Task scheduling methods, devices, equipment and storage media
By selecting the target execution node and transmission path in the edge computing system, the problem of resource occupation and coordinated scheduling of transmission resources in dynamic workflow execution scenarios of task scheduling methods is solved, and the energy consumption of the edge computing system is optimized.
Patent Information
- Application Number
- CN202611182703.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-05
- Publication Date
- 2026-11-17
AI Technical Summary
Existing task scheduling methods cannot effectively characterize the resource consumption of task execution within nodes and the resource consumption of data transmission dependencies between nodes in dynamic workflow execution scenarios. In particular, they cannot adapt to the limitations of dynamic resource sharing and non-fully connected topologies in real edge networks.
By obtaining the remaining capacity of edge nodes and the remaining bandwidth of physical links, the target execution node and the target transmission path of dependent data are selected, so that the energy consumption of the edge computing system meets the preset optimization target and realizes the collaborative scheduling of computing network.
A method is provided that enables joint decision-making on task execution and data transmission under resource constraints, avoiding resource conflicts and improving the energy consumption optimization effect of edge computing systems.
Smart Images

Figure CN123123350A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of edge computing technology, and in particular to a task scheduling method, apparatus, device and storage medium. Background Technology
[0002] Multi-access edge computing provides low-latency services for latency-sensitive and compute-intensive applications by deploying computing, storage, and network resources closer to the user. Many edge applications consist of multiple subtasks with predecessor-successor relationships, which can be represented as a directed acyclic graph. For these workflow applications, a successor task can only begin execution after its predecessor task has completed and all relevant intermediate data transmissions have been completed. Therefore, task scheduling not only needs to determine the execution order and execution nodes of tasks, but also needs to handle the dynamic occupancy of computing and communication resources during task execution.
[0003] Existing task scheduling methods typically model task execution time as the relationship between task computation and the processing power or CPU frequency of the selected server, and inter-task data transmission as the communication time, energy consumption, or cost between selected execution nodes. While these models can describe basic task dependencies and node selection problems, they still have shortcomings in computing-network collaborative modeling. First, many models abstract inter-node data transmission as direct communication between any two nodes, equivalent to assuming a fully connected network topology or direct reachability between any two execution nodes. However, in real edge networks, edge nodes are typically constrained by physical deployment location, access network structure, link capacity, and routing constraints. Direct communication links between any two nodes are not necessarily present, and cross-node dependent data transmission often requires multiple hops. Second, many models treat task execution as exclusive use of the selected computing node or processor, with task execution time determined by task computation and the overall computing power of the node, rarely depicting CPU allocation, cache usage, and resource release processes when multiple tasks are executed concurrently within the same node. Therefore, for dynamic workflow execution scenarios, a computing-network collaborative scheduling method is needed that can simultaneously characterize the resource consumption of task execution within a node and the resource consumption of inter-node dependent data transmission. Summary of the Invention
[0004] The main objective of this application is to provide a task scheduling method, apparatus, device, and storage medium, which aims to solve the technical problem of how to provide a computing network collaborative scheduling method that can simultaneously characterize the resource consumption of task execution within a node and the resource consumption of data transmission dependencies between nodes in a dynamic workflow execution scenario.
[0005] To achieve the above objectives, this application provides a task scheduling method applied to an edge computing system, the edge computing system including multiple edge nodes and physical links connecting the edge nodes, the method comprising the following steps: If it is detected that there are tasks to be scheduled in the workflow of the edge computing system, the remaining capacity of each edge node and the remaining bandwidth of each physical link are obtained; Based on the remaining capacity, the target execution node corresponding to the task to be scheduled is selected from each of the edge nodes; When the target execution node is different from the node where the predecessor task of the task to be scheduled is located, the target transmission path corresponding to the dependent data of the task to be scheduled is selected from each of the physical links according to the remaining bandwidth; The task to be scheduled is scheduled based on the target execution node and the target transmission path. The target execution node and the target transmission path work together to ensure that the energy consumption of the edge computing system meets the preset optimization target.
[0006] In one embodiment, the remaining capacity includes remaining CPU capacity and remaining cache capacity; The step of selecting the target execution node corresponding to the task to be scheduled from each of the edge nodes based on the remaining capacity includes: From each of the edge nodes, candidate nodes are selected whose remaining CPU capacity is greater than or equal to the minimum CPU requirement of the task to be scheduled, and whose remaining cache capacity is greater than or equal to the cache requirement of the task to be scheduled. Based on the remaining CPU capacity, remaining cache capacity, estimated execution time of the task to be scheduled on each candidate node, and estimated data transfer time from the node containing the predecessor task of the task to be scheduled to each candidate node, the target execution node corresponding to the task to be scheduled is selected from the candidate nodes.
[0007] In one embodiment, the step of selecting the target transmission path corresponding to the dependent data of the task to be scheduled from each of the physical links based on the remaining bandwidth includes: Obtain the data volume of the dependent data of the task to be scheduled, and determine the bandwidth requirement based on the data volume of the dependent data; From each of the physical links, select available links whose remaining bandwidth is greater than or equal to the bandwidth requirement, and construct a link subgraph based on the available links; In the link subgraph, the available path with the fewest hops from the node where the predecessor task is located to the target execution node is selected as the target transmission path corresponding to the dependent data.
[0008] In one embodiment, the step of scheduling the task to be scheduled based on the target execution node and the target transmission path includes: The CPU allocation amount for the scheduled task is determined based on the CPU requirement type of the scheduled task, and the CPU allocation amount is used as the execution resource allocation for the scheduled task on the target execution node. In the case of cross-node dependent data transmission, the bandwidth occupancy status of the physical link on the target transmission path is updated to occupied status, and the occupancy duration is recorded as the transmission time of the dependent data. After the dependent data transmission is completed, the bandwidth resources of the physical link on the target transmission path are released; When the task to be scheduled is accepted by the target execution node, the cache occupancy status of the task to be scheduled is updated to occupied status. After the dependent data transmission is completed and the target execution node is able to allocate the CPU allocation to the task to be scheduled, the task to be scheduled is executed on the target execution node. After the scheduled task is completed, the CPU and cache resources occupied by the scheduled task on the target execution node are released.
[0009] In one embodiment, the step of determining the CPU allocation amount for the task to be scheduled based on the CPU requirement type of the task to be scheduled includes: When the CPU requirement type is frequency insensitive, the CPU allocation of the task to be scheduled is determined as the minimum CPU requirement of the task to be scheduled. When the CPU demand type is unlimited frequency sensitive, the CPU allocation of the task to be scheduled is determined to be a value that is greater than or equal to the minimum CPU demand and less than or equal to the CPU capacity of the target execution node. When the CPU demand type is threshold frequency sensitive, the CPU allocation amount of the task to be scheduled is determined to be a value that is greater than or equal to the minimum CPU demand and less than or equal to the maximum effective CPU demand of the task to be scheduled.
[0010] In one embodiment, the task scheduling method further includes: Construct the first group corresponding to the task selection rule and the second group corresponding to the node selection rule; Select a first representative individual from the first population and a second representative individual from the second population, and combine the first representative individual and the second representative individual into a complete scheduler; Run the full scheduler in an event-driven simulator to perform scheduling simulation on the dynamically arriving workflows and determine the fitness of the full scheduler; Evolutionary operations are performed on the first population and the second population based on the fitness until a preset termination condition or the maximum number of generations is reached, at which point the optimal task selection rule and the optimal node selection rule are output. The task to be scheduled is determined from each ready task according to the optimal task selection rule, and the target execution node is selected from each edge node according to the optimal node selection rule.
[0011] In one embodiment, after the step of scheduling the task to be scheduled based on the target execution node and the target transmission path, the method further includes: When the target execution node cannot accept the task to be scheduled, the task to be scheduled is stored in the global cache queue; wherein, the inability to accept includes: the remaining cache capacity of the target execution node is less than the cache requirement of the task to be scheduled, or the remaining CPU capacity of the target execution node is less than the minimum CPU requirement of the task to be scheduled and the task to be scheduled cannot enter the local cache queue to wait. Determine the scheduling priority of each cache task in the global cache queue, and reselect execution nodes for each cache task in descending order of scheduling priority; When the scheduled task fails to execute or cannot be accepted by the target execution node, the global cache reallocation process of the workflow is triggered.
[0012] Furthermore, to achieve the above objectives, this application also proposes a task scheduling device, which includes: The data acquisition module is used to acquire the remaining capacity of each edge node and the remaining bandwidth of each physical link when the workflow of the edge computing system detects that there are tasks to be scheduled. The first selection module is used to select the target execution node corresponding to the task to be scheduled from each of the edge nodes according to the remaining capacity, so that the computing energy consumption of the edge computing system meets the preset optimization target. The second selection module is used to select the target transmission path corresponding to the dependent data of the task to be scheduled from each of the physical links according to the remaining bandwidth when the target execution node is different from the node where the predecessor task of the task to be scheduled is located, so that the transmission energy consumption of the edge computing system meets the preset optimization target. The scheduling and execution module is used to schedule the task to be scheduled based on the target execution node and the target transmission path.
[0013] In addition, to achieve the above objectives, this application also proposes a task scheduling device, the device comprising: a memory, a processor, and a task scheduler stored in the memory and executable on the processor, the task scheduler being configured to implement the steps of the task scheduling method described above.
