Method for unloading computing task by edge server with timeliness perception

Abstract

The invention discloses a method for unloading a computing task from an edge server by user mobile equipment. The method comprises the following steps of: S1, reading equipment information, task information on equipment and related information of the edge server; s2, the read task information and the edge server information are input into a task distribution algorithm to obtain a task distribution scheme, the task distribution method comprises a long-term average AoI optimization algorithm based on dynamic adjustment, and finally the task distribution scheme and system long-term average AoI are obtained; and S3, the equipment unloads the task to the edge server for execution according to the allocation scheme. According to the method, the load state of the edge server can be balanced, the task unloading strategy is dynamically adjusted by preferentially selecting the edge node which is light in load and provides accurate service in combination with the task calculation requirement, the storage requirement and the load change of the edge server, and single-point overload is avoided.

Description

Technical Field

[0001] The present invention relates to the field of Mobile Edge Computing (MEC), and discloses a timeliness-aware method for offloading computing tasks to multiple edge servers and ensuring timeliness. Background Art

[0002] In recent years, with the rapid development of mobile internet and the Internet of Things (IoT) technologies, the massive amount of data generated by terminal devices has posed a severe challenge to traditional cloud computing models. Data processing based on centralized cloud servers is prone to network congestion and high latency due to long-distance transmission. MEC has emerged as a reliable solution to these problems. By offloading tasks to edge servers deployed near the source, MEC effectively alleviates traffic pressure and significantly reduces service latency. However, existing MEC task scheduling methods still have the following key limitations: MEC systems typically have limited resources, which are shared and competed for by multiple tasks. Scheduling computing tasks within the system has become a challenging problem and has attracted widespread attention. Existing research has primarily focused on computing resource allocation and optimizing task completion time, but has overlooked the core impact of information timeliness on system performance. Outdated information not only reduces service quality but can also lead to significant decision-making errors. Age of Information (AoI), an important metric for measuring information freshness in real-time-sensitive scenarios, directly impacts decision accuracy and system reliability in real-time scenarios such as intelligent driving and the Industrial Internet of Things. Therefore, considering AoI in solving the information timeliness problem in the process of user edge computing task offloading is of great significance and value.

[0003] Numerous researchers have made significant progress in task offloading and resource allocation for mobile edge computing and proposed numerous solutions. Most work focuses on performance optimization (e.g., latency, energy consumption, etc.) in single-edge server scenarios. While scheduling issues in multi-edge server systems are gaining increasing attention, existing research still lacks universal modeling and dynamic optimization of information age. Most work focuses on specific scenarios (e.g., two-server heterogeneous systems) or static allocation strategies, failing to comprehensively consider practical constraints (e.g., service caching, resource dynamics) and the coordinated optimization of long-term average accessibility (AoI). For example, the paper "Data Freshness Optimization Under CAA in the UAV-Aided MECN: A Potential Game Perspective" investigates data freshness optimization in UAV-aided MEC networks from a game-theoretic perspective, experimentally verifying the existence of a Nash equilibrium. However, its game model is limited to single-server scenarios and fails to extend to AoI optimization in multi-server architectures. Furthermore, while AoI is considered, it lacks an effective response mechanism to the dynamics of real-time tasks, resulting in insufficient timeliness guarantees for high-priority tasks.

[0004] The paper "On the Age of Information of a Queuing System with Heterogeneous Servers" rigorously derived a closed-form AoI expression for a two-server queuing system for heterogeneous servers for the first time, providing a theoretical foundation for subsequent research. However, this paper is still limited to the scenario, applicable only to two-server systems and not generalized to multi-server scenarios (N ≥ 3). Although it uses a method combining random geometry and queuing theory to analyze the impact of service rate differences on AoI, it ignores server resource constraints (such as cache and VM resources) and the dependencies between task services. Therefore, the timeliness of task offloading to multiple servers in the MEC environment remains to be solved. Summary of the Invention

[0005] This invention addresses the following issues: research on mobile edge computing in the existing technology is still limited to single-server or simple dual-server scenarios; traditional scheduling algorithms cannot take into account both AoI optimization and resource constraints under the dynamic load of multiple servers, which not only causes delays in the response of high-time-efficiency tasks, but also leads to low utilization of edge resources; existing AoI optimization solutions are mostly limited to specific server configurations (such as a two-server heterogeneous system) and are difficult to generalize to complex multi-server scenarios.

