3 to try.

[0057] if When , so it is necessary to satisfy Container The resources are , ,in For the task The correlation constant, Calculate the complexity and amount of task data for the task itself , and the actual allocated computing resources Related functions, so It can be further written as only about the variables in the form of , due to the computation time With computing resources decreases with the increase of The inverse function of Existence. ,in The redundancy guarantee coefficient set by the user to ensure that the task will not time out. The maximum amount of resources that the cloud center node can currently allocate to a single container .like , then create a new container for the cloud center node The amount of computing resources allocated , and the task Dispatch to The head of the task queue; when When , it means that the current computing resources of the cloud center node are insufficient to support the task It is completed before the deadline, so the process goes to step S3 to try again.

[0058] Step S3: Use the edge server node as the task computing node to recalculate the sum of the task scheduling time, data upload time, and result return time. :

[0059] If for all edge server nodes, there are , then the expected completion time from task initiation to result reception for all edge server nodes as task computing nodes exceeds the task deadline , that is, the expected completion time of all edge server nodes exceeds the task deadline, so the task is abandoned And end the scheduling;

[0060] If there is an edge server node, so that the expected completion time from task initiation to result reception does not exceed the task deadline when the task is calculated, the edge server nodes that meet the conditions are scheduled in turn using the edge server node task queue insertion strategy. If successful, the scheduling is terminated. If all edge server nodes that meet the conditions fail, the edge server node task queue creation strategy is used to schedule the task. If successful, the scheduling is terminated. If failed, the task is abandoned. And end the scheduling.

[0061] For tasks that cannot find a suitable scheduling strategy in the cloud center node, we take advantage of the low transmission delay of the edge server node and further try to schedule tasks on the edge server node. Specifically, it includes the edge server node task queue insertion strategy and the edge server node task queue creation strategy. The former aims to try to insert the computing task into the existing task queue of the edge server node and ensure that the insertion position and subsequent tasks will not time out; the latter, when there is no suitable insertion position, tries to complete the computing task before the deadline by creating a new container and a new task queue in the edge server node.

[0062] In a further embodiment, the goal of using the edge server node task queue insertion strategy for task scheduling is to try to insert tasks into the queue in sequence. Dispatch to The existing task queues of edge server nodes are searched until a task with the expected completion time is found. Less than the deadline Target edge server node , the target container on the node , and the target insertion position of the container task queue When the above target node, container, and insertion position can be found, the task is completed. The actual scheduling plan.

[0063] Since the scheduling target is , so it is necessary to satisfy For edge server nodes ( ) can handle computing tasks The set of all containers is recorded as ,in is the total number of containers in the collection. For each container Task queue Try inserting a task (Note that for each given container , are known constants related to tasks and containers).

[0064] For all edge server nodes, , try steps ① and ②; if for a certain edge server node If you succeed in ① or ②, the task is completed. Scheduling; if for all edge server nodes If both attempts ① and ② fail, the new strategy of the edge server node task queue is used to try to Perform scheduling. The specific steps are as follows:

[0065] ① Separate order ,like , so that , explain the task Even if inserted at the end of the task queue, it can still meet the deadline requirements and will not affect the expected completion time of the original tasks in the queue. Insert cloud center node The task queue is at the end of the task, and the task scheduling ends. Then the That is , End of the queue That is ;like If all attempts to insert fail, proceed to step ②.

[0066] ② Separate orders , , compare in turn and Size until the task queue is found ( ) bit task, so that , and for . Description of the task Insert into The task queue of The position after will not cause The following task times out. Queued to the edge server node currently being tried superior Task queue No. The task scheduling ends. The edge server node number at this time is That is , edge server node middle That is , The task queue of That is If for all ,as well as , no results were found that met the above conditions. and , then it means that for this edge server node The attempt failed, so , return to step ① and try the next edge server node.

[0067] In a further embodiment, the goal of using the edge server node task queue new creation strategy for task scheduling is to try to sequentially schedule tasks in Edge Server Nodes Create a new container To perform tasks In the edge server node Create a new container on , and the container The resources are , which enables the task exist Expected completion time of the execution Less than the deadline . Then at this time when the above target node ,container , allocate resources That is the task The actual scheduling plan.

