Performance evaluation method of MEC task offloading strategy based on double-layer edge structure

Abstract

The present invention provides a performance evaluation method for an MEC task offloading strategy based on a double-layer edge structure, which includes: solving the average delay and average battery power of delayed tasks in the local and edge layers respectively; using the full probability formula to calculate the average battery power of delayed tasks in the MEC system, and then obtaining the average battery power of the MEC system; weighing the average delay of delayed tasks and the average battery power of the MEC system, and constructing a MEC system evaluation function with multiple inequality constraints; using the general constraint multiplier method PHR based on Lagrange multipliers, using MATLAB to solve the strategy optimization result that makes the MEC system evaluation function reach the minimum value, and obtaining the optimal task offloading probability combination, and the delayed tasks include real-time tasks and non-real-time tasks. The present invention is oriented to real-time tasks and non-real-time tasks, and makes reasonable allocation decisions on the tasks of the MEC system based on the double-layer edge structure, and has the characteristics of low delay, low battery power and efficient offloading performance.

Description

Technical Field

[0001] The present invention belongs to the technical field of multi-access edge computing and task offloading, and in particular to a performance evaluation method for a MEC task offloading strategy based on a double-layer edge structure. Background Art

[0002] With the rapid development of the Internet of Things and 5G communication technology, the penetration rate of mobile devices (MD) is increasing, and the computing power and storage capacity of devices such as smartphones, smart bracelets and shared bicycles are constantly being updated. In addition to conventional non-real-time tasks, more and more real-time tasks have emerged, such as e-health, face recognition, and augmented reality / virtual realization. These applications not only bring a large amount of computing, but also often put forward higher requirements on latency and device battery power. However, these diverse tasks are not matched with the progress of the battery and computing power of mobile devices, and the large amount of computing requirements far exceeds the service capabilities that local mobile devices can provide.

[0003] To solve the above problems, cloud computing has emerged as a new computing paradigm. It concentrates computing, storage, and network management in the cloud and provides elastic computing and storage capabilities. Therefore, offloading tasks to cloud servers can reduce the load on MDs and save critical battery power evaluation. However, traditional cloud computing has many computing limitations: on the one hand, the transmission of large amounts of data requires a lot of bandwidth, which can easily cause congestion with the current capacity of the transmission network; on the other hand, the long distance between MDs and cloud servers can lead to high latency and a lot of jitter; at the same time, MDs, new network applications, and services are developing rapidly, and users have higher requirements for data transmission rate, low latency, and quality of service (QoS). These have led to traditional cloud computing being difficult to meet the needs of many technologies and scenarios. To alleviate the limitations of cloud computing and extend the service life of MDs, mobile edge computing (MEC) is a new computing paradigm that adopts a distributed model, integrates cloud computing and mobile networks, and provides considerable computing resources at the edge of the network. Task offloading in MEC system networks to serve computing-intensive and real-time applications has become a major trend in the development of future communication networks. In the process of task offloading, although MEC has advantages such as low latency, high bandwidth, high energy efficiency and computing sensitivity, the computing resources of MEC servers at the edge of the network are limited compared with cloud computing. However, once large-scale task offloading occurs in the edge computing system, the MEC servers at the edge will be overloaded. Accordingly, the QoS of mobile users will be greatly deteriorated due to service interruption, waiting and extended execution time. To alleviate the pressure and resource constraints at the edge, most edge service providers usually deploy two types of base stations: micro base stations (SBS) and macro base stations (MBS). More importantly, in order to achieve the proximity of mobile users, MEC servers are deployed on MBS and SBS. Therefore, based on the above research background, it is urgent and necessary to seek a performance evaluation method for MEC task offloading strategy based on a double-layer edge structure for real-time and non-real-time tasks, and make reasonable allocation decisions for tasks in the MEC system. Summary of the invention

