Edge computing adaptive network dynamic scheduling method

By transforming edge computing network scheduling problems into geometric problems and applying dynamic principles, the problem of inefficient scheduling in traditional methods is solved, and efficient and intelligent task management and resource allocation are achieved, which is suitable for large-scale edge computing networks.

CN119011503APending Publication Date: 2024-11-22TSINGHUA SHENZHEN INTERNATIONAL GRADUATE SCHOOL
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411125689.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-16
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The traditional edge computing network scheduling methods have problems such as algorithm complexity, slow convergence speed, inability to find the optimal solution within a limited time, and insufficient adaptability, making it difficult to efficiently manage and schedule tasks of large-scale edge computing nodes.

Method used

Adaptive network dynamic scheduling method is adopted to transform task scheduling problems into geometric problems, and build a model using dynamic principles. Through the area and mass of task blocks, the dynamic allocation of task requests between edge nodes is set, and constraints are set for simulation and prediction. , realize dynamic network scheduling.

Benefits of technology

It significantly improves the efficiency and performance of task scheduling, can quickly adapt to different network application scenarios, reduces energy consumption and costs, and realizes optimized resource allocation, which is especially suitable for time-sensitive tasks.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119011503A_ABST
    Figure CN119011503A_ABST
Patent Text Reader

Abstract

An edge computing adaptive network dynamic scheduling method comprises the following steps: S1, collecting end equipment information, and determining an edge node arrangement scheme; s2, determining the jurisdiction range of each edge node, mapping the jurisdiction range to a virtual scheduling layer in equal proportion, and converting a dynamic scheduling optimization problem into a geometric problem; s3, mapping the task request into a task block with quality and area; and S4, setting constraint conditions, and establishing a dynamic model to simulate and predict edge nodes and task requests. And S5, according to a simulation result, determining a network dynamic scheduling scheme to perform edge computing network dynamic scheduling. According to the method, the dynamics principle is applied to the field of edge calculation and forward dynamics simulation thinking solution for the first time, repeated iterative optimization or long-time training is not needed, complexity explosion and calculation duration sudden increase caused by increase of the number of end-side nodes are effectively solved, the limitation of the applicable scale is broken through, adaptability is extremely high, and the method is particularly suitable for time-sensitive tasks and has good application prospects. And a scheduling scheme can be quickly determined.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of edge computing and network scheduling, and in particular to an edge computing adaptive network dynamic scheduling method. Background Art

[0002] With the rapid development of information technology and the Internet of Things, the number of IoT devices is growing exponentially. Traditional cloud computing architectures have gradually revealed drawbacks such as high data transmission latency and high network bandwidth pressure. Edge computing pushes computing and data processing capabilities to the edge of the network closer to users, achieving faster and more real-time data processing and response, making it possible to meet large-scale network environments and complex task scheduling.

[0003] However, with the expansion of edge computing networks and the diversification of application scenarios, how to efficiently manage and schedule tasks on large-scale edge computing nodes has become a key challenge. Traditional methods, such as machine learning, convex optimization, and heuristic algorithms, are widely used, but they often have problems such as complex algorithms, slow convergence or inability to converge, inability to find the optimal solution or near-optimal solution that meets specific optimization goals within a limited time, insufficient adaptability, and low efficiency in task scheduling.

[0004] It should be noted that the information disclosed in the above background technology section is only used for understanding the background of the present application, and therefore may include information that does not constitute prior art known to ordinary technicians in the field. Summary of the invention

[0005] The main purpose of the present invention is to overcome the defects of the above-mentioned background technology and provide an edge computing adaptive network dynamic scheduling method.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] An edge computing adaptive network dynamic scheduling method comprises the following steps:

[0008] S1: Collect terminal device information and determine the edge node layout plan;

[0009] S2: Determine the jurisdiction of each edge node and map it proportionally to the virtual scheduling layer, transforming the dynamic scheduling optimization problem into a geometric problem;

[0010] S3: Map the task request into a task block with mass and area. The area of ​​the task block represents the required amount of resources. The mass of the task block is used to enable the task block to exhibit motion behavior in the dynamic model, thereby simulating the dynamic allocation of task requests among edge nodes.

[0011] S4: Set constraints and build a dynamic model to simulate and predict edge nodes and task requests.

[0012] S5: Based on the simulation results, determine the network dynamic scheduling scheme to perform dynamic scheduling of the edge computing network.

[0013] In some optional embodiments, in step S2, the virtual scheduling layer G includes a computation-dynamics layer G c , Communication-Dynamics Layer G m and storage-dynamic layer G s , corresponding to the optimal scheduling of computing resources, communication resources and storage resources, respectively. Indicates that the resources of edge nodes include computing resources, communication resources, storage resources, etc. Indicates that at time t, the position P of the edge node j =(x j (t),y j (t)),

[0014] In some optional implementations, on the edge layer, the resource mapping maps the resources of the edge node to a specific area on the dynamic layer, and the area of ​​the area reflects the amount of resources corresponding to the edge node, wherein the two-dimensional plane graph of the edge node mapped on the computing-dynamic layer is G cj , and for the computational-dynamics layer G c Any two different edge nodes in the mapping area G cj and G cj′ , the following conditions are met:

[0015] (1)A j (G cj )=r cj α c , that is, the mapping area G cj Area A j (G cj ) is equal to the computing resource amount r of edge node j cj Multiply by α c , α c It is a coefficient used to convert the computing resources of edge nodes into the corresponding geometric area;

[0016] (2) That is, the mapping area G of any two edge nodes cj and G cj′ No overlap;

[0017] (3)P j ∈G cj , that is, the position P of edge node j at time t jLocated in its computational-dynamic layer G c The mapped area G cj Inside

[0018] Among them, A j (G cj ) reflects the amount of computing resources of the edge node, α c The conversion factor representing the actual computing resources and the corresponding dynamic region area.