[0014] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a task scheduler is stored, and when the task scheduler is executed by a processor, it implements the steps of the task scheduling method described above.
[0015] In addition, to achieve the above objectives, the present invention also provides a computer program product, the computer program product including a task scheduler, which, when executed by a processor, implements the steps of the task scheduling method as described above.
[0016] One or more technical solutions proposed in this application have at least the following technical effects: The task scheduling method of this application includes: when it is detected that there are tasks to be scheduled in the workflow of the edge computing system, obtaining the remaining capacity of each edge node and the remaining bandwidth of each physical link; selecting a target execution node corresponding to the task to be scheduled from each edge node according to the remaining capacity; when the target execution node is different from the node where the predecessor task of the task to be scheduled is located, selecting a target transmission path corresponding to the dependent data of the task to be scheduled from each physical link according to the remaining bandwidth; scheduling the task to be scheduled based on the target execution node and the target transmission path, wherein the target execution node and the target transmission path work together to make the energy consumption of the edge computing system meet a preset optimization target.
[0017] This application selects the target execution node corresponding to the scheduled task based on the remaining capacity of each edge node. This node selection depends on the current available resource status of the edge node, rather than a pre-defined exclusive node allocation. This allows the same edge node to accept multiple tasks within its remaining capacity, reflecting the dynamic occupancy of computing and storage resources during task execution within a node. Simultaneously, it selects the target transmission path corresponding to the dependent data based on the remaining bandwidth of each physical link. This path selection depends on the current available bandwidth of the link, rather than assuming a direct communication link between any two nodes. This allows for the selection of a practically feasible transmission route for cross-node data transmission under physical link constraints. The determined target execution node and target transmission path are then used together to schedule the task. This allows for joint decision-making of resource consumption during task execution and data transmission under the constraint of a pre-defined energy consumption optimization objective. This avoids resource conflicts or inefficient allocation caused by independent computing and communication resource scheduling. Therefore, it provides a computing network collaborative scheduling method that can simultaneously characterize the resource occupancy of task execution within a node and the resource occupancy of data transmission between nodes. This solves the technical problem of existing methods, which, due to the assumption of direct communication between nodes and task exclusive node CPU execution, cannot adapt to dynamic resource sharing and non-fully connected topologies in real edge networks. Attached Figure Description
[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.
[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0020] Figure 1 This is a flowchart illustrating the first embodiment of the task scheduling method of this application; Figure 2 This is a flowchart illustrating the second embodiment of the task scheduling method of this application; Figure 3 This is a flowchart illustrating the third embodiment of the task scheduling method of this application; Figure 4 This is a diagram of the multi-access edge computing system architecture for the task scheduling method of this application; Figure 5 This is a structural block diagram of the first embodiment of the task scheduling device of this application; Figure 6 This is a schematic diagram of the task scheduling device of this application.
[0021] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0022] To make the above-mentioned objectives, features, and advantages of this application more apparent and understandable, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.
[0023] This application provides a task scheduling method, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the task scheduling method of this application.
[0024] In this embodiment, the task scheduling method is applied to an edge computing system, which includes multiple edge nodes and physical links connecting the edge nodes. The method includes the following steps: Step S1: If it is detected that there are tasks to be scheduled in the workflow of the edge computing system, obtain the remaining capacity of each edge node and the remaining bandwidth of each physical link.
[0025] It should be noted that the executing entity of this application embodiment can be a task scheduling device capable of executing the task scheduling method of this application, and this embodiment does not limit this. The following uses a task scheduling device (hereinafter referred to as scheduling device) as an example to describe this embodiment and the following embodiments.
[0026] It should also be noted that the aforementioned edge computing system can be a computing network system composed of multiple edge nodes and the physical links connecting them. It is used to deploy computing, storage, and network resources closer to the user, providing low-latency services for latency-sensitive and compute-intensive applications. The aforementioned workflow can be an application consisting of multiple subtasks with predecessor-successor relationships. The dependencies between these subtasks can be described using a directed acyclic graph. A subsequent task can only begin execution after its predecessor task has completed and all relevant intermediate data transmissions have been completed.
[0027] It should be understood that the aforementioned tasks awaiting scheduling can be tasks that, during workflow execution, have met the conditions of completed predecessor tasks and ready dependent data, and are waiting to be assigned to an execution node and started execution. The aforementioned remaining capacity can refer to the computing and storage resources of the edge node that are not currently occupied, including remaining CPU capacity and remaining cache capacity. The remaining CPU capacity characterizes the computing power that the edge node can currently allocate to new tasks, and the remaining cache capacity characterizes the cache space that the edge node can currently use to store task-related data and execution status. The aforementioned remaining bandwidth can refer to the transmission bandwidth resources of the physical link that are not currently occupied, used to measure the maximum transmission rate of new data transmission tasks that the physical link can still support.
[0028] Step S2: Select the target execution node corresponding to the task to be scheduled from each of the edge nodes according to the remaining capacity.
[0029] It should be noted that the aforementioned target execution node can refer to the edge node ultimately selected by the scheduling device for the task to be scheduled, used to execute the task. In an edge computing system, the various subtasks of a workflow can be assigned to different edge nodes for execution, and the target execution node is the node where the computing resources allocated to the currently scheduled task are located during the scheduling decision process.
[0030] In its implementation, after obtaining the remaining capacity of each edge node and the remaining bandwidth of each physical link, the scheduling device can determine whether each edge node has the capacity to accept tasks to be scheduled based on its current remaining CPU and cache capacity. The scheduling device comprehensively evaluates the remaining capacity of each edge node. In cases where multiple tasks may execute concurrently on the same node, the scheduler needs to determine whether the node's remaining CPU and cache are sufficient to accept new tasks. Specifically, the scheduling device compares the CPU and cache resource requirements of the task to be scheduled with the remaining CPU and cache capacity of each edge node, and selects the edge node that best meets the task's resource requirements and has the optimal overall conditions as the target execution node for the task to be scheduled.
[0031] Step S3: When the target execution node is different from the node where the predecessor task of the task to be scheduled is located, select the target transmission path corresponding to the dependent data of the task to be scheduled from each of the physical links according to the remaining bandwidth.
[0032] It should be noted that the aforementioned predecessor task can refer to a task in the directed acyclic graph structure of the workflow that precedes the currently scheduled task and whose execution result serves as the input dependency of the currently scheduled task. The aforementioned dependency data can refer to intermediate computation results generated after the predecessor task is completed and need to be passed to the successor task as input data. In cross-node transmission scenarios, dependency data needs to be transmitted from the node containing the predecessor task of the currently scheduled task to the target execution node via one or more physical links. The aforementioned target transmission path can refer to the sequence of physical links selected by the scheduling device in the physical link topology of the edge computing system for the data transmission from the node containing the predecessor task to the target execution node. In non-fully connected edge network topologies, there may not be a direct communication link between any two nodes, and cross-node dependency data transmission often requires multiple hops.
[0033] In practical implementation, if the predecessor and successor tasks are located on different nodes, a feasible transmission path can be selected based on the physical link connection relationship and the remaining bandwidth of the links. When the target execution node is different from the node where the predecessor task is located, the scheduling device determines that a cross-node transmission path needs to be selected for the dependent data transmission. At this time, the scheduling device obtains the node where the predecessor task is located as the transmission source node, the target execution node as the transmission destination node, and the data volume information of the dependent data of the task to be scheduled. Subsequently, the scheduling device determines the bandwidth requirement required for the transmission process based on the data volume of the dependent data, and, combined with the remaining bandwidth of each physical link at the current time, selects a transmission path for the cross-node dependent data on a non-fully connected topology. If the predecessor and successor tasks are assigned to the same node, no cross-node transmission time is generated, the scheduling device does not need to perform path selection operation, and directly records the transmission time as zero.
[0034] Step S4: Schedule the task to be scheduled based on the target execution node and the target transmission path. The target execution node and the target transmission path work together to ensure that the energy consumption of the edge computing system meets the preset optimization target.
[0035] It should be noted that the aforementioned preset optimization objective can refer to the optimization direction or standard that the scheduling device expects to achieve when making scheduling decisions for each task in the workflow. In edge computing systems, the preset optimization objective is usually to minimize the total system energy consumption, including the sum of the computing energy consumption generated by all edge nodes executing tasks and the transmission energy consumption generated by cross-node dependent data transmission. By rationally selecting the execution nodes and data transmission paths for each task, the scheduling device enables task execution and dependent data transmission to work together to reduce the total system energy consumption while satisfying resource constraints and dependency constraints.
[0036] In its implementation, the scheduling device integrates task execution and dependent data transmission into a single scheduling framework. Based on the target execution node and target transmission path selected in the preceding steps, it initiates a complete scheduling process for the task to be scheduled, continuously updating the system status during task execution and dependent data transmission. Regarding computing resources, it supports concurrent execution of multiple tasks within the same edge node, allocating CPU resources according to task requirements and explicitly considering task cache usage. In terms of network resources, it supports parallel transmission of multiple dependent data under link bandwidth constraints. Throughout the scheduling process, CPU sharing, cache usage, routing feasibility, link bandwidth usage, computing energy consumption, and transmission energy consumption are all included in the evaluation process. Each decision-making stage is guided by a preset optimization objective: minimizing the sum of computing energy consumption and dependent data transmission energy consumption across all edge nodes. The selection of the target execution node and target transmission path jointly determines the task's computation location and data transmission path, thus jointly affecting the total system energy consumption.
