Computing resource matching method and device based on edge computing platform

Abstract

The invention relates to a computing resource matching method and device based on an edge computing platform, computer equipment, a computer readable storage medium and a computer program product. The method comprises the steps of obtaining a target computing task of user equipment and uploading the target computing task to an edge computing platform; determining a scheduling mode of the target calculation task according to the state information of each edge server and the task attribute of the target calculation task; determining a target edge server according to the scheduling mode; obtaining a to-be-processed calculation task list of the target edge server; and by taking maximization of the processing benefit as a constraint target, allocating computing resources to all the computing tasks in the to-be-processed computing task list according to the to-be-allocated computing resources of the edge computing platform. By the adoption of the method, task scheduling and resource allocation can be optimized in the edge computing environment, efficient matching of the task and the server is achieved, and the resource utilization efficiency is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of edge computing, and in particular to a computing resource matching method and device based on an edge computing platform, a computer device, a computer readable storage medium and a computer program product. BACKGROUND

[0002] With the rapid development of Internet of Things (IoT), big data, artificial intelligence (AI) and other technologies, the service content and application scenarios of user terminals are greatly enriched, and the demand for computing resources is also rapidly growing. Due to the limitations of centralized architecture, the traditional cloud computing model is difficult to meet the needs of low-latency and real-time scenarios such as autonomous driving and industrial Internet of Things. The edge computing model emerges as the times require, which deploys computing resources at the edge of the network to realize the decentralization of data processing, effectively alleviate the burden of cloud computing, reduce latency and improve user experience.

[0003] However, the edge computing environment faces many challenges. The edge devices have strong resource heterogeneity and various task types, resulting in complex resource management and task scheduling. Existing task scheduling and resource allocation methods mainly focus on single target optimization, lacking the balance between resource utilization efficiency and task scheduling flexibility from a global perspective. In particular, in a dynamic task request environment, how to achieve fast and accurate matching between tasks and servers while ensuring the real-time and efficiency of resource allocation has become a technical problem to be solved.

[0004] Specifically, traditional task offloading research mainly focuses on target device selection and offloading optimization, but lacks research on resource allocation management of target devices after task offloading, which may lead to insufficient resource utilization or bottlenecks of target devices. On the other hand, existing resource allocation management is mainly for single-server scenarios, and there is a lack of research on resource collaborative allocation in multi-server scenarios. In the edge computing environment, how to efficiently allocate resources among multiple edge servers to meet different task requirements is a key problem that needs to be broken through in current technology.

[0005] Therefore, a computing resource matching method and device based on an edge computing platform, a computer device, a computer readable storage medium and a computer program product are needed, which can optimize task scheduling and resource allocation in an edge computing environment, achieve efficient matching between tasks and servers, and improve resource utilization efficiency. SUMMARY

[0006] Therefore, it is necessary to provide a computing resource matching method and device based on an edge computing platform, a computer device, a computer readable storage medium and a computer program product, which can optimize task scheduling and resource allocation in an edge computing environment, achieve efficient matching between tasks and servers, and improve resource utilization efficiency.

[0007] In a first aspect, the application provides a computing resource matching method based on an edge computing platform, comprising:

[0008] obtaining a target computing task of a user device and uploading the target computing task to an edge computing platform;

[0009] obtaining state information of each edge server in the edge computing platform and task attributes of the target computing task;

[0010] determining a scheduling mode of the target computing task according to the state information of each edge server and the task attributes of the target computing task;

[0011] determining a target edge server for processing the target computing task according to the scheduling mode of the target computing task;

[0012] obtaining a list of to-be-processed computing tasks of the target edge server, the list of to-be-processed computing tasks including the target computing task and other computing tasks in a hosting state;

[0013] maximizing processing benefits as a constraint target, and allocating computing resources to all computing tasks in the list of to-be-processed computing tasks according to to-be-allocated computing resources of the edge computing platform.

[0014] In one embodiment, the determining of the scheduling mode of the target computing task according to the state information of each edge server and the task attributes of the target computing task comprises:

[0015] constructing a deep reinforcement learning model, the deep reinforcement learning model including a state space, an action space and a reward function of the edge computing platform;

[0016] The state information of the edge server includes a data upload rate of a communication link between a user device generating the target computing task and the edge server, and a computing resource vector, a real-time load condition vector, a pricing variable vector and a cost variable vector of the edge server.

[0017] The state space and the reward function are determined according to the state information of the edge server;

[0018] The action space is determined according to the task attributes of the target computing task;

[0019] The target edge server is matched as the scheduling mode by using the deep reinforcement learning model, and the target computing task is scheduled to the target edge server.

[0020] In one of the embodiments, the matching the target edge server as the scheduling mode and scheduling the target computing task to the target edge server by using the deep reinforcement learning model comprises:

[0021] obtaining a weight factor of a reward generated when each target computing task is scheduled to a corresponding edge server;

[0022] determining a scheduling strategy of the target computing task according to the reward function and the weight factor;

[0023] traversing a plurality of state transition sequences in the state space, and determining a highest cumulative reward corresponding to the scheduling strategy of the target computing task according to the action space and the state space;

[0024] matching the target edge server as the scheduling mode and scheduling the target computing task to the target edge server according to the highest cumulative reward.

[0025] In one of the embodiments, the traversing a plurality of state transition sequences in the state space, and determining a highest cumulative reward corresponding to the scheduling strategy of the target computing task according to the action space and the state space comprises:

[0026] traversing a plurality of state transition sequences, and calculating an advantage function value corresponding to each server state and a corresponding scheduling action in the action space and the state space;

[0027] iteratively updating the scheduling strategy of the target computing task by a gradient ascent method according to the advantage function value to achieve the highest cumulative reward.

[0028] In one of the embodiments, the taking the processing revenue maximization as the constraint target and allocating computing resources to all computing tasks in the to-be-processed computing task list according to the to-be-allocated computing resources of the edge computing platform comprises:

[0029] obtaining a task delay of the target computing task and an energy consumption of the target edge server in processing the task in the case that the target computing task is scheduled to the target edge server, the task delay comprising a transmission delay, a computing delay, and a total task processing delay;

[0030] obtaining an operation cost and an operation revenue of the edge computing platform, wherein the operation cost comprises an energy consumption cost and a resource occupation cost of task processing, and the operation revenue comprises a charging of the edge computing platform to a user and a penalty cost of service quality conversion;

[0031] The target computing task is referenced in terms of task latency, energy consumption of the edge server processing the task, operation cost and operation benefit of the edge computing platform, processing benefit maximization is taken as a constraint target, and computing resources are allocated to all computing tasks in the list of computing tasks to be processed according to the computing resources to be allocated of the edge computing platform.

[0032] In one of the embodiments, the target computing task is referenced in terms of task latency, energy consumption of the edge server processing the task, operation cost and operation benefit of the edge computing platform, processing benefit maximization is taken as a constraint target, and computing resources are allocated to all computing tasks in the list of computing tasks to be processed according to the computing resources to be allocated of the edge computing platform, including:

[0033] The target computing task is referenced in terms of task latency, energy consumption of the edge server processing the task, operation cost and operation benefit of the edge computing platform, and operation benefit maximization and preset latency expectation value are taken as constraint targets.

[0034] The power resource allocation strategy based on the particle swarm algorithm calculates the optimal solution of the constraint target by simulating the movement and cooperation of individual particles in the search space.

