Methods, systems, storage media, and devices for scheduling edge heterogeneous computing devices

Abstract

This invention provides a method, system, storage medium, and device for scheduling edge heterogeneous computing power devices. The method includes: scanning edge heterogeneous computing power assets, identifying and registering each edge computing power device and its computing power resources to form a computing power asset list; based on the computing power asset list, analyzing business scenarios and business needs, and predicting changes in business needs of the edge heterogeneous computing power; collecting edge heterogeneous computing power status parameters, and assigning values ​​to the collected status parameters based on changes in business needs to obtain a comprehensive evaluation value; and, based on the obtained comprehensive evaluation value, using the RSDL scheduling algorithm and a red-black tree structure, collaboratively scheduling each edge heterogeneous computing power device. This invention dynamically allocates computing resources based on the real-time business needs of the task and the load status of the edge computing power devices, assigning tasks to nodes with lower loads for computation, thereby reducing overall energy consumption.

Description

Technical Field

[0001] This invention belongs to the field of heterogeneous computing technology, and in particular relates to a method, system, storage medium and device for scheduling edge heterogeneous computing power devices. Background Technology

[0002] With the booming development of edge computing, we have witnessed the emergence of edge computing power in various forms, such as GPU servers, embedded AI boxes, and FPGAs. These diverse computing power structures have brought unprecedented flexibility and efficiency to data processing and computing tasks. However, this increased heterogeneity has also brought huge challenges to resource management and scheduling, making efficient allocation and utilization of computing power an urgent problem to be solved.

[0003] In existing technologies, the existence of heterogeneous computing resources renders traditional resource management and scheduling methods inapplicable. Each type of heterogeneous computing resource has its unique performance and characteristics, making it a technical challenge to quickly and accurately select the most suitable computing resources based on application requirements. Furthermore, the communication and collaborative work between heterogeneous resources require additional mechanisms, which also increases the complexity of management and scheduling.

[0004] Secondly, the needs of edge computing users are showing a trend of diversification and rapid growth. From simple data processing to complex AI calculations, from real-time video monitoring to remote industrial control, various application scenarios have different computing power requirements. However, the current heterogeneous computing resources at the edge and device sides are still in an "island" state, lacking collaborative management, making it difficult to effectively integrate and utilize these resources, thus failing to meet the needs of complex application scenarios. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method for scheduling edge heterogeneous computing devices, the method comprising:

[0006] Perform edge heterogeneous computing power asset scanning, identify and register each edge computing power device and its computing power resources, and form a computing power asset list;

[0007] Based on the computing power asset inventory, analyze business scenarios and business needs to predict changes in business needs for edge heterogeneous computing power.

[0008] Collect edge heterogeneous computing power status parameters, and assign values ​​to the collected status parameters based on changes in business needs to obtain a comprehensive evaluation value;

[0009] Based on the obtained comprehensive evaluation value, and using the RSDL scheduling algorithm and red-black tree structure, the heterogeneous computing power devices at the edge are coordinated and scheduled.

[0010] Furthermore, the collection of edge heterogeneous computing power status parameters includes: collection using an edge computing power gateway and collection using an edge AI computing power server.

[0011] Furthermore, the step of assigning values ​​to the collected state parameters to obtain a comprehensive evaluation value includes: using the entropy weight assignment method to calculate the entropy weight of the state parameters and obtain a comprehensive evaluation value.

[0012] Furthermore, the method of calculating the entropy weights of the state parameters using the entropy weight assignment method includes:

[0013] The collected state parameters are processed using the range standardization method, and the weight of the state parameters is calculated.

[0014] Calculate the entropy value of the state parameters based on the calculated parameter weights;

[0015] Based on the calculated entropy value, calculate the coefficient of variation, and based on the coefficient of variation, calculate the entropy weights of the state parameters.

[0016] Based on the calculated entropy weights and state parameter proportions, the comprehensive evaluation value of the edge heterogeneous computing power device is obtained.