[0019] In some optional embodiments, in step S2, the jurisdiction of each edge node is mapped into a series of two-dimensional plane graphics with different areas and connected but non-overlapping edges. The area of ​​the graphics represents the computing power of the edge node; the boundaries between the graphics represent the connections for task transfer and offloading, and the boundaries of the plane represent the upper limit of the computing power of the edge node. The mapping areas of any two edge nodes do not overlap and can only be adjacent.

[0020] In some optional implementations, in step S3, the task request is doubly modeled as a two-dimensional plane figure with area and a three-dimensional geometric object with mass and volume.

[0021] In some optional implementations, in step S3, the resource requirement s of the task request is determined ci , based on the conversion factor α between the computing resources of the edge node and the area of ​​its corresponding dynamic region c and resource demand ci Calculate the radius of the task block: Generate position Q with task i =(x i (t),y i (t)) is the center of the circle and the radius is r ci A circular task block with a certain area is drawn on a two-dimensional plane, and the area of ​​the task block represents the amount of required resources; and corresponding to the two-dimensional circular task block, it is modeled as a three-dimensional ball with a certain volume and mass in three-dimensional space, and the volume and mass of the ball are associated with the area of ​​the two-dimensional circular task block and the resource allocation strategy of the edge node.

[0022] In some optional implementations, in step S3, the resource requirement of the task request is determined, and the conversion coefficient α between the resource requirement and the computing resource of the edge node and the area of ​​the corresponding dynamic region is calculated. c Calculate the area of ​​the task block; generate the position Q based on the task i =(x i (t),y i(t)) is the centroid, and the task is mapped into an irregular two-dimensional plane shape with a certain area and a changeable graphic boundary; and corresponding to the two-dimensional irregular shape task block, it is modeled in three-dimensional space as a fluid geometry with a certain volume and mass, and the volume and mass of the fluid geometry are associated with the area of ​​the two-dimensional irregular shape task block and the resource allocation strategy of the edge node.

[0023] In some optional embodiments, in step S4, the constraints include edge node computing resource constraints, edge node bandwidth constraints, edge node storage constraints, and one or more constraints set for priority processing of time-sensitive tasks, offloading tasks beyond the coverage of edge nodes to cloud processing, node failure processing, and automatic shutdown of some edge nodes when there are fewer task requests.

[0024] In some optional implementations, in step S4, one or more of the following processing strategies are adopted:

[0025] Implement the task processing strategy nearby. For each task request, first determine whether the edge node in the area where it is generated has sufficient resources to process the task, that is, verify whether the available resources of the edge node are greater than the resources required by the task; if the resources are sufficient, no external force is applied to keep the task processed at the current node; if the resources are insufficient, an initial force is applied to move the task to the jurisdiction of other edge nodes;

[0026] The dynamic model is used to simulate the dynamic allocation process of task requests by calculating the gravitational force between the edge node and the ball representing the task request. The gravitational formula of the ball is: The net force on the ball Where ζ is the gravity adjustment coefficient according to different types of tasks, and are the edge node quality and task request quality, r (j) is the distance from the location where the task request is generated to the edge node; the ball will move in the direction of the resultant force;

[0027] For time-sensitive tasks, a larger force Fmax and a larger mass m are applied to the ball. max , put aside other tasks that do not need to be processed in priority and match them to edge nodes for processing more quickly;

[0028] For time-sensitive tasks, priority processing is achieved by increasing the gravity adjustment coefficient and task quality;

[0029] When all edge node resources are occupied, the task requirements beyond the coverage range are offloaded to the cloud for processing;

[0030] When an edge node fails, the mass of the failed node is adjusted to zero, and the task request is redistributed to the adjacent healthy node. The way to achieve this in the simulation is: provide a force pointing to other edge nodes for the ball in the failed edge node area, generate acceleration for the corresponding ball, and make the ball move to the adjacent edge node area;

[0031] When the edge node occupancy rate is lower than the preset threshold, some edge nodes are automatically selected to be shut down to reduce energy consumption and save costs. The quality of the shut-down nodes is set to zero to guide task requests to be allocated to open nodes.

[0032] In some optional implementations, in step S5, one or more of the following scheduling strategies are adopted:

[0033] Define the total momentum of the system p as a measure of the network state, which represents the dynamic sum of all task requests and interactions between edge nodes; set the threshold of the total momentum of the system p θ , as the basis for judging whether the scheduling plan needs to be updated; during the simulation process, when the total momentum of the system p(t) is less than the threshold p θ When , the network scheduling scheme at time t is updated;

[0034] The dynamic model is used to simulate the interaction between task requests and edge nodes, including attraction, repulsion and resistance, to simulate the dynamic allocation of task requests;

[0035] For newly generated task requests or those that cannot be processed at the current node, the dynamic model is used to predict their impact on the existing task allocation and reallocate them to edge nodes with sufficient resources.

