MEC task offloading strategy and optimization method with task retry

Abstract

The present invention provides a MEC task offloading strategy and optimization method with task retry, which includes: solving the first average delay of the task receiving service on the local processor, solving the second average delay of the task offloading to the edge server to receive service, and then solving the average delay of the task, abstracting the compromise optimization problem of multiple performance indicators into a multi-objective optimization problem, using the NSGA-III algorithm, using MATLAB software to solve the multi-objective optimization problem, and obtaining the Pareto frontier point that meets the conditions. Aiming at the limitations of edge network equipment, the present invention introduces a task retry mechanism at the edge layer under the premise of ensuring the service quality of end users, which can effectively reduce the task abandonment rate, improve the throughput of the edge system, and assist in selecting a suitable optimal solution to improve system utilization.

Description

Technical Field

[0001] The present invention belongs to the field of edge computing and task offloading technology, and in particular to a MEC task offloading strategy and an optimization method with task retry. Background Art

[0002] In recent years, with the popularity of mobile smart devices and the development of wireless communication technologies such as 5G, Mobile Edge Computing (MEC) has been proposed as an emerging computing paradigm as an extension and supplement to the traditional cloud computing model. The basic idea of ​​mobile edge computing is to offload the computing tasks generated on mobile devices from the cloud to the network edge layer, thereby meeting the low latency requirements of computing-intensive applications such as real-time online games and augmented reality.

[0003] Virtualization technology is one of the key technologies for the realization of mobile edge computing, which can make the allocation of mobile edge computing resources more flexible and improve resource utilization. At the same time, virtualization technology is mature and has little impact on upper-layer applications. Virtualization technology mainly uses two core technologies: server virtualization and application virtualization. Through server virtualization technology, each physical machine with computing power at the edge of the network is virtualized into multiple virtual servers, so that different tasks can run on different virtual servers. Therefore, in the virtualized environment of mobile edge computing, the granularity of management and scheduling becomes virtual machines. There have been many studies on quality of service (QoS) guarantee in MEC, especially the waiting time guarantee of applications or tasks (which can also be regarded as the response time of applications or tasks, referred to as delay) is usually the focus of consideration. Meeting the delay can not only improve the user experience satisfaction, but also improve the efficiency of the application, making it faster and more real-time data processing. Compared with large cloud computing centers, MEC has limited computing resources. Therefore, while meeting the delay as much as possible, the throughput of the edge should also be improved. For delay-sensitive tasks and delay-tolerant tasks, tasks are divided into real-time tasks and non-real-time tasks. For example, in the fields of Internet of Vehicles services and autonomous driving, vehicles need to perceive road conditions, obstacles, driving information of surrounding vehicles, etc. in real time. This information can be quickly calculated and transmitted through MEC computing offloading technology to predict how to drive next. Therefore, these computationally intensive tasks can be defined as real-time tasks. However, tasks in some application scenarios can tolerate longer delays, such as email systems, wildlife tracking networks, and deep space communication networks, so they can be defined as non-real-time tasks. In this way, the quality of service for users is guaranteed based on the characteristics of each type of task.

[0004] A single sequential control flow in a process is a thread, and multiple threads can run concurrently in a process. Multiple threads can complete work according to their own lines and control concurrently. These threads are multiple independent executables in the program. The use of multi-threading mechanism is not to improve operating efficiency, nor is it simply to complete multiple tasks synchronously, but to better improve the efficiency of computer system use and maximize resource utilization efficiency. However, so far, there is a lack of research on the task offloading strategy of mobile edge computing by combining task retry mechanism with multi-threading mechanism. Therefore, based on the above research background, in order to improve the throughput of the edge system, under the premise of ensuring the quality of service for end users, by introducing the task retry mechanism in the edge layer, based on task classification, seek a mobile edge computing task offloading strategy with edge-end collaboration and task retry, and by constructing multiple queuing models to characterize the service process of tasks, it is very urgent and necessary to optimize the system performance by using multi-objective optimization. Summary of the invention

[0005] In view of the defects in the above-mentioned prior art, the present invention proposes a MEC task offloading strategy and optimization method with task retry. The method includes solving the first average delay of the task receiving service on the local processor, solving the second average delay of the task offloading to the edge server to receive service, and then solving the average delay of the task, abstracting the compromise optimization problem of multiple performance indicators into a multi-objective optimization problem, using the NSGA-III algorithm, and using MATLAB software to solve the multi-objective optimization problem to obtain a Pareto frontier point that meets the conditions. In view of the limitations of edge network equipment, the present invention introduces a task retry mechanism at the edge layer while ensuring the quality of service for end users, which can effectively reduce the task abandonment rate, improve the throughput of the edge system, and assist in selecting a suitable optimal solution to improve system utilization.