[0006] The present invention proposes a time-sensitive method for offloading computing tasks to edge servers. By constructing a multi-edge server queuing model, the AoI (AoI) metric is deeply integrated into the task scheduling decision-making process, achieving efficient offloading of computing tasks and resource allocation. This method includes reading user mobile device information, task information on the device, and related information about the edge server, inputting the read task information and edge server information into a task allocation method to obtain a task allocation scheme. The task allocation method includes a long-term average AoI optimization algorithm based on dynamic adjustment, and the device offloads tasks to the edge server for execution according to the allocation scheme. The present invention can effectively offload computing tasks from mobile devices to the edge server that provides the services required for the task, and by achieving load balancing between edge servers, minimize the system's long-term average AoI.

[0007] Technical solution:

[0008] A method for offloading computing tasks from a user's mobile device to an edge server is disclosed. The method uses a task allocation algorithm to obtain a task allocation plan, and then offloads the task to the edge server for execution according to the obtained allocation plan. The main steps include:

[0009] Step S1, read device information, task information on the device and related information of the edge server;

[0010] Step S2: Input the read task information and edge server information (including at least the computing and resource requirements corresponding to the services required by the task, the mapping relationship between the edge server and cached services, and the remaining computing and storage resources of the edge server) into a task allocation method to obtain a task allocation plan. The task allocation method includes an optimization algorithm based on dynamically adjusting the long-term average AoI. Finally, the task allocation plan and the long-term average AoI of the system are obtained.

[0011] In step S3, the device offloads the task to the edge server for execution according to the allocation plan.

[0012] The device information and related information of the tasks on the device read in step S1 include but are not limited to the number of tasks, the amount of data in the tasks, the average workload required to execute the tasks (in units of CPU cycles / bit), the service type required by the tasks, etc.; k independent computing tasks to be processed by the user's mobile device are obtained, each of which can be offloaded and scheduled, specifically described as Task = {t1, t2, ..., t k There are N edge servers in the system, and each edge server has a storage capacity of P. n and computing power R n Associated, where n∈{1,2,…,N}; each task can be scheduled to a single container. There are a total of S types of containers in the system, using Container jRepresents the j-th type container, j∈{1,2,…,S}, where r j and p j Respectively represent the computing requirements and storage requirements of this type of container, f j Indicates the operating frequency of this type of container.

[0013] Step S2 describes a long-term average AoI optimization algorithm based on dynamic adjustment. The algorithm inputs a set of tasks and a set of edge servers, and outputs a task allocation plan and the system's long-term average AoI. The algorithm in step S2 includes the following main steps:

[0014] Step S21, initializing the task allocation set Allocation, the current load of each edge server λ_n=[0]*N, and the current service rate of each edge server μ_n=[0]*N;

[0015] Step S22: for each group of tasks that require the same type of service, find candidate edge servers (candidate_servers). If the edge server resources are sufficient (server_n.R n ≥task_i.r and server_n.P n ≥task_i.p), then add it to the candidate edge server list; if there is no available edge server, throw an exception. Among them, task_i.r represents the computing requirements of this type of container, and task_i.p represents the storage requirements of this type of container. Define a set of variables to indicate whether the container type of service s is cached on the edge server n, and introduce the variable x nj ∈{0,1}. Specifically, the binary variable x nj =1 indicates the container instance Container j Deployed on Server n Up, on the contrary x nj = 0. Define a binary variable y nj ∈{0,1} indicates whether the container instance on the edge server is scaled up, where y nj =1 indicates the container instance Container on edge server n j To achieve balanced service rate, vertical scaling was performed, and the number of cores c j Make adjustments, otherwise y nj =0;

[0016] Step S23: assign tasks and adjust load balancing. For each task, select the edge server with the smallest current load in the candidate edge server list and assign the task, and update the edge server resource R n and P n , update the current load of the edge server λ_n+=1;

[0017] Step S24, obtaining a task allocation plan;

[0018] Step S25: After the arrival rates of the edge servers are balanced, the service rates of the edge servers are calculated. For all tasks assigned to each edge server server_n, the processing time is calculated: process_time_n = (task_i.D × task_i.cl) / (f j ×c j ), task_i.D and task_i.cl represent the amount of data input for the task and the average computational workload required to execute the task, respectively. The average processing time of the task on the edge server is defined as avg_process_time_n, and the service rate of each edge server is μ_n = 1 / avg_process_time_n. Calculate the average service rate μ_avg of all edge servers.