[0035] Based on the optimal solution, computing resources are allocated to all computing tasks in the list of computing tasks to be processed according to the computing resources to be allocated of the edge computing platform.

[0036] In the case where there is a completed target computing task on the edge server, the power resource allocation strategy based on the particle swarm algorithm re-allocates computing resources to all computing tasks in the list of computing tasks to be processed.

[0037] In a second aspect, the application further provides a computing resource matching device based on an edge computing platform, including:

[0038] An acquisition module is configured to acquire a target computing task of a user equipment and upload the target computing task to an edge computing platform.

[0039] The acquisition module is further configured to acquire state information of each edge server in the edge computing platform and task attributes of the target computing task.

[0040] A task scheduling module is configured to determine a scheduling mode of the target computing task according to the state information of each edge server and the task attributes of the target computing task.

[0041] The task scheduling module is further configured to determine a target edge server for processing the target computing task according to the scheduling mode of the target computing task.

[0042] The acquisition module is further configured to acquire a to-be-processed computing task list of the target edge server, the to-be-processed computing task list including the target computing task and other computing tasks in a hosting state.

[0043] The resource allocation module is configured to maximize processing benefits as a constraint target, and allocate computing resources to all computing tasks in the to-be-processed computing task list according to to-be-allocated computing resources of the edge computing platform.

[0044] In a third aspect, the present application further provides a computer device, comprising a memory and a processor, the memory stores a computer program, and the processor realizes the following steps when executing the computer program:

[0045] acquiring a target computing task of a user device and uploading the target computing task to an edge computing platform;

[0046] acquiring state information of each edge server in the edge computing platform and task attributes of the target computing task;

[0047] determining a scheduling mode of the target computing task according to the state information of each edge server and the task attributes of the target computing task;

[0048] determining a target edge server for processing the target computing task according to the scheduling mode of the target computing task;

[0049] acquiring a to-be-processed computing task list of the target edge server, the to-be-processed computing task list including the target computing task and other computing tasks in a hosting state;

[0050] maximizing processing benefits as a constraint target, and allocating computing resources to all computing tasks in the to-be-processed computing task list according to to-be-allocated computing resources of the edge computing platform.

[0051] In a fourth aspect, the present application further provides a computer readable storage medium, which stores a computer program, and the computer program realizes the following steps when executed by a processor:

[0052] acquiring a target computing task of a user device and uploading the target computing task to an edge computing platform;

[0053] acquiring state information of each edge server in the edge computing platform and task attributes of the target computing task;

[0054] determining a scheduling mode of the target computing task according to the state information of each edge server and the task attributes of the target computing task;

[0055] determine a target edge server for processing the target computing task according to a scheduling manner of the target computing task;

[0056] obtain a to-be-processed computing task list of the target edge server, the to-be-processed computing task list containing the target computing task and other computing tasks in a hosting state;

[0057] maximize processing benefits as a constraint target, and allocate computing resources to all computing tasks in the to-be-processed computing task list according to to-be-allocated computing resources of the edge computing platform.

[0058] In a fifth aspect, the present application further provides a computer program product comprising a computer program which, when executed by a processor, implements the following steps:

[0059] obtain a target computing task of a user equipment, and upload the target computing task to an edge computing platform;

[0060] obtain state information of each edge server in the edge computing platform and task attributes of the target computing task;

[0061] determine a scheduling manner of the target computing task according to the state information of each edge server and the task attributes of the target computing task;

[0062] determine a target edge server for processing the target computing task according to the scheduling manner of the target computing task;

[0063] obtain a to-be-processed computing task list of the target edge server, the to-be-processed computing task list containing the target computing task and other computing tasks in a hosting state;

[0064] maximize processing benefits as a constraint target, and allocate computing resources to all computing tasks in the to-be-processed computing task list according to to-be-allocated computing resources of the edge computing platform.

[0065] The computing resource matching method and device based on the edge computing platform, the computer device, the computer readable storage medium and the computer program product can more accurately allocate tasks to appropriate servers, avoid resource waste and improve overall resource utilization efficiency by determining the scheduling mode according to the state information of the edge server and the task attribute of the target computing task. The computing resource is allocated to all computing tasks in the list of computing tasks to be processed, with the constraint target of processing revenue maximization, which can ensure that each task can obtain the optimal processing effect under limited resources, thereby improving the overall task processing revenue. The target computing task of the user equipment and the state information of each edge server in the edge computing platform are dynamically acquired, so that the task scheduling can adapt to the changes of platform resources in real time, and the flexibility and adaptability of the task scheduling are enhanced. Through accurate task scheduling and efficient resource allocation, the user's computing task can be completed faster, the waiting time of task processing is reduced, and the user experience is improved. The optimized resource allocation strategy can reduce unnecessary energy consumption and resource occupation, and reduce the operating cost of the edge computing platform. By reasonably allocating computing resources, system instability caused by resource overload or deficiency is avoided, and the overall stability of the system is improved. BRIEF DESCRIPTION OF DRAWINGS

[0066] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related art, the drawings needed to be used in the description of the embodiments of the present application or the related art will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other related drawings can be obtained by those skilled in the art without creative labor.

[0067] Figure 1 An application environment diagram of the computing resource matching method based on the edge computing platform in one embodiment;

[0068] Figure 2 A flowchart of the computing resource matching method based on the edge computing platform in one embodiment;

[0069] Figure 3 A schematic diagram of uploading the target computing task to the edge computing platform in one embodiment;

[0070] Figure 4 A flowchart of the computing resource matching method based on the edge computing platform in another embodiment;

[0071] Figure 5 A structural block diagram of the computing resource matching device based on the edge computing platform in one embodiment;

[0072] Figure 6 An internal structure diagram of the computer device in one embodiment. DETAILED DESCRIPTION

[0073] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and not used to limit the present application.

[0074] The computing resource matching method based on the edge computing platform provided by the embodiments of the present application can be applied to an application environment as shown in Figure 1 . The terminal 102 communicates with the server 104 through a network. The data storage system can store data that needs to be processed by the server 104. The data storage system can be integrated on the server 104, or placed on a cloud or other network server.

[0075] The server 104 controls the terminal 102 to obtain a target computing task of a user device, and uploads the target computing task to an edge computing platform; obtains state information of each edge server in the edge computing platform and a task attribute of the target computing task; the server 104 determines a scheduling mode of the target computing task according to the state information of each edge server and the task attribute of the target computing task; determines a target edge server for processing the target computing task according to the scheduling mode of the target computing task; obtains a to-be-processed computing task list of the target edge server, the to-be-processed computing task list including the target computing task and other computing tasks in a hosting state; and takes maximizing processing benefits as a constraint target, and allocates computing resources to all computing tasks in the to-be-processed computing task list according to the to-be-allocated computing resources of the edge computing platform.

[0076] The terminal 102 can be, but is not limited to, various personal computers, notebook computers, smart phones, tablet computers, Internet of Things devices and portable wearable devices. The Internet of Things device can be a smart speaker, a smart television, a smart air conditioner, a smart vehicle device, a projection device, etc. The portable wearable device can be a smart watch, a smart bracelet, a head-mounted device, etc. The head-mounted device can be a virtual reality (VR) device, an augmented reality (AR) device, smart glasses, etc. The server 104 can be a standalone physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0077] In an exemplary embodiment, as shown in Figure 2 , a computing resource matching method based on an edge computing platform is provided. Taking the server 104 in Figure 1 as an example, the method includes the following steps S202 to S212.