[0037] For example, suppose the system topology is represented as a graph. ,in Represents the set of edge nodes. This represents the set of physical communication links. If there is no direct link between two nodes, cross-node data transmission must be completed via a multi-hop path. Each edge node... With the largest CPU capacity and cache capacity Each link With maximum transmission bandwidth .node The static power consumption is The power consumption coefficient is dynamically calculated as follows: When node When acting as a transmitting node for data transmission, its transmission power is .
[0038] The set of dynamically arriving workflows in the system is denoted as Each workflow From task set The workflow is structured and described using a directed acyclic graph (DAG). The arrival time is The deadline is .Task Described by the following attribute: CPU requirement type Calculation workload caching requirements Minimum CPU requirements Maximum effective CPU requirements Precursor Mission Collection and subsequent task set If the task It is a task The subsequent task is... Indicates from Transmit to The amount of dependent data.
[0039] Task CPU Requirement Type Includes three categories: when When the task is frequency-insensitive, adding more CPUs after the minimum CPU allocation has been reached will not shorten the execution time; when At that time, the task is frequency-sensitive with no upper limit, and the execution speed increases with the increase of CPU allocation; when At that time, the task is threshold frequency sensitive, and CPU allocation exceeds... It will no longer bring any additional benefits.
[0040] Task The execution node is denoted as The task start time and completion time are recorded as follows: and Assigned to task The CPU frequency is denoted as It remains unchanged during task execution. Indicator variable. Indicates task At any moment Is it currently being executed? Indicates time At the node A set of tasks waiting to be executed or currently being executed in the cache. For tasks... To the mission Dependence on data transmission, Indicates link Whether it is selected into the transmission path, This indicates the time of transmission. Bandwidth used.
[0041] Based on the preceding text, the aforementioned preset optimization objective can be to minimize the sum of the computational energy consumption of all edge nodes and the energy consumption of dependent data transmission, as shown in the following expression: .
[0042] In addition, the preset optimization objective may include, but is not limited to, the following constraints.
[0043] Tasks in a workflow cannot begin execution earlier than the arrival time of the workflow to which they belong. ; Tasks in a workflow must be completed before the deadline of their respective workflows: ; Subsequent tasks must wait for the predecessor task to complete and for dependent data to be transferred before they can begin execution. ; Task CPU allocation needs to meet the feasible range corresponding to different CPU demand types: ; The amount of CPU service obtained during task execution should be equal to the amount of computation it performs. ; At any given time, the total CPU usage of all tasks executing on the same node cannot exceed the CPU capacity of that node. ; At any given time, the total cache usage of all allocated tasks in a node's cache cannot exceed the node's cache capacity. ; At any given time, the total bandwidth used by all dependent data transmissions through the same physical link cannot exceed the bandwidth capacity of that link. ; Cross-node dependent data transmission needs to satisfy the flow conservation constraint to form a feasible transmission path from the predecessor task execution node to the successor task execution node: ; If the predecessor and successor tasks are assigned to the same node, no cross-node transmission time is incurred; otherwise, the transmission time is determined by the amount of dependent data and the bottleneck bandwidth on the selected path. ; The CPU utilization of node n at time t is defined as the ratio of the total CPU usage of tasks being executed on that node to the node's CPU capacity: ; node At any moment The computing power consumption consists of static power consumption and dynamic power consumption that varies with CPU utilization. ; Dependence data from task Transmit to task The resulting transmission energy consumption is the sum of the products of the transmission power consumption of each sending node on the selected path and the transmission time: ; The above model (i.e., the preset optimization objective and corresponding constraints) simultaneously characterizes the task execution order, execution node selection, CPU sharing within a node, cache usage, cross-node path selection, link bandwidth usage, transmission completion time, and computation and transmission energy consumption.
[0044] This embodiment selects the target execution node corresponding to the scheduled task based on the remaining capacity of each edge node. This node selection depends on the current available resource status of the edge node, rather than a pre-defined exclusive node allocation. This allows the same edge node to accept multiple tasks within its remaining capacity, reflecting the dynamic occupancy of computing and storage resources during task execution within a node. Simultaneously, it selects the target transmission path corresponding to the dependent data based on the remaining bandwidth of each physical link. This path selection depends on the current available bandwidth of the link, rather than assuming a direct communication link between any two nodes. This allows for the selection of a practically feasible transmission route for cross-node data transmission under physical link constraints. The determined target execution node and target transmission path are then used together to schedule the task. This allows for joint decision-making of resource consumption during task execution and data transmission under the constraint of a pre-defined energy consumption optimization objective. This avoids resource conflicts or inefficient allocation caused by independent computing and communication resource scheduling. Therefore, it provides a computing network collaborative scheduling method that can simultaneously characterize the resource occupancy of task execution within a node and the resource occupancy of data transmission between nodes. This solves the technical problem of existing methods, which, due to the assumption of direct communication between nodes and task exclusive node CPU execution, cannot adapt to dynamic resource sharing and non-fully connected topologies in real edge networks.
[0045] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to the above embodiment can be referred to the above description, and will not be repeated hereafter. Based on this, a second embodiment of the task scheduling method of this application is proposed; please refer to... Figure 2 , Figure 2 This is a flowchart illustrating the second embodiment of the task scheduling method of this application.
[0046] Furthermore, the remaining capacity includes remaining CPU capacity and remaining cache capacity. In order to distinguish the differences between different nodes in terms of comprehensive resource matching degree and communication cost, and thus accurately select the target execution node corresponding to the task to be scheduled from the candidate nodes, step S2 may include: Step S21: Select candidate nodes from each edge node whose remaining CPU capacity is greater than or equal to the minimum CPU requirement of the task to be scheduled, and whose remaining cache capacity is greater than or equal to the cache requirement of the task to be scheduled.
[0047] It should be noted that the aforementioned candidate nodes refer to edge nodes selected by the scheduling device from all edge nodes in the edge computing system, which possess the capability to accept tasks to be scheduled in terms of both CPU and cache resources. The aforementioned minimum CPU requirement refers to the minimum amount of CPU computing resources necessary for the scheduled task to execute normally. The aforementioned cache requirement refers to the size of the edge node cache space required by the scheduled task during execution, used to store input data, intermediate variables, and output data required during task execution.
[0048] It should be understood that for each edge node, the scheduling device compares the node's remaining CPU capacity with the minimum CPU requirement of the task to be scheduled, and simultaneously compares the node's remaining cache capacity with the cache requirement of the task to be scheduled. At any given time, the total CPU usage of all tasks executing on the same node cannot exceed the node's CPU capacity; at any given time, the total cache usage of all allocated tasks in the node's cache cannot exceed the node's cache capacity. Therefore, edge nodes with remaining CPU capacity greater than or equal to the minimum CPU requirement and remaining cache capacity greater than or equal to the cache requirement can be retained as candidate nodes; edge nodes that do not meet either of the above conditions are excluded. After traversing and filtering all edge nodes, a candidate node set is generated. If the candidate node set is empty, it is determined that no edge node can currently accept the task to be scheduled, and a global cache reallocation process needs to be triggered or the task needs to be temporarily stored and scheduled after subsequent resource release.
[0049] Step S22: Based on the remaining CPU capacity, remaining cache capacity, estimated execution time of the task to be scheduled on each candidate node, and estimated data transfer time from the node where the predecessor task of the task to be scheduled is located to each candidate node, select the target execution node corresponding to the task to be scheduled from each candidate node.
[0050] It should be noted that the above communication time estimate may refer to the estimated transmission time required for the predecessor task dependency data required by the task to be scheduled to be assigned to a candidate node for execution, from the nodes where the predecessor tasks are located to the candidate node.
[0051] It should be understood that the scheduling device can evaluate candidate nodes based on their status characteristics, such as node CPU, cache, task execution time estimation, and communication time estimation. For each candidate node in the candidate node set, the scheduling device obtains the remaining CPU capacity and remaining cache capacity of that node as basic indicators for evaluating the sufficiency of computing resources. Simultaneously, a path search process is performed for that candidate node to obtain a communication time estimate: the scheduling device searches for a feasible path in the physical link topology of the edge computing system, using the node containing the predecessor task of the task to be scheduled as the source node and the candidate node as the destination node. Based on the current remaining bandwidth of each physical link along this path, the estimated transmission time of dependent data is calculated. This process is only for estimation and does not actually occupy link bandwidth. Using the remaining CPU capacity, remaining cache capacity, and communication time estimate as three evaluation dimensions, if there are nodes that meet the time feasibility requirements, node selection rules are applied to comprehensively evaluate each candidate node among these nodes. After the evaluation is completed, the scheduling device selects the candidate node with the best overall performance as the target execution node for the task to be scheduled.