[0019] Step S26: For all edge servers, adjust the container computing resources to make the service rate of each edge server balanced;

[0020] In step S27, after the load of each edge server is balanced, the long-term average AoI of the single edge server is calculated. For each edge server, the long-term average AoI of each edge server is calculated:

[0021]

[0022] Where, ρ n =λ_n / μ_n, μ_n represents the service rate of each edge server, ρ n represents the utilization rate of each edge server, and λ_n represents the current load of each edge server, that is, the arrival rate;

[0023] Step S28, using the Gamma distribution approximation calculation, the long-term average AoI distribution is approximated to the Gamma distribution Gamma (k = 2, θ i ), the long-term average AoI events of each edge server are independent of each other, and the joint survival function of the system is:

[0024]

[0025] Using polynomial expansion we can get:

[0026]

[0027] The expression of E[Δ] can be obtained by integrating the above formula:

[0028] E[Δ]=∫0 ∞ P(Δ(t)>x)dx

[0029]

[0030] Using the properties of the Gamma distribution, through the integral formula:

[0031]

[0032] Calculate item by item:

[0033]

[0034] definition Then the expression of E[Δ] of edge multi-server is:

[0035]

[0036] In the symmetric case, all edge servers have the same task arrival rate. It is easy to see that E[Δ] can take an extreme value at this time, and θ1=θ2=…=θ N =θ. We can get:

[0037]

[0038] Substituting into the long-term average AoI expression, we can get:

[0039]

[0040] Since all θ i Similarly, the coefficients of the polynomial expansion are the number of combinations:

[0041]

[0042] Δ1 represents the long-term average AoI of the server under the M / M / 1 queue. We can conclude that when the task arrival rate of all servers is the same, the size of the long-term average AoI of the system is the long-term average AoI of a single server. times.

[0043] Step S29: Output the final task allocation solution and the system's long-term average AoI.

[0044] Beneficial effects of the present invention

[0045] 1) The method described in the present invention can balance the load status of the edge server, combine the task computing requirements and storage requirements with the changes in the edge server load, and dynamically adjust the task offloading strategy by giving priority to edge nodes with lighter loads and providing accurate services to avoid single point overload.

[0046] 2) The method described in the present invention achieves service rate balance by vertically scaling containers, improves system throughput and overall resource utilization, reduces the system's long-term average AoI, and ensures system stability and information validity.

[0047] 3) The method described in this paper models the long-term average AoI based on the Gamma distribution and, combined with the M / M / 1 queuing model, derives a formula for calculating the long-term average AoI for a multi-server system. This method overcomes the limitations of traditional models, more accurately characterizing the timeliness of information in complex scenarios and providing universal optimization theory support for real-time-sensitive systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 The figure is a flow chart of the overall method of the present invention.

[0049] Figure 2 The figure is a detailed flow chart of the method of the present invention. DETAILED DESCRIPTION

[0050] The technical solution of the present invention is described in detail below with reference to the accompanying drawings:

[0051] Combine Figure 1 The present invention proposes a method for offloading computing tasks from a time-sensitive edge server, comprising the following steps:

[0052] Step S1, read device information, task information on the device and related information of the edge server;

[0053] Step S2: Input the read task information and edge server information (including at least the computing and resource requirements corresponding to the services required by the task, the mapping relationship between the edge server and the cached services, and the remaining computing and storage resources of the edge server) into a task allocation method to obtain a task allocation plan. The task allocation method includes a long-term average AoI optimization algorithm based on dynamic adjustment. Finally, the task allocation plan and the long-term average AoI of the system are obtained.

[0054] In step S3, the device offloads the task to the edge server for execution according to the allocation plan.

