High-efficiency scheduling method, system and medium for heterogeneous computing resources of edge servers

Abstract

The present invention discloses a high-efficiency scheduling method, system and medium for heterogeneous computing resources of edge servers, and the method includes the following steps: by prior calculation, computing tasks are summarized into n types of tasks; the state of each server of the edge server is initialized; a cache area is added in the process of transmitting computing tasks between the terminal and the edge server, and the computing tasks transmitted to the edge server are queued and analyzed; the computing task W of the current cache channel and the current operation and storage state of the server are used as input to construct an IPQC problem, and then the particle swarm optimization method is used to find the optimal solution. The present invention realizes a high-efficiency scheduling method for heterogeneous computing resources of edge servers, which can utilize the limited heterogeneous computing resources of edge servers, greatly improve the utilization rate of computing resources of edge servers, and accelerate the processing speed of edge servers at the same time, so as to more efficiently complete the computing tasks transmitted from terminals to edge servers.

Description

Technical Field

[0001] The present invention belongs to the technical field of cloud computing, and in particular relates to a high-efficiency scheduling method, system and medium for heterogeneous computing resources of an edge server. Background Art

[0002] At present, cloud computing can be said to be a new innovation in the era of data and information after computers and the Internet. It can be said that it is a leap forward in today's information age. In general, cloud computing has strong scalability and demand, and can provide users with a brand new experience. Its core is to coordinate many computer resources together, so that users can obtain unlimited resources through the network, and the resources obtained are not limited by time and space.

[0003] However, as people's demand for services and needs increases, cloud computing services will become less effective in providing services for some time-sensitive needs. Due to the geographical location of cloud servers, services with high timeliness, such as TV series, will have different service speeds in different regions. With the advent of the Internet of Things and 5G, the service model of cloud computing will gradually fail to keep up with the pace of development. The concept of edge computing, also known as fog computing, was proposed immediately. Its purpose is to improve the computing power of servers and deploy servers on the "edge" of the "cloud". When users use the resources of cloud servers, they can directly extract resources from edge servers. Cloud servers mainly play the role of coordinating and planning multiple edge servers. They can also collect some logs and other data from all edge servers for analysis, build models, and play a supervisory role.

[0004] However, in the current cloud-edge-end collaborative integration structure, in order to meet the real-time requirements of intelligent applications, there will be a great waste of heterogeneous computing resources with edge servers. At the same time, as terminals become more and more diverse, heterogeneous computing tasks are also increasing. If these heterogeneous computing tasks cannot be handled well, they may even lead to the inability to complete the tasks in time and feedback to the terminals. In the traditional cloud-edge-end collaborative integration structure, the communication process between the terminal and the edge node, the computing tasks are transmitted from the terminal to the edge node for computing in the order of "first send, first process", which will lead to the waste of computing resources without the best utilization, and sometimes even cause the subsequent tasks to fail to be completed on time. Because only considering the allocation of a single task in the current server state, it will always be impossible to judge whether the subsequent computing tasks require more computing resources of some current servers than the current computing tasks, and it is also impossible to guarantee that all computing tasks can meet the real-time requirements. If all computing tasks are to be considered, it will fall into the difficult problem of hidden Markov chains. Summary of the invention

[0005] The main purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art and to provide a high-efficiency scheduling method, system and medium for heterogeneous computing resources of edge servers, which can utilize the limited heterogeneous computing resources of edge servers, greatly improve the utilization rate of computing resources of edge servers, and accelerate the processing speed of edge servers, so as to more efficiently complete the computing tasks transmitted from terminals to edge servers.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] On the one hand, the present invention proposes a high-efficiency scheduling method for heterogeneous computing resources of edge servers, comprising the following steps:

[0008] Through a priori calculation, the computing tasks are classified into n types of tasks, and the average computing time and energy consumption index required for each task on each edge server are calculated, and the weighted addition is obtained to get the total k;

[0009] Initialize the status of each server on the edge server;

[0010] Add a cache area during the transmission of computing tasks between the terminal and the edge server to queue and analyze the computing tasks transmitted to the edge server;

[0011] Taking the computing task W of the current cache channel and the current operation and storage status of the server as input, the IPQC problem is constructed, and then the particle swarm optimization method is used to find the optimal solution.