[0036] The task offloading strategy is considered in the simulation to ensure that the task request can only be completely assigned to one edge node and avoid cross-node task allocation.

[0037] A computer program product includes a computer program, and when the computer program is executed by a processor, the edge computing adaptive network dynamic scheduling method is implemented.

[0038] The present invention has the following beneficial effects:

[0039] The edge computing adaptive network dynamic scheduling method of the present invention solves the challenges faced by traditional methods in large-scale network environments, such as algorithm complexity, slow convergence speed, and difficulty in finding the optimal solution within a limited time, by innovatively converting the task scheduling problem into a geometric problem and applying the principle of dynamics to simulate it. This method has significant technical advantages, including:

[0040] -Strong adaptability, able to cope with task management and scheduling needs in daily and emergency situations;

[0041] -For the first time, the principle of dynamics is applied to the field of edge computing, and forward dynamics simulation thinking is used to solve the problem, avoiding multiple iterations and long-term training in traditional inverse optimization problems, and effectively solving the complexity explosion problem caused by the increase in the number of edge nodes;

[0042] - Able to make autonomous decisions to shut down some edge nodes, thereby reducing energy consumption and costs and achieving optimal resource allocation;

[0043] - Particularly suitable for time-sensitive tasks, it ensures the rapidity of solution and can provide scheduling solutions in real time, significantly improving the efficiency and performance of task scheduling.

[0044] In summary, the present invention provides an efficient, intelligent and adaptable edge computing network dynamic scheduling solution, which provides an innovative optimization approach for task management and resource allocation in large-scale network environments.

[0045] Other beneficial effects of the embodiments of the present invention will be further described below. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 is a flow chart of an edge computing adaptive network dynamic scheduling method according to an embodiment of the present invention;

[0047] Figure 2 is a schematic diagram of an edge node scheduling layer model according to an embodiment of the present invention;

[0048] Figure 3 is a schematic diagram of a task application scheduling layer model according to an embodiment of the present invention;

[0049] Figure 4 This is a diagram demonstrating the effect of the edge computing adaptive network dynamic scheduling method according to an embodiment of the present invention.

[0050] The marks in the figure are as follows:

[0051] 1 edge node, 2-5 different task request mappings. DETAILED DESCRIPTION

[0053] The following is a detailed description of the embodiments of the present invention. It should be emphasized that the following description is only exemplary and is not intended to limit the scope and application of the present invention.

[0054] An edge computing adaptive network dynamic scheduling method can be used to solve but not limited to problems in task scheduling, dynamic networking, resource allocation, etc. The technical key points of this method include: a) mapping edge nodes and task requests to a virtual scheduling layer respectively, and converting the dynamic scheduling optimization problem into a geometric problem. b) using the principle of dynamics, establishing a dynamic model to simulate edge nodes and task requests, and determining the network dynamic scheduling scheme based on the subordinate range of task nodes.

[0055] See also Figure 1 , an embodiment of the present invention provides an edge computing adaptive network dynamic scheduling method, comprising the following steps:

[0056] S1: Collect terminal device information and determine the edge node layout plan;

[0057] S2: Determine the jurisdiction of each edge node and map it proportionally to the virtual scheduling layer, transforming the dynamic scheduling optimization problem into a geometric problem;

[0058] S3: Map the task request into a task block with mass and area. The area of ​​the task block represents the required amount of resources. The mass of the task block is used to enable the task block to exhibit motion behavior in the dynamic model, thereby simulating the dynamic allocation of task requests among edge nodes.

[0059] S4: Set constraints and build a dynamic model to simulate and predict edge nodes and task requests.

[0060] S5: Based on the simulation results, determine the network dynamic scheduling scheme to perform dynamic scheduling of the edge computing network.

[0061] The edge computing adaptive network dynamic scheduling method of the present invention innovatively introduces the kinematic principle into edge computing simulation solution, is suitable for adaptive network dynamic scheduling, can quickly adapt to and optimize different network application scenarios, has a fast solution speed, does not require multiple iterative optimizations or long-term training, and effectively solves the complexity explosion and computing time surge caused by the increase in the number of edge nodes, and realizes fast and intelligent task scheduling, which is especially suitable for time-sensitive tasks.

[0062] In some embodiments, a method for scheduling edge computing tasks based on the principle of dynamics mainly considers the principle of spatial proximity to select edge servers, referring to Figures 1 to 4 , the method specifically comprises the following steps:

[0063] S1: According to the edge node deployment plan, determine the edge node computing power, storage capacity, network bandwidth and other information.

[0064] S2: Determine the jurisdiction of each edge node and map it proportionally to the virtual scheduling layer, converting the dynamic scheduling optimization problem into a geometric problem.

[0065] G represents the resource scheduling layer, which may include the computation-dynamics layer G c , Communication-Dynamics Layer G m and storage-dynamic layer G s , corresponding to the optimal scheduling of computing resources, communication resources, and storage resources. Indicates that the resources of edge nodes include computing resources, communication resources, storage resources, etc. At time t, the position of the edge node is represented by P j =(x j (t),y j (t)).