[0052] In this implementation, the remaining capacity of edge nodes is divided into remaining CPU capacity and remaining cache capacity. This allows for the verification of whether the computational and storage resource requirements of the scheduled task are simultaneously met when selecting candidate nodes. Then, the target execution node is selected by comprehensively considering the remaining CPU capacity, remaining cache capacity, and estimated communication time between the scheduled task and each candidate node. This ensures that the node selection process is not based on isolated sorting of resource availability, but rather on the current node's computational and storage resource availability, as well as cross-node data transmission overhead due to node location differences, all within the same decision framework. Therefore, even when multiple candidate nodes meet the resource acceptance criteria, the differences in overall resource matching and communication costs among different nodes can be distinguished, thus accurately selecting the target execution node corresponding to the scheduled task from the candidate nodes.
[0053] Furthermore, in order to reduce the total number of link resources occupied by dependent data during cross-node transmission and reduce the cumulative transmission time caused by multi-hop forwarding, step S3 may include: Step S31: Obtain the data volume of the dependent data of the task to be scheduled, and determine the bandwidth requirement based on the data volume of the dependent data.
[0054] It should be noted that the aforementioned dependent data volume can refer to the size of the intermediate computation results generated after the predecessor task required by the task to be scheduled has been completed. The aforementioned bandwidth requirement can refer to the transmission rate that the scheduling device expects to reserve on the transmission path to complete the transmission of dependent data.
[0055] It should be understood that when determining bandwidth requirements, the scheduling device can use the following methods: divide the amount of dependent data by the preset expected transmission time to obtain the required transmission rate, which is the bandwidth requirement; or directly determine the corresponding bandwidth requirement level according to the amount of dependent data and preset mapping rules.
[0056] Step S32: Select available links from each of the physical links whose remaining bandwidth is greater than or equal to the bandwidth requirement, and construct a link subgraph based on the available links.
[0057] It should be noted that the aforementioned available links refer to physical links whose remaining bandwidth at the current moment is greater than or equal to the bandwidth requirements of the data transmission depended on by the task to be scheduled. The aforementioned link subgraph can be a graph structure consisting of the set of available links and their associated nodes remaining after removing links that do not meet the bandwidth conditions from the original physical network topology.
[0058] In the specific implementation, for data transmission from the source node to the destination node with a required transmission rate of r, the simulator of the scheduling device can filter available links based on the remaining bandwidth of the links at the current moment, forming a set of feasible links. Specifically, each physical link in the edge computing system can be traversed, its remaining bandwidth at the current moment can be read, and this remaining bandwidth can be compared with the previously calculated bandwidth requirement. If the remaining bandwidth of a physical link is greater than or equal to the bandwidth requirement, the physical link is marked as a usable link; if the remaining bandwidth is less than the bandwidth requirement, the physical link does not meet the transmission conditions and is excluded from the available links. After traversing and filtering all physical links, a set of all available links is obtained. Subsequently, the scheduling device constructs a link subgraph based on these available links: the two endpoint nodes of each available link are included as vertices in the link subgraph, and the available link is included as an edge connecting these two vertices. If there is currently no path that satisfies the remaining bandwidth constraint, the transmission process waits for the link resources to be released and then re-attempts path selection.
[0059] Step S33: Select the available path with the fewest hops from the node where the predecessor task is located to the target execution node in the link subgraph, and use it as the target transmission path corresponding to the dependent data.
[0060] It should be noted that the hop count mentioned above can refer to the number of physical links traversed by the transmission path. Each physical link traversed from the source node to the destination node is counted as one hop. Minimum hop count means finding the path from the source node to the destination node with the fewest physical links in the link subgraph composed of available links that meet bandwidth constraints. This path corresponds to the shortest transmission path.
[0061] In its implementation, this embodiment determines bandwidth requirements based on the amount of dependent data of the task to be scheduled. This ensures that the bandwidth conditions for path selection are directly related to the traffic characteristics of the data to be transmitted, rather than applying a uniform bandwidth judgment standard to all cross-node transmissions. This avoids the problems of over-reserving bandwidth for small data transmissions or underestimating bandwidth for large data transmissions. Furthermore, the available path with the fewest hops from the node where the predecessor task is located to the target execution node is selected as the target transmission path in the link subgraph. While ensuring that each hop on the path meets bandwidth constraints, the path with the fewest intermediate nodes is prioritized. This reduces the total number of link resources occupied by dependent data during cross-node transmission and lowers the cumulative transmission time caused by multi-hop forwarding.
[0062] Furthermore, to address the difficulty in coordinating the management of the timing of occupancy and release of computing and communication resources during the scheduling process, step S4 may include: Step S41: Determine the CPU allocation amount of the task to be scheduled based on the CPU requirement type of the task to be scheduled, and use the CPU allocation amount as the execution resource allocation of the task to be scheduled on the target execution node.
[0063] It should be noted that the CPU requirement type mentioned above can refer to different categories used to describe the allocation of CPU computing resources for the scheduled task, and the CPU allocation amount mentioned above can refer to the specific CPU frequency value allocated by the scheduling device to the scheduled task on the target execution node.
[0064] Step S42: In the case of cross-node dependent data transmission, update the bandwidth occupancy status of the physical link on the target transmission path to the occupied status, and record the occupancy duration as the transmission time of the dependent data.
[0065] It should be noted that the bandwidth occupancy status mentioned above refers to the identification information indicating whether the transmission bandwidth resources of the physical link are currently occupied by a certain data transmission. The occupancy duration mentioned above refers to the length of time a data transmission takes from start to finish. The transmission time of dependent data mentioned above refers to the actual time required for dependent data to be transmitted from the source node to the destination node along the target transmission path, and its magnitude depends on the amount of dependent data and the bottleneck bandwidth on the target transmission path. In cross-node transmission scenarios, transmission time directly affects the start execution time of subsequent tasks.
[0066] Step S43: After the dependent data transmission is completed, release the bandwidth resources of the physical link on the target transmission path.
[0067] It should be understood that the aforementioned bandwidth resources may refer to the maximum data transmission rate capacity of a physical link per unit time. During dependent data transmission, a portion of the bandwidth resources of each physical link on the target transmission path is occupied by that transmission. Once the transmission is complete, these occupied bandwidth resources need to be released back to the link resource pool for use by other data transmissions.
[0068] Step S44: When the task to be scheduled is accepted by the target execution node, the cache occupancy status of the task to be scheduled is updated to occupied status. After the dependent data transmission is completed and the target execution node is able to allocate the CPU allocation to the task to be scheduled, the task to be scheduled is executed on the target execution node.
[0069] In its implementation, the scheduling device first determines whether the remaining CPU capacity of the target execution node meets the minimum CPU requirements of the task to be scheduled, and simultaneously determines whether the remaining cache capacity of the target execution node meets the cache requirements of the task to be scheduled. If the CPU resources meet the requirements, the task begins execution and occupies the corresponding CPU; if the CPU resources are temporarily insufficient but the node cache still has space, the task enters the local cache of that node and waits for subsequent CPU release. When it is determined that the target execution node can accept the task to be scheduled, the cache occupancy status of the task to be scheduled is updated to occupied, indicating that the task occupies the corresponding amount of cache space in the cache of the target execution node, and this cache occupancy status will continue until the task is completed. Subsequently, the scheduling device continuously monitors the transmission status of dependent data and the preparation status of CPU allocation. When the dependent data transmission completion event is triggered, and the scheduling device has reserved the CPU allocation for the task to be scheduled on the target execution node, it can be determined that the task to be scheduled has all the conditions to start execution. At this time, the task to be scheduled can be submitted to the task execution queue of the target execution node, and the task begins execution and occupies the corresponding CPU.
[0070] Step S45: After the scheduled task is completed, release the CPU resources and cache resources occupied by the scheduled task on the target execution node.
[0071] It should be noted that the CPU resources mentioned above refer to the processing power resources in the target execution node used to execute the task's computational logic. During execution, a scheduled task will occupy a portion of the CPU resources; after execution, these resources need to be released for use by other tasks on the same node. The cache resources mentioned above refer to the storage space resources in the target execution node used to store task-related data, input files, and execution status information. During execution, a scheduled task will occupy a portion of the cache resources; after execution, these resources also need to be released for use by subsequent tasks.
[0072] In its implementation, this embodiment determines the CPU allocation based on the CPU requirement type of the task to be scheduled. After the target transmission path is selected, the bandwidth occupancy status of the physical link on the path is updated to occupied. After transmission is completed, the bandwidth resources are released. This incorporates the transmission of dependent data as a resource occupancy event with a clear start and end time into the scheduling execution process, ensuring that link bandwidth is explicitly reserved during transmission and released promptly after transmission. Simultaneously, after the dependent data transmission is completed and the CPU allocation is ready, the task to be scheduled is executed, and the cache occupancy status is updated to occupied. After execution, the occupied CPU and cache resources are released, ensuring that the task's readiness condition depends on both data transmission completion and CPU resource allocation. Cache resources are also included in the management scope of synchronous occupation with CPU resources. Through the above-described phased occupation and release process, this embodiment manages the occupation and release of computing and cache resources within a node and the occupation and release of link bandwidth resources between nodes in a unified scheduling execution sequence, thereby solving the problem of difficulty in coordinating the occupation and release sequence of computing and communication resources during scheduling.