[0012] Preferably, the server includes a CPU, a GPU and an FPGA, each server is of the same type, and the score of each task running on each server is expressed as k n1 , k n2 , k n3 , where n1 is the CPU, n2 is the GPU, and n3 is the FPGA. Then the applications of all M terminals are composed of F types of tasks and can be expressed as:

[0013] F={(c1, d1), (c2, d2), (c3, d3), (c4, d4),...(c m , d m );}

[0014] Where c and d represent the calculation and data required for a certain task respectively.

[0015] Preferably, the prior calculation includes labeling the task type and performing normalization and structuring preprocessing operations on the corresponding data.

[0016] Preferably, the size of the cache area is W, and the calculation formula of W is as follows:

[0017] First, calculate the average task expectations of three types of servers: CPU, GPU, and FPGA: L c , L g , L f ,

[0018]

[0019] in, μ i is the service capability, k i is the number of channels, i∈(g,c,f);

[0020] The size of the buffer is:

[0021]

[0022] W is an integer and is rounded down, satisfying 0<W<min(T g , T c , T f ), where T i , i∈(g,c,f) is the maximum capacity of each server;

[0023] When W does not satisfy the constraint condition, W=W*θ, where 0<θ<1 until the constraint is satisfied.

[0024] Preferably, the initialization is specifically:

[0025] The edge manager node randomly assigns temporary tasks to the servers to establish communication between the servers. The temporary tasks will be replaced by the tasks transmitted from the terminals. The number of temporary tasks set for the CPU, GPU, and FPGA servers is G respectively. c , G g , G f :

[0026] G c =(T c -W)*γ c ;

[0027] G g =(T g -W)*γ g ;

[0028] G f =(T f -W)*γ f ;

[0029] Here is the T i, i∈(g, c, f) is the maximum capacity of each server, and γ is a learnable and optimized learning rate.

[0030] Preferably, when there are no external computing tasks, temporary tasks need to be randomly assigned to establish communications between different servers in advance.

[0031] Preferably, when all tasks in the cache area are cleared and no computing tasks are temporarily received from the terminal, the edge server will return to its initial state and transmit the operation log to the cloud, and start initializing the status of each server until a new task is received from the cache area. This cycle continues until the edge server is shut down to end the entire process.

[0032] Preferably, the particle swarm optimization method is used to find the optimum, specifically:

[0033] The initial population is calculated by the traversal scanning method; each candidate for resource allocation is regarded as a particle; the fitness of each particle is calculated; the fitness is calculated as follows:

[0034]

[0035] Where n is the total number of computing tasks in the cache, and k is the score of tasks running on the corresponding server, specifically:

[0036] k m =a m ·communicate(Cost)+b m ·execution(Time)+c m ·execution(Consumption)

[0037] Among them, a m +b m +c m =1 and 0<a m <1,0<b m <1,0<c m <1, k m is the score of the task running in m servers; communicate(Cost) is the time consumption of communication between servers, execution(Time) is the execution time of the task, and execution(Consumption) is the energy consumption of the task running in m servers;

[0038] The particles are screened by fitness, the best particles are selected and the top 90% of the particles are saved for the next cycle;

[0039] The speed and position update formula of the particle group is used to update the speed and position. The specific formula is: (x+1)d =v xd +c1(P id -P xd )+c2(P gd -P xd )

[0040] P (x+1)d =P xd +v (x+1)d

[0041] Where v is the speed, c1, c2 are the individual learning factor and social learning factor in the range of [0, 4]. id is the best past position of the current particle in dimension d, P gd is the optimal position of the entire population in dimension d, P xd is the position of the current particle in the d dimension;

[0042] The speed and position update operations of the optimized particles are iterated until the preset conditions are met, and finally the optimal particles and the optimal solution are obtained.