[0006] The present invention provides a MEC task offloading strategy and optimization method with task retry, which comprises the following steps:

[0007] S1, the first average delay T of the task being served by the local processor loc :According to the analysis results of the M / M / 1 queuing model, the first average delay T of the local processor receiving the service is obtained loc for:

[0008]

[0009] Among them, μ loc The parameters representing the exponential distribution of the time taken by the local processor to service real-time and non-real-time tasks; represents the probability of the task being executed locally and p represents the probability of offloading tasks to the edge layer and 0≤p≤1; λ represents the parameter of the Poisson process to which the arrival of tasks obeys;

[0010] S2, the second average delay T of the task being offloaded to the edge server to receive service edg :When the task is offloaded to the edge layer to receive service, the second average delay of the task is T edg Including the time T of the local send port transmission task trans and the time T that the task receives service from the edge server serve ;

[0011] S21. According to the analysis result of the M / M / 1 queuing model, the time T of the local sending port transmission task is obtained. trans for:

[0012]

[0013] Among them, μ trans The parameter indicating the exponential distribution of the time for the local sending port to transmit real-time tasks and non-real-time tasks;

[0014] S22: According to the ratio of real-time tasks to non-real-time tasks, the time T for the task to be served by the edge server is obtained. serve for

[0015] T serve =α×T nonreal +(1-α)×T real (3)

[0016] Among them, T nonreal It represents the average time required for non-real-time tasks to complete the service and leave from the edge layer; T real represents the average time required for a real-time task to complete the service and leave from the edge layer; α represents the proportion of non-real-time tasks in the arriving task flow 0≤α≤1;

[0017] S23, obtain the second average delay T edg for:

[0018] T edg =T trans +T serve (3);

[0019] S3, average delay T for solving tasks:

[0020]

[0021] Where ρ represents the edge server throughput, and:

[0022] ρ=(pλ+λ′)×(1-P serious) (17)

[0023] Where λ′ represents the retry module throughput; P serious Indicates the probability that the system is heavily loaded, and has:

[0024]

[0025] Among them, π i,j,k It is expressed as the probability distribution of the number of tasks in the retry module being i, the number of real-time tasks in the edge server being j, and the number of non-real-time tasks being k in the steady state;

[0026] S4. The trade-off optimization problem of multiple performance indicators is abstracted into a multi-objective optimization problem: Combining the expectation of maximizing the edge server throughput ρ(p), minimizing the average delay T(p) of the task and the average number of task retries N(p), the multi-objective optimization problem is designed as follows:

[0027]

[0028] Where c represents the c virtual machines divided by the edge layer physical machine through virtualization technology and c≥1; μ1 represents the service rate of non-real-time tasks; d represents the number of retry threads in the retry module and d≥1; θ represents the retry rate of the retry task in the retry thread and 0<θ<+∞;

[0029] S5. Use the NSGA-III algorithm and MATLAB software to solve the multi-objective optimization problem in step S4 and obtain the Pareto frontier point that meets the conditions.

[0030] Furthermore, the average time T required for the non-real-time task in step S22 to arrive at the edge layer and leave after the service is completed is nonreal The calculation of includes the following steps:

[0031] S2211, for non-real-time tasks, when the system load is low, the non-real-time tasks are assigned by the load balancer to the server to receive services and return to the local end after the service is completed. Under low load, the average time T required for non-real-time tasks to arrive at the edge layer and leave after the service is completed nonreal1 Time T′ for server service only nonreal :

[0032]

[0033] S2212: When the system load is heavy, non-real-time tasks will be assigned to the retry module by the load balancer. Under heavy load, the average time T required for non-real-time tasks to complete the service from reaching the edge layer to leaving the service is nonreal2 Including server service time T' nonreal and the time T of the non-real-time task in the retry moduleretry :

[0034] T nonreal2 =T′ nonreal +T retry (4);

[0035] S2213: For non-real-time tasks, the average time T required for the non-real-time tasks to arrive at the edge layer and leave after the service is completed nonreal for:

[0036] T nonreal =(1-P serious )×T nonreal1 +P serious ×T nonreal2 (13).