[0073] Furthermore, to address the problem that existing methods struggle to rationally allocate node CPU resources due to insufficient characterization of the relationship between task execution time and CPU allocation, step S41 may include: Step S411: When the CPU requirement type is frequency insensitive, the CPU allocation of the task to be scheduled is determined as the minimum CPU requirement of the task to be scheduled.
[0074] It should be understood that when the CPU requirement is frequency-insensitive, adding more CPUs after the task's minimum CPU allocation has been reached will not shorten the execution time. For such tasks, allocating additional computing resources beyond the minimum CPU requirement will not improve execution speed, therefore there is no need to allocate more CPU resources to them.
[0075] Step S412: When the CPU demand type is unlimited frequency sensitive, the CPU allocation of the task to be scheduled is determined to be a value that is greater than or equal to the minimum CPU demand and less than or equal to the CPU capacity of the target execution node.
[0076] It should be understood that when the CPU demand type is unlimited frequency sensitive, the scheduling device can read the minimum CPU requirement of the task to be scheduled from the attribute information of the task to be scheduled, and read the CPU capacity of the target execution node from the attribute information of the target execution node. The range of CPU allocation is greater than or equal to the minimum CPU requirement and less than or equal to the CPU capacity of the target execution node. Within this range, the scheduling device can select a specific value as the CPU allocation based on the current system resource status and optimization objectives: for example, selecting the minimum CPU allocation that can meet the task deadline requirement to save resources, or selecting a larger CPU allocation to shorten the task execution time, but the selected value cannot exceed the CPU capacity of the target execution node.
[0077] Step S413: When the CPU demand type is threshold frequency sensitive, the CPU allocation amount of the task to be scheduled is determined to be a value that is greater than or equal to the minimum CPU demand and less than or equal to the maximum effective CPU demand of the task to be scheduled.
[0078] It should be understood that when the CPU demand type is threshold frequency sensitive, allocating CPU beyond the maximum effective CPU demand will no longer bring additional benefits. The aforementioned maximum effective CPU demand may refer to the upper limit of CPU allocation set for threshold frequency sensitive tasks. When the CPU frequency allocated to this task exceeds this upper limit, the execution speed of the task will no longer increase with the increase of CPU allocation, and the excess CPU resources will not shorten the task execution time.
[0079] In its implementation, this embodiment determines the CPU allocation based on the CPU requirement type of the task to be scheduled: when the CPU requirement type is frequency-insensitive, the CPU allocation is set to the minimum CPU requirement to avoid allocating CPU resources beyond the effective utilization range to this type of task, thus avoiding waste of computing resources; when the CPU requirement type is unlimited frequency-sensitive, the CPU allocation is set to a value between the minimum CPU requirement and the CPU capacity of the target execution node, allowing this type of task to obtain more computing resources within the remaining CPU capacity of the node to shorten execution time; when the CPU requirement type is threshold frequency-sensitive, the CPU allocation is set to a value between the minimum CPU requirement and the maximum effective CPU requirement to avoid the CPU allocation exceeding the effective utilization limit of the task and occupying ineffective computing resources. By refining the allocation criteria of computing resources within a node among tasks with different CPU requirement types, the problem of existing methods being unable to reasonably allocate node CPU resources due to insufficient characterization of the relationship between task execution time and CPU allocation is solved.
[0080] Based on the first and / or second embodiments of this application, the content that is the same as or similar to the first and second embodiments described above in the third embodiment of this application can be referred to the above description and will not be repeated hereafter. On this basis, a third embodiment of the task scheduling method of this application is proposed; please refer to... Figure 3 , Figure 3 This is a flowchart illustrating the third embodiment of the task scheduling method of this application.
[0081] Furthermore, to address the issue of balancing decision quality and computational efficiency in online scheduling within dynamic workflow scenarios, the task scheduling method may further include: Step A1: Construct the first group corresponding to the task selection rule and the second group corresponding to the node selection rule.
[0082] It should be noted that the task selection rule mentioned above can refer to a decision rule used to prioritize the ready tasks in the current ready task set, determining which tasks will be scheduled for execution first during the scheduling process. Similarly, the node selection rule mentioned above can refer to a decision rule used to select the best execution node from candidate edge nodes for a chosen task, determining which edge node the task will be assigned to for execution during the scheduling process.
[0083] In its implementation, the scheduling device employs a co-evolutionary genetic programming approach to construct two independent rule subpopulations: a first population for evolving task selection rules and a second population for evolving node selection rules. Individuals in each subpopulation cannot independently complete a workflow scheduling task; they must be combined with a task selection rule and a node selection rule to form a complete scheduler, which is then evaluated in an event-driven simulator. This is because the task scheduling order and execution node selection jointly affect task execution, dependent data transmission, node buffer usage, link bandwidth usage, and subsequent rescheduling behavior. The scheduling device uses a ramped-half-and-half method to initialize the genetically programmed individuals in both subpopulations. Each individual is a tree-like expression composed of a set of terminals and a set of functions. The terminals of the task selection rules primarily describe the task and workflow-related states, such as task CPU requirements, task type, reference execution time, sub-deadline, uplink sorting value, communication time estimate, ready queue size, number of remaining tasks, and workflow weight. The node selection rules primarily describe the node, cache, and communication-related states, such as node CPU capacity, remaining CPU, cache capacity, remaining cache, estimated execution time of tasks in the node cache, static power consumption, communication time, estimated execution time, dynamic power consumption, actual node availability time, and task relaxation time. The function set includes addition, subtraction, multiplication, protected division, maximum and minimum value operations, where protected division returns 1 when the denominator is zero.
[0084] Step A2: Select a first representative individual from the first population and a second representative individual from the second population, and combine the first representative individual and the second representative individual into a complete scheduler.
[0085] It should be noted that the aforementioned first representative individual can refer to a task selection rule individual selected from the first population, used to represent the current optimal or typical rule of that population during the fitness evaluation process of co-evolution. The aforementioned second representative individual can refer to a node selection rule individual selected from the second population, used to represent the current optimal or typical rule of that population during the fitness evaluation process of co-evolution. The aforementioned complete scheduler can refer to a complete scheduling heuristic composed of a task selection rule and a node selection rule. The task selection rule and the node selection rule are responsible for decision-making in two different dimensions during the scheduling process, and their combination forms a complete scheduler capable of independently completing a full scheduling simulation process.
[0086] In its implementation, after initializing the two subpopulations, the scheduling device enters the preparation phase for cooperative evaluation. The scheduling device selects one individual from the first subpopulation as the first representative individual and one individual from the second subpopulation as the second representative individual. During the first generation of evolution, the representative individuals are randomly selected from their respective subpopulations. In subsequent generations, the representative individuals are randomly selected from the elite individuals of their respective subpopulations, where elite individuals are those with superior fitness evaluation results. The scheduling device then combines the first and second representative individuals into a complete scheduler.
[0087] Step A3: Run the full scheduler in the event-driven simulator to perform scheduling simulation on the dynamically arriving workflows and determine the fitness of the full scheduler.
[0088] It should be understood that an event-driven simulator can refer to a simulation execution environment built on a discrete event-driven mechanism. The simulation process does not proceed according to fixed time slices, but is triggered by event types in the system state. Event types include workflow submission events, data transmission completion events, task execution completion events, task completion collection events, local cache reactivation events, global cache reallocation events, and simulation termination events. The aforementioned fitness can refer to a quantitative indicator used to evaluate the scheduling quality of the complete scheduler. The fitness value is calculated based on the total system energy consumption and constraint satisfaction recorded during the scheduling simulation.
[0089] In its implementation, the scheduling device takes the pre-assembled complete scheduler as input, loads it into the event-driven simulator, and submits a set of dynamically arriving workflows to be scheduled as training instances. The simulator starts and runs, executing scheduling according to the processes of dynamic workflow submission, task dispatch, data transmission, task execution, local cache reactivation, global cache reallocation, and subsequent task activation. During simulation, when a task needs to be selected from the ready task set or the global cache, the simulator invokes the task selection rules in the complete scheduler; when an execution node needs to be selected for a chosen task, the simulator invokes the node selection rules in the complete scheduler; when a path needs to be selected for cross-node dependent data transmission, the simulator invokes the minimum hop count path rule under capacity constraints. The simulator continuously updates the system status, including task start time, completion time, node CPU usage, cache usage, link bandwidth usage, transmission path, and workflow completion time. After all workflows are executed, the simulator calculates the total energy consumption of this simulation based on preset optimization objectives and checks whether each workflow meets the deadline constraints. The scheduling device calculates the fitness of the complete scheduler based on the simulation results and uses this fitness value as the fitness record for the individual participating in the evaluation.
[0090] Step A4: Perform evolutionary operations on the first population and the second population according to the fitness until the preset termination condition or the maximum number of generations is reached, and output the optimal task selection rule and the optimal node selection rule.
[0091] It should be noted that the aforementioned evolutionary operations can refer to genetic operators used in genetic programming to generate a new generation of individuals, including four types: elite preservation, replication, crossover, and mutation. The aforementioned preset termination condition can refer to the conditions pre-set by the scheduling device to stop the evolutionary process, such as the fitness value converging to a preset threshold or the number of generations reaching a preset maximum value. The aforementioned maximum number of generations can refer to the maximum number of iterations pre-set by the scheduling device for the evolutionary process.