[0017] Furthermore, the step of coordinating the scheduling of various edge heterogeneous computing devices based on the obtained comprehensive evaluation value and using the RSDL scheduling algorithm and red-black tree structure includes:

[0018] A static priority parameter is configured for each edge heterogeneous computing device, wherein the magnitude of the static priority parameter is determined by a comprehensive evaluation value;

[0019] Configure virtual runtime parameters for each edge heterogeneous computing device, wherein the virtual runtime parameters are determined by static priority parameters;

[0020] A red-black tree is used as the queue structure, and each edge heterogeneous computing device is sorted according to the virtual runtime parameter and inserted into the edge heterogeneous computing device ready queue.

[0021] The RSDL scheduling algorithm is used to maintain the ready queue of edge heterogeneous computing devices and coordinate the scheduling of each edge heterogeneous computing device.

[0022] An edge heterogeneous computing power device scheduling system, the system comprising: a computing power device scanning module, a computing power demand prediction module, a computing power device status monitoring module, and a computing power device scheduling module;

[0023] The computing power device scanning module is used to scan edge heterogeneous computing power assets, identify and register each edge computing power device and its computing power resources, and form a computing power asset list.

[0024] The computing power demand prediction module is used to analyze business scenarios and business needs based on the computing power asset list, and predict changes in business demand for edge heterogeneous computing power.

[0025] The computing power equipment status monitoring module is used to collect edge heterogeneous computing power status parameters, and assign values ​​to the collected status parameters based on changes in business needs to obtain a comprehensive evaluation value.

[0026] The computing power device scheduling module is used to coordinate the scheduling of each edge heterogeneous computing power device based on the obtained comprehensive evaluation value and using the RSDL scheduling algorithm and red-black tree structure.

[0027] Furthermore, the computing power device status monitoring module is specifically used to collect data using an edge computing power gateway and an edge AI computing power server.

[0028] Furthermore, the computing power device status monitoring module is specifically used to calculate the entropy weight of the collected status parameters using the entropy weight assignment method, and obtain a comprehensive evaluation value.

[0029] Furthermore, the computing power device status monitoring module is specifically used for

[0030] The collected state parameters are processed using the range standardization method, and the weight of the state parameters is calculated.

[0031] Calculate the entropy value of the state parameters based on the calculated parameter weights;

[0032] Based on the calculated entropy value, calculate the coefficient of variation, and based on the coefficient of variation, calculate the entropy weights of the state parameters.

[0033] Based on the calculated entropy weights and state parameter proportions, the comprehensive evaluation value of the edge heterogeneous computing power device is obtained.

[0034] Furthermore, the computing power device scheduling module is specifically used for:

[0035] A static priority parameter is configured for each edge heterogeneous computing device, wherein the magnitude of the static priority parameter is determined by a comprehensive evaluation value;

[0036] Configure virtual runtime parameters for each edge heterogeneous computing device, wherein the virtual runtime parameters are determined by static priority parameters;

[0037] A red-black tree is used as the queue structure, and each edge heterogeneous computing device is sorted according to the virtual runtime parameter and inserted into the edge heterogeneous computing device ready queue.

[0038] The RSDL scheduling algorithm is used to maintain the ready queue of edge heterogeneous computing devices and coordinate the scheduling of each edge heterogeneous computing device.

[0039] A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of any of the above-described edge heterogeneous computing device scheduling methods.

[0040] An electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0041] Memory, used to store computer programs;

[0042] When a processor executes a program stored in memory, it implements the steps of any of the above-described edge heterogeneous computing device scheduling methods.

[0043] A computer program product includes a computer program / instructions that, when executed by a processor, implement the steps of any of the edge heterogeneous computing device scheduling methods described above.

[0044] Compared with the prior art, the present invention has the following advantages:

[0045] 1. This invention proposes a scheduling method for edge heterogeneous computing devices. By dynamically allocating computing resources based on the real-time business needs of tasks and the load status of edge computing devices, tasks are assigned to nodes with lower loads for computation, thereby reducing overall energy consumption. Furthermore, energy optimization methods, such as task migration, sleep / wake-up, and dynamic voltage / frequency adjustment, can be employed to further reduce energy consumption.

[0046] 2. By employing methods such as task partitioning, task scheduling, and resource management, efficient video intelligent analysis task computation can be achieved on edge nodes, significantly reducing overall computation time and latency.