[0043] On the other hand, the present invention proposes a high-efficiency scheduling system for heterogeneous computing resources of edge servers, which is applied to the high-efficiency scheduling method for heterogeneous computing resources of edge servers, including a priori calculation module, initialization module, cache module and solution module;

[0044] The a priori calculation module is used to classify the computing tasks into n types of tasks, and calculate the average computing time and energy consumption index required for each task on each server of various edge servers, and the weighted addition is used to obtain a total of k;

[0045] The initialization module is used to initialize the status of each server of the edge server;

[0046] The cache module is used to add a cache area during the transmission of computing tasks between the terminal and the edge server, and queue and analyze the computing tasks transmitted to the edge server;

[0047] The solution module is used to construct the IPQC problem with the computing task W of the current cache channel and the current operation and storage status of the server as input, and then use the particle swarm optimization method to find the optimal solution.

[0048] Another aspect of the present invention provides a storage medium storing a program, which, when executed by a processor, implements the high-efficiency scheduling method for heterogeneous computing resources of an edge server.

[0049] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0050] (1) The present invention proposes a high-efficiency scheduling method for heterogeneous computing resources of edge servers for cloud-edge-end integration. In the traditional cloud-edge-end collaborative integration structure, a cache area is established during the communication process between the terminal and the edge node. The computing tasks transmitted from the terminal to the edge node are cached according to time periods. The cloud-side scheduling algorithm sorts and schedules all computing tasks in the cache area according to the heterogeneous computing resource status of the current edge node.

[0051] (2) In the present invention, a default maximum value is set for the size of the cache area. As the actual situation changes, the throughput of the cache area will also be adaptively scaled according to the arrangement of the scheduling algorithm. This ensures that the edge server can be maximized in each time period and avoids the problem of hidden Markov chains.

[0052] (3) The data preprocessing of the present invention also allows the heterogeneity of computing tasks to be taken into account by the scheduling algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a flow chart of a high-efficiency scheduling method for heterogeneous computing resources of edge servers of the present invention;

[0054] Figure 2 It is a schematic diagram of the overall structure of the present invention;

[0055] Figure 3 It is a score diagram of each task on each server of the present invention;

[0056] Figure 4 (a)- Figure 4 (c) is a graph of three edge servers randomly allocating temporary tasks according to the present invention;

[0057] Figure 5 It is a schematic diagram of data throughput of the cache channel of the present invention;

[0058] Figure 6 It is a schematic diagram of the cloud system structure of the present invention;

[0059] Figure 7 It is a structural schematic diagram of a high-efficiency scheduling system for heterogeneous computing resources of edge servers of the present invention;

[0060] Figure 8 It is a schematic diagram of the structure of the storage medium of the present invention. DETAILED DESCRIPTION

[0061] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present application.

[0062] Example

[0063] The high-efficiency scheduling method for heterogeneous computing resources of edge servers of the present invention comprehensively considers the heterogeneity of edge servers when considering scheduling strategies and constructing scheduling models, and considers heterogeneous computing resources as a core influencing feature, so that the scheduling method is more efficient when facing heterogeneous computing tasks. Secondly, a dynamic queuing buffer area is designed to allow the heterogeneous computing resources of the edge server to process tasks and data more saturatedly and faster in each time period. When the buffer area is put aside and the tasks are viewed individually, the edge server calculates the incoming task data in the best posture in each time period.

[0064] like Figure 1 , Figure 2 As shown, the high-efficiency scheduling method for heterogeneous computing resources of edge servers in this embodiment specifically includes the following steps:

[0065] Step 1: Before building the model, you need to run all tasks and data or select the most representative tasks and data on various types of servers on the edge server, and record the various operating indicators. Then, based on the weighted combination of various indicators, form a score for each task on each server.

[0066] Furthermore, through a priori calculation, the computing tasks are classified into n types of tasks; and the average computing time, energy consumption and other indicators required for each task on each edge server are calculated, and the weighted addition is used to obtain the total score k. Here, three types of servers, CPU, GPU, and FPGA, are used as examples, and each server is of the same type. Then, the score of each task running on each server can be expressed as k n1 , k n2 , k n3 , where n1 is the CPU, n2 is the GPU, and n3 is the FPGA. Then the applications of all M terminals can be composed of F types of tasks and can be expressed as:

[0067] F={(c1, d1), (c2, d2), (c3, d3), (c4, d4),...(c m , d m );}

[0068] Where c and d represent the calculation and data required for a task, respectively. Figure 3 shown.