[0068] The condition for using the new strategy of edge server node task queue for task scheduling is that there is no suitable position to insert the task in the task queue of the existing container in the edge server node. Therefore, you need to try to use the edge server node ( ) Try to execute the task by creating a new container . Note the edge server node ( ) The current container number is Then, the newly created container is recorded as At this time, according to formula 2, the task The expected completion time is:

[0069]

[0070] can be approximated as a constant, is a known constant related to tasks and nodes, so let Also a constant. The fixed time overhead of the new strategy for the task queue is adopted. If the fixed time overhead is greater than or equal to the task deadline, the task deadline requirement cannot be met no matter how it is scheduled.

[0071] if , it means that even if the task execution time is not considered, the task cannot be completed before the deadline. Therefore, the task will time out no matter how it is scheduled in the system, so the task is directly abandoned. .

[0072] if When , so it is necessary to satisfy Container The resources are , ,in For the task The correlation constant is thus It can be further written as only about the variables in the form of , due to the computation time With computing resources decreases with the increase of The inverse function of Existence. ,in The redundancy guarantee coefficient set by the user to ensure that the task will not time out. The maximum amount of resources that can currently be allocated to a single container .like , then it is the edge server node Create a new container The amount of computing resources allocated , and the task Dispatch to The task queue is headed by the edge server node number. That is , create a new container The amount of computing resources allocated That is ;when , it indicates that the edge server node The current computing resources are insufficient to support the task Completed before the deadline, so Retry to schedule to the next edge server node; if If all attempts fail, it means that the task will time out no matter how it is scheduled in the system, so the task is directly abandoned. .

[0073] For the task If the scheduling is completed by using the cloud center node task queue insertion strategy, it means that the task is assigned to the target container of the cloud node. , and the target insertion position of the container task queue ; If the new strategy of the cloud center node task queue is used to complete the scheduling, it means that the task is assigned to the cloud node ( ) ,and The resources allocated are ; If the scheduling is completed using the edge server node task queue insertion strategy, it means that the task is assigned to the target edge server node , the target container on the node , and the target insertion position of the container task queue ; If the new strategy of the edge server node task queue is used to complete the scheduling, it means that the task is assigned to the edge server node New container on ,and The resources allocated are ; If the new strategy is executed until the edge server node task queue is completed, the task If the scheduling position is still not found, it means that no matter how the task is scheduled, will time out, so the task will be abandoned .

Claims

1. A time-sensitive task scheduling method for edge-cloud systems, characterized in that: The specific steps include: Step S1: Use the cloud center node as the task computing node to calculate the sum of the task scheduling time, data upload time, and result return time. ; like , then the expected completion time from task initiation to result reception is Exceeding the task deadline , proceed to step S3, otherwise proceed to step S2; Step S2, use the cloud center node task queue insertion strategy to schedule the task, if successful, end the scheduling, otherwise, use the cloud center node task queue new creation strategy to schedule the task, if successful, end the scheduling, if failed, proceed to step S3; Step S3: Use the edge server node as the task computing node to recalculate the sum of the task scheduling time, data upload time, and result return time. : If for all edge server nodes, there are , then the expected completion time from task initiation to result reception for all edge server nodes as task computing nodes exceeds the task deadline , that is, the expected completion time of all edge server nodes exceeds the task deadline, so the task is abandoned And end the scheduling; If there is an edge server node, so that the expected completion time from task initiation to result reception does not exceed the task deadline when the task is calculated, the edge server nodes that meet the conditions are scheduled in turn using the edge server node task queue insertion strategy. If successful, the scheduling is terminated. If all edge server nodes that meet the conditions fail, the edge server node task queue creation strategy is used to schedule the task. If successful, the scheduling is terminated. If failed, the task is abandoned. And end the scheduling.

2. The time-sensitive task scheduling method for edge-end cloud system according to claim 1 is characterized in that: The specific method of using the cloud center node task queue insertion strategy for task scheduling is: Tasks Schedule to the existing task queue of the cloud center node until a task with the expected completion time is found Less than the deadline The target container on the cloud center node , and the target insertion position of the target container task queue , specifically including: ① Separate order , , The existing cloud center nodes can handle computing tasks The total number of containers, if any , so that , Representation container The current task queue The expected completion time of a task, For container The number of tasks currently in the task queue, Representation container The expected completion time of all tasks in the current task queue; then the task Even if inserted at the end of the task queue, it can still meet the deadline requirements and will not affect the expected completion time of the original tasks in the queue. Insert cloud center node The task queue is at the end of the task, and the task scheduling ends; at this time, the cloud center node That is , End of the queue That is ;like If all attempts to insert fail, proceed to step ②; ② Separate orders , , compare in turn and Size until you find The first task in the task queue bit task, so that , and for , both have ,in Indicates the first The remaining deadline margin of the task, that is, The difference between the deadline and the expected completion time of a task indicates that the task Insert into Task queue No. The position after will not cause The following task times out, so the task Queued to cloud center node The task queue of bit, and end this task scheduling, then the cloud center node That is , The queue of The position is , if for all ,as well as , no results were found that met the above conditions. and , it means that the scheduling failed.