[0055] Example

[0056] The following combination Figure 2 Let's elaborate on this example. In this example, a user's mobile device needs to process nine tasks. Considering the case where there are three edge servers providing only one type of service, we need to offload the following tasks to the edge servers to achieve the optimal solution of minimizing the system's long-term average AoI. The task information is as follows:

[0057]

[0058] Step S1: Read the device information and related information of the tasks on the device, including the data volume of the task, the average workload required to execute the task (in CPU cycles / bit), etc.; obtain 9 independent computing tasks to be processed by the user's mobile device. Each task can be offloaded and scheduled, specifically described as Task = {t1, t2, t3, t4, t5, t6, t7, t8, t9,}, and each task can be scheduled to a single container. There are N = 3 edge servers in the system, and each edge server has a storage capacity P n and computing power R n , where n∈{1,2,3}; each edge server is configured with 16 CPUs and 32GB of memory. In this system, each edge server caches only one type of container to serve tasks, with the container type's computational requirement (CPU count) r set to 2 and its storage requirement (memory) p set to 4GB. The container's operating frequency f is set to 2.5GHz.

[0059] Step S21, initializing the task allocation set Allocation, the current load of each edge server λ_n = [0, 0, 0], the current service rate of each edge server μ_1 = μ_2 = μ_3 = 0;

[0060] Step S22: In this system, each edge server only caches one type of container to provide services for the task. Then, for the task group tasks: find candidate edge servers candidate_servers. If the server resources are sufficient (server_n.R n ≥task_i.r and server_n.P n ≥task_i.p), then add it to the candidate edge server list; if there is no available edge server, throw an exception. Among them, task_i.r represents the computing requirements of this type of container, and task_i.p represents the storage requirements of this type of container. It is easy to see that at this time all edge servers have sufficient resources to cache the containers corresponding to the services required by the task, so all edge servers are added to candidate_servers;

[0061] Define a set of variables to indicate whether the container type of service s is cached on the edge server n, and introduce the variable x nj ∈{0,1}. Specifically, the binary variable x nj =1 indicates the container instance Container j Deployed on Server n Up, on the contrary x nj = 0. Define a binary variable y njy e {0,1} indicates whether the container instance on the server is vertically scaled, where y nj = 1 indicates that the container instance Container j is vertically scaled to achieve service rate balance, the number of cores c j is adjusted, otherwise y nj = 0; at this time, x nj of each server is 1;

[0062] Step S23, allocate tasks and adjust load balancing, for each task task, select the edge server with the smallest current load in the candidate edge server list and allocate the task, and update the edge server resource R n = 16-2 = 14 and P n = 32-4 = 28GB, update the current load of the edge server λ_n = [3,3,3];

[0063] Step S24, the task allocation scheme is as follows:

[0064]

[0065] Step S25: After the arrival rates of the edge servers are balanced, the service rate of each edge server is calculated. For all tasks assigned to each edge server server_n, the processing time is calculated: process_time_i = (task_i.D × task_i.cl) / (f × r), where task_i.D and task_i.cl represent the amount of data input for the task and the average computing workload required to execute the task, respectively. The average processing time of the task on the edge server is defined as avg_process_time_n, and the processing time of each edge server is obtained. The service rate is μ_n = 1 / avg_process_time_n. For Server_1, the processing time of task_1 is (5Mb×250cycles / bit) / (2.5GHz×2)=0.25s, the processing time of task_4 is (4Mb×200cycles / bit) / (2.5GHz×2)=0.16s, and the processing time of task_7 is (30Mb×1000cycles / bit) / (2.5GHz×2)=6s. The average processing time of this edge server is about 2.14s. For Server_2, the processing time of task_2 is (8Mb×300cycles / bit) / (2.5GHz×2)=0.48s, the processing time of task_5 is (25Mb×750cycles / bit) / (2.5GHz×2)=3.75s, and the processing time of task_8 is (10Mb×350cycles / bit) / (2.5GHz×2)=0.7s. The average processing time of this edge server is about 1.64s. For Server_3, the processing time of task_3 is (12Mb×400cycles / bit) / (2.5GHz×2)=0.96s, the processing time of task_6 is (15Mb×500cycles / bit) / (2.5GHz×2)=1.5s, and the processing time of task_9 is (7Mb×275cycles / bit) / (2.5GHz×2)=0.385s. The average processing time of this edge server is approximately 0.95s. The average service rate of all edge servers is calculated from step S24, and μ_1≈0.47, μ_2≈0.61, μ_3≈1.05, and μ_avg=0.71.