[0092] In its implementation, after evaluating the fitness of all individuals in the current population, the scheduling device determines whether a preset termination condition has been met or the maximum number of generations has been reached. If the current population has not yet met the termination condition and the number of generations has not reached its maximum, the scheduling device uses a tournament selection method to select parent individuals from the population. Tournament selection is based on individual fitness competition, with individuals having higher fitness having a higher probability of being selected to participate in subsequent genetic operations. Subsequently, the scheduling device uses four types of genetic operations—elite retention, replication, crossover, and mutation—to generate the next generation population. Elite retention directly replicates several individuals with better fitness to the next generation; replication selects individuals through a tournament and directly preserves their tree structure; crossover swaps subtrees between parent individuals, thereby generating new rule expressions; mutation randomly selects a subtree from the parent individuals and replaces it with a new subtree generated according to the initialization syntax. These operations can introduce structural changes while preserving high-quality rules, continuing the search for better task selection and node selection rules.
[0093] Furthermore, the scheduling device repeatedly executes the combined complete scheduler, event-driven simulation evaluation, and genetic operations on the two newly generated subpopulations until a preset termination condition is met or the maximum number of generations is reached. When evolution terminates, the scheduling device outputs the currently obtained optimal task selection rule and optimal node selection rule. In the output, the scheduling device selects the individual with the best fitness from the first population as the optimal task selection rule and the individual with the best fitness from the second population as the optimal node selection rule, using these two rules as the learning results of the offline evolutionary phase.
[0094] Step A5: Determine the task to be scheduled from each ready task according to the optimal task selection rule, and select the target execution node from each edge node according to the optimal node selection rule.
[0095] It should be noted that the aforementioned ready tasks refer to tasks in the workflow execution process where all predecessor tasks have been completed and all dependent data generated by the predecessor tasks have arrived at the current node, satisfying all the prerequisites required for start-up execution. The ready task set is the set of tasks that the scheduling device selects from all incomplete tasks at the scheduling decision point and that can be scheduled for immediate execution.
[0096] It should be understood that the online scheduling phase no longer executes genetic evolution; instead, these two rules are directly invoked at the corresponding decision points in the event-driven scheduling process. After obtaining the optimal task selection rule and the optimal node selection rule during the offline evolution phase, the scheduling device deploys these two rules to the online scheduling environment. When a decision point arises during system operation requiring the selection of tasks from the set of ready tasks for scheduling, the scheduling device obtains all currently ready tasks, uses the attribute characteristics and workflow status characteristics of each ready task as input, substitutes them into the tree expression corresponding to the optimal task selection rule to calculate the priority score, and sorts the ready tasks according to the calculation results, determining the ready task with the highest priority as the current task to be scheduled.
[0097] Subsequently, the scheduling device needs to select an execution node for the task to be scheduled, and obtain the current status characteristics of each edge node, including node CPU capacity, remaining CPU capacity, cache capacity, remaining cache capacity, static power consumption, estimated communication time, and dynamic power consumption. Using the status characteristics of each edge node as input, the device substitutes them into the tree expression corresponding to the optimal node selection rule to calculate the priority score of each node, and sorts the edge nodes according to the calculation results. The scheduling device selects the edge node with the highest priority as the target execution node. Since only the rule tree expression needs to be calculated during the online phase, and path selection is performed using the deterministic capacity-constrained minimum hop count path rule, low decision-making overhead can be maintained even when dynamic workflows are constantly arriving and node resources and link bandwidth are continuously changing.
[0098] In this implementation, a first population corresponding to the task selection rule and a second population corresponding to the node selection rule are constructed. Representative individuals selected from the two populations are combined into a complete scheduler and run in an event-driven simulator to determine fitness. This couples the task selection rule and the node selection rule into a unified whole during the fitness evaluation phase, forcing the optimization of any rule to consider the resource state changes brought about by the other's decision. Based on this, evolutionary operations are performed on the two populations according to fitness until the termination condition is met, at which point the optimal task selection rule and the optimal node selection rule are output. In the online scheduling phase, the output optimal rules are directly used to determine the tasks to be scheduled and the target execution nodes. At this point, only the value of the rule expression needs to be calculated to complete the decision, without re-executing the search or optimization at each scheduling moment. Through the above method, the task ranking rule and the node selection rule can be adapted to each other through a co-evolutionary mechanism in the offline training phase, while retaining the decision characteristics of low computational overhead in the online phase. This solves the technical problem of balancing decision quality and computational efficiency in online scheduling in dynamic workflow scenarios.
[0099] Furthermore, to avoid the task being blocked for an extended period due to a single acceptance failure, step S4 may further include: Step S5: When the target execution node cannot accept the task to be scheduled, the task to be scheduled is stored in the global cache queue; wherein, the inability to accept includes: the remaining cache capacity of the target execution node is less than the cache requirement of the task to be scheduled, or the remaining CPU capacity of the target execution node is less than the minimum CPU requirement of the task to be scheduled and the task to be scheduled cannot enter the local cache queue to wait.
[0100] It should be noted that the aforementioned local cache can refer to a waiting queue set up internally within the edge node, used to temporarily store tasks that have been allocated to that node but cannot start execution immediately due to insufficient CPU resources. The aforementioned global cache queue can refer to a unified waiting queue set up by the scheduling device in the system, used to temporarily store tasks that need to be rescheduled because the target execution node cannot accept them.
[0101] In the specific implementation, if the remaining cache capacity is less than the cache requirement of the task to be scheduled, the scheduling device directly determines that it cannot accept the task. If the remaining cache capacity meets the cache requirement but the remaining CPU capacity is less than the minimum CPU requirement, the scheduling device further checks whether there is any remaining space in the local cache queue of the target execution node to accommodate the task. If the local cache queue is full or the task cannot enter the local cache queue to wait for other reasons, the scheduling device also determines that it cannot accept the task. When it is determined that it cannot be accepted, the task to be scheduled is stored in the global cache queue. The task no longer occupies the target execution node and waits for the subsequent global cache reallocation process to reselect an execution node for it. If the cache requirement of the task can be met and the CPU resources are insufficient, it can enter the local cache queue to wait. The scheduling device determines that the task can be accepted, the task is temporarily stored in the local cache queue, and is reactivated for execution after the CPU resources are released.
[0102] Step S6: Determine the scheduling priority of each cache task in the global cache queue, and reselect the execution node for each cache task in descending order of scheduling priority.
[0103] In the specific implementation, after storing the tasks to be scheduled into the global cache queue, the scheduling device triggers the scheduling process of the global cache queue. The scheduling device first retrieves all currently existing cached tasks in the global cache queue and determines the corresponding scheduling priority for each cached task. Specifically, the tasks in the global cache are reordered according to the task selection rules. The scheduling device inputs the status characteristics of each cached task into the tree expression corresponding to the optimal task selection rule to calculate a priority score. Based on the calculation results, the cached tasks are sorted, with higher scores indicating higher scheduling priority.
[0104] After determining the scheduling priority of each cached task, the scheduling device reselects an execution node for each cached task in descending order of priority. For each selected cached task, the node selection rules reselect candidate execution nodes; during the node evaluation process, the path selection rules are only used to estimate communication time and do not actually occupy link bandwidth. If a task is successfully reassigned, it is removed from the global cache, and the dispatch process is initiated based on the estimated availability time of the selected node; if a cached task cannot be accepted after reselecting an execution node, it is kept in the global cache queue to await subsequent rescheduling opportunities. The scheduling device processes tasks one by one according to priority until the global cache queue is empty or all remaining tasks cannot be reassigned in the current state.
[0105] Step S7: When the scheduled task fails to execute or cannot be accepted by the target execution node, the global cache reallocation process of the workflow is triggered.
[0106] It should be noted that the above-mentioned global cache reallocation process can refer to the system process by which the scheduling device reschedules and allocates a task when the task needs to be rescheduled because the selected node cannot accept it or the execution fails. This process includes clearing the previous node allocation of the task, updating the temporary source position, putting the task into the global cache queue, reordering and reselecting the execution node, etc.
[0107] It should be understood that when a task needs to be rescheduled due to the selected node's inability to accept it or its execution failure, the system clears its previous node allocation, updates its temporary source location, and adds the task body and required file size to the task size calculation before placing it in the global cache. Tasks in the global cache are reordered according to the task selection rules. For each selected cached task, the node selection rules reselect candidate execution nodes; during node evaluation, the path selection rules are only used to estimate communication time and do not actually consume link bandwidth. If the task is successfully reassigned, it is removed from the global cache, and the dispatch process is initiated based on the estimated availability time of the selected node.
[0108] In this implementation, when the target execution node cannot meet the minimum CPU and / or cache requirements of the task to be scheduled, and the task cannot enter the local cache to wait, the task to be scheduled is stored in a global cache queue. This ensures that tasks that cannot be executed due to insufficient resources on the current node are not immediately discarded or wait indefinitely, but are uniformly aggregated in the global cache to await subsequent rescheduling opportunities. When a task to be scheduled fails to execute or cannot be accepted by the target execution node, the global cache reallocation process of the workflow is triggered, thereby avoiding the situation where the task is blocked for a long time due to a single failure to accept it.