3. The time-sensitive task scheduling method for edge-to-edge cloud systems according to claim 1, characterized in that: The specific method of using the new strategy of the cloud center node task queue to schedule tasks is: Create a new container on the cloud center node To perform tasks , If you create a new container on a cloud center node , and the container The resources are , which enables the task exist Expected completion time of the execution Less than the deadline , the container at this time , allocate resources That is the task The actual scheduling plan is as follows: make , is the fixed time cost of adopting the task queue creation strategy, The time to start the execution container; if , even if the task execution time is not considered, the task cannot be completed before the deadline, and the process goes to step S3; if When , so it is necessary to satisfy ,container The resources are , , Calculate the complexity and amount of task data for the task itself , and the actual allocated computing resources Related function; let ; Due to the calculation time With computing resources decreases with the increase of The inverse function of Existence; order ,in The redundancy guarantee factor set by the user to ensure that the task will not time out, ; Record the maximum amount of resources that the cloud center node can currently allocate to a single container ,like , then create a new container for the cloud center node The amount of computing resources allocated , and the task Dispatch to The head of the task queue; when When , it means that the current computing resources of the cloud center node are insufficient to support the task Complete before the deadline and proceed to step S23.

4. The time-sensitive task scheduling method for edge-to-edge cloud systems according to claim 1, characterized in that: The specific method of using the edge server node task queue insertion strategy for task scheduling is: Initially, ; ① Separate order ,like , so that , explain the task If the task can still meet the deadline even if it is inserted at the end of the task queue and does not affect the expected completion time of the original task in the queue, the task will be Insert cloud center node The task queue is at the end of the task, and the task scheduling ends. At this time, the cloud center node That is , End of the queue That is ;like If all attempts to insert fail, proceed to step ②; ② Separate orders , , compare in turn and Size until you find The first task in the task queue bit task, so that , and for , both , explain the task Insert into Task queue No. The position after will not cause If the subsequent task times out, the task Queued to the edge server node currently being tried of The task queue of The task scheduling ends. The edge server node number at this time That is , edge server node middle That is , The task queue of That is ; If for all ,as well as , no results were found that met the above conditions. and , then it means that for this edge server node The attempt failed, so , return to step ① and try the next edge server node.

5. The time-sensitive task scheduling method for edge-end cloud system according to claim 1, characterized in that: Edge server nodes The current container number is , then the newly created container is recorded as , ; make , The fixed time overhead of creating a new strategy for the task queue; if , it means that even if the task execution time is not considered, the task cannot be completed before the deadline, so the task is directly abandoned. ; if When the container The resources are , , Calculate the complexity and amount of task data for the task itself and the actual computing resources allocated Related function, let ; Due to the calculation time With computing resources decreases with the increase of The inverse function of Existence; order ,in The redundancy guarantee coefficient set by the user to ensure that the task will not time out, record the edge server node The maximum amount of resources that can currently be allocated to a single container ,like , then it is the edge server node Create a new container The amount of computing resources allocated , and the task Dispatch to The task queue is headed by the edge server node number. That is , create a new container The amount of computing resources allocated That is ;when , it indicates that the edge server node The current computing resources are insufficient to support the task Completed before the deadline, so Retry to schedule to the next edge server node; if If all attempts fail, the task will be abandoned. .

6. The time-sensitive task scheduling method for edge-end cloud system according to claim 1, characterized in that: The sum of task scheduling time, data upload time, and result return time Specifically: ； In the formula, Schedule time for tasks, To calculate the data upload time, The time when the calculation result is returned.

7. The time-sensitive task scheduling method for edge-end cloud system according to claim 6 is characterized in that: Data upload time Specifically: ， in, is the task data volume, For Link The propagation delay of For Link bandwidth; The specific time for the result return is: 。

Images