[0004] In view of the defects in the above-mentioned prior art, the present invention proposes a performance evaluation method for a MEC task offloading strategy based on a double-layer edge structure. The method includes solving the average delay and average battery power of the delayed task in the local and edge layer processing respectively; using the full probability formula to calculate the average battery power of the delayed task in the MEC system, and then obtaining the average battery power of the MEC system; weighing the average delay of the delayed task and the average battery power of the MEC system, and constructing a MEC system evaluation function with multiple inequality constraints; using the general constraint multiplier method PHR based on Lagrange multipliers, using MATLAB to solve the strategy optimization result that makes the MEC system evaluation function reach the minimum value, and obtain the optimal task offloading probability combination, and the delayed task includes real-time tasks and non-real-time tasks. The present invention is oriented to real-time tasks and non-real-time tasks, and makes reasonable allocation decisions on the tasks of the MEC system based on a double-layer edge structure, with the characteristics of low latency, low battery power and efficient offloading performance.

[0005] The present invention provides a performance evaluation method for a MEC task offloading strategy based on a double-layer edge structure, which comprises the following steps:

[0006] S1. Calculate the average delay and average battery power of the delayed tasks processed locally, where the delayed tasks include real-time tasks and non-real-time tasks;

[0007] S2, calculate the average delay and average battery power of the delay task processed in the edge layer;

[0008] S3. Use the full probability formula to calculate the average battery power of the delayed task in the MEC system, and then obtain the average battery power E of the MEC system mec ;

[0009] S4. Weigh the average latency of the delayed tasks and the average battery power of the MEC system, and construct a MEC system evaluation function with multiple inequality constraints

[0010]

[0011] in, It represents the probability of non-real-time tasks being offloaded to the edge layer SBS through the local output port for processing; represents the probability of real-time tasks being offloaded to the edge layer MBS cluster II through the local output port for processing; f 1 represents the impact factor of the average delay of real-time tasks on the system evaluation function; f 2 represents the influence factor of the average battery power of the MEC system on the system evaluation function; f 3 represents the impact factor of the average delay of non-real-time tasks on the system evaluation function, and satisfies f 1 +f2 +f 3 =1; They represent the average delay of real-time tasks and non-real-time tasks in the MEC system respectively; E mec represents the average battery power of the MEC system; 1 ,λ 2 Respectively represent the arrival rates of non-real-time tasks and real-time tasks in the system, and satisfy λ 1 ≥0 and λ 2 ≥0;μ lc represents the service rate of the local server CPU and satisfies μ lc ≥0;μ lo represents the service rate of the local output port and satisfies μ lo ≥0;μ 1 represents the service rate of a virtual machine on SBS and satisfies μ 1 ≥0;μ 2 represents the service rate of a virtual machine on MBS cluster I, and satisfies μ 2 ≥μ 1 ;μ 3 represents the service rate of a virtual machine on MBS cluster II, and satisfies μ 3 ≥μ 2 ;n 1 ,n 2 ,n 3 are all constants;

[0012] S5, using the general constraint multiplier method PHR based on Lagrange multipliers, using MATLAB to solve the MEC system evaluation function in S4 The strategy optimization result that reaches the minimum value obtains the optimal task offloading probability combination

[0013] Further, the step S1 specifically includes the following steps:

[0014] S11, average latency of real-time tasks processed by the local server CPU and average battery level They are:

[0015]

[0016]

[0017] in, Indicates the probability of real-time tasks being processed by the local server CPU; and They represent the execution and idling power of the local server CPU respectively; Indicates the circulation intensity of real-time tasks in the local server CPU;

[0018] S12, average transmission delay of local output port of real-time task and average battery level They are:

[0019]

[0020]

[0021] in, It represents the probability of real-time tasks being transmitted to the edge layer MBS for processing via the local output port; and denote the execution and idling power of the local output port, respectively; is the traffic intensity of the real-time task at the local output port;

[0022] S13, average latency of non-real-time tasks processed by the local server CPU and average battery level for:

[0023]

[0024]

[0025] in, Indicates the probability of non-real-time tasks being processed by the local server CPU; Indicates the circulation intensity of non-real-time tasks in the local server CPU;

[0026] S14, average transmission delay of local output port of non-real-time task and average battery level for:

[0027]

[0028]

[0029] in, It represents the probability of non-real-time tasks being transmitted to the edge layer SBS for processing via the local output port; Indicates the traffic intensity of non-real-time tasks at the local output port.