[0047] 3. Dynamic scheduling can balance the load on nodes, preventing some nodes from becoming overloaded and causing system performance degradation. By optimizing resource allocation and task scheduling, the system throughput can be improved, allowing it to handle more tasks.

[0048] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This invention relates to an edge computing network.

[0051] Figure 2 A flowchart of a method for scheduling edge heterogeneous computing devices according to the present invention is shown.

[0052] Figure 3 A schematic diagram of an edge heterogeneous computing power device scheduling system module of the present invention is shown.

[0053] Figure 4 The red-black tree queue structure set up in one embodiment of the present invention is shown.

[0054] Figure 5 A table showing the relationship between the vruntime parameter and the static_prio parameter of this invention is provided. Detailed Implementation

[0055] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] The edge computing network involved in this invention, such as Figure 1 As shown, its structure includes an edge heterogeneous computing power collaborative scheduling system, an edge computing power gateway, and edge computing power facilities. The edge computing power gateway is an intelligent device integrating high-performance computing, network communication, and data storage functions. It is located at the edge of the network, near the source of data generation, and aims to improve data processing efficiency and response speed. This box possesses powerful computing capabilities, enabling it to perform localized data analysis and processing tasks. It can process and analyze data collected by devices such as sensors and cameras in real time, providing precise data support for fields such as intelligent security, industrial manufacturing, and environmental monitoring.

[0057] Edge computing facilities are AI computing server devices deployed privately at the edge. By integrating high-performance graphics processing units (GPUs), they provide powerful parallel computing capabilities, enabling rapid processing and analysis of large-scale datasets at the network edge. In particular, for complex computing tasks such as deep learning, computer vision, and image processing, they achieve low-latency, high-efficiency intelligent decision-making and data processing.

[0058] The edge heterogeneous computing power collaborative scheduling system is used for unified management and efficient scheduling of edge computing resources such as edge gateway boxes and edge computing servers. Through intelligent computing power allocation and task scheduling strategies, it optimizes resource utilization, improves data processing speed and system responsiveness, and ensures the stable and efficient operation of various edge intelligent applications. This platform can monitor the operating status of edge devices in real time and dynamically adjust computing resources to meet the computing power, latency, and bandwidth requirements of different application scenarios. Through the edge computing power collaborative management platform, users can achieve centralized management, flexible configuration, and intelligent scheduling of edge computing resources, improving business operation efficiency, reducing maintenance costs, and providing strong support for digital transformation and intelligent upgrades.

[0059] In one embodiment of the present invention, such as Figure 2 As shown, a method for scheduling edge heterogeneous computing devices includes the following steps:

[0060] S1. Scan edge heterogeneous computing assets, identify and register each edge computing device and its computing resources, and form a computing asset list.

[0061] S2. Based on the computing power asset list, analyze business scenarios and business needs, predict changes in business needs of edge heterogeneous computing power, and ensure the rational allocation and efficient utilization of computing power resources.

[0062] S3. Collect edge heterogeneous computing power status parameters, and assign values ​​to the collected status parameters based on changes in business needs to obtain a comprehensive evaluation value, thereby monitoring and recording edge heterogeneous computing power issues.

[0063] S3.1 Collect edge computing power status parameters.

[0064] Optionally, status parameters collected by an edge AI computing server can be used, including: device performance parameters, network status parameters, task execution parameters, and energy consumption parameters.

[0065] Status parameters collected using the edge computing gateway include: device connection parameters, data traffic parameters, protocol conversion parameters, and local processing parameters.

[0066] Optionally, device performance parameters include CPU utilization, memory usage, and disk space usage, which reflect the overall performance status of the server.

[0067] Network status parameters, including network bandwidth usage, network latency, and packet loss rate, help to understand the server's network connectivity status.

[0068] Task execution parameters include the number of running tasks, task execution progress, and task resource consumption. These parameters are used to monitor and manage tasks on the server.

[0069] Energy consumption parameters, including the energy consumption of servers, are also important data collection targets, which help achieve green and energy-saving operation and maintenance goals.

[0070] Device connection parameters include the connection status, device type, and device identifier of various devices (such as sensors, PLCs, etc.) that the gateway collects.