[0066] Specifically, one representative implementation method is to map the jurisdiction of each edge node into a series of two-dimensional plane graphs with different areas and connected but non-overlapping edges. The area of ​​the graph represents the computing power of the edge node; the boundaries between the graphs represent the connection of task transfer and unloading, and the boundaries of the plane represent the upper limit of the computing power of the edge node. The mapping areas of any two edge nodes do not overlap, but can only be adjacent.

[0067] S3: Receive a task request and map the task request to a graphic task block with a certain mass and a certain area in a two-dimensional plane. The task size represents the amount of resources required for the task request.

[0068] Various IoT devices will generate tasks, and tasks are used Indicates that tasks will have corresponding resource requirements, and the resources they occupy are The edge node is only one hop away from the terminal device, so the transmission delay is low. Therefore, when the edge node resources are sufficient, the tasks generated by the terminal device are preferentially assigned to the nearby edge nodes for processing to further reduce the transmission delay.

[0069] At time t, the task generation location is denoted by Q i =(x i (t),y i (t)). The edge nodes are mapped to the two-dimensional plane graph range of the virtual scheduling layer, represented by G j If Q i ∈G j , then task i is processed by edge node j.

[0070] S4: Set constraints and build a dynamic model to simulate and predict edge nodes and task requests. Mainly consider the principle of spatial proximity. For time-sensitive tasks, decide whether to prioritize them based on the task type variable. Select the force, which can be gravitational repulsion or resistance. If the node is damaged, the task that the edge node is responsible for will be handed over to the neighboring node. When there are fewer task requests, some nodes will be automatically shut down. When the edge node resources are insufficient, the task needs to be offloaded to the cloud for processing.

[0071] The constraints set in S4 include but are not limited to edge node computing resource constraints; edge node bandwidth constraints; and edge node storage constraints.

[0072] The selection of some scheduling strategies and simulation of emergencies can also be achieved by setting constraints, including but not limited to: giving priority to time-sensitive tasks; offloading tasks beyond the coverage of edge nodes to the cloud for processing; node failure handling; and automatically shutting down some edge nodes when there are fewer task requests.

[0073] Furthermore, a dynamic model is established to simulate edge nodes and task requests to achieve intelligent dynamic scheduling.

[0074] S5: Determine a network dynamic scheduling solution based on the simulation.

[0075] The dynamic scheduling scheme of the edge computing adaptive network is determined through the simulation results of the dynamic model.

[0076] Specific embodiments of the present invention are further described below.

[0077] refer to Figures 1 to 3 , the present application provides an edge computing adaptive network dynamic scheduling method based on the principle of dynamics, comprising the following steps:

[0078] S1: According to the edge node deployment plan, determine the edge node computing power, storage capacity, network bandwidth and other information.

[0079] Collect end device information, including device type, device model, and geographic location, and determine the edge node layout plan, including the edge node setting geographic location, resource type, and resource size, specifically including computing resources, communication resources, and storage resources.

[0080] S2: Determine the jurisdiction of each edge node and map it proportionally to the virtual scheduling layer, converting the dynamic scheduling optimization problem into a geometric problem.

[0081] G represents the resource scheduling layer, which may include the computation-dynamics layer G c , Communication-Dynamics Layer G m and storage-dynamic layer G s, corresponding to the optimal scheduling of computing resources, communication resources, and storage resources. Indicates that the resources of edge nodes include computing resources, communication resources, storage resources, etc. At time t, the position of the edge node is represented by P j =(x j (t),y j (t)).

[0082] Specifically, one representative implementation method is to map the jurisdiction of each edge node into a series of two-dimensional plane graphs with different areas and connected but non-overlapping edges. The area of ​​the graph represents the computing power of the edge node; the boundaries between the graphs represent the connection of task transfer and unloading, and the boundaries of the plane represent the upper limit of the computing power of the edge node. The mapping areas of any two edge nodes do not overlap, but can only be adjacent.

[0083] This step is illustrated and described by taking the example that communication resources and storage resources are sufficient and only computing resources need to be considered.

[0084] Its resource scheduling layer should be the computation-dynamic layer G c , the two-dimensional plane graph of the edge nodes mapped on the computational-dynamics layer is G cj , then for any G cj and G cj′ i c , the following conditions should be met:

[0085] (1)A j (G cj )=r cj α c

[0086] (2)

[0087] (3)P j ∈G cj

[0088] On the edge layer, resource mapping maps a certain resource of the edge node to a specific area on the dynamic layer, and the area of ​​the area reflects the amount of resources corresponding to the edge node, such as A j (G cj ) reflects the amount of computing resources of the edge node, α c The conversion factor representing the actual computing resources and the corresponding dynamic region area.

[0089] S3: Receive a task request and map the task request to a graphic task block with a certain mass and a certain area in a two-dimensional plane. The size of the task block represents the amount of resources required for the task request.

[0090] Various IoT devices will generate tasks, and tasks are used Indicates that tasks will have corresponding resource requirements, and the resources they occupy are The edge node is only one hop away from the terminal device, so the transmission delay is low. Therefore, when the edge node resources are sufficient, the tasks generated by the terminal device are preferentially assigned to the nearby edge nodes for processing to further reduce the transmission delay.