[0078] wherein:

[0079] Step S202, obtaining the target computing task of the user device and uploading the target computing task to the edge computing platform.

[0080] Specifically, as shown in Figure 3 the target computing task of the user device is obtained, wherein the user device can be any terminal device that needs to perform a computing task, such as a smartphone, a tablet computer, an IoT device, etc.

[0081] The target computing task refers to the specific computing work that the user device needs to complete, such as video processing, data analysis, machine learning inference, etc. These tasks may not be suitable for direct execution on the user device due to large computing volume or the need for real-time processing.

[0082] uploading to the edge computing platform, wherein the edge computing platform is a distributed computing infrastructure located at the edge of the network, closer to the user device. It is composed of multiple edge servers that can process data physically closer to the data source. The user device sends the target computing task to the edge computing platform. This usually involves transmitting task data through network connections (such as Wi-Fi, 5G, etc.) to edge servers. The purpose is to execute the computing task on the edge computing platform, which can reduce latency, improve response speed, and take advantage of the stronger computing power of edge servers. In addition, offloading computing tasks to the edge computing platform can also reduce the burden on the user device, save power, and improve device performance.

[0083] Step S204, obtaining the state information of each edge server in the edge computing platform and the task attributes of the target computing task.

[0084] Specifically, the state information of the edge server includes the current condition of each edge server in the edge computing platform, such as:

[0085] Computing power: the processor performance of the server, available CPU and GPU resources, etc.

[0086] Storage status: storage space usage on the server, including used space and available space.

[0087] Network condition: the network connection quality of the server, including bandwidth, latency and data transmission rate.

[0088] Load condition: the number and type of tasks currently being executed on the server, and the occupancy of these tasks on server resources.

[0089] Energy consumption status: the energy consumption of the server, including current energy consumption and energy efficiency.

[0090] Fault and maintenance information: whether the server has failed, whether it needs maintenance, etc.

[0091] Task attributes of the target computing task involve the characteristics of the specific computing task that needs to be executed on the edge computing platform, such as:

[0092] Computing requirements: the amount of computing resources required by the task, including CPU, GPU, and memory, etc.

[0093] Data requirements: the size and type of data required for task processing.

[0094] Time sensitivity: the latency requirement of the task, i.e., how long the task needs to be completed.

[0095] Priority: the importance or urgency of the task, which may affect the scheduling order of the task.

[0096] Dependency relationship: whether the task depends on the completion of other tasks, or whether multiple tasks need to be processed in parallel.

[0097] Step S206, according to the state information of each edge server and the task attributes of the target computing task, determine the scheduling mode of the target computing task.

[0098] Specifically, evaluate the current load, available resources (such as CPU, memory, storage, network bandwidth, etc.), energy consumption, and possible failure state of each edge server. Determine which servers have enough resources to handle new computing tasks, and which servers may need to offload part of the tasks to avoid overload.

[0099] Analyze the attributes of the target computing task, considering the computing requirements, data size, expected completion time (latency requirement), priority, and any specific resource requirements of the task. Determine whether the task requires specific hardware support, such as GPU acceleration, or has strict requirements on network latency.

[0100] According to the state of the server and the attributes of the task, decide which server is most suitable for executing a specific task. This process may involve considering the geographical location of the task to reduce data transmission distance and time. Develop a scheduling strategy that takes into account server load balancing, task priority, optimal resource utilization, and user service level agreement (SLA) requirements. The scheduling strategy may include task pre-allocation, dynamic migration, load balancing, or failover, etc.

[0101] Step S208, according to the scheduling mode of the target computing task, determine the target edge server for processing the target computing task.

[0102] Specifically, the scheduling manner refers to a series of rules and strategies formulated according to the attributes of the tasks (such as computing requirements, data size, latency requirements, etc.) and the state information of the edge servers (such as current load, available resources, geographical location, etc.), which are used to guide how to allocate computing tasks to edge servers.

[0103] The target edge server refers to the edge server selected according to the above-mentioned scheduling manner that is most suitable for processing a specific computing task, with the purpose of ensuring that the task can be completed efficiently, quickly and reliably.

[0104] For example, check the processing capacity of each edge server, including CPU, GPU, memory and storage, to determine whether they meet the computing requirements of the task. Evaluate the network connection quality between user devices and edge servers, including bandwidth, latency and stability, to ensure the efficiency and reliability of data transmission. Consider the specific requirements of the task, such as real-time performance, data security and privacy protection, which may affect the selection of servers. In order to optimize resource utilization and avoid overload, it is necessary to balance the load between servers and select servers with lower current load to process new tasks. In possible cases, select servers with lower energy consumption or higher cost-effectiveness to reduce operating costs.

[0105] Step S210, obtaining the list of to-be-processed computing tasks of the target edge server, the list of to-be-processed computing tasks including the target computing task and other computing tasks in a hosting state.

[0106] Specifically, the list of to-be-processed computing tasks is a dynamically updated list that records all computing tasks that have not been completed on the target edge server, and each task in the list is in a waiting execution or execution state.

[0107] The list of to-be-processed computing tasks not only includes the newly determined target computing task, but also includes other computing tasks in a hosting state that have been allocated to the server but have not been completed. The other computing tasks in a hosting state refer to tasks that have been executed on the target edge server, but may be temporarily suspended or waiting for execution due to resource limitations, priority adjustments or other reasons. These tasks may need to wait for more computing resources to become available, or wait for the completion of the currently executing high-priority task before continuing execution.

[0108] Step S212, taking the maximization of processing benefits as a constraint target, allocating computing resources to all computing tasks in the list of to-be-processed computing tasks according to the to-be-allocated computing resources of the edge computing platform.

[0109] Specifically, the processing revenue maximization is the main goal in the resource allocation process, that is, through a reasonable resource allocation strategy, the total revenue obtained by the edge computing platform from processing the computing tasks is maximized. The revenue can be multi-faceted, including but not limited to direct economic revenue (such as service fees), improved system efficiency, enhanced user experience, reduced energy consumption cost, etc.

[0110] In the resource allocation process, in addition to maximizing revenue, other constraints such as task latency requirements, system stability, resource availability, etc. need to be considered. These constraints ensure that the resource allocation scheme not only improves revenue, but also meets the basic requirements of system operation and user service level agreement (SLA).

[0111] The computing resources to be allocated refer to the computing resources in the edge computing platform that have not been allocated to any task, including CPU, memory, storage, network bandwidth, etc. These resources need to be reasonably allocated according to the needs and priorities of the tasks to ensure that all tasks can be effectively processed.

[0112] Allocating computing resources to all computing tasks involves allocating appropriate resources to each task from the computing resources to be allocated according to the needs and priorities of each task. Resource allocation needs to consider multiple factors such as task urgency, resource utilization efficiency, system load balancing, etc. to ensure optimal utilization of resources. Various algorithms and strategies can be used to achieve this goal, such as linear programming, integer programming, heuristic algorithms, machine learning, etc. These methods can help the platform find the optimal or near-optimal solution to resource allocation under the premise of meeting the constraints.