[0037] Furthermore, the average time T required for the real-time task in step S22 to arrive at the edge layer and leave after the service is completed is real The calculation of includes the following steps:

[0038] S2221. For real-time tasks, when the real-time tasks enter the edge layer, when the system load is low, the real-time tasks are assigned by the load balancer to the server to receive services and return to the local end after the services are completed. The average time T′ required for the real-time tasks to leave the edge layer after the services are completed is low. real for:

[0039]

[0040] Among them, μ2 represents the service rate of real-time tasks;

[0041] S2222. When the system load is heavy, the real-time task is discarded by the server. When the load is low, the average time T″ required for the real-time task to complete the service from reaching the edge layer to leaving the server is real =0;

[0042] S2223: For real-time tasks, the average time T required for the real-time task to arrive at the edge layer and leave after the service is completed real for:

[0043] T real =(1-P serious )×T′ real +P serious ×T″ real (15).

[0044] Preferably, the time T of the non-real-time task in the retry module in step S2212 is retry The calculation of includes the following steps:

[0045] S22121. According to the geometric distribution, the average number of task retries N in the retry module is:

[0046]

[0047] S22122. Consider the retry module as an M / M / c queuing model. The arrival rate of new tasks in the retry module is retry for:

[0048] λ retry =αpλP serious (7);

[0049] S22123. Considering the situation where a task fails to be retried and returns to the retry queue, the expression is listed according to the in-and-out balance principle under the steady state of the system:

[0050] λ′=λ retry +λ′P serious (8);

[0051] Then substitute it into formula (7) to obtain the actual arrival rate of tasks in the retry module, that is, the retry module throughput λ′:

[0052]

[0053] S22124. According to the analysis results of the M / M / c queuing model, the average queue length L in the retry module is obtained. retry for:

[0054]

[0055] S22125. Get the task's stay time T in the retry module sojourn for:

[0056]

[0057] S22126. Obtain the time T of the non-real-time task in the retry module. retry for:

[0058] T retry =N×T sojourn (12).

[0059] Preferably, the arrival of tasks is set to follow a Poisson process with parameter λ (0<λ<+∞), and the time for local processors to serve real-time tasks and non-real-time tasks is set to follow a Poisson process with parameter μ loc (0<μ loc <+∞), the time for local sending port to transmit real-time tasks and non-real-time tasks is subject to the parameter μ trans (0<μ loc<μ trans <+∞) exponential distribution.

[0060] Preferably, considering the different tolerance of tasks to delay, the edge server sets different service rates for real-time tasks and non-real-time tasks, where the service rate for non-real-time tasks is μ1, and the service rate for real-time tasks is μ2 (0<μ1<μ2<+∞).

[0061] Compared with the prior art, the technical effects of the present invention are:

[0062] 1. The present invention designs a MEC task offloading strategy and optimization method with task retry. Aiming at the limitations of edge network equipment, in order to improve the throughput of the edge system, a task retry mechanism is introduced in the edge layer under the premise of ensuring the service quality of end users, and a mobile edge computing task offloading strategy with edge-end collaboration and task retry is proposed; and according to different offloading strategies for different tasks, multiple queuing models are constructed to characterize the service process of the task. By deploying a retry module in the edge layer, seeking services when the system load is heavy can effectively improve the throughput of the edge system.

[0063] 2. The present invention designs a MEC task offloading strategy and optimization method with task retry. By performing steady-state analysis on the proposed method, performance indicators such as the average delay of real-time tasks and non-real-time tasks, the number of non-real-time task retries, and edge server throughput are obtained. The compromise optimization problem of multiple performance indicators is abstracted into a multi-objective optimization problem. The NSGA-III algorithm is used to obtain the Pareto frontier point that meets the conditions. Experiments have proved that introducing a task retry mechanism at the edge layer can effectively reduce the task abandonment rate and improve the edge system throughput and system utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Other features, objects and advantages of the present application will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings.

