A collaborative scheduling method for edge computing tasks
By dividing resource request points and supply points in the edge computing system, and optimizing resource allocation using utility functions and Nash equilibrium model, the problem of edge computing resource optimization in the Internet of Things is solved, and low-latency and high-efficiency task collaborative scheduling is achieved, improving the Resource utilization and system performance.
Patent Information
- Application Number
- CN202111658416.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2021-12-30
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2041-12-30
AI Technical Summary
In the Internet of Things scenario, the optimization of limited communication resources and computing resources in the edge computing system is difficult to meet the task coordination requirements between multiple perception nodes, resulting in waste of computing resources and high energy consumption, and the time limit requirements and data coupling are high, making it difficult to Implement effective allocation of resources.
A collaborative scheduling method for edge computing tasks is adopted. By dividing the edge computing system into terminal network layer, edge computing layer and cloud layer, using a Raspberry Pi to form an edge server, divide resource request points and resource supply points according to time slots, and use utility functions Optimize resource allocation with the Nash equilibrium model to achieve task decomposition and reasonable resource allocation.
It effectively reduces task delay and overall system power consumption, improves resource utilization and task completion rate, realizes the reasonable allocation of computing resources and tasks, and optimizes the overall performance and cost of edge computing systems.
Smart Images

Figure CN114398162B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of edge computing technology, and in particular relates to a method for collaborative scheduling of edge computing tasks. Background Art
[0002] With the rapid development of IoT applications, the key centralized data processing technologies represented by cloud computing models can no longer efficiently and timely process the data generated by IoT sensing nodes. To address this problem, the edge computing model with the core concept of "data processing should be closer to the source of data" has emerged. By deploying edge computing servers at the edge of the network, low-latency communication resources and relatively abundant computing resources can be provided for sensing nodes.
[0003] For sensor nodes in the Internet of Things, their battery life and computing power are severely limited due to constraints such as size and application environment. Therefore, it is a feasible and effective solution to offload certain computing tasks to edge servers that are physically close to the sensor nodes. Through edge computing, sensor nodes can reduce local computing energy consumption and increase execution speed at the cost of transmission energy consumption and transmission time of offloading computing tasks to edge servers.
[0004] The completion of an IoT service often requires data from multiple sensor nodes to be calculated in coordination. There is even a situation where a sensor node needs the calculation result of another sensor node to start its own computing task. This type of IoT service belongs to the task coordination scenario, which increases the degree of data coupling and greatly increases the difficulty of optimizing edge computing resources. Due to the complexity of the edge environment and the heterogeneity of the sensor nodes, the computing power, amount of data generated, and computing complexity of each device are different; the data of the sensor nodes are usually real-time and have the time limit requirements of the IoT service itself; the communication link between the sensor node and the edge server has bandwidth limitations and noise interference; the computing power of the edge server is far less than that of the cloud server. Therefore, it is necessary to coordinate and optimize the limited communication resources and computing resources in the edge computing system. The optimization of the limited communication resources and computing resources in the edge computing system is an optimization problem with the goal of minimizing the total energy consumption of the sensor nodes in the edge computing system under the constraints of the time limit requirements of the IoT service itself and the task coordination requirements between the data of each sensor node. Summary of the invention
[0005] In order to solve the above problems, the purpose of the present invention is to provide a method for collaborative scheduling of edge computing tasks.