[0113] In the above method of matching computing resources based on an edge computing platform, the scheduling mode is determined according to the state information of the edge servers and the task attributes of the target computing tasks, which can more accurately allocate tasks to appropriate servers, avoid resource waste, and improve overall resource utilization efficiency. Taking the maximization of processing revenue as a constraint target, computing resources are allocated to all computing tasks in the list of computing tasks to be processed, which can ensure that each task can obtain the optimal processing effect under limited resources, thereby improving the overall task processing revenue. Dynamically obtaining the target computing tasks of user devices and the state information of edge servers in the edge computing platform enables the task scheduling to adapt to changes in platform resources in real time, enhancing the flexibility and adaptability of task scheduling. Through precise task scheduling and efficient resource allocation, user computing tasks can be completed faster, reducing the waiting time for task processing and improving user experience. Optimized resource allocation strategies can reduce unnecessary energy consumption and resource occupation, reducing the operating cost of the edge computing platform. By reasonably allocating computing resources, system instability caused by resource overload or shortage is avoided, improving the overall stability of the system.

[0114] In one exemplary embodiment, as shown in Figure 4 determining a scheduling manner of the target computing task according to the state information of each edge server and the task attribute of the target computing task, including:

[0115] Step S402, constructing a deep reinforcement learning model, the deep reinforcement learning model including a state space of the edge computing platform, an action space and a reward function;

[0116] Step S404, determining the state space and the reward function according to the state information of the edge server; wherein the state information of the edge server includes a data upload rate of a communication link between a user equipment generating the target computing task and the edge server, and a computing resource vector, a real-time load condition vector, a pricing variable vector and a cost variable vector of the edge server;

[0117] Step S406, determining the action space according to the task attribute of the target computing task;

[0118] Step S408, matching the target edge server as the scheduling manner by using the deep reinforcement learning model, and scheduling the target computing task to the target edge server.

[0119] Specifically, a model capable of learning the optimal task scheduling strategy, i.e. a deep reinforcement learning model, is created. The components include:

[0120] State space: describes the global state of the edge computing platform, including the state of all edge servers and the state of all pending tasks.

[0121] Action space: defines all possible actions that the model can take, i.e. all possible ways of assigning tasks to servers.

[0122] Reward function: design a function to evaluate the effect of each task assignment to guide the model to learn how to make better decisions.

[0123] Based on the state information of the edge server, the state space is defined, including: the data upload rate of the user equipment to the edge server, the computing resources (such as CPU, memory) of the edge server, the real-time load condition of the edge server, the pricing and cost variables such as service fee and energy consumption cost.

[0124] Based on the benefits and costs of task execution, the reward function is designed, including: the benefits brought by task completion, the QoS penalty caused by task processing delay, and the energy consumption and resource occupation cost.

[0125] Based on the attributes of the target computing task, the action space is defined, including: the computing requirements of the task, the data size and transmission requirements, the delay expectation and priority of the task, and specific resource requirements such as GPU acceleration.

[0126] The trained deep reinforcement learning model is used to evaluate different server options and select the server that is best suited to handle the target computing task, and then the target computing task is scheduled to be executed on the selected edge server.

[0127] In this embodiment, the scheduling of computing tasks is dynamically optimized by the deep reinforcement learning model, enabling the edge computing platform to intelligently determine the best edge server for task processing based on real-time server status and task attributes. This approach can improve resource utilization efficiency, reduce costs, and enhance user experience. In this way, the edge computing platform can implement more flexible and efficient task scheduling strategies, thereby improving overall operational revenue.

[0128] In one exemplary embodiment, the deep reinforcement learning model is used to match the target edge server as the scheduling method, and the target computing task is scheduled to the target edge server, including:

[0129] Obtaining a weight factor for the reward generated when each target computing task is scheduled to the corresponding edge server;

[0130] Determining the scheduling strategy for the target computing task according to the reward function and the weight factor;

[0131] Traversing multiple state transition sequences in the state space, and determining the highest cumulative reward corresponding to the scheduling strategy for the target computing task according to the action space and the state space;

[0132] According to the highest cumulative reward, the target edge server is matched as the scheduling method, and the target computing task is scheduled to the target edge server.

[0133] The weight factor is related to the QoS (Quality of Service) penalty cost considered when defining the operational revenue of the edge computing platform. The weight factor can reflect the degree of influence of different tasks on the platform's operational revenue, for example, for time-sensitive tasks, their weight factor may be higher to ensure that these tasks are given priority. The reward function reflects the revenue obtained after a task is executed, and should guide the agent to make scheduling decisions that yield higher revenue while ensuring the QoS of the services provided. Combining the reward function and the weight factor, a scheduling strategy can be determined that aims to maximize the platform's operational revenue while meeting the latency requirements of tasks.

[0134] By traversing the state transition sequence, since the state space includes the communication link data upload rate between the user equipment and the edge server, the server computing resources, the load situation, the pricing variables and the cost variables, etc., the action space represents the scheduling strategy of the task, i.e. the task is scheduled to be executed on a specific server. In the state space, multiple possible state transitions are simulated, and the effects of different scheduling strategies are evaluated to determine the strategy that can bring the highest cumulative reward.

[0135] Through simulation and learning, it is determined which scheduling strategy can bring the highest cumulative reward, i.e. which strategy can most effectively utilize resources and maximize revenue. According to the highest cumulative reward, the most suitable edge server is selected to process the target computing task, i.e. the task scheduling decision is made by using the deep reinforcement learning method.

[0136] In this embodiment, deep reinforcement learning is used to dynamically optimize the scheduling of computing tasks in the edge computing platform. By simulating different scheduling strategies and evaluating their effects, the model can learn an optimal scheduling strategy that can maximize the operating revenue of the edge computing platform while meeting the task delay requirements. This method can adapt to the dynamic changes in the edge computing environment, improve resource utilization efficiency, reduce costs, and improve user experience.

[0137] In one exemplary embodiment, a plurality of state transition sequences are traversed in the state space, and according to the action space and the state space, the highest cumulative reward corresponding to the scheduling strategy of the target computing task is determined, including:

[0138] By traversing a plurality of state transition sequences, the advantage function value corresponding to each server state and the corresponding scheduling action in the action space and the state space is calculated;

[0139] According to the advantage function value, the scheduling strategy of the target computing task is iteratively updated by the gradient ascent method to achieve the highest cumulative reward.

[0140] Specifically, by traversing the state transition sequence, the advantage function value is calculated, wherein the advantage function refers to the measure of the pros and cons of taking a certain action in a specific state relative to the average action, which is closely related to the reward function, and the reward function reflects the revenue after the task is executed and considers the QoS (Quality of Service) penalty.

[0141] For each server state and corresponding scheduling action, the advantage function value is calculated to evaluate the pros and cons of the action relative to other actions. With the advantage function value, the scheduling strategy is iteratively updated by the gradient ascent method. The Proximal Policy Optimization (PPO) algorithm is used, which improves training stability by limiting the step size of policy update. According to the collected data, the advantage function is calculated, and a clipping mechanism is introduced to maintain the "locality" of policy update. Then, according to the advantage function calculation value, the gradient ascent updates the policy parameters, prompting the policy to evolve towards a higher cumulative reward. Through the above iterative process, the scheduling strategy is continuously optimized until the strategy that maximizes the cumulative reward is found. This means pursuing the maximization of platform operating revenue while trying to meet the latency expectations of the task as much as possible. According to the optimized scheduling strategy, the target edge server is matched as the scheduling method, and the target computing task is scheduled to the target edge server.