[0030] Preferably, the step S2 specifically includes the following steps:

[0031] S21, average latency of real-time tasks processed on edge layer MBS and average battery level for:

[0032]

[0033]

[0034] in, represents the probability of real-time tasks being offloaded to the edge layer MBS cluster II through the local output port for processing; and They represent the execution and idling power of a virtual machine on the edge layer MBS, respectively; represents the flow intensity of real-time tasks in cluster II on the MBS; k represents the number of real-time tasks on the edge layer MBS cluster II; L R represents the loss rate of delay-sensitive tasks on edge layer MBS cluster II, and Indicates that the number of real-time tasks on MBS cluster II is n 3 probability;

[0035] S22, average latency of non-real-time tasks processed on edge layer SBS and average battery level for:

[0036]

[0037]

[0038] in, It represents the probability of non-real-time tasks being offloaded to the edge layer SBS through the local output port for processing; and They represent the execution and idling power of a virtual machine on the edge layer SBS, respectively; represents the flow intensity of real-time tasks in MEC servers on SBS; π i,j represents the steady-state probability when the number of non-real-time tasks on the edge layer MBS cluster I is i and the number of non-real-time tasks on the edge layer SBS is j; K represents the system capacity of the edge layer SBS, and the system capacity of the edge layer SBS includes the total number of virtual machines and the cache capacity on the edge layer SBS;

[0039] S23, the probability of non-real-time tasks being offloaded to edge layer MBS cluster I for the second time when edge layer SBS is overloaded for:

[0040]

[0041] Among them, π i,Krepresents the steady-state probability when the number of non-real-time tasks on edge layer MBS cluster I is i and the number of non-real-time tasks on edge layer SBS is K, that is, the steady-state probability that all virtual machines on edge layer SBS are fully running and the cache on SBS is also fully occupied by non-real-time tasks;

[0042] S24, average latency of non-real-time tasks processed on edge layer MBS and average battery level for:

[0043]

[0044]

[0045] in, It represents the probability of non-real-time tasks being offloaded to the edge layer SBS through the local output port for processing; It represents the flow intensity of non-real-time tasks in cluster I on MBS.

[0046] Preferably, the step S3 specifically includes the following steps:

[0047] S31. Use the total probability formula to obtain the average battery power of real-time tasks and non-real-time tasks in the MEC system.

[0048]

[0049]

[0050] S32: Obtain the average battery power E of the MEC system by weighting mec :

[0051]

[0052] Preferably, the average delay of the delay task in step S4 includes the average delay of real-time tasks and non-real-time tasks in the MEC system. With the help of the total probability formula, we can get:

[0053]

[0054]

[0055] Preferably, the MEC system includes a local layer and an edge layer, a task scheduler, a local server CPU and a local output port are deployed in the local layer, and two types of base stations, SBS and MBS, are deployed in the edge layer; a preemption priority strategy is introduced in each of the local server CPU and the local output port, and an edge secondary unloading strategy is introduced in the edge layer.

[0056] Compared with the prior art, the technical effects of the present invention are:

[0057] 1. The present invention designs a performance evaluation method for a MEC task offloading strategy based on a double-layer edge structure. Aiming at the scenario where a large number of tasks are offloaded to the edge layer, and taking into account the pressure of the edge layer, a MEC system architecture based on a double-layer edge structure is built; based on the MEC system architecture, a flexible task offloading strategy is proposed for heterogeneous task requirements, which has the characteristics of low latency, low battery power and efficient offloading performance.