[0109] In one feasible implementation, the scheduling device of this application may include a multi-access edge computing system. Please refer to [link / reference needed] for details. Figure 4 , Figure 4 This diagram illustrates the system architecture of a multi-access edge computing system using the task scheduling method described in this application. It showcases the system architecture for dynamic workflow collaborative scheduling within a resource-constrained multi-access edge computing (MEC) environment. The system employs a layered design, consisting of an orchestration management layer and a multi-access edge computing network layer, supporting the collaborative execution of dynamic workflows across heterogeneous edge nodes. In this scenario, dynamic workflows arrive continuously, and different tasks have dependencies based on a directed acyclic graph (DAG). Edge nodes are limited by cache and CPU capacity, and inter-node links are constrained by bandwidth. To address this, the system jointly considers computational and network resources, optimizing the energy consumption of dynamic workflows through task scheduling, node selection, and path planning, with the goal of minimizing total energy consumption (TEC).
[0110] The orchestration management layer, controlled by the MEC controller, is responsible for global workflow management, resource awareness, and scheduling decisions. First, dynamic workflow requests from different users or terminal devices are submitted to the system, where the workflow analysis and request management module performs DAG structure parsing, task dependency analysis, and execution status tracking. Simultaneously, the MEC resource awareness and status acquisition module collects real-time status information from edge nodes and links to assess the availability of node computing resources and link resources. Based on this, the system automatically generates scheduling rules using the scheduler and executes three key decision-making processes: task selection, node selection, and shortest path selection based on capacity constraints. This jointly optimizes task execution locations and data transmission paths, achieving compute-network collaborative scheduling.
[0111] The multi-access edge computing network layer consists of multiple heterogeneous and not fully connected edge nodes. Each node can receive dynamic workflow requests and process them collaboratively through constrained links. Each edge node has limited cache and CPU capacity constraints. Tasks first enter a local cache queue, and then CPU allocation is performed within resource limits, thus supporting the parallel execution of multiple tasks. Meanwhile, data transmission between edge nodes is limited by link bandwidth. Multiple workflow data streams can be transmitted in parallel on the same link, but the overall throughput must not exceed the link capacity limit. The diagram shows nodes... To the node Taking the link as an example, the data transmission process under limited bandwidth is shown in detail, thus reflecting the coupling relationship between computing and communication resources in the dynamic workflow computing network collaborative scheduling process.
[0112] It should be noted that the above examples are for understanding this application only and do not constitute a limitation on the task scheduling method of this application. Any simple modifications based on this technical concept are within the scope of protection of this application. All actions in this application that may involve privacy are performed in compliance with the relevant data protection regulations of the country where the application is located and with authorization from the owner of the corresponding device.
[0113] This application also provides a task scheduling device, please refer to... Figure 5 , Figure 5 This is a structural block diagram of the first embodiment of the task scheduling device of this application.
[0114] like Figure 5 As shown, the task scheduling device proposed in this application includes: The data acquisition module 501 is used to acquire the remaining capacity of each edge node and the remaining bandwidth of each physical link when it is detected that there are tasks to be scheduled in the workflow of the edge computing system. The first selection module 502 is used to select the target execution node corresponding to the task to be scheduled from each of the edge nodes according to the remaining capacity, so that the computing energy consumption of the edge computing system meets the preset optimization target. The second selection module 503 is used to select the target transmission path corresponding to the dependent data of the task to be scheduled from each of the physical links according to the remaining bandwidth when the target execution node is different from the node where the predecessor task of the task to be scheduled is located, so as to make the transmission energy consumption of the edge computing system meet the preset optimization target. The scheduling execution module 504 is used to schedule the task to be scheduled based on the target execution node and the target transmission path.
[0115] This embodiment selects the target execution node corresponding to the scheduled task based on the remaining capacity of each edge node. This node selection depends on the current available resource status of the edge node, rather than a pre-defined exclusive node allocation. This allows the same edge node to accept multiple tasks within its remaining capacity, reflecting the dynamic occupancy of computing and storage resources during task execution within a node. Simultaneously, it selects the target transmission path corresponding to the dependent data based on the remaining bandwidth of each physical link. This path selection depends on the current available bandwidth of the link, rather than assuming a direct communication link between any two nodes. This allows for the selection of a practically feasible transmission route for cross-node data transmission under physical link constraints. The determined target execution node and target transmission path are then used together to schedule the task. This allows for joint decision-making of resource consumption during task execution and data transmission under the constraint of a pre-defined energy consumption optimization objective. This avoids resource conflicts or inefficient allocation caused by independent computing and communication resource scheduling. Therefore, it provides a computing network collaborative scheduling method that can simultaneously characterize the resource occupancy of task execution within a node and the resource occupancy of data transmission between nodes. This solves the technical problem of existing methods, which, due to the assumption of direct communication between nodes and task exclusive node CPU execution, cannot adapt to dynamic resource sharing and non-fully connected topologies in real edge networks.
[0116] Based on the first embodiment of the task scheduling device described in this application, a second embodiment of the task scheduling device of this application is proposed.
[0117] In this embodiment, as one implementation method, the remaining capacity includes the remaining CPU capacity and the remaining cache capacity. The first selection module 502 is further configured to filter candidate nodes from each of the edge nodes whose remaining CPU capacity is greater than or equal to the minimum CPU requirement of the task to be scheduled and whose remaining cache capacity is greater than or equal to the cache requirement of the task to be scheduled; and select the target execution node corresponding to the task to be scheduled from each of the candidate nodes based on the remaining CPU capacity, the remaining cache capacity, the estimated execution time of the task to be scheduled on each of the candidate nodes, and the estimated dependent data transfer time from the node where the predecessor task of the task to be scheduled is located to each of the candidate nodes.
[0118] In one implementation, the second selection module 503 is further configured to obtain the data volume of the dependent data of the task to be scheduled, and determine the bandwidth requirement based on the data volume of the dependent data; filter available links from each of the physical links whose remaining bandwidth is greater than or equal to the bandwidth requirement, and construct a link subgraph based on the available links; select the available path with the fewest hops from the node where the predecessor task is located to the target execution node in the link subgraph as the target transmission path corresponding to the dependent data.
[0119] In one implementation, the scheduling execution module 504 is further configured to: determine the CPU allocation amount of the task to be scheduled based on the CPU requirement type of the task to be scheduled; and allocate the CPU allocation amount as the execution resource allocation for the task to be scheduled on the target execution node; in the case of cross-node dependent data transmission, update the bandwidth occupancy status of the physical link on the target transmission path to an occupied state, and record the occupancy duration as the transmission time of the dependent data; after the dependent data transmission is completed, release the bandwidth resources of the physical link on the target transmission path; when the task to be scheduled is accepted by the target execution node, update the cache occupancy status of the task to be scheduled to an occupied state; after the dependent data transmission is completed and the target execution node is able to allocate the CPU allocation amount to the task to be scheduled, execute the task to be scheduled on the target execution node; and after the task to be scheduled is completed, release the CPU resources and cache resources occupied by the task to be scheduled on the target execution node.
[0120] In one implementation, the scheduling execution module 504 is further configured to: when the CPU demand type is frequency-insensitive, determine the CPU allocation amount of the task to be scheduled as the minimum CPU demand of the task to be scheduled; when the CPU demand type is unlimited frequency-sensitive, determine the CPU allocation amount of the task to be scheduled as a value greater than or equal to the minimum CPU demand and less than or equal to the CPU capacity of the target execution node; and when the CPU demand type is threshold frequency-sensitive, determine the CPU allocation amount of the task to be scheduled as a value greater than or equal to the minimum CPU demand and less than or equal to the maximum effective CPU demand of the task to be scheduled.
[0121] In one implementation, the first selection module 502 is further configured to construct a first population corresponding to the task selection rule and a second population corresponding to the node selection rule; select a first representative individual from the first population and a second representative individual from the second population, and combine the first representative individual and the second representative individual into a complete scheduler; run the complete scheduler in an event-driven simulator to perform scheduling simulation on the dynamically arriving workflow, and determine the fitness of the complete scheduler; perform evolutionary operations on the first population and the second population according to the fitness until a preset termination condition or the maximum number of generations is reached, and output the optimal task selection rule and the optimal node selection rule; determine the task to be scheduled from each ready task according to the optimal task selection rule, and select the target execution node from each edge node according to the optimal node selection rule.
[0122] In one implementation, the scheduling execution module 504 is further configured to store the task to be scheduled in a global cache queue when the target execution node cannot accept the task to be scheduled; wherein, the inability to accept includes: the remaining cache capacity of the target execution node is less than the cache requirement of the task to be scheduled, or the remaining CPU capacity of the target execution node is less than the minimum CPU requirement of the task to be scheduled and the task to be scheduled cannot enter the local cache queue to wait; determine the scheduling priority of each cache task in the global cache queue, and reselect an execution node for each cache task in descending order of scheduling priority; when the task to be scheduled fails to execute or cannot be accepted by the target execution node, trigger the global cache reallocation process of the workflow.
[0123] The task scheduling device provided in this application, employing the task scheduling method described in the above embodiments, can solve the technical problem of how to provide a computing network collaborative scheduling method that can simultaneously characterize the resource consumption of task execution within a node and the resource consumption of data transmission dependencies between nodes in a dynamic workflow execution scenario. Compared with the prior art, the beneficial effects of the task scheduling device provided in this application are the same as those of the task scheduling method provided in the above embodiments, and other technical features in the task scheduling device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.