[0069] Step 2: Initialize the status of each server on the edge server, such as Figure 4 (a)- Figure 4 As shown in (c): Taking CPU, GPU, and FPGA as examples, the edge manager node randomly assigns temporary tasks to these three servers to establish communication between the servers. The temporary tasks will be replaced by the tasks transmitted from the terminal. The number of temporary tasks set for CPU, GPU, and FPGA servers is G respectively. c , G g , G f :

[0070] G c =(T c -W)*γ c ;

[0071] G g =(T g -W)*γ g ;

[0072] G f =(T f -W)*γ f .

[0073] Here is the T i , i∈(g, c, f) is the maximum capacity of each server, and γ is a learnable and optimized learning rate.

[0074] Step 3: Add a cache area during the transmission from the terminal to the edge server to queue and analyze the computing tasks transmitted to the edge server, so as to speed up the efficiency of task execution and determine the size of the cache area W. Here, taking three types of servers, CPU, GPU, and FPGA, as examples, the first step is to calculate the average task expectation of the three types of servers: L c , L g , L f , calculated as follows:

[0075]

[0076] in μ i is the service capability, k i is the number of channels;

[0077] The size of the cache area is:

[0078]

[0079] W is an integer and is rounded down, satisfying 0<W<min(T g , T c , T f ), where T i , i∈(g,c,f) is the maximum capacity of each server.

[0080] Furthermore, when W does not satisfy the constraint condition, W=W*θ, where 0<θ<1 until the constraint is satisfied.

[0081] Step 4: Determine the size of the cache. This part uses batch processing instead of queuing individual tasks, which helps the edge node computing model to efficiently calculate the optimal resource allocation plan for each time period, so that the task completion time for each time period is the shortest and the real-time requirements of intelligent applications are met. Determine that the computing tasks that can be processed by the remaining processing space of each server in the current time period are optimal, such as Figure 5 shown.

[0082] The size of the throughput computing tasks in the cache area in each time period is different. The edge node may not complete all the tasks in the cache area in each time period. In the next time period, the size of the throughput computing tasks in the cache area will be determined based on the amount of computing tasks completed in the previous time period, and the size of the computing tasks completed in each time period is determined based on the optimal decision of the edge server.

[0083] The current task scheduling model of the edge server will allocate tasks for the current computing tasks and the status of each server, and give the optimal solution through the scheduling algorithm. When a computing task that needs to be waited for appears, the corresponding processing priority will be added according to the return time required by the current terminal, and then stored in the cache together with the tasks input from the terminal in the next time period. The edge server completes computing tasks in different time periods, and is executed according to the optimal allocation plan for the current time period. Therefore, from a macro perspective, all time periods are combined into one time period, and the cache will form a dynamic ranking channel.

[0084] Step 5: Take the computing task W of the current cache channel, the current operation and storage status of the three server groups of CPU, GPU, and FPGA as input, construct the IPQC problem, and then use the particle swarm optimization method to find the optimal solution, such as Figure 6 shown.

[0085] Furthermore, when all the tasks in the cache are cleared and no computing tasks are received from the terminal, the edge server returns to the initial state of step 2, transmits the operation log to the cloud, and starts to initialize the status of each server until a new task is received from the cache. This cycle continues until the edge server is shut down to end the entire process.

[0086] Furthermore, the steps to obtain the optimal solution are as follows:

[0087] Step 5.1: The initial population is calculated by a traversal scanning method, wherein each candidate solution for resource allocation is regarded as a particle.

[0088] Step 5.2: Calculate the fitness of each particle; the fitness is calculated as follows:

[0089]

[0090] Where n is the total number of computing tasks in the cache, and k is the score of the tasks running on the corresponding server as described in claim 1, specifically:

[0091] k m =a m ·communicate(Cost)+b m ·execution(Time)+c m ·execution(Consumption)

[0092] Among them, a m +b m +c m =1 and 0<a m <1,0<b m <1,0<c m <1. k m is the score obtained by running the task in m servers; communicate(Cost) is the time consumption for communication between servers, execution(Time) is the execution time of the task, and execution(Consumption) is the energy consumption of the task running in m servers.