[0065] Figure 1 It is a flow chart of the MEC task offloading strategy and optimization method with task retry of the present invention;

[0066] Figure 2 It is a schematic diagram of the MEC system architecture with task retry of the present invention;

[0067] Figure 3 This is the variation trend of the edge server throughput of the present invention. DETAILED DESCRIPTION

[0068] The present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It should also be noted that, for ease of description, only the parts related to the relevant invention are shown in the accompanying drawings.

[0069] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0070] Aiming at the energy waste problems in cloud data centers, such as a large amount of idle energy consumption due to idle computing nodes and a large amount of luxury energy consumption due to unreasonable task scheduling, an energy-saving strategy based on double-threshold hysteresis cluster scheduling mechanism is proposed.

[0071] Figure 1 and Figure 2 The MEC task offloading strategy and optimization method with task retry of the present invention are shown, and the method includes the following steps:

[0072] Assume that the arrival of tasks follows a Poisson process with parameter λ (0<λ<+∞), and the time for local processors to serve real-time tasks and non-real-time tasks follows a parameter μ loc (0<μ loc <+∞), the time for local sending port to transmit real-time tasks and non-real-time tasks is subject to the parameter μ trans (0<μ loc <μ trans <+∞). Considering the different tolerance of tasks to delay, the edge server sets different service rates for real-time tasks and non-real-time tasks, where the service rate of non-real-time tasks is μ1 and the service rate of real-time tasks is μ2 (0<μ1<μ2<+∞).

[0073] S1, the first average delay T of the task being served by the local processor loc :According to the analysis results of the M / M / 1 queuing model, the first average delay T of the local processor receiving the service is obtained loc for:

[0074]

[0075] Among them, μ loc The parameters representing the exponential distribution of the time taken by the local processor to service real-time and non-real-time tasks; represents the probability of the task being executed locally and p represents the probability of offloading a task to the edge layer and is 0≤p≤1; λ represents the parameter of the Poisson process to which the arrival of the task obeys.

[0076] S2, the second average delay T of the task being offloaded to the edge server to receive service edg :When the task is offloaded to the edge layer to receive service, the second average delay of the task is T edg Including the time T of the local send port transmission task trans and the time T that the task receives service from the edge server serve .

[0077] S21. According to the analysis result of the M / M / 1 queuing model, the time T of the local sending port transmission task is obtained. trans for:

[0078]

[0079] Among them, μ trans Parameters that indicate the exponential distribution of the time taken by the local send port to transmit real-time tasks and non-real-time tasks.

[0080] S22: According to the ratio of real-time tasks to non-real-time tasks, the time T for the task to be served by the edge server is obtained. serve for

[0081] T serve =α×T nonreal +(1-α)×T real (3)

[0082] Among them, T nonreal It represents the average time required for non-real-time tasks to complete the service and leave from the edge layer; T real It represents the average time required for a real-time task to complete the service and leave from the time it arrives at the edge layer; α represents the proportion of non-real-time tasks in the arriving task flow, 0≤α≤1.

[0083] The average time T required for a non-real-time task to arrive at the edge layer and leave after the service is completed nonreal The calculation of includes the following steps:

[0084] S2211, for non-real-time tasks, when the system load is low, the non-real-time tasks are assigned by the load balancer to the server to receive services and return to the local end after the service is completed. Under low load, the average time T required for non-real-time tasks to arrive at the edge layer and leave after the service is completed nonreal1 Time T′ for server service only nonreal :

[0085]

[0086] S2212: When the system load is heavy, non-real-time tasks will be assigned to the retry module by the load balancer. Under heavy load, the average time T required for non-real-time tasks to complete the service from reaching the edge layer to leaving the service is nonreal2 Including server service time T' nonreal and the time T of the non-real-time task in the retry module retry :

[0087] T nonreal2 =T′ nonreal +T retry (10).

[0088] Time T of non-real-time task in the retry module retry The calculation of includes the following steps:

[0089] S22121. According to the geometric distribution, the average number of task retries N in the retry module is:

[0090]

[0091] S22122. Consider the retry module as an M / M / c queuing model. The arrival rate of new tasks in the retry module is retry for:

[0092] λ retry =αpλP serious (15).