[0006] In order to achieve the above-mentioned invention object, the present invention adopts the following technical scheme:
[0007] A method for collaborative scheduling of edge computing tasks, comprising the following steps:
[0008] S1. Modeling an edge computing system, wherein the edge computing system is divided into three layers, from bottom to top: terminal network layer, edge computing layer, and cloud layer; the terminal network layer includes multiple sensors for sensing original events; the edge computing layer includes multiple Raspberry Pis, and multiple Raspberry Pis form an edge server layer with relatively powerful computing capabilities, and each Raspberry Pi of the edge server layer is set as edge server RPI1, edge server RPI2, edge server RPI3, ..., edge server RPI n , multiple Raspberry Pis are mainly used to receive data collected by multiple sensors connected to them respectively, and perform data calculation on some of the data; the cloud layer is equipped with an adaptation strategy controller, which manages and schedules the edge nodes of the edge computing layer through control strategies; the time in the system is divided into multiple time slots with a duration of T, let t represent the index of the time slot, t=0,1,...,T;
[0009] S2. All the above Raspberry Pi edge servers are divided into resource request points and resource provision points according to whether there are computing resources remaining in time slot t;
[0010] Determine the request point and provision point to which all nodes belong: According to the formula The edge servers are divided into resource request points (RPIs) i and Resource Supply Point RPI j ,in, RPI at time slot t i The maximum amount of tasks that can be accepted, λ i is the terminal user's request to RPI at time slot t i The total amount of unloaded tasks; when α i >0, indicating the RPI of this edge server i There are resources remaining, that is, the RPI of the edge server i Provides a resource point; when α i When <0, it means the RPI of this edge server i No computing resources are left, and other edge server RPIs are needed B Provide resources for collaborative computing to complete tasks, that is, the edge server RPI i It is the resource request point;
[0011] S3. Model the demand point and find its utility function. The steps are as follows:
[0012] Resource Request Point (RPI) i The utility function of is mainly composed of three parts: collaborative computing overhead, local computing overhead, and incentive cost;
[0013] 3.1. Local computing is divided into computing delay and computing energy consumption. The computing delay is modeled according to the M / M / 1 queuing system. The computing delay is calculated as
[0014] in, Indicates on RPI i The size of the workload being processed,
[0015] The computing energy consumption of the edge server for processing each task is proportional to the square of the CPU speed, which means that the energy consumption is
[0016] Among them, κ represents a constant, which depends on the CPU architecture;
[0017] 3.2. The average computational delay of collaborative computing is
[0018] in, For RPi j The total amount of tasks on For RPi j The sum of the terminal offloading tasks and the collaborative tasks;
[0019] The collaborative computation cost is:
[0020] 3.3. Combine the bids of the supply points in the network model to get the RPI i B To RPI j S The amount of collaborative computing tasks requested is d ij When you need to send an RPI j S The price paid is S ij =q ij p j .
[0021] RPI i B To RPI j S The total fee paid is
[0022]
[0023] 3.4. The utility function is defined as F i B (p,d i ,d -i ), expressed as
[0024] S4. Resource supply point RPIj Carry out modeling and find its utility function;
[0025] The resource supply point model is its revenue function G j (p j ,q),G j (p j ,q)=Q j p j ,
[0026] in, Indicates the request point to RPI j The total number of collaborative computing tasks requested; p j Indicates each RPI J S 's bid;
[0027] S5. Find the Nash equilibrium of non-cooperative game points:
[0028] The Nash equilibrium of the non-cooperative game point is the RPI of any request point i The utility function of the first-order partial derivative is Taking the second-order partial derivative, we get: This shows that Nash equilibrium exists.
[0029] Due to the adoption of the above-mentioned technical solution, the present invention has the following advantages:
[0030] This edge computing task collaborative scheduling method decomposes large-scale, complex tasks into small-scale, simple, fine-grained task component sets, thereby reducing the computing load of a single task and optimizing the overall performance and cost of computing task processing; it has lower task delays, higher task completion rates, and lower overall system power consumption; it realizes the reasonable allocation of computing resources and tasks in the edge computing system, reduces task completion delays, and improves resource utilization. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 A model diagram of an edge computing system of an edge computing task collaborative scheduling method of the present invention;
[0032] Figure 2 It is a time series diagram of the collaborative scheduling method of edge computing tasks of the present invention;
[0033] Figure 3 This is an algorithm flow chart of the edge computing task collaborative scheduling method of the present invention. DETAILED DESCRIPTION
[0034] The technical solution of the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.