[0142] In this embodiment, deep reinforcement learning is used to dynamically optimize the scheduling of computing tasks in the edge computing platform. By simulating different scheduling strategies and evaluating their effects, the model can learn an optimal scheduling strategy that maximizes the operating revenue of the edge computing platform while meeting the latency requirements of the task. This method can adapt to the dynamic changes in the edge computing environment, improve resource utilization efficiency, reduce costs, and improve user experience.

[0143] In an exemplary embodiment, the processing revenue maximization is taken as a constraint target, and the computing resources of the edge computing platform are allocated to all computing tasks in the to-be-processed computing task list, including:

[0144] In the case where the target computing task is scheduled to the target edge server, the task latency of the target computing task and the energy consumption of the target edge server processing the task are obtained, and the task latency includes transmission latency, computation latency, and total task processing latency.

[0145] The operating cost and operating revenue of the edge computing platform are obtained, wherein the operating cost includes the energy consumption cost and resource occupation cost of task processing, and the operating revenue includes the charging of the edge computing platform to users and the penalty cost of service quality conversion;

[0146] Referring to the task latency of the target computing task, the energy consumption of the edge server processing the task, and the operating cost and operating revenue of the edge computing platform, the processing revenue maximization is taken as a constraint target, and the computing resources of the edge computing platform are allocated to all computing tasks in the to-be-processed computing task list.

[0147] Task latency includes transmission latency (data transmission time from user devices to edge servers), computation latency (task processing time on the server), and total task processing latency (total time from task start to completion). These latency metrics are key indicators for evaluating quality of service (QoS). The energy consumed by edge servers when processing tasks directly affects operating costs.

[0148] Obtain operating costs and operating revenue. Operating costs include the energy and resource costs of task processing. These costs must be minimized to improve the economic efficiency of the platform. Operating revenue includes user fees and penalties for quality of service (QoS) conversions. Revenue must be maximized to improve the economic efficiency of the platform.

[0149] Taking into account task latency, energy consumption, operating costs, and revenue, maximizing processing revenue is set as a constraint objective. This means maximizing the economic benefits of the platform while meeting QoS requirements. Heuristic algorithms (such as particle swarm optimization) are used to allocate computing resources. This process involves optimizing resource allocation to maximize revenue while meeting task latency requirements. The computing resource matching problem for edge computing platforms can be formalized as a mixed integer nonlinear programming problem. The computing resource allocation problem needs to be solved under multiple constraints, including task latency, energy consumption, and cost.

[0150] In this embodiment, a resource allocation strategy that comprehensively considers task attributes, server status, and economic benefits is implemented. By optimizing the allocation of computing resources, the edge computing platform can improve resource utilization efficiency, reduce operating costs, and maximize operating returns while ensuring service quality. This approach helps achieve efficient operation of the edge computing platform, improve user satisfaction, and enhance the platform's market competitiveness.

[0151] In an exemplary embodiment, referring to the task latency of the target computing task, the energy consumption of the edge server processing task, the operating cost and operating income of the edge computing platform, maximizing the processing income is used as a constraint goal, and computing resources are allocated to all computing tasks in the list of pending computing tasks based on the computing resources to be allocated on the edge computing platform, including:

[0152] Considering the task latency of the target computing task, the energy consumption of the edge server processing task, and the operating cost and operating income of the edge computing platform, the maximum operating income and the preset expected delay value are set as the constraint objectives.

[0153] The computing resource allocation strategy based on the particle swarm algorithm simulates the movement and cooperation of individual particles in the search space to calculate the optimal solution for the constraint target;

[0154] Based on the optimal solution, all computing tasks in the to-be-processed computing task list are allocated computing resources according to the to-be-allocated computing resources of the edge computing platform.

[0155] In the case that there is a target computing task with execution completed on the edge server, the computing resource allocation strategy based on the particle swarm algorithm is used to re-allocate computing resources for all computing tasks in the to-be-processed computing task list.

[0156] Specifically, the maximization of operating income and the preset delay expectation value are taken as constraint targets, which involves improving the economic benefits of the edge computing platform as much as possible while meeting the task delay requirement.

[0157] The computing resource allocation strategy based on the particle swarm algorithm (PSORA) is used to solve this optimization problem. The particle swarm algorithm is a global optimization algorithm that finds the optimal solution by simulating the movement and cooperation of individual particles in the search space, and is used to re-allocate computing resources for new tasks or completed tasks within the edge server to improve the execution income of the tasks. Experimental results show that the particle swarm method has a calculation time of 1s-3.3s when the number of tasks is between 20 and 100; when the number of tasks is 100, the energy consumption is only 4107J; when the task reach rate reaches 100%, the time delay is only 12.5ms; the particle swarm method has short task allocation calculation time, low energy consumption, and short data transmission time delay, and can better meet the needs of practical applications.

[0158] After finding the optimal solution, all computing tasks in the to-be-processed computing task list are allocated computing resources according to the to-be-allocated computing resources of the edge computing platform. This includes re-allocating computing resources for all computing tasks in the to-be-processed computing task list in the case that there is a target computing task with execution completed on the edge server. And when the task execution is completed, a heuristic algorithm is executed to re-adjust the computing resource allocation strategy of the server.

[0159] Wherein, under the condition that the task quantity remains unchanged, the cost decreases first and then increases with the increase of the proportion of tasks allocated to the cloud computing node. The reason for this phenomenon is that although the transmission delay of the edge computing node is small, its processing capacity is limited, so in the task quantity allocation scheme, the cost-effectiveness is not optimal. Therefore, when scheduling computing tasks, the processing capacity and transmission delay of the cloud and edge computing nodes must be considered comprehensively, and by reasonably allocating computing tasks, the advantages of cloud-edge collaboration can be fully utilized, the resource utilization rate can be improved, and the energy consumption can be reduced.

[0160] In this embodiment, by adopting a computing resource allocation strategy based on a particle swarm algorithm, the edge computing platform can effectively allocate computing resources to all computing tasks in the pending computing task list to achieve the constraint objectives of maximizing operational benefits and meeting the preset latency expectations. This strategy calculates the optimal solution to the constraint objectives by simulating the movement and cooperation of individual particles in the search space. Its task allocation calculation time is short, energy consumption is low, and data transmission latency is short, which can well meet the needs of practical applications. This method can dynamically adapt to changes in the edge computing environment, improve resource utilization efficiency, reduce costs, and enhance user experience.

[0161] The most detailed embodiment of this application is:

[0162] like Figure 3 As shown, Figure 3 This patent describes a computing resource matching system architecture for edge computing platforms. Multiple heterogeneous edge servers are deployed on the edge computing platform. User devices within the edge computing platform's service area can request computing services from the edge computing platform for a fee. User devices generate computing tasks and deliver them in full to the edge computing platform for processing. After receiving the computing tasks uploaded by the user devices, the edge computing platform allocates computing resources for the tasks based on the method described in this patent.

[0163] The computing resource matching method for edge computing platforms is as follows:

[0164] (1) Model the computing resource matching problem for edge computing platforms.