[0071] Data traffic parameters, including the number of data packets sent and received, data transmission rate, etc., help to understand the gateway's data processing capabilities;

[0072] Protocol conversion parameters include the fact that gateways typically support the conversion of multiple communication protocols, so it is necessary to collect detailed information on protocol conversion to ensure correct data transmission;

[0073] Local processing parameters include information such as processing results, processing time, and resource consumption collected by the gateway when processing data locally, in order to evaluate the efficiency and effectiveness of local processing.

[0074] S3.2 Assign values ​​to the collected status parameters to obtain a comprehensive evaluation value, thereby monitoring edge computing devices and discovering and recording edge heterogeneous computing power issues.

[0075] In another embodiment of the present invention, the entropy weight of the collected edge computing power state parameters is calculated using the entropy weight assignment method to obtain a comprehensive evaluation value. The steps include:

[0076] S3.2.1 Assume that the edge computing power collects n state parameters every m periods. Its matrix X is represented as:

[0077]

[0078] S3.2.2 The audit standardization method is used to process the state parameters and calculate the parameter weights. For example, the parameter weight P of the j-th parameter state of the i-th edge heterogeneous computing device is calculated. ij Its formula is expressed as:

[0079]

[0080] S3.2.3. Based on the calculated parameter weights, calculate the entropy value of the j-th state parameter, where the entropy value e is calculated. j The formula is expressed as follows:

[0081]

[0082] in, , is the normalization constant.

[0083] S3.2.4 Calculate the coefficient of variation d based on the calculated entropy value. j Its formula is expressed as follows:

[0084]

[0085] S3.2.5 Calculate the entropy weight w based on the coefficient of variation. j Its formula is expressed as follows:

[0086]

[0087] S3.2.6. Based on the calculated entropy weight and parameter proportions, calculate the comprehensive evaluation value Z of the i-th edge computing device (server facility or edge gateway). i , which is expressed as its formula,

[0088]

[0089] S4. Based on the obtained comprehensive evaluation value, and using the RSDL scheduling algorithm and red-black tree structure, coordinate the scheduling of each edge heterogeneous computing device.

[0090] Alternatively, the core of the RSDL algorithm lies in its unique priority handling mechanism and the combination of "small polling" and "large rotation".

[0091] In another embodiment of the present invention, firstly, a static priority parameter (static_prio parameter) is configured for each edge heterogeneous computing power device, and the scheduling platform allows it to execute at this priority, wherein the size of the static_prio parameter is determined by the state parameter entropy weight comprehensive evaluation value Z.

[0092] Then, configure a virtual runtime parameter (vruntime parameter) for each edge heterogeneous computing device to record the cumulative call duration of that node.

[0093] Finally, an Edge Heterogeneous Computing Device Ready Queue (MECQ) is set up in the global system, using red-black trees as the data structure. Red-black trees sacrifice the superior condition of strict height balance, only requiring partial balance, reducing the requirements for rotation, thereby improving performance. Red-black trees can perform search, insertion, and deletion operations in O(log₂n) time complexity.

[0094] Optionally, the edge heterogeneous computing device with the smallest vruntime parameter value will be the next to run, in order to ensure that the cumulative runtime of each task is distributed as evenly as possible.

[0095] In another embodiment of the invention, a typical round-robin algorithm is used to rotate among the highest-priority edge heterogeneous computing devices. When an edge heterogeneous computing device exhausts its vruntime parameter quota at a given priority level, it will be assigned a new quota, such as the quota assigned to the i-th edge heterogeneous computing device for its next virtual runtime, expressed by the formula:

[0096] vruntime_i = vruntime_i × (weight = 1024) / static_prio_i

[0097] The quota for the `vruntime` parameter is determined by the static priority parameter `static_prio`, such as... Figure 5 As shown, high-priority edge heterogeneous computing devices have a higher entropy weight comprehensive evaluation value Z, a relatively slower growth of the vruntime parameter, a shorter single run time, and a higher call frequency; low-priority edge heterogeneous computing devices have a lower entropy weight comprehensive evaluation value Z, a relatively faster growth of the vruntime parameter, a longer single run time, and a higher call frequency, thus ensuring that even the lowest-priority edge heterogeneous computing devices can eventually be utilized.