[0093] Step 5.3: Screen each particle by its fitness, select the best particle and save the top 90% of the particles for the next cycle.

[0094] Step 5.4: The speed and position update formula of the particle group is used to update the speed and position. The specific formula is:

[0095] v (x+1)d =vxd +c1(P id -P xd )+c2(P gd -P xd )

[0096] P (x+1)d =P xd +v (x+1)d

[0097] Where v is the speed, c1, c2 are the individual learning factor and social learning factor in the range of [0, 4]. id is the best past position of the current particle in dimension d, P gd is the optimal position of the entire population in dimension d, P xd is the position of the current particle in the d dimension.

[0098] Step 5.5: Iterate the speed update and position update operations on the optimized particles until the preset conditions are met, and finally obtain the optimal particle and the optimal solution.

[0099] The present invention optimizes each server as a computing resource to maximize the development and utilization of various computing resources of edge servers. In addition, a flexible and scalable cache channel is designed to optimize the utilization of edge servers in each time period, accelerate and enhance the utilization of computing resources of edge servers.

[0100] like Figure 7 As shown, in another embodiment, a high-efficiency scheduling system for heterogeneous computing resources of edge servers is applied to the high-efficiency scheduling method for heterogeneous computing resources of edge servers, including a priori calculation module, initialization module, cache module and solution module;

[0101] The a priori calculation module is used to classify the computing tasks into n types of tasks, and calculate the average computing time and energy consumption index required for each task on each server of various edge servers, and the weighted addition is used to obtain a total of k;

[0102] The initialization module is used to initialize the status of each server of the edge server;

[0103] The cache module is used to add a cache area during the transmission of computing tasks between the terminal and the edge server, and queue and analyze the computing tasks transmitted to the edge server;

[0104] The solution module is used to construct the IPQC problem with the computing task W of the current cache channel and the current operation and storage status of the server as input, and then use the particle swarm optimization method to find the optimal solution.

[0105] like Figure 8 As shown, in another embodiment of the present application, a storage medium is further provided, storing a program, and when the program is executed by a processor, a high-efficiency scheduling method for heterogeneous computing resources of an edge server is implemented, specifically:

[0106] Through a priori calculation, the computing tasks are classified into n types of tasks, and the average computing time and energy consumption index required for each task on each edge server are calculated, and the weighted addition is obtained to get the total k;

[0107] Initialize the status of each server on the edge server;

[0108] Add a cache area during the transmission of computing tasks between the terminal and the edge server to queue and analyze the computing tasks transmitted to the edge server;

[0109] Taking the computing task W of the current cache channel and the current operation and storage status of the server as input, the IPQC problem is constructed, and then the particle swarm optimization method is used to find the optimal solution.

[0110] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0111] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.