[0165] Define the edge server set deployed in the edge computing platform as S = {S1, S2, ..., S n The system works in time slot mode, and the time set is represented by T = {1,2,...,T}. The task set from the user equipment is represented by R = {r1,r2,...,r m} indicates that these tasks are sorted in ascending order according to the generation time. i You can use tuples <C i ,D i ,t i ,τ i > indicates that C i represents the computational load of the task (quantified using CPU cycles), D i Indicates the data size (measured in bits), t i represents the generation time of the task, τ i Represents the expected delay of the task.

[0166] Upon arrival of a computing task, the edge computing platform needs to assign an edge server for the task and allocate computing resources on the server to execute the task. Define binary variable x i,j to represent whether task r i is scheduled to edge server S j for processing. x i,j = 1 means that task r i is scheduled to edge server S j for processing.

[0167] When the edge server responsible for executing the computing task is determined, the user device offloads the full amount of the computing task to the edge server for execution through a cellular link. Assuming that task r i is scheduled to edge server S j for processing, the transmission delay of computing task r i can be represented as:

[0168]

[0169] where d i,j represents the data rate of the communication link between the user device generating task r i and the target edge server S j , and the value is calculated according to the Shannon formula. Because the edge server is close to the terminal, and it is assumed that the signal propagates in the medium at the speed of light, the propagation delay of the signal in the medium is ignored.

[0170] Define the computing start deadlines of task r i as k i,j and h i,j respectively, and the computing delay can be represented as T com,ij = h i,j -k i,j . When a new task is assigned to edge server S j or a task processing is completed, S j will adaptively adjust the computing resources allocated to each task, so during the execution of task r i , the computing resources allocated to task r j by edge server S i may change, and define to represent the computing resources allocated to task r j by edge server S i at time slot t, and there is:

[0171]

[0172] The total processing delay of task r i can be represented as Td ij = hi,j -t i , which will be consistent with the expected delay τ of the task i The comparison is used as a reference for service quality measurement.

[0173] Edge Server S j Processing Task i The energy consumption can be expressed as:

[0174]

[0175] Where α is the energy loss coefficient, which represents the effective exchange capacitance that depends on the chip architecture.

[0176] The operating costs of edge computing systems mainly include the energy consumption cost and resource occupation cost of task processing. i Dispatch to S j The cost of processing can be expressed as:

[0177] I i,j =i e,j *E i,j +i t,j *T com,ij ;

[0178] The first term on the right side of the formula represents task r i The energy cost of processing, the second item on the right represents the energy consumption of task r i The resource occupation cost of processing, where i e,j Indicates server S j The cost per unit of energy consumption, i t,j Represents the unit time cost of other tasks hosted on edge servers.

[0179] The edge computing platform earns revenue by selling computing services to users. j Computational tasks performed r i , the platform fee can be calculated as follows:

[0180]

[0181] The first term on the right side of the equation represents the server S j The second item on the right represents the basic charge. When the task load is larger or the task is expected to be completed within a shorter delay, a higher additional fee needs to be paid.

[0182] The following definition of the operating revenue of an edge computing platform takes into account not only the platform's operating costs and user fees, but also the limited resources of edge servers, which means the computing services provided by the edge computing platform may not always meet the latency expectations of tasks. To improve user experience, poor QoS is converted into a penalty cost when defining the platform's operating revenue.

[0183] Task r i Scheduling to S j Penalty cost v of processing i,j which can be expressed as:

[0184]

[0185] where v m is the maximum value of the penalty cost, and p, b are parameters to control the penalty cost. When the processing delay of a task is less than the delay expectation of the task, the penalty cost added to the platform operation revenue is 0, while when the processing delay of a task exceeds the delay expectation of the task, the penalty cost added to the platform operation revenue will grow rapidly with the increase of the task processing time until reaching the maximum value.

[0186] In order to maximize the platform operation revenue while meeting the delay expectation of the task as much as possible, the computing resource matching problem for the edge computing platform can be defined as:

[0187]

[0188] s.t.C1:x i,j ∈{0,1};

[0189]

[0190] The computing resource matching problem for the edge computing platform is a mixed integer nonlinear programming problem. In the present application, the PRMRM algorithm is designed to solve the problem.

[0191] (2) According to the computing task attributes and the edge server attributes, a PRMRM two-stage algorithm is designed to solve the computing resource matching optimization problem.

[0192] Since the scheduling strategy of the task is a discrete variable, and the computing resource allocated by the server for the task is a continuous variable, it is challenging to solve them together. The original problem is divided into two sub-problems of task offloading decision and computing resource allocation, and a PRMRM two-stage algorithm is proposed to solve the above two sub-problems in stages. PRMRM performs computing resource matching of computing tasks through two stages. Specifically, when a task arrives, in the first stage, a deep reinforcement learning method is used to make a task scheduling decision task scheduling strategy. In the second stage, when the server executing the task is determined, there may be other tasks that have not been executed on the server. According to the current state of the server and the state of other computing tasks hosted on the server, a heuristic algorithm is used to solve a reasonable computing resource allocation strategy to maximize the processing revenue of the computing tasks hosted on the server. When the task execution on the server is completed, the heuristic algorithm is also executed to re-adjust the computing resource allocation strategy of the server.

[0193] The first stage of the PRMRM algorithm is based on reinforcement learning, which requires pre-training of the reinforcement learning model. Reinforcement learning is a kind of unsupervised machine learning method, in which an agent learns in the interaction with the environment, and obtains the optimal strategy by maximizing the cumulative return.

[0194] The scheduling problem of tasks is modeled as a Markov process (MDP), which can be represented as a 5-tuple where is the state space, is the action space, is the state transition function, is the reward function, and γ is the discount factor.

[0195] In the invention, the state space s i when task r i arrives is defined as s i = (v com , w i , z basic , p unit , p e , i t , i i , C i , D i , t i , τ i ), where v i is the data upload rate of the communication link between the user device generating task r com and all edge servers, w i is a vector representing all server computing resources, k i is a vector describing the load of all servers when r basic arrives, p unit , p e , i t are vectors representing all server pricing variables and cost variables, respectively, and the load of server S i when r j arrives is defined as:

[0196]

[0197] R j represents the set of tasks hosted on server S j , and the server load reflects the load of tasks hosted on the server that have not been completed and the urgency of these tasks to some extent. The computing resources of a server with high load are more strained, and scheduling computing tasks to these servers will further exacerbate the burden on the server, and the execution time and execution cost of the tasks are also likely to be higher.

[0198] The action space of the system is defined as a i = (x i,1 ,…,x i,n ), x i,j ∈{0,1}, represents the scheduling strategy of the task, x i,j =1 indicates that the task is scheduled to the server S j for execution, and vice versa, indicates that the task is not scheduled to S j for execution, and each task will only be scheduled to one server for execution.

[0199] The reward function guides the behavior of the agent, and in the present application, the reward function reflects the income obtained after the completion of a task execution. The reward function should guide the agent to make scheduling decisions with higher income, and at the same time, the reward function should constrain the agent to ensure the QoS of the provided service as much as possible when making scheduling decisions. The design of the reward function is as follows:

[0200] r i = c i,j -I i,j -v i,j ;

[0201] The above execution action is from a strategy π, π is a mapping from the state space to the action space, that is:

[0202] a i = π(s i );

[0203] The goal of the MDP model is to obtain an optimized strategy, that is, after taking the corresponding action according to the strategy in the corresponding state, the expectation of the cumulative reward of the reinforcement learning goal is maximized, that is, to solve:

[0204]

[0205] γ i is the discount factor, indicating the weight of the reward generated by the scheduling decision for each task.