[0091] At time t, the task generation location is denoted by Q i =(x i (t),y i (t)). The edge nodes are mapped to the two-dimensional plane graph range of the virtual scheduling layer, represented by G j If Q i ∈G j , then task i is processed by edge node j.

[0092] The task blocks are modeled as three-dimensional geometric objects with different volumes and two-dimensional plane figures with different areas.

[0093] Further, there are two modeling options:

[0094] Option 1: Q i =(x i (t),y i (t)) is the center of the circle, mapping the task into a circle with a certain area in the two-dimensional plane, and a small ball with a certain volume and mass in the three-dimensional plane.

[0095] Option 2: Q i =(x i (t),y i (t)) is the centroid, mapping the task into an irregular shape with a certain area and variable graphic boundaries. The three-dimensional corresponding fluid has a certain volume and mass.

[0096] Assume that communication resources and storage resources are sufficient and only computing resources need to be considered, and take Solution 1 as an example to illustrate and describe this step.

[0097] The size of the task block is determined by the conversion factor α between the computing resource and the area of ​​the corresponding dynamic region. c and resource demand ci Decision. The task block is based on r i A circle with a radius of r i is a small ball with a radius of Since the size of a task block is determined by its resource requirements, its radius can be determined by the following formula:

[0098]

[0099] Figure 1-4 Taking Solution 1 as an example, different task blocks are modeled as circles with different areas (small balls with different volumes).

[0100] S4: Set constraints and build a dynamic model to simulate and predict edge nodes and task requests.

[0101] The constraints set in S4 include but are not limited to: edge node computing resource constraints; edge node bandwidth constraints; edge node storage constraints.

[0102] The selection of some scheduling strategies and simulation of emergencies can also be achieved by setting constraints, including but not limited to: giving priority to time-sensitive tasks; offloading tasks beyond the coverage of edge nodes to the cloud for processing; node failure handling; and automatically shutting down some edge nodes when there are fewer task requests.

[0103] Furthermore, a dynamic model is established to simulate edge nodes and task requests to achieve intelligent dynamic scheduling.

[0104] The first solution described in S3 is taken as an example for detailed explanation.

[0105] Edge node computing resource constraints: Set the boundaries of the area corresponding to each edge node to provide a reliable basis for matching task requests with edge nodes.

[0106] Task processing at the nearest location: First, determine whether the edge node corresponding to the task generation area is capable of processing, that is, the resources of the edge node - the resources occupied by the task requirements being processed > the resources required for the newly generated task requirements. If so, no external force is applied to the ball, F = 0; if not, an initial force F is applied to the ball to make it move to the jurisdiction of other edge nodes.

[0107] There are many ways to select F, such as attraction, repulsion, and resistance. The following is only one possible implementation plan.

[0108] Considering the computing power and spatial location distribution of edge nodes, the edge nodes are given quality Task request quality Referring to the law of universal gravitation, the jth edge node gives the ball a gravitational force

[0109]

[0110] The net force on the ball is Among them, r is the distance from the location where the task request is generated to the edge node, ζ is the gravity adjustment coefficient, and ζ takes different values ​​according to different types of tasks.

[0111] Therefore, the ball will move in the direction of the resultant force.

[0112] Prioritize time-sensitive tasks: For time-sensitive tasks, the way to achieve it during simulation is: define β to describe the task type. When the task is time-sensitive, β = 1, otherwise, β = 0. If β = 1, a larger force Fmax and a larger mass m are applied to the ball. max , put aside other tasks that do not need to be processed as a priority and match them to edge nodes for processing more quickly.

[0113] Specifically, only one feasible implementation plan is given below, referring to the solution given for the selection of F in the task nearby processing.

[0114] For time-sensitive tasks, the gravity adjustment coefficient ζ is increased, and the edge node will produce a greater gravitational force on the ball. At the same time, the time-sensitive task is given a larger mass. According to the momentum theorem, the task has a larger impulse, thereby excluding the surrounding tasks that do not need to be processed as a priority and quickly matching them to the edge node.

[0115] Tasks beyond the coverage of edge nodes need to be offloaded to the cloud for processing: If the resources of all edge nodes are occupied, the newly added task requirements need to be offloaded to the cloud for processing.

[0116] Specifically, the example is given below, assuming that communication resources and storage resources are sufficient and only computing resources need to be considered.

[0117] When the edge node resources are less than the resources required by the task, that is, When the number of tasks exceeding the upper bound of the edge node is exceeded, they need to be offloaded to the cloud for processing.

[0118] Node fault handling: If an edge node fails, the task that the edge node is responsible for handling needs to be handed over to the adjacent node. The way to achieve this in the simulation is: provide a force F pointing to other edge nodes for the ball in the area of ​​the faulty edge node, and the corresponding ball will generate acceleration Make the ball move toward the area near the edge node.

[0119] Specifically, only one feasible implementation plan is given below, referring to the solution given for the selection of F in the task nearby processing.

[0120] use Indicates the situation of edge node strike or damage. Assuming that the jth edge node fails, the quality of the other jth edge node is: However, the ball will also be affected by the gravity of other nodes, so the ball will move to the edge node area without faults.