[0035] like Figure 1 ,3 As shown, a method for collaborative scheduling of edge computing tasks includes the following steps:
[0036] S1. Modeling an edge computing system, wherein the edge computing system is divided into three layers, from bottom to top: terminal network layer, edge computing layer, and cloud layer; the terminal network layer includes multiple sensors for sensing raw events; the edge computing layer includes multiple Raspberry Pis with weak computing power, which are insufficient to complete complex tasks by themselves. Multiple Raspberry Pis form an edge server layer with relatively strong computing power, and each Raspberry Pi of the edge server layer is set as edge server RPI1, edge server RPI2, edge server RPI3, ..., edge server RPI n , multiple Raspberry Pis are mainly used to receive data collected by multiple sensors connected to them respectively, and perform data calculation on some of the data; the cloud layer is equipped with an adaptation strategy controller, which manages and schedules the edge nodes of the edge computing layer through control strategies; the time in the system is divided into multiple time slots with a duration of T, let t represent the index of the time slot, t=0,1,...,T;
[0037] S2. All the above Raspberry Pi edge servers are divided into resource request points and resource provision points according to whether there are computing resources remaining in time slot t;
[0038] Determine the request point and provision point to which all nodes belong: According to the formula The edge servers are divided into resource request points (RPIs) i and Resource Supply Point RPI j , where i represents the edge server name belonging to the resource request point, i=0,1,...,n; j represents the edge server name number of the resource supply point, j=0,1,...,m; RPI at time slot t i The maximum amount of tasks that can be accepted, λ i is the terminal user's request to RPI at time slot t i The total amount of unloaded tasks; when α i >0, it means that this edge server has surplus resources, that is, the edge server is a resource provision point; when α i When <0, it means that this edge server has no computing resources remaining and needs other edge servers to provide resources for collaborative computing to complete the task, that is, this edge server is a resource request point;
[0039] S3. Model the demand point and find its utility function. The steps are as follows:
[0040] Resource Request Point (RPI) iThe utility function of is mainly composed of three parts: collaborative computing overhead, local computing overhead, and incentive cost;
[0041] 3.1. Local computing is divided into computing delay and computing energy consumption. The computing delay is modeled according to the M / M / 1 queuing system. The computing delay is calculated as
[0042] in, Indicates on RPI i The size of the workload being processed,
[0043] The computing energy consumption of the edge server for processing each task is proportional to the square of the CPU speed, which means that the energy consumption is
[0044] Among them, κ represents a constant, which depends on the CPU architecture;
[0045] 3.2. The average computational delay of collaborative computing is
[0046] in, For RPi j The total amount of tasks on For RPi j The sum of the terminal offloading tasks and the collaborative tasks;
[0047] The collaborative computation cost is:
[0048] 3.3. Combine the bids of the supply points in the network model to get the RPI i B To RPI j S The amount of collaborative computing tasks requested is d ij When you need to send an RPI j S The price paid is S ij =q ij p j .
[0049] RPI i B To RPI j S The total fee paid is
[0050]
[0051] 3.4. The utility function is defined as F i B (p,d i ,d -i ), expressed as
[0052] S4. Resource supply point RPI j Carry out modeling and find its utility function;
[0053] The resource supply point model is its revenue function G j (p j ,q),G j (p j ,q)=Q j p j ,
[0054] in, Indicates the request point to RPI j The total number of collaborative computing tasks requested; p j Indicates each RPI J S Bid price; RPI j S It is a resource supply point;
[0055] S5. Find the Nash equilibrium of non-cooperative game points:
[0056] The Nash equilibrium of the non-cooperative game point is the RPI of any request point i The utility function of the first-order partial derivative is Taking the second-order partial derivative, we get: This shows that Nash equilibrium exists.
[0057] Reference Figure 2 , the time series diagram of the edge computing task collaborative scheduling method of the present invention, specifically, for each Δt gap in the adjustment selling price of the resource supply point, the resource request point is divided into Δτ for iterative solution.