[0206] (3) Training a deep reinforcement learning model to make task offloading decisions.

[0207] In the first stage of the PRMRM algorithm, when the edge computing receives the computing task uploaded by the user equipment, first, according to the task attributes and the current state of each edge server, a deep reinforcement learning model is constructed using the proximal policy optimization (PPO) algorithm, and then the completed deep reinforcement learning model is used to offload the task to a suitable edge server.

[0208] The proximal policy optimization algorithm is an advanced model-free deep reinforcement learning method that uses a clipping mechanism to limit the update step size during policy updates to ensure that the new policy does not deviate too far from the old policy, thereby greatly improving training stability while maintaining learning efficiency. In each round of proximal policy optimization, the current policy π is first used to θ Execute multiple environment interaction sequences and record the state transition sequence i-1 ,a i-1 ,r i-1 ,s i >, where s i-1 Indicates the state when the i-1th task arrives, a i-1 For the action taken, r i11 The reward for executing the i-1th task, until the sequence terminates naturally, is used for policy evaluation and optimization. The advantage function is then calculated based on the collected data. A pruning mechanism is introduced during this process to maintain the locality of the policy update. Gradient ascent is then performed based on the calculated advantage function to update the policy parameters θ, driving the policy toward higher cumulative rewards.

[0209] (4) When a task arrives, computing resources are allocated to the task according to the computing resource allocation strategy based on particle swarm algorithm (PSORA).

[0210] In the second stage of the PRMRM algorithm, after the offloading decision of the computing task is obtained according to the trained proximal policy optimization model, the computing resource allocation strategy based on particle swarm optimization (PSORA) is used to allocate computing resources to the task.

[0211] When tasks are scheduled to a server, the list of tasks hosted on the server changes, and the allocation of computing resources needs to be adjusted to increase the benefits generated by task processing.

[0212] The computing resource allocation problem can be modeled as a constrained optimization problem. The solution space of the optimization problem, i.e., the computing resources allocated by the server to the task, is defined as Among them F ji Represents the computing node S j For task r i The allocated computing resources,are continuous variables, and K represents the maximum number of tasks offloaded to the current server.,The optimization problem is defined as,.

[0213]

[0214]

[0215] ​The optimization problem is a nonlinear programming problem, in the present application, the problem is solved based on a particle swarm optimization algorithm resource allocation strategy (PSORA). The particle swarm optimization algorithm is inspired by the behavior of social groups such as bird flocks and fish schools. It is a global optimization algorithm used to find the optimal solution in a multidimensional space. The basic idea of PSO is to simulate the movement and cooperation of individual particles in the search space to find the optimal solution. It should be noted that when a task on the server is completed, in order to avoid the idle of the computing resources allocated to the computing task, the PSORA needs to be executed to adjust the allocation of computing resources.

[0216] It should be understood that, although each step in the flowchart involved in each embodiment as described above is shown in sequence according to the arrow, these steps are not necessarily executed in the order indicated by the arrow. Unless otherwise specified herein, the execution of these steps is not strictly limited in sequence, and these steps can be executed in other orders. Moreover, at least part of the steps in the flowchart involved in each embodiment as described above can include multiple steps or stages, which are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily sequential, but can be alternately executed with at least part of other steps or steps or stages in other steps.

[0217] Based on the same inventive concept, the embodiments of the present application also provide an edge computing platform-based computing resource matching device for implementing the above-mentioned edge computing platform-based computing resource matching method. The problem-solving implementation scheme provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more edge computing platform-based computing resource matching device embodiments provided below can refer to the limitations of the edge computing platform-based computing resource matching method in the above text, which will not be repeated here.

[0218] In one exemplary embodiment, as shown in Figure 5 An edge computing platform-based computing resource matching device is provided, comprising:

[0219] The acquisition module 502 is configured to acquire a target computing task of a user equipment and upload the target computing task to an edge computing platform.

[0220] The acquisition module 502 is further configured to acquire state information of each edge server in the edge computing platform and a task attribute of the target computing task.

[0221] The task scheduling module 504 is configured to determine a scheduling mode of the target computing task according to the state information of each edge server and the task attribute of the target computing task.

[0222] The task scheduling module 504 is further configured to determine a target edge server for processing a target computing task according to a scheduling manner of the target computing task.

[0223] The obtaining module 502 is further configured to obtain a list of to-be-processed computing tasks of the target edge server, the list of to-be-processed computing tasks including the target computing task and other computing tasks in a hosting state.

[0224] The resource allocation module 506 is configured to maximize processing benefits as a constraint target, and allocate computing resources of an edge computing platform to all computing tasks in the list of to-be-processed computing tasks according to the computing resources.

[0225] In an exemplary embodiment, the task scheduling module 504 is further configured to construct a deep reinforcement learning model, the deep reinforcement learning model including a state space, an action space and a reward function of the edge computing platform; wherein the state information of the edge server includes a data upload rate of a communication link between a user equipment generating the target computing task and the edge server, and a computing resource vector, a real-time load condition vector, a pricing variable vector and a cost variable vector of the edge server; the state space and the reward function are determined according to the state information of the edge server; the action space is determined according to the task attribute of the target computing task; the target edge server is matched as the scheduling manner by using the deep reinforcement learning model, and the target computing task is scheduled to the target edge server.

[0226] In an exemplary embodiment, the task scheduling module 504 is further configured to obtain a weight factor of a reward generated when each target computing task is scheduled to a corresponding edge server; determine a scheduling strategy of the target computing task according to the reward function and the weight factor; traverse a plurality of state transition sequences in the state space, and determine a highest cumulative reward corresponding to the scheduling strategy of the target computing task according to the action space and the state space; and match the target edge server as the scheduling manner according to the highest cumulative reward, and schedule the target computing task to the target edge server.

[0227] In an exemplary embodiment, the task scheduling module 504 is further configured to traverse a plurality of state transition sequences, and calculate an advantage function value corresponding to each server state and a corresponding scheduling action in the action space and the state space; and update the scheduling strategy of the target computing task by a gradient ascent method according to the advantage function value to achieve the highest cumulative reward.

[0228] In an example embodiment, the resource allocation module 506 is configured to, in a case where the target computing task is scheduled to the target edge server, acquire a task latency of the target computing task and an energy consumption of the target edge server for processing the task, the task latency including a transmission latency, a computing latency, and a total task processing latency; acquire an operation cost and an operation benefit of the edge computing platform; the operation cost including an energy consumption cost and a resource occupation cost for processing the task, and the operation benefit including a charging of the edge computing platform to a user and a penalty cost for quality of service conversion; and refer to the task latency of the target computing task, the energy consumption of the edge server for processing the task, the operation cost and the operation benefit of the edge computing platform, maximize the processing benefit as a constraint target, and allocate computing resources to all computing tasks in the to-be-processed computing task list according to the to-be-allocated computing resources of the edge computing platform.