[0058] 2. The present invention designs a performance evaluation method for a MEC task offloading strategy based on a double-layer edge structure. To ensure the real-time performance of real-time tasks, a preemption priority strategy is introduced; to ensure the throughput of non-real-time tasks, an edge secondary offloading strategy is introduced; queuing theory is used to establish a calculation model of tasks at the local layer and the edge layer, and a steady-state analysis is performed on the system, solving performance indicators such as the average delay of tasks in the steady state and the average battery power of the MEC system; a MEC system evaluation function with multiple inequality constraints is constructed, and an optimization scheme for the offloading probability of two types of tasks is given using the PHR algorithm based on Lagrange multipliers. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] 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.

[0060] Figure 1 It is a flow chart of a performance evaluation method of a MEC task offloading strategy based on a double-layer edge structure of the present invention;

[0061] Figure 2 It is a schematic diagram of the structure of the MEC system of the present invention;

[0062] Figure 3 It is a relationship diagram between the average delay of real-time tasks in the MEC system and the unloading probability of real-time tasks under different real-time task arrival rates of the present invention;

[0063] Figure 4 It is the relationship between the average delay of non-real-time tasks in the MEC system and the offloading probability of non-real-time tasks under different non-real-time task arrival rates of the present invention;

[0064] Figure 5a is the lambda of the present invention 1 =0.9,λ 2 =0.9, the battery power of the MEC system changes with the probability of offloading real-time tasks and the probability of offloading non-real-time tasks;

[0065] Figure 5b is the lambda of the present invention1 =0.9,λ 2 =1.0, the battery power of the MEC system changes with the probability of offloading real-time tasks and the probability of offloading non-real-time tasks;

[0066] Figure 5c is the lambda of the present invention 1 =1.0,λ 2 =1.0, the battery power of the MEC system changes with the probability of offloading real-time tasks and the probability of offloading non-real-time tasks;

[0067] Figure 5d is the lambda of the present invention 1 =1.0,λ 2 =1.1, the battery power of the MEC system changes with the probability of offloading real-time tasks and the probability of offloading non-real-time tasks;

[0068] Figure 6 It is the variation trend of the MEC system evaluation function of the present invention along with the offloading probability of real-time tasks and the offloading probability of non-real-time tasks. DETAILED DESCRIPTION

[0069] 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.

[0070] 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.

[0071] Figure 1 The performance evaluation method of the MEC task offloading strategy based on the double-layer edge structure of the present invention is shown, and the method includes the following steps:

[0072] like Figure 2 As shown in the figure, the MEC system includes a local layer and an edge layer. In the local layer, a task scheduler, a local server CPU, and a local output port are deployed. On the local server CPU port and the local output port, a sufficiently large cache is deployed for real-time tasks and non-real-time tasks. The task scheduler is responsible for making offloading decisions. Under the coordination of the task scheduler, newly generated tasks are assigned to the local layer or offloaded to the edge layer through the local output port. To ensure the real-time nature of real-time tasks, a preemption priority strategy is introduced in the local server CPU and the local output port. Real-time tasks can preempt the computing resources of non-real-time tasks. Once the computing resources are preempted by real-time tasks, non-real-time tasks will return to the corresponding cache and queue again.

[0073] In order to reduce the load at the edge layer, two types of base stations, micro base stations SBS and macro base stations MBS, are deployed. On the SBS, a MEC server is deployed to process non-real-time tasks. Through virtualization technology, the MEC server on the SBS is virtualized into multiple virtual machines. All virtual machines in the MEC server share a cache with limited capacity. On the MBS, multiple MEC servers are deployed, and the MEC servers on the MBS are virtualized into two virtual machine groups: Cluster I and Cluster II. All virtual machines in Cluster I share a sufficiently large cache. Real-time tasks offloaded to the MBS are only processed by virtual machines in Cluster II. If all virtual machines in Cluster II are busy, newly arrived real-time tasks will be discarded. In order to ensure the throughput of non-real-time tasks, an edge secondary offloading strategy is introduced. Non-real-time tasks offloaded to the edge layer are first processed in the SBS. If the SBS is overloaded, they will be offloaded to the virtual machines in Cluster I on the MBS for processing.