[0121] Automatically shut down some edge nodes when there are fewer task requests: If there are more idle edge nodes (node ​​occupancy rate η i =∑ i s ij / Ei If the number of task requests is less than the preset threshold, some edge nodes will be shut down autonomously to reduce energy consumption and save costs.

[0122] The specific implementation plan is similar to the node failure handling plan, defining the nodes that need to be shut down as For nodes that need to be closed, let their mapping quality Then the ball will be affected by the gravitational force of the unclosed nodes and move toward the unclosed nodes.

[0123] S5: Determine a network dynamic scheduling solution based on the simulation.

[0124] The dynamic scheduling scheme of the edge computing adaptive network is determined through the simulation results of the dynamic model.

[0125] Specifically, one possible implementation is to generate a network scheduling plan at time t when the total system momentum drops within a set threshold. Define the total system momentum as p, and set the threshold of the total system momentum as p θ When p(t) <p θ When the network scheduling plan is generated immediately.

[0126] Taking the solution 1 described in S3 as an example, assuming that communication resources and storage resources are sufficient, and only considering computing resources as an example, combined with Figure 2-4 , the specific implementation method is explained in detail.

[0127] Figure 2 This is a schematic diagram of the edge node scheduling layer model of the present application, where 1 is an edge node. After a certain area is mapped to the virtual scheduling layer, the edge node is divided into eight small areas according to the edge node layout plan and edge node information, and each small area has a matching edge node. Specifically, the area is mapped into a 100×60 rectangular area (the length and area units are not very meaningful, only the scale needs to be unified, so it is expressed as a dimensionless number), and eight edge nodes are arranged. The area of ​​each area represents the size of the edge node computing resources. According to the mapping, A 1(G c1 )=900,A2(G c2 )=675,A3(G c3 )=600,A4(G c4 )=875,A5(G c5 )=400,A6(G c6 )=600,A7(G c7 )=1350,A8(G c8 )=600. In the subsequent simulation process, the edge node position and its computing resource amount remain unchanged.

[0128] Figure 3This is a schematic diagram of the task application scheduling layer model of this application. 2-5 are task blocks mapped to different task requests. The size of the circle area represents the amount of computing resources required for the task. The larger the amount of resources required and the larger the radius, the larger the area. According to the jurisdiction of the edge node, determine the edge node that processes the task request. In the figure, task request 2 is processed by edge node ①, task request 3 is processed by edge node ②, task request 4 is processed by edge node ⑥, and task request 5 is processed by edge node ⑦.

[0129] Figure 3 This is a demonstration diagram of the effect of an adaptive network dynamic scheduling method based on dynamic principles proposed in this application.

[0130] At time t, the network dynamic scheduling scheme is as follows Figure 4 As shown in (a): Obviously, the edge node ⑤ needs to process a large number of tasks. The radii of the mapped task blocks are: r c1 =4, r c2 =4, r c3 =2.5, r c4 =4, r c5 =2.5, r c6 =2.2, the coordinates of the center of the task block mapped are

[0131] (38, 28.1), (53, 33), (48.6, 15), (50, 23), (44.6, 19), (43.6, 25).

[0132] If a new task request is generated in this area, it is very likely that the edge node will have insufficient processing capacity, while other edge nodes have more idle resources. Therefore, when the processing capacity of edge node ⑤ is insufficient, some tasks can be handed over to other edge nodes for processing.

[0133] like Figure 4 As shown in (b): At time t+1, a new task request is generated in area ⑤ at coordinates (48.6, 15). According to the size of computing resources required by the task request, it is mapped into a radius r c6 =6 circular task blocks.

[0134] like Figure 4 (c) shows that the behavior is modeled as a dynamic model. It is assumed that task offloading adopts binary offloading, so a task request can only correspond to one edge node. In the dynamic model, it is expressed as: the task block representing the task request must be completely within the range of a certain edge node and cannot span two or more edge nodes. Since edge node ⑤ does not have enough remaining resources to process the newly generated tasks Some tasks need to be handed over to other edge nodes for processing. The new network scheduling solution is solved by simulating the dynamic model. With the coordinates (48.6, 15) as the center, a small ball with a mass of 6 is built to simulate the task request. It represents the task The ball will collide with the balls representing other tasks, thus exerting a force on the balls representing other tasks. Furthermore, the ball will produce The acceleration, according to The speed of the ball can be calculated.

[0135] like Figure 4 (d) shows: When p(t) <p θ When t+1 is reached, a dynamic network scheduling plan is generated. Processed by edge node ③, Processed by edge node④, Processed by edge node⑤, Processed by edge node⑧.

[0136] In summary, the edge computing adaptive network dynamic scheduling method proposed in this invention combines the dynamic principle with the geometric mapping technology to propose an innovative solution to cope with the complexity of task scheduling, dynamic networking and resource allocation in the field of edge computing. The main technical advantages of this method are reflected in the following aspects:

[0137] It can adapt to a variety of daily and emergency situations, has strong adaptability and flexibility, and can meet the needs of different application scenarios.