[0093] S22123. Considering the situation where a task fails to be retried and returns to the retry queue, the expression is listed according to the in-and-out balance principle under the steady state of the system:

[0094] λ′=λ retry +λ′P serious (16).

[0095] Then substitute it into formula (7) to obtain the actual arrival rate of tasks in the retry module, that is, the retry module throughput λ′:

[0096]

[0097] S22124. According to the analysis results of the M / M / c queuing model, the average queue length L in the retry module is obtained. retry for:

[0098]

[0099] Among them, π i,j,k It is expressed as the probability distribution of the number of tasks in the retry module being i, the number of real-time tasks in the edge server being j, and the number of non-real-time tasks being k in the steady state.

[0100] S22125. Get the task's stay time T in the retry module sojourn for:

[0101]

[0102] S22126. Get the time T of the non-real-time task in the retry module retry for:

[0103] T retry =N×T sojourn (20).

[0104] Therefore, substituting formula (20) into formula (10) can further obtain the average time T required for non-real-time tasks to arrive at the edge layer and complete the service and leave under heavy load: nonreal2 :

[0105]

[0106] S2213: For non-real-time tasks, the average time T required for non-real-time tasks to arrive at the edge layer and leave after service completion nonreal for:

[0107] T nonreal =(1-P serious )×T nonreal1 +P serious ×T nonreal2 (11).

[0108] The average time T required for a real-time task to complete service and leave from the time it arrives at the edge layer real The calculation of includes the following steps:

[0109] S2221. For real-time tasks, when the real-time tasks enter the edge layer, when the system load is low, the real-time tasks are assigned by the load balancer to the server to receive services and return to the local end after the services are completed. The average time T′ required for the real-time tasks to leave the edge layer after the services are completed is low. real for:

[0110]

[0111] Where μ2 represents the service rate of the real-time task.

[0112] S2222. When the system load is heavy, the real-time task is discarded by the server. When the load is low, the average time T″ required for the real-time task to complete the service from reaching the edge layer to leaving the server is real =0.

[0113] S2223: For real-time tasks, the average time T required for the real-time task to arrive at the edge layer and leave after the service is completed real for:

[0114] T real =(1-P serious )×T′ real +P serious ×T″ real (13).

[0115] S23, obtain the second average delay T edg for:

[0116] T edg =T trans +T serve (4).

[0117] S3, average delay T for solving tasks:

[0118]

[0119] Where ρ represents the edge server throughput, and:

[0120] ρ=(pλ+λ′)×(1-P serious ) (6)

[0121] Where λ′ represents the retry module throughput; P serious Indicates the probability that the system is heavily loaded, and has:

[0122]

[0123] S4. The trade-off optimization problem of multiple performance indicators is abstracted into a multi-objective optimization problem: Combining the expectation of maximizing the edge server throughput ρ(p), minimizing the average delay T(p) of the task and the average number of task retries N(p), the multi-objective optimization problem is designed as follows:

[0124]

[0125] Wherein, c represents the c virtual machines divided by the edge layer physical machine through virtualization technology and c≥1; μ1 represents the service rate of non-real-time tasks; d represents the number of retry threads in the retry module and d≥1; θ represents the retry rate of the retry task in the retry thread and 0<θ<+∞.

[0126] S5. Using the NSGA-III algorithm, use MATLAB software to solve the multi-objective optimization problem in step S4 and obtain the Pareto frontier points that meet the conditions. Since the objectives are contradictory, decision makers can choose the optimal solution configuration from the Pareto frontier points according to their preferences. For example, consider the different preferences between end users and MEC service providers and select the appropriate optimal solution.

[0127] In a specific embodiment, Figure 2 As shown in Figure 1, the MEC system architecture consists of three parts: local end, base station, and edge layer. The edge layer is equipped with a load balancer to execute task offloading strategies.

[0128] On the local side, consider a single user generating a task. The local sending port and local processor are used for task transmission offload and execution processing functions respectively. They each have an infinite capacity cache space. When a task occupies a port or processor, the newly arrived task will queue in the cache space.

[0129] A single physical machine is deployed at the edge layer. Through virtualization technology, the single physical machine is divided into multiple virtual machines to process the incoming tasks. The load balancer in the edge server will distribute the tasks reasonably according to the load of the server. A retry module is deployed in the physical machine to allow tasks to re-seek services when the server load is heavy. The retry module consists of a retry thread and a cache space with unlimited capacity. Tasks will wait for retry in the retry thread. When all retry threads are occupied, newly arrived tasks will wait in the buffer.