[0098] In another embodiment of the invention, such as Figure 4 As shown, firstly, a red-black tree sorted by the vruntime parameter value is maintained through RSDL. All runnable scheduling entities are inserted into the red-black tree in order of vruntime parameter value. Scheduling edge heterogeneous computing power devices with smaller vruntime parameter values ​​are arranged on the left side of the red-black tree. In the next task scheduling, the scheduling entity with the shorter virtual runtime is selected to run.

[0099] In another embodiment of the invention, such as Figure 3 As shown, an edge heterogeneous computing power device scheduling system includes: a computing power device scanning module, a computing power demand prediction module, a computing power device status monitoring module, and a computing power device scheduling module.

[0100] Optionally, the computing power scanning module is used to scan edge heterogeneous computing power assets, identify and register each edge computing power device and its computing power resources, and form a computing power asset list;

[0101] The computing power demand forecasting module is used to analyze business scenarios and business needs based on the computing power asset list, and to predict changes in business demand for edge heterogeneous computing power.

[0102] The computing power equipment detection module is used to collect edge heterogeneous computing power status parameters, and assign values ​​to the collected status parameters based on changes in business needs to obtain a comprehensive evaluation value, thereby monitoring and recording edge heterogeneous computing power issues.

[0103] The computing power device scheduling module is used to coordinate the scheduling of various edge heterogeneous computing power devices based on the obtained comprehensive evaluation value and using the RSDL scheduling algorithm and red-black tree structure.

[0104] Based on the above disclosure, the present invention also provides an electronic device. The electronic device of this embodiment includes at least one processor and at least one storage medium electrically connected to the processor. The storage medium is electrically connected to the processor, wherein the storage medium stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described above.

[0105] Based on the same inventive concept, the present invention also provides a storage medium storing instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method as described above.

[0106] The foregoing description and accompanying drawings fully illustrate embodiments of the invention to enable those skilled in the art to practice them. Other embodiments may include structural and other changes. The embodiments represent only possible variations. Individual components and functions are optional unless explicitly required, and the order of operation may vary. Some portions and features of some embodiments may be included or substituted for portions and features of other embodiments. Embodiments of the invention are not limited to the structures described above and shown in the accompanying drawings, and various modifications and changes may be made without departing from their scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for scheduling edge heterogeneous computing devices, characterized in that, The method includes, Perform edge heterogeneous computing power asset scanning, identify and register each edge computing power device and its computing power resources, and form a computing power asset list; Based on the computing power asset inventory, analyze business scenarios and business needs to predict changes in business needs for edge heterogeneous computing power. Collect edge heterogeneous computing power status parameters, and assign values ​​to the collected status parameters based on changes in business needs to obtain a comprehensive evaluation value; Based on the obtained comprehensive evaluation value, and using the RSDL scheduling algorithm and red-black tree structure, the heterogeneous computing power devices at the edge are coordinated and scheduled. The step of assigning values ​​to the collected state parameters to obtain a comprehensive evaluation value includes: using the entropy weight assignment method to calculate the entropy weight of the state parameters and obtain a comprehensive evaluation value, specifically including: Assuming the edge computing power collects n state parameters every m cycles... Its matrix X is represented as: The range normalization method is used to process the state parameters, and the parameter weight P of the j-th parameter state of the i-th edge heterogeneous computing power device is calculated. ij Its formula is expressed as: in, =1, 2, 3, ..., m; =1, 2, 3, ..., n; Calculate the entropy value e of the j-th state parameter based on the calculated parameter weights. j Among them, the entropy value e is calculated. j The formula is expressed as follows: ； in, , is the normalization constant; Calculate the coefficient of variation d based on the calculated entropy value. j Its formula is expressed as follows: ； Calculate the entropy weight w based on the coefficient of variation. j Its formula is expressed as follows: Based on the calculated entropy weight and parameter proportions, calculate the comprehensive evaluation value Z of the i-th edge computing device. i Its formula is expressed as follows: ； The process of coordinating the scheduling of heterogeneous edge computing devices based on the obtained comprehensive evaluation value and using the RSDL scheduling algorithm and red-black tree structure includes: A static priority parameter is configured for each edge heterogeneous computing device, wherein the magnitude of the static priority parameter is determined by a comprehensive evaluation value; Configure virtual runtime parameters for each edge heterogeneous computing device, wherein the virtual runtime parameters are determined by static priority parameters; A red-black tree is used as the queue structure, and each edge heterogeneous computing device is sorted according to the virtual runtime parameter and inserted into the edge heterogeneous computing device ready queue. The RSDL scheduling algorithm is used to maintain the ready queue of edge heterogeneous computing devices and coordinate the scheduling of each edge heterogeneous computing device.