[0138] For the first time, the principle of dynamics is applied to the field of edge computing. Compared with traditional optimization methods, this method adopts forward dynamics simulation thinking to solve the problem, avoiding the complex reverse optimization process.

[0139] The state of edge nodes and task requests can be modeled and predicted through the dynamic model, which can quickly solve the problem, especially in time-sensitive tasks, and can quickly determine the scheduling plan.

[0140] It effectively solves the problems of complexity explosion and increased computing time caused by the growth in the number of edge nodes, breaking the limitations of the applicability of traditional methods under large-scale network topology structures.

[0141] It can make autonomous decisions to shut down some edge nodes, thereby reducing energy consumption and costs, and achieving optimal resource allocation and energy-saving effects.

[0142] It can provide scheduling solutions in real time, improve the efficiency and performance of task scheduling, and ensure the efficient operation of the edge computing network.

[0143] In general, the edge computing adaptive network dynamic scheduling method of the present invention, with its unique technical advantages, provides an efficient, intelligent and cost-effective solution for large-scale and diversified edge computing task management and scheduling.

[0144] An embodiment of the present invention further provides a storage medium for storing a computer program, which at least performs the above method when executed.

[0145] An embodiment of the present invention further provides a control device, comprising a processor and a storage medium for storing a computer program; wherein the processor is configured to execute at least the method described above when executing the computer program.

[0146] An embodiment of the present invention further provides a processor, wherein the processor executes a computer program and at least executes the method described above.

[0147] The storage medium can be implemented by any type of non-volatile storage device, or a combination thereof. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a ferromagnetic random access memory (Flash Memory), a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM); the magnetic surface memory can be a disk memory or a tape memory. The storage medium described in the embodiments of the present invention is intended to include, but is not limited to, these and any other suitable types of memory.

[0148] In the several embodiments provided by the present invention, it should be understood that the disclosed systems and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0149] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units; some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0150] In addition, all functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may be separately used as a unit, or two or more units may be integrated into one unit; the above-mentioned integrated units may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0151] A person skilled in the art can understand that: all or part of the steps of implementing the above method embodiment can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiment; and the aforementioned storage medium includes: mobile storage devices, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), disks or optical disks, etc. Various media that can store program codes.

[0152] Alternatively, if the above-mentioned integrated unit of the present invention is implemented in the form of a software function module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiment of the present invention can be essentially or partly reflected in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as mobile storage devices, ROM, RAM, magnetic disks or optical disks.

[0153] The methods disclosed in the several method embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method embodiments.

[0154] The features disclosed in several product embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new product embodiments.

[0155] The features disclosed in several method or device embodiments provided by the present invention can be arbitrarily combined without conflict to obtain new method embodiments or device embodiments.

[0156] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art of the present invention, several equivalent substitutions or obvious variations can be made without departing from the concept of the present invention, and the performance or use is the same, which should be regarded as belonging to the protection scope of the present invention.

Claims

1. An edge computing adaptive network dynamic scheduling method, characterized in that: The steps include: S1: Collect terminal device information and determine the edge node layout plan; S2: Determine the jurisdiction of each edge node and map it proportionally to the virtual scheduling layer, transforming the dynamic scheduling optimization problem into a geometric problem; S3: Map the task request into a task block with mass and area. The area of ​​the task block represents the amount of resources required. The mass of the task block is used to enable the task block to exhibit motion behavior in the dynamic model to simulate the dynamic allocation of task requests among edge nodes. S4: Set constraints and build a dynamic model to simulate and predict edge nodes and task requests. S5: Based on the simulation results, determine the network dynamic scheduling scheme to perform dynamic scheduling of the edge computing network.

2. The edge computing adaptive network dynamic scheduling method according to claim 1, characterized in that: In step S2, the virtual scheduling layer G includes the computation-dynamics layer G c , Communication-Dynamics Layer G m and storage-dynamic layer G s , corresponding to the optimal scheduling of computing resources, communication resources and storage resources, respectively. Indicates that the resources of edge nodes include computing resources, communication resources, storage resources, etc. Indicates that at time t, the position P of the edge node j =(x j (t), y j (t)), On the edge layer, resource mapping maps the resources of edge nodes to a specific area on the dynamic layer. The area of ​​the area reflects the amount of resources corresponding to the edge node. The two-dimensional plane graph of the edge node mapped on the computing-dynamic layer is G cj , and for the computational-dynamics layer G c Any two different edge nodes in the mapping area G cj and G cj′ , the following conditions are met: (1)A j (G cj )=r cj α c , that is, the mapping area G cj Area A j (G cj ) is equal to the computing resource amount r of edge node j cj Multiply by α c , α c It is a coefficient used to convert the computing resources of edge nodes into the corresponding geometric area; (2) That is, the mapping area G of any two edge nodes cj and G cj′ No overlap; (3)P j ∈G cj , that is, the position P of edge node j at time t j Located in its computational-dynamic layer G c The mapped area G cj Inside Among them, A j (G cj ) reflects the amount of computing resources of the edge node, α c The conversion factor representing the actual computing resources and the corresponding dynamic region area.