[0130] In order to verify the effectiveness of the MEC task offloading strategy and optimization method with task retry proposed in this invention, numerical experiments and simulation experiments are carried out. The experimental results are as follows: Figure 3 As shown in the figure. By comparing the numerical experiment with the simulation experiment, the effectiveness of the strategy and the rationality of the queuing model are verified. By comparing with and without the retry module, it can be seen that the introduction of the retry mechanism can effectively improve the throughput of the edge system.

[0131] The present invention designs a MEC task offloading strategy and optimization method with task retry. Aiming at the limitations of edge network equipment, in order to improve the throughput of the edge system, a task retry mechanism is introduced in the edge layer under the premise of ensuring the service quality of terminal users, and a mobile edge computing task offloading strategy with edge-end collaboration and task retry is proposed; and according to different offloading strategies for different tasks, multiple queuing models are constructed to characterize the service process of the task, and by deploying a retry module in the edge layer, the throughput of the edge system can be effectively improved by seeking services when the system load is heavy; by performing a steady-state analysis on the proposed method, performance indicators such as the average delay of real-time tasks and non-real-time tasks, the number of non-real-time task retries, and the edge server throughput are obtained, and the compromise optimization problem of multiple performance indicators is abstracted into a multi-objective optimization problem. Using the NSGA-III algorithm, the Pareto frontier point that meets the conditions is obtained. Experiments have proved that introducing a task retry mechanism in the edge layer can effectively reduce the task abandonment rate and improve the edge system throughput and system utilization.

[0132] Finally, it should be noted that the above embodiments are only intended to illustrate rather than limit the technical solutions of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the present invention can still be modified or replaced by equivalents. Any modification or partial replacement that does not depart from the spirit and scope of the present invention should be included in the scope of the claims of the present invention.

Claims

1. A MEC task offloading strategy and optimization method with task retry, characterized in that: It includes the following steps: S1, the first average delay T of the task being served by the local processor loc :According to the analysis results of the M / M / 1 queuing model, the first average delay T of the local processor receiving the service is obtained loc for: Among them, μ loc The parameters representing the exponential distribution of the time taken by the local processor to service real-time and non-real-time tasks; represents the probability of the task being executed locally and p represents the probability of offloading tasks to the edge layer and 0≤p≤1; λ represents the parameter of the Poisson process to which the arrival of tasks obeys; S2, the second average delay T of the task being offloaded to the edge server to receive service edg :When the task is offloaded to the edge layer to receive service, the second average delay of the task is T edg Including the time T of the local send port transmission task trans and the time T that the task receives service from the edge server serve ; S21. According to the analysis result of the M / M / 1 queuing model, the time T of the local sending port transmission task is obtained. trans for: Among them, μ trans The parameter indicating the exponential distribution of the time for the local sending port to transmit real-time tasks and non-real-time tasks; S22: According to the ratio of real-time tasks to non-real-time tasks, the time T for the task to be served by the edge server is obtained. serve for T serve =α×T nonreal +(1-a)×T real (3) Among them, T nonreal It represents the average time required for non-real-time tasks to complete the service and leave from the edge layer; T real represents the average time required for a real-time task to complete the service and leave from the edge layer; α represents the proportion of non-real-time tasks in the arriving task flow 0≤α≤1; S23, obtain the second average delay T edg for: T edg =T trans +T serve (4); S3, average delay T for solving tasks: Where ρ represents the edge server throughput, and: ρ=(pλ+λ′)×(1-P serious ) (6) Where λ′ represents the retry module throughput; P serious Indicates the probability that the system is heavily loaded, and has: Among them, π i,j,k It is expressed as the probability distribution of the number of tasks in the retry module being i, the number of real-time tasks in the edge server being j, and the number of non-real-time tasks being k in the steady state; S4. The trade-off optimization problem of multiple performance indicators is abstracted into a multi-objective optimization problem: Combining the expectation of maximizing the edge server throughput ρ(p), minimizing the average delay T(p) of the task and the average number of task retries N(p), the multi-objective optimization problem is designed as follows: Where c represents the c virtual machines divided by the edge layer physical machine through virtualization technology and c≥1; μ1 represents the service rate of non-real-time tasks; d represents the number of retry threads in the retry module and d≥1; θ represents the retry rate of the retry task in the retry thread and 0<θ<+∞; S5. Use the NSGA-III algorithm and MATLAB software to solve the multi-objective optimization problem in step S4 and obtain the Pareto frontier point that meets the conditions.