[0066] Step S26: For all edge servers, adjust the container computing resources to make the service rate of each edge server balanced; at this time, Variance (μ n)>0.01, the container of Server_1 is vertically expanded, the number of CPU cores is increased by 1, and the remaining computing resources of the edge server are reduced by 1. The processing time of task_1 is (5Mb×250cycles / bit) / (2.5GHz×3)≈0.167s, the processing time of task_4 is (4Mb×200cycles / bit) / (2.5GHz×3)≈0.107s, and the processing time of task_7 is (30Mb×1 000cycles / bit) / (2.5GHz×3)=4s, the average processing time of the edge server is about 1.42s, μ_1≈0.70; the container of Server_2 is vertically expanded, the number of CPU cores is increased by 1, and the remaining computing resources of the server are reduced by 1. The processing time of task_2 is (8Mb×300cycles / bit) / (2.5GHz×3)=0.32s, and the processing time of task_5 is (25Mb×750c Cycles / bit) / (2.5GHz×3)=2.5s, the processing time of task_8 is (10Mb×350cycles / bit) / (2.5GHz×3)=0.467s, the average processing time of the edge server is about 1.10s, μ_2≈0.91; the container of Server_3 is scaled down vertically, the number of CPU cores is reduced by 1, and the remaining computing resources of the edge server are increased by 1. The processing time of task_3 is (12Mb×40 0 cycles / bit) / (2.5GHz×1)=1.92s, the processing time of task_6 is (15Mb×500cycles / bit) / (2.5GHz×1)=3s, and the processing time of task_9 is (7Mb×275cycles / bit) / (2.5GHz×1)=0.77s. The average processing time of the edge server is about 1.90s, μ_3≈0.53; at this time, μ_avg=0.71;

[0067] Through continuous adjustment, it is found that when the number of CPUs allocated to the Server_1 container is 3, the number of CPUs allocated to the Server_2 container is 2, and the number of CPUs allocated to the Server_3 container is 1, μ_1≈0.70, μ_2≈0.61, μ_3≈0.53, and μ_avg≈0.61. It is found that the condition of Variance(μ_n)<0.01 is met, and step S28 is executed;

[0068] Step S27: Calculate the long-term average AoI of a single edge server. The arrival rate of each edge server is 0.3 tasks / sec. For each edge server, calculate the utilization rate ρ n =λ_n / μ_n, the utilization rate of each edge server is: ρ1≈0.4286, ρ2≈0.4918, ρ3≈0.5660. Using the formula: The long-term average AoI of each edge server is calculated as: Δ1≈5.22, Δ2≈5.75, Δ3≈6.61;

[0069] Step S28, using the Gamma distribution approximation calculation, the long-term average AoI distribution is approximated to the Gamma distribution Gamma (k = 2, θ i ), at this time, N = 3, and the final expression of the long-term average AoI is:

[0070]

[0071] in,

[0072]

[0073] Substituting the data obtained in the previous step, we can get θ1=2.61, θ2=2.875, θ3=3.305, and we can calculate θ0≈0.9676. Calculation shows that the long-term average AoI E[Δ] of the system is ≈2.79.

[0074] It is not difficult to see that when When , E[Δ] takes an extreme value, and the corresponding information ages of the three queues are the same (θ1=θ2=θ3=θ). Simplified, we can get:

[0075]

[0076] Now calculate the long-term average AoI of the three edge servers, Δ=(5.22+5.75+6.61) / 3=5.86, and we can conclude that E[Δ]≈2.79 is approximately Δ=5.86. times, reverse reasoning is used to obtain the minimum value of the system's long-term average AoI;

[0077] Step S29: Output the final task allocation solution and the minimum long-term average AoI of the system.

[0078] The specific embodiments described herein are merely illustrative of the spirit of the present invention. Persons skilled in the art may make various modifications, additions, or substitutions to the described specific embodiments without departing from the spirit of the present invention or exceeding the scope of the appended claims.

Claims

1. A method for offloading computing tasks from a user's mobile device to an edge server, characterized by: A task allocation scheme is obtained by using a task allocation algorithm, and tasks are offloaded to edge servers for execution according to the obtained allocation scheme to minimize the long-term average AoI of the system. The steps include: Step S1, read device information, task information on the device and related information of the edge server; Step S2: Inputting the read task information and edge server information into a task allocation algorithm to obtain a task allocation plan. The task allocation algorithm includes a long-term average AoI optimization algorithm based on dynamic adjustment, and finally obtaining a task allocation plan and the system's long-term average AoI. The edge server information includes at least a mapping relationship between tasks and required service types, a mapping relationship between edge servers and cached services, and remaining computing resources and storage resources of the edge servers. In step S3, the device offloads the task to the edge server for execution according to the allocation plan.