3. The edge computing adaptive network dynamic scheduling method as described in claim 2 is characterized in that: In step S2, the jurisdiction of each edge node is mapped into a series of two-dimensional plane figures with different areas and connected but non-overlapping edges. The area of ​​the figure represents the computing power of the edge node; the boundary between the figures represents the connection for task transfer and offloading, and the boundary of the plane represents the upper limit of the computing power of the edge node. The mapping areas of any two edge nodes do not overlap and can only be adjacent.

4. The edge computing adaptive network dynamic scheduling method according to any one of claims 1 to 3, characterized in that: In step S3, the task request is doubly modeled as a two-dimensional plane figure with area and a three-dimensional geometric object with mass and volume.

5. The edge computing adaptive network dynamic scheduling method as described in claim 4 is characterized in that: In step S3, the resource requirement s of the task request is determined ci , based on the conversion factor α between the computing resources of the edge node and the area of ​​its corresponding dynamic region c and resource demand ci Calculate the radius of the task block: Generate position Q with task i =(x i (t), y i (t)) is the center of the circle, with radius r ci A circular task block with an area is drawn on a two-dimensional plane, and the area of ​​the task block represents the amount of required resources; and corresponding to the two-dimensional circular task block, it is modeled as a three-dimensional sphere with volume and mass in three-dimensional space, and the volume and mass of the sphere are associated with the area of ​​the two-dimensional circular task block and the resource allocation strategy of the edge node.

6. The edge computing adaptive network dynamic scheduling method as claimed in claim 4, characterized in that: In step S3, the resource requirement of the task request is determined, and the conversion coefficient α between the resource requirement and the computing resources of the edge node and the area of ​​the corresponding dynamic region is calculated. c Calculate the area of ​​the task block; generate the position Q based on the task i =(x i (t), y i (t)) is the centroid, and the task is mapped into an irregular two-dimensional plane shape with an area and a changeable graphic boundary; and corresponding to the two-dimensional irregular shape task block, it is modeled in three-dimensional space as a fluid geometry with volume and mass, and the volume and mass of the fluid geometry are associated with the area of ​​the two-dimensional irregular shape task block and the resource allocation strategy of the edge node.

7. The edge computing adaptive network dynamic scheduling method according to any one of claims 1 to 6, characterized in that: In step S4, the constraints include edge node computing resource constraints, edge node bandwidth constraints, edge node storage constraints, and one or more of the constraints set for priority processing of time-sensitive tasks, offloading tasks beyond the coverage of edge nodes to the cloud for processing, node failure processing, and automatic shutdown of some edge nodes when there are fewer task requests.

8. The edge computing adaptive network dynamic scheduling method according to any one of claims 1 to 7, characterized in that: In step S4, one or more of the following processing strategies are adopted: Implement the task processing strategy nearby. For each task request, first determine whether the edge node in the area where it is generated has sufficient resources to process the task, that is, verify whether the available resources of the edge node are greater than the resources required by the task; if the resources are sufficient, no external force is applied to keep the task processed at the current node; if the resources are insufficient, an initial force is applied to move the task to the jurisdiction of other edge nodes; The dynamic model is used to simulate the dynamic allocation process of task requests by calculating the gravitational force between the edge node and the ball representing the task request. The gravitational formula of the ball is: The net force on the ball in It is the gravity adjustment coefficient based on different types of tasks. and are the edge node quality and task request quality, n is the number of edge nodes, r (j) is the distance from the location where the task request is generated to the edge node; the ball will move in the direction of the resultant force; For time-sensitive tasks, a larger force and a larger mass are applied to the ball, and other tasks that do not need to be processed in priority are excluded so that they can be matched to edge nodes for processing more quickly; For time-sensitive tasks, priority processing is achieved by increasing the gravity adjustment coefficient and task quality; When all edge node resources are occupied, the task requirements beyond the coverage range are offloaded to the cloud for processing; When an edge node fails, the mass of the failed node is adjusted to zero, and the task request is redistributed to the adjacent healthy node. The way to achieve this in the simulation is: provide a force pointing to other edge nodes for the ball in the failed edge node area, generate acceleration for the corresponding ball, and make the ball move to the adjacent edge node area; When the edge node occupancy rate is lower than the preset threshold, some edge nodes are automatically selected to be shut down to reduce energy consumption and save costs. The quality of the shut-down nodes is set to zero to guide task requests to be allocated to open nodes.

9. The edge computing adaptive network dynamic scheduling method according to any one of claims 1 to 8, characterized in that: In step S5, one or more of the following scheduling strategies are adopted: Define the total momentum of the system p as a measure of the network state, which represents the dynamic sum of all task requests and interactions between edge nodes; set the threshold of the total momentum of the system p θ , as the basis for judging whether the scheduling plan needs to be updated; during the simulation process, when the total momentum of the system p(t) is less than the threshold p θ When , the network scheduling scheme at time t is updated; The dynamic model is used to simulate the interaction between task requests and edge nodes, including attraction, repulsion and resistance, to simulate the dynamic allocation of task requests; For newly generated task requests or those that cannot be processed at the current node, the dynamic model is used to predict their impact on the existing task allocation and reallocate them to edge nodes with sufficient resources. The task offloading strategy is considered in the simulation to ensure that the task request can only be completely assigned to one edge node and avoid cross-node task allocation.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the edge computing adaptive network dynamic scheduling method as described in any one of claims 1 to 9 is implemented.