[0229] In an example embodiment, the resource allocation module 506 is configured to, refer to the task latency of the target computing task, the energy consumption of the edge server for processing the task, the operation cost and the operation benefit of the edge computing platform, maximize the operation benefit and a preset latency expectation value as constraint targets; based on a computing power resource allocation strategy of a particle swarm algorithm, calculate the constraint targets to obtain an optimal solution by simulating movement and cooperation of individual particles in a search space; based on the optimal solution, allocate computing resources to all computing tasks in the to-be-processed computing task list according to the to-be-allocated computing resources of the edge computing platform; and in a case where there is a completed target computing task on the edge server, re-allocate computing resources to all computing tasks in the to-be-processed computing task list based on the computing power resource allocation strategy of the particle swarm algorithm.

[0230] The above various modules in the computing resource matching device based on the edge computing platform can be all or partially implemented by software, hardware, and a combination thereof. The above various modules can be embedded in or independent of a processor in a computer device in a hardware form, or can be stored in a memory in the computer device in a software form, so as to be called and executed by a processor to perform operations corresponding to the above various modules.

[0231] In an example embodiment, a computer device is provided, which can be a server, and an internal structure diagram of the computer device can be as shown in Figure 6As shown in the figure. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through the system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capability. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store target computing task data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through network connection. The computer program is executed by the processor to implement a computing resource matching method based on an edge computing platform.

[0232] Those skilled in the art can understand that, Figure 6 The structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different component arrangement.

[0233] In one exemplary embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the above method.

[0234] In one embodiment, a computer readable storage medium is provided, having a computer program stored thereon, and the computer program is executed by the processor to implement the steps of the above method.

[0235] In one embodiment, a computer program product is provided, including a computer program, and the computer program is executed by the processor to implement the steps of the above method.

[0236] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant regulations.

[0237] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. The non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. The volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, the RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., without being limited thereto.

[0238] The technical features of the above embodiments can be combined arbitrarily. In order to make the description simple, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combinations of the technical features do not exist contradictory, they should be considered as the scope of the present application.

[0239] The above-described embodiments are merely illustrative of several embodiments of the present application, and the description is relatively specific and detailed, but should not be understood as a limitation on the scope of the patent. It should be noted that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which are all within the scope of the present application. Therefore, the scope of protection of the present application should be subject to the appended claims.

Claims

1. A computing resource matching method based on an edge computing platform, characterized in that: The method comprises: Obtain the target computing task of the user device and upload the target computing task to the edge computing platform; Obtaining status information of each edge server in the edge computing platform and task attributes of the target computing task; Determining a scheduling method for the target computing task based on the status information of each edge server and the task attributes of the target computing task; Determining a target edge server for processing the target computing task according to a scheduling method of the target computing task; Obtain a list of pending computing tasks of the target edge server, where the list of pending computing tasks includes the target computing task and other computing tasks in a managed state; Taking the maximization of processing benefits as the constraint goal, computing resources are allocated to all computing tasks in the list of pending computing tasks according to the available computing resources of the edge computing platform.

2. The method according to claim 1, characterized in that The determining, based on the status information of each edge server and the task attributes of the target computing task, a scheduling method for the target computing task includes: Building a deep reinforcement learning model, which includes the state space, action space, and reward function of the edge computing platform; The state information of the edge server includes the data upload rate of the communication link between the user device generating the target computing task and the edge server, as well as the computing resource vector, real-time load status vector, pricing variable vector and cost variable vector of the edge server; Determining the state space and the reward function according to the state information of the edge server; determining the action space according to the task attributes of the target calculation task; The deep reinforcement learning model is used to match the target edge server as a scheduling method, and the target computing task is scheduled to the target edge server.

3. The method according to claim 2, characterized in that The using the deep reinforcement learning model to match the target edge server as a scheduling method, and scheduling the target computing task to the target edge server, includes: Obtain the weight factor of the reward generated when each target computing task is dispatched to the corresponding edge server; Determining a scheduling strategy for the target computing task based on the reward function and the weight factor; Traversing a plurality of state transition sequences in the state space, and determining a maximum cumulative reward corresponding to a scheduling strategy for a target computing task based on the action space and the state space; According to the highest cumulative reward, a target edge server is matched as a scheduling mode, and the target computing task is scheduled to the target edge server.

4. The method according to claim 3, characterized in that The traversing multiple state transition sequences in the state space and determining the highest cumulative reward corresponding to the scheduling strategy of the target computing task according to the action space and the state space includes: Traversing multiple state transition sequences, in the action space and the state space, calculating the advantage function value corresponding to each server state and the corresponding scheduling action; According to the advantage function value, the scheduling strategy of the target computing task is iteratively updated through the gradient ascent method to achieve the highest cumulative reward.

5. The method according to claim 1, wherein The method uses maximizing processing revenue as a constraint objective and allocates computing resources to all computing tasks in the list of pending computing tasks based on the pending computing resources of the edge computing platform, including: When the target computing task is scheduled to the target edge server, obtaining a task delay of the target computing task and energy consumption of the target edge server processing the task, wherein the task delay includes transmission delay, computing delay, and total task processing delay; Obtaining the operating costs and operating income of the edge computing platform; wherein the operating costs include the energy consumption cost and resource occupation cost of task processing, and the operating income includes the fees charged by the edge computing platform to users and the penalty cost of service quality conversion; Referring to the task delay of the target computing task, the energy consumption of the edge server processing task, the operating cost and operating income of the edge computing platform, maximizing the processing income is taken as the constraint goal, and computing resources are allocated to all computing tasks in the list of pending computing tasks based on the computing resources to be allocated on the edge computing platform.

6. The method according to claim 5, characterized in that The reference to the task delay of the target computing task, the energy consumption of the edge server processing task, the operating cost and operating income of the edge computing platform, taking the maximization of processing income as a constraint goal, and allocating computing resources to all computing tasks in the list of pending computing tasks according to the to-be-allocated computing resources of the edge computing platform, includes: Taking into account the task latency of the target computing task, the energy consumption of the edge server processing task, the operating cost and operating income of the edge computing platform, maximizing the operating income and the preset delay expectation value are used as constraint objectives; A computing resource allocation strategy based on a particle swarm algorithm calculates the optimal solution for the constraint target by simulating the movement and cooperation of individual particles in the search space; Based on the optimal solution, allocate computing resources to all computing tasks in the pending computing task list according to the computing resources to be allocated on the edge computing platform; When there are completed target computing tasks on the edge server, computing resources are reallocated to all computing tasks in the pending computing task list based on the computing power resource allocation strategy of the particle swarm algorithm.

7. A computing resource matching device based on an edge computing platform, characterized in that: The device comprises: An acquisition module is used to acquire the target computing task of the user device and upload the target computing task to the edge computing platform; The acquisition module is further used to obtain the status information of each edge server in the edge computing platform and the task attributes of the target computing task; A task scheduling module, configured to determine a scheduling method for the target computing task based on the status information of each edge server and the task attributes of the target computing task; The task scheduling module is further used to determine a target edge server for processing the target computing task according to the scheduling mode of the target computing task; The acquisition module is further used to obtain a list of pending computing tasks of the target edge server, where the list of pending computing tasks includes the target computing task and other computing tasks in a managed state; The resource allocation module is used to maximize the processing benefit as a constraint target and allocate computing resources to all computing tasks in the list of pending computing tasks based on the computing resources to be allocated on the edge computing platform.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images