2. The edge heterogeneous computing power device scheduling method according to claim 1, characterized in that, The collection of edge heterogeneous computing power status parameters includes: collecting edge heterogeneous computing power status parameters using an edge computing power gateway and collecting edge heterogeneous computing power status parameters using an edge AI computing power server.

3. An edge heterogeneous computing power device scheduling system, characterized in that, The system includes: a computing power device scanning module, a computing power demand prediction module, a computing power device status monitoring module, and a computing power device scheduling module; The computing power device scanning module is used to scan edge heterogeneous computing power assets, identify and register each edge computing power device and its computing power resources, and form a computing power asset list. The computing power demand prediction module is used to analyze business scenarios and business needs based on the computing power asset list, and predict changes in business demand for edge heterogeneous computing power. The computing power equipment status monitoring module is used to collect edge heterogeneous computing power status parameters, and assign values ​​to the collected status parameters based on changes in business needs to obtain a comprehensive evaluation value. The computing power device scheduling module is used to coordinate the scheduling of each edge heterogeneous computing power device based on the obtained comprehensive evaluation value and using the RSDL scheduling algorithm and red-black tree structure. Specifically, the computing power equipment status monitoring module is used to calculate the entropy weight of the collected status parameters using the entropy weight assignment method to obtain a comprehensive evaluation value, specifically including: Assuming the edge computing power collects n state parameters every m cycles... Its matrix X is represented as: The range normalization method is used to process the state parameters, and the parameter weight P of the j-th parameter state of the i-th edge heterogeneous computing power device is calculated. ij Its formula is expressed as: in, =1, 2, 3, ..., m; =1, 2, 3, ..., n; Calculate the entropy value e of the j-th state parameter based on the calculated parameter weights. j Among them, the entropy value e is calculated. j The formula is expressed as follows: ； in, , is the normalization constant; Calculate the coefficient of variation d based on the calculated entropy value. j Its formula is expressed as follows: ； Calculate the entropy weight w based on the coefficient of variation. j Its formula is expressed as follows: Based on the calculated entropy weight and parameter proportions, calculate the comprehensive evaluation value Z of the i-th edge computing device. i Its formula is expressed as follows: ； The computing power device scheduling module is specifically used for, A static priority parameter is configured for each edge heterogeneous computing device, wherein the magnitude of the static priority parameter is determined by a comprehensive evaluation value; Configure virtual runtime parameters for each edge heterogeneous computing device, wherein the virtual runtime parameters are determined by static priority parameters; A red-black tree is used as the queue structure, and each edge heterogeneous computing device is sorted according to the virtual runtime parameter and inserted into the edge heterogeneous computing device ready queue. The RSDL scheduling algorithm is used to maintain the ready queue of edge heterogeneous computing devices and coordinate the scheduling of each edge heterogeneous computing device.

4. The edge heterogeneous computing power device scheduling system according to claim 3, characterized in that, The computing power device status monitoring module is specifically used to collect edge heterogeneous computing power status parameters using an edge computing power gateway and to collect edge heterogeneous computing power status parameters using an edge AI computing power server.

5. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the edge heterogeneous computing device scheduling method according to any one of claims 1-2.

6. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a program stored in memory, it implements the steps of the edge heterogeneous computing device scheduling method according to any one of claims 1-2.

7. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instruction is executed by the processor, it implements the steps of the edge heterogeneous computing device scheduling method as described in any one of claims 1-2.

Images