[0074] Considering that the arrival rates of non-real-time tasks and real-time tasks in the system are λ 1 ,λ 2 , and satisfies λ 1 ≥0 and λ 2 ≥0; the service rate of the local server CPU is μ lc , and satisfy μ lc ≥0; the service rate of the local output port is μ lo , and satisfy μ lo ≥0; the service rate of a virtual machine on SBS is μ 1 , and satisfy μ 1 ≥0; the service rate of a virtual machine on MBS cluster I is μ 2 , and satisfy μ 2 ≥μ 1 ; The service rate of a virtual machine on MBS cluster II is μ 3 , and satisfy μ 3 ≥μ 2 .

[0075] S1. Calculate the average delay and average battery power of the delayed tasks processed locally. The delayed tasks include real-time tasks and non-real-time tasks.

[0076] S11, average latency of real-time tasks processed by the local server CPU and average battery level They are:

[0077]

[0078]

[0079] in, Indicates the probability of real-time tasks being processed by the local server CPU; and They represent the execution and idling power of the local server CPU respectively; Indicates the circulation intensity of real-time tasks on the local server CPU.

[0080] S12, average transmission delay of local output port of real-time task and average battery level They are:

[0081]

[0082]

[0083] in, It represents the probability of real-time tasks being transmitted to the edge layer MBS for processing via the local output port; and denote the execution and idling power of the local output port, respectively; is the traffic intensity of the real-time task at the local output port.

[0084] S13, average latency of non-real-time tasks processed by the local server CPU and average battery level for:

[0085]

[0086]

[0087] in, Indicates the probability of non-real-time tasks being processed by the local server CPU; Indicates the flow intensity of non-real-time tasks on the local server CPU.

[0088] S14, average transmission delay of local output port of non-real-time task and average battery level for:

[0089]

[0090]

[0091] in, It represents the probability of non-real-time tasks being transmitted to the edge layer SBS for processing via the local output port; Indicates the traffic intensity of non-real-time tasks at the local output port.

[0092] S2. Calculate the average latency and average battery power of latency tasks processed at the edge layer.

[0093] S21, average latency of real-time tasks processed on edge layer MBS and average battery level for:

[0094]

[0095]

[0096] in, represents the probability of real-time tasks being offloaded to the edge layer MBS cluster II through the local output port for processing; and They represent the execution and idling power of a virtual machine on the edge layer MBS, respectively; represents the flow intensity of real-time tasks in cluster II on the MBS; k represents the number of real-time tasks on the edge layer MBS cluster II; L R represents the loss rate of delay-sensitive tasks on edge layer MBS cluster II, and Indicates that the number of real-time tasks on MBS cluster II is n 3 probability.

[0097] S22, average latency of non-real-time tasks processed on edge layer SBS and average battery level for:

[0098]

[0099]

[0100] in, It represents the probability of non-real-time tasks being offloaded to the edge layer SBS through the local output port for processing; and They represent the execution and idling power of a virtual machine on the edge layer SBS, respectively; represents the flow intensity of real-time tasks in MEC servers on SBS; π i,j represents the steady-state probability when the number of non-real-time tasks on the edge layer MBS cluster I is i and the number of non-real-time tasks on the edge layer SBS is j; K represents the system capacity of the edge layer SBS, which includes the total number of virtual machines and cache capacity on the edge layer SBS.

[0101] S23, the probability of non-real-time tasks being offloaded to edge layer MBS cluster I for the second time when edge layer SBS is overloaded for:

[0102]

[0103] Among them, π i,K It represents the steady-state probability when the number of non-real-time tasks on edge layer MBS cluster I is i and the number of non-real-time tasks on edge layer SBS is K, that is, the steady-state probability that all virtual machines on edge layer SBS are fully running and the