[0058] Reference Figure 3 , a process of an implementation method of the edge computing task collaborative scheduling method of the present invention includes the following steps:
[0059] (1) Each RPI counts the amount of terminal task offloads it receives;
[0060] (2) Classify RPIs using the above formula;
[0061] (3) The resource request point sends out a request for help to the surrounding areas;
[0062] (4) Resource supply points send information about their remaining resources and initial resource prices to the surrounding areas;
[0063] (5) Iterate to find the solution of Nash equilibrium. The operation is:
[0064] according to Right now So the price equation for any resource supply point is: Iterate and solve according to the above formula.
[0065] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.
Claims
1. A method for collaborative scheduling of edge computing tasks, characterized by: It includes the following steps: S1. Modeling an edge computing system, wherein the edge computing system is divided into three layers, from bottom to top: terminal network layer, edge computing layer, and cloud layer; the terminal network layer includes multiple sensors for sensing original events; the edge computing layer includes multiple Raspberry Pis, and the multiple Raspberry Pis form an edge server layer with computing capabilities, and each Raspberry Pi of the edge server layer is set as edge server RPI1, edge server RPI2, edge server RPI3, ..., edge server RPI n , multiple Raspberry Pis are used to receive data collected by multiple sensors connected to them respectively, and perform data calculation on some of the data; the cloud layer is equipped with an adaptation strategy controller, which manages and schedules the edge nodes of the edge computing layer through control strategies; the time in the system is divided into multiple time slots with a duration of T, let t represent the index of the time slot, t=0,1,...,T; S2, divide all Raspberry Pi edge servers into resource request points and resource provision points according to whether there are computing resources remaining in time slot t; Determine the request point and provision point to which all nodes belong: According to the formula The edge servers are divided into resource request points (RPIs) i and Resource Supply Point RPI j ,in, RPI at time slot t i The maximum amount of tasks that can be accepted, λ i is the terminal user's request to RPI at time slot t i The total amount of unloaded tasks; when α i >0, indicating the RPI of this edge server i There are resources remaining, that is, the RPI of the edge server i Provides a resource point; when α i When <0, it means the RPI of this edge server i No computing resources are left, and other edge server RPIs are needed B Provide resources for collaborative computing to complete tasks, that is, the edge server RPI i It is the resource request point; S3. Model the demand point and find its utility function. The steps are as follows: Resource Request Point (RPI) i The utility function of is composed of three parts: collaborative computing overhead, local computing overhead, and incentive cost; 3.
1. Local computing is divided into computing delay and computing energy consumption. The computing delay is modeled according to the M / M / 1 queuing system. The computing delay is calculated as in, Represents RPI i The amount of tasks that can be completed per second; Indicates on RPI i The size of the workload being processed, Indicates that the terminal user sends RPI i The total amount of unloaded tasks, For the original RPI i The amount of tasks that need to be solved; The computing energy consumption of the edge server for processing each task is proportional to the square of the CPU speed, which means that the energy consumption is Among them, κ represents a constant, which depends on the CPU architecture; 3.
2. The average computational delay of collaborative computing is in, For RPi j The total amount of tasks on For RPi j The sum of the terminal offloading tasks and the collaborative tasks; The collaborative computation cost is: 3.
3. Combine the bids of the supply points in the network model to get the RPI i B RPI j S The amount of collaborative computing tasks requested is d ij When you need to send an RPI j S The price paid is S ij =q ij p j; RPI i B RPI j S The total fee paid is 3.
4. The utility function is defined as Expressed as S4. Resource supply point RPI j Carry out modeling and find its utility function; The resource supply point model is its revenue function G j (p j ,q),G j (p j ,q)=Q j p j , in, Indicates the request point to RPI j The total number of collaborative computing tasks requested, p j Indicates each RPI J S 's bid; S5. Find the Nash equilibrium of non-cooperative game points: The Nash equilibrium of the non-cooperative game point is the RPI of any request point i The utility function of the first-order partial derivative is Taking the second-order partial derivative, we get: This shows that Nash equilibrium exists.
Citation Information
Cited By
Industrial software practical training edge computing collaborative scheduling system and evaluation method
CN122884674A