2. The method according to claim 1, characterized in that The device information and related information of the tasks on the device are read, including but not limited to the number of tasks, the amount of data in the tasks, the average workload required to execute the tasks, and the service type required by the tasks; k independent computing tasks to be processed by the user's mobile device are obtained, and each task can be offloaded and scheduled; there are N edge servers in the system, and each edge server has a storage capacity P n and computing power R n associated, where n∈{1,2,…,N}; each task can be scheduled to a single container; there are a total of S types of containers in the system, using Container j Represents the j-th type container, j∈{1,2,…,S}, where r j and p j Respectively represent the computing requirements and storage requirements of this type of container, f j Indicates the operating frequency of this type of container.

3. The method according to claim 1, characterized in that The long-term average AoI optimization algorithm based on dynamic adjustment takes a set of tasks and a set of edge servers as input and outputs a task allocation plan and the long-term average AoI of the system. The algorithm in step S2 includes the following steps: Step S21, initializing the task allocation set Allocation, the current load of each edge server λ_n=[0]*N, and the current service rate of each edge server μ_n=[0]*N; Step S22: For each group of tasks that require the same type of service, find candidate edge servers candidate_servers. If the edge server has sufficient resources, add it to the candidate edge server list. If there is no available edge server, throw an exception. Define a set of variables to indicate whether the container type of service s is cached on edge server n. Introduce variable x nj ∈{0,1}; Specifically, the binary variable x nj =1 indicates the container instance Container j Deployed on Server n Up, on the contrary x nj =0; defines a binary variable y nj ∈{0,1} indicates whether the container instance on the edge server is scaled up, where y nj =1 indicates the container instance Container on edge server n j To achieve balanced service rate, vertical scaling was performed, and the number of cores c j Make adjustments, otherwise y nj =0; Step S23: assign tasks and adjust load balancing. For each task, select the edge server with the smallest current load in the candidate edge server list and assign the task, and update the edge server resource R n and P n , update the current load of the edge server λ_n+=1; Step S24, obtaining a task allocation plan; Step S25: After the arrival rates of the edge servers are balanced, the service rates of the edge servers are calculated. For all tasks assigned to each edge server server_n, the processing time process_time_n is calculated. process_time_n = (task_i.D × task_i.cl) / (f j ×c j ), task_i.D and task_i.cl represent the amount of data input for the task and the average computational workload required to execute the task, respectively. The average processing time of the task on the edge server is defined as avg_process_time_n, and the service rate of each edge server is μ_n = 1 / avg_process_time_n. Calculate the average service rate μ_avg of all edge servers. Step S26: For all edge servers, adjust the container computing resources to make the service rate of each edge server balanced; Step S27: After the load of each edge server is balanced, the long-term average AoI of the single edge server is calculated and recorded as Δ n represents the long-term average AoI of the nth single edge server; the solution formula is: Where, ρ n =λ_n / μ_n, μ_n represents the service rate of each edge server, ρ n represents the utilization rate of each edge server, and λ_n represents the current load of each edge server, that is, the arrival rate; Step S28: Use the Gamma distribution to approximate the long-term average AoI distribution to be approximated as: i Gamma distribution, Gamma(k=2,θ i ), the long-term average AoI of a single edge server i is expressed as: E[D i ]=2θ i The long-term average AoI events of each edge server Δ i (t)>x are independent of each other, and the expression of the long-term average AoI of the system is: in, In the symmetric case, the task arrival rate of all edge servers is the same, then E[Δ] can take an extreme value, and θ1=θ2=…=θ N =θ; we get: Substituting the long-term average AoI expression into the equation, we can obtain: When the task arrival rate of all edge servers is the same, the long-term average AoI of the system is: E[Δ] represents the long-term average AoI of the system, and Δ1 represents the long-term average AoI of a single edge server; Step S29: Output the task allocation plan and the system's long-term average AoI.

4. The method according to claim 3, characterized in that Sufficient edge server resources mean: server_n.R n ≥task_i.r and server_n.P n ≥task_i.p In the formula, server_n.R n Represents the computing resources of edge server n; task_i.r represents the computing requirements of this type of container; server_n.P n It represents the storage resources of edge server n; task_i.p represents the storage requirements of this type of container.

Images