[0124] This application provides a task scheduling device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the task scheduling method in the first embodiment described above.
[0125] The following reference Figure 6 The diagram illustrates a structural schematic of a task scheduling device suitable for implementing embodiments of this application. The task scheduling device in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 6 The task scheduling device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.
[0126] like Figure 6As shown, the task scheduling device may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in the read-only memory 1002 or a program loaded from the storage device 1003 into the random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the task scheduling device. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, a touch screen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. Communication device 1009 allows the task scheduling device to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows task scheduling devices with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.
[0127] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.
[0128] The task scheduling device provided in this application, employing the task scheduling method described in the above embodiments, can solve the technical problem of how to provide a computing network collaborative scheduling method that can simultaneously characterize the resource consumption of task execution within a node and the resource consumption of data transmission dependencies between nodes in a dynamic workflow execution scenario. Compared with the prior art, the beneficial effects of the task scheduling device provided in this application are the same as those of the task scheduling method provided in the above embodiments, and other technical features of this task scheduling device are the same as those disclosed in the previous embodiment method, and will not be repeated here.
[0129] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.
[0130] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
[0131] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the task scheduling method described in the above embodiments.
[0132] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.
[0133] The aforementioned computer-readable storage medium may be included in the task scheduling device; or it may exist independently and not be assembled into the task scheduling device.
[0134] The aforementioned computer-readable storage medium carries one or more programs that, when executed by a task scheduling device, enable the task scheduling device to write computer program code for performing the operations of this application in one or more programming languages or a combination thereof. These programming languages include object-oriented programming languages such as Java, Smalltalk, and C++; and also conventional procedural programming languages such as C or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, such as a local area network (LAN) or a wide area network (WAN), or connected to an external computer (e.g., via the Internet using an Internet service provider).
[0135] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.
[0136] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.
[0137] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described task scheduling method. This solves the technical problem of how to provide a computer network collaborative scheduling method that can simultaneously characterize the resource consumption of task execution within a node and the resource consumption of data transmission dependencies between nodes in a dynamic workflow execution scenario. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the task scheduling method provided in the above embodiments, and will not be repeated here.
[0138] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the task scheduling method described above.
[0139] The computer program product provided in this application solves the technical problem of how to provide a computer network collaborative scheduling method that can simultaneously characterize the resource consumption of task execution within a node and the resource consumption of data transmission dependencies between nodes in a dynamic workflow execution scenario. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the task scheduling method provided in the above embodiments, and will not be repeated here.
[0140] The above description is only a part of the embodiments of this application and does not limit the scope of protection of this application. All equivalent structural transformations made under the technical concept of this application and using the content of this application specification and drawings, or direct / indirect applications in other related technical fields, are included in the scope of protection of this application.
Claims
1. A task scheduling method, characterized by, The method is applied to an edge computing system, which includes multiple edge nodes and physical links connecting the edge nodes. The method includes the following steps: If it is detected that there are tasks to be scheduled in the workflow of the edge computing system, the remaining capacity of each edge node and the remaining bandwidth of each physical link are obtained; Based on the remaining capacity, the target execution node corresponding to the task to be scheduled is selected from each of the edge nodes; When the target execution node is different from the node where the predecessor task of the task to be scheduled is located, the target transmission path corresponding to the dependent data of the task to be scheduled is selected from each of the physical links according to the remaining bandwidth; The task to be scheduled is scheduled based on the target execution node and the target transmission path. The target execution node and the target transmission path work together to ensure that the energy consumption of the edge computing system meets the preset optimization target.
2. The task scheduling method as described in claim 1, characterized in that, The remaining capacity includes the remaining CPU capacity and the remaining cache capacity; The step of selecting the target execution node corresponding to the task to be scheduled from each of the edge nodes based on the remaining capacity includes: From each of the edge nodes, candidate nodes are selected whose remaining CPU capacity is greater than or equal to the minimum CPU requirement of the task to be scheduled, and whose remaining cache capacity is greater than or equal to the cache requirement of the task to be scheduled. Based on the remaining CPU capacity, remaining cache capacity, estimated execution time of the task to be scheduled on each candidate node, and estimated data transfer time from the node containing the predecessor task of the task to be scheduled to each candidate node, the target execution node corresponding to the task to be scheduled is selected from the candidate nodes.
3. The task scheduling method of claim 1, wherein, The step of selecting the target transmission path corresponding to the dependent data of the task to be scheduled from each of the physical links based on the remaining bandwidth includes: Obtain the data volume of the dependent data of the task to be scheduled, and determine the bandwidth requirement based on the data volume of the dependent data; From each of the physical links, select available links whose remaining bandwidth is greater than or equal to the bandwidth requirement, and construct a link subgraph based on the available links; In the link subgraph, the available path with the fewest hops from the node where the predecessor task is located to the target execution node is selected as the target transmission path corresponding to the dependent data.
4. The task scheduling method of claim 1, wherein, The step of scheduling the task to be scheduled based on the target execution node and the target transmission path includes: The CPU allocation amount for the scheduled task is determined based on the CPU requirement type of the scheduled task, and the CPU allocation amount is used as the execution resource allocation for the scheduled task on the target execution node. In the case of cross-node dependent data transmission, the bandwidth occupancy status of the physical link on the target transmission path is updated to occupied status, and the occupancy duration is recorded as the transmission time of the dependent data. After the dependent data transmission is completed, the bandwidth resources of the physical link on the target transmission path are released; When the task to be scheduled is accepted by the target execution node, the cache occupancy status of the task to be scheduled is updated to occupied status. After the dependent data transmission is completed and the target execution node is able to allocate the CPU allocation to the task to be scheduled, the task to be scheduled is executed on the target execution node. After the scheduled task is completed, the CPU and cache resources occupied by the scheduled task on the target execution node are released.
5. The task scheduling method of claim 4, wherein, The step of determining the CPU allocation amount for the task to be scheduled based on its CPU requirement type includes: When the CPU requirement type is frequency insensitive, the CPU allocation of the task to be scheduled is determined as the minimum CPU requirement of the task to be scheduled. When the CPU demand type is unlimited frequency sensitive, the CPU allocation of the task to be scheduled is determined to be a value that is greater than or equal to the minimum CPU demand and less than or equal to the CPU capacity of the target execution node. When the CPU demand type is threshold frequency sensitive, the CPU allocation amount of the task to be scheduled is determined to be a value that is greater than or equal to the minimum CPU demand and less than or equal to the maximum effective CPU demand of the task to be scheduled.
6. The task scheduling method as described in claim 1, characterized in that, The method further includes: Construct the first group corresponding to the task selection rule and the second group corresponding to the node selection rule; Select a first representative individual from the first population and a second representative individual from the second population, and combine the first representative individual and the second representative individual into a complete scheduler; Run the full scheduler in an event-driven simulator to perform scheduling simulation on the dynamically arriving workflows and determine the fitness of the full scheduler; Evolutionary operations are performed on the first population and the second population based on the fitness until a preset termination condition or the maximum number of generations is reached, at which point the optimal task selection rule and the optimal node selection rule are output. The task to be scheduled is determined from each ready task according to the optimal task selection rule, and the target execution node is selected from each edge node according to the optimal node selection rule.
7. The task scheduling method of claim 1, wherein, After the step of scheduling the task to be scheduled based on the target execution node and the target transmission path, the method further includes: When the target execution node cannot accept the task to be scheduled, the task to be scheduled is stored in the global cache queue; wherein, the inability to accept includes: the remaining cache capacity of the target execution node is less than the cache requirement of the task to be scheduled, or the remaining CPU capacity of the target execution node is less than the minimum CPU requirement of the task to be scheduled and the task to be scheduled cannot enter the local cache queue to wait. Determine the scheduling priority of each cache task in the global cache queue, and reselect execution nodes for each cache task in descending order of scheduling priority; When the scheduled task fails to execute or cannot be accepted by the target execution node, the global cache reallocation process of the workflow is triggered.
8. A task scheduling apparatus characterized by comprising: The task scheduling device includes: The data acquisition module is used to acquire the remaining capacity of each edge node and the remaining bandwidth of each physical link when the workflow of the edge computing system detects that there are tasks to be scheduled. The first selection module is used to select the target execution node corresponding to the task to be scheduled from each of the edge nodes according to the remaining capacity, so that the computing energy consumption of the edge computing system meets the preset optimization target. The second selection module is used to select the target transmission path corresponding to the dependent data of the task to be scheduled from each of the physical links according to the remaining bandwidth when the target execution node is different from the node where the predecessor task of the task to be scheduled is located, so that the transmission energy consumption of the edge computing system meets the preset optimization target. The scheduling and execution module is used to schedule the task to be scheduled based on the target execution node and the target transmission path.
9. A task scheduling device, characterized by The device includes: a memory, a processor, and a task scheduler stored in the memory and executable on the processor, the task scheduler being configured to implement the steps of the task scheduling method as described in any one of claims 1 to 7.
10. A storage medium, characterized by The storage medium is a computer-readable storage medium, and a task scheduler is stored on the storage medium. When the task scheduler is executed by a processor, it implements the steps of the task scheduling method as described in any one of claims 1 to 7.