Claims

1. A high-efficiency scheduling method for heterogeneous computing resources of edge servers, characterized in that: The steps include: Through a priori calculation, the computing tasks are classified into multiple task types, and the average computing time and energy consumption index required for each task on each edge server are calculated, and the weighted addition is the sum k; Initialize the status of each server on the edge server; A cache area is added during the transmission of computing tasks between the terminal and the edge server to queue and analyze the computing tasks transmitted to the edge server; Taking the computing task of the current cache channel and the current operation and storage status of the server as input, the IPQC problem is constructed, and then the particle swarm optimization method is used to find the optimal solution; The size of the cache area is W, and the calculation formula of W is as follows: First, calculate the average task expectations of three types of servers: CPU, GPU, and FPGA: L c ,L g ,L f , in, μ i is the service capability, k i is the number of channels, i∈(g,c,f); The size of the cache area is: Will Round down to get W, satisfying 0 <W<min(T g ,T c ,T f ), where T i is the maximum capacity of each server, i∈(g,c,f); When W does not satisfy the constraint, W = W*θ, where 0<θ<1 until the constraint is satisfied; The particle swarm optimization method is used to find the optimal solution, specifically: The initial population is calculated by the traversal scanning method, wherein each candidate solution for resource allocation is regarded as a particle; The fitness of each particle is calculated as follows: Where n is the total number of computing tasks in the cache area, and k is the score of tasks running on the corresponding server, specifically: k m =a m ·communicate(Cost)+b m ·execution(Time)+c m ·execution(Consumption) where a m +b m +c m =1 and 0 m <1,0 m <1,0 <c m <1, k m is the score obtained by running the task in m servers; communicate(Cost) is the time consumption of communication between servers, execution(Time) is the execution time of the task, and execution(Consumption) is the energy consumption of the task executed in m servers;​​ The particles are screened by fitness, the best particles are selected and the top 90% of the particles are saved for the next cycle; The speed and position update formula of the particle group is used to update the speed and position. The specific formula is: v (x+1)d =v xd +c1(P id -P xd )+c2(P gd -P xd ) P (x+1)d =P xd +v (x+1)d c1, c2 are individual learning factors and social learning factors, ranging from [0, 4]; P id is the best past position of the current particle in dimension d, P gd is the optimal position of the entire population in dimension d, P xd is the position of the current particle in the d dimension; The speed and position update operations of the optimized particles are iterated until the preset conditions are met, and finally the optimal particles and the optimal solution are obtained.

2. The high-efficiency scheduling method for heterogeneous computing resources of edge servers according to claim 1 is characterized in that: Each server is a CPU, GPU or FPGA, and each server is of the same type. The score of each task running on each server is expressed as k n1 ,k n2 ,k n3 , where n1 is the CPU, n2 is the GPU, and n3 is the FPGA.

3. The high-efficiency scheduling method for heterogeneous computing resources of edge servers according to claim 1 is characterized in that: The prior calculation includes labeling the task type and performing normalization and structuring preprocessing operations on the corresponding data.

4. The high-efficiency scheduling method for heterogeneous computing resources of edge servers according to claim 1, characterized in that: The initialization is specifically as follows: The edge manager node randomly assigns temporary tasks to the servers to establish communication between the servers. The temporary tasks will be replaced by the tasks transmitted from the terminals. The number of temporary tasks set for the CPU, GPU, and FPGA servers is G respectively. c ,G g ,G f : G c =(T c -W)*γ c ; G g =(T g -W)*γ g ; G f =(T f -W)*γ f ; Here is the T i is the maximum capacity of each server, i∈(g,c,f), γ c , γ g , γ f is the learning rate that can be optimized.

5. The high-efficiency scheduling method for heterogeneous computing resources of edge servers according to claim 1, characterized in that: When there are no external computing tasks, temporary tasks need to be randomly assigned to establish communication between different servers in advance.

6. The high-efficiency scheduling method for heterogeneous computing resources of edge servers according to claim 1, characterized in that: When all the tasks in the cache area are cleared and no computing tasks are received from the terminal, the edge server returns to the initial state and transmits the operation log to the cloud to initialize the status of each server until a new task is received from the cache area. This cycle continues until the edge server is shut down to end the entire process.

7. A high-efficiency scheduling system for heterogeneous computing resources of edge servers, characterized in that: A high-efficiency scheduling method for heterogeneous computing resources of edge servers applied to any one of claims 1-6, comprising a priori computing module, an initialization module, a cache module, and a solution module; The a priori calculation module is used to classify the computing tasks into multiple task types, and calculate the average computing time and energy consumption index required for each task on each server of various edge servers, and the weighted addition is used to obtain the sum k; The initialization module is used to initialize the status of each server of the edge server; The cache module is used to add a cache area during the transmission of computing tasks between the terminal and the edge server, and queue and analyze the computing tasks transmitted to the edge server; The solution module is used to construct the IPQC problem using the computing task of the current cache channel and the current operation and storage status of the server as input, and then use the particle swarm optimization method to find the optimal solution.

8. A storage medium storing a program, characterized in that: When the program is executed by a processor, the high-efficiency scheduling method for heterogeneous computing resources of an edge server as described in any one of claims 1 to 6 is implemented.

Images