2. The MEC task offloading strategy and optimization method with task retry according to claim 1 is characterized in that: The average time T required for the non-real-time task in step S22 to arrive at the edge layer and leave after the service is completed nonreal The calculation of includes the following steps: S2211, for non-real-time tasks, when the system load is low, the non-real-time tasks are assigned by the load balancer to the server to receive services and return to the local end after the service is completed. Under low load, the average time T required for non-real-time tasks to arrive at the edge layer and leave after the service is completed nonreal1 Time T′ for server service only nonreal : S2212: When the system load is heavy, non-real-time tasks will be assigned to the retry module by the load balancer. Under heavy load, the average time T required for non-real-time tasks to complete the service from reaching the edge layer to leaving the service is nonreal2 Including server service time T' nonreal and the time T of the non-real-time task in the retry module retry : T nonreal2 =T′ nonreal +T retry (10); S2213: For non-real-time tasks, the average time T required for the non-real-time tasks to arrive at the edge layer and leave after the service is completed nonreal for: T nonreal =(1-P serious )×T nonreal1 +P serious ×T nonreal2 (11)。 3. The MEC task offloading strategy and optimization method with task retry according to claim 1 is characterized in that: The average time T required for the real-time task in step S22 to arrive at the edge layer and leave after the service is completed real The calculation of includes the following steps: S2221. For real-time tasks, when the real-time tasks enter the edge layer, when the system load is low, the real-time tasks are assigned by the load balancer to the server to receive services and return to the local end after the services are completed. The average time T required for the real-time tasks to leave the edge layer after the services are completed is r ' eal for: Among them, μ2 represents the service rate of real-time tasks; S2222. When the system load is heavy, the real-time task is discarded by the server. When the load is low, the average time T″ required for the real-time task to complete the service from reaching the edge layer to leaving the server is real =0; S2223: For a real-time task, the average time T required for the real-time task to complete service and leave from the edge layer real for: T real =(1-P serious )×T′ real +P serious ×T″ real (13)。 4. The MEC task offloading strategy and optimization method with task retry according to claims 1 and 2 is characterized in that: The time T of the non-real-time task in the retry module in step S2212 retry The calculation of includes the following steps: S22121. According to the geometric distribution, the average number of task retries N in the retry module is: S22122. Consider the retry module as an M / M / c queuing model. The arrival rate of new tasks in the retry module is retry for: l retry =αρλP serious (15); S22123. Considering the situation where a task fails to be retried and returns to the retry queue, the expression is listed according to the in-and-out balance principle under the steady state of the system: λ′=λ retry +λ′P serious (16); Then substitute it into formula (7) to obtain the actual arrival rate of tasks in the retry module, that is, the retry module throughput λ′: S22124. According to the analysis results of the M / M / c queuing model, the average queue length L in the retry module is obtained. retry for: S22125. Get the task's stay time T in the retry module sojourn for: S22126. Obtain the time T of the non-real-time task in the retry module. retry for: T retry =N×T sojourn (20)。 5. The MEC task offloading strategy and optimization method with task retry according to claim 1 is characterized in that: Assume that the arrival of tasks follows a Poisson process with parameter λ (0<λ<+∞), and the time for local processors to serve real-time tasks and non-real-time tasks follows a parameter μ loc (0<μ loc <+∞), the time for local sending port to transmit real-time tasks and non-real-time tasks is subject to the parameter μ trans (0<μ loc <μ trans <+∞) exponential distribution.

6. The MEC task offloading strategy and optimization method with task retry according to claim 1 is characterized in that: Taking into account the different tolerance of tasks to delay, the edge server sets different service rates for real-time tasks and non-real-time tasks, where the service rate of non-real-time tasks is μ1, and the service rate of real-time tasks is μ2 (0<μ1<μ2<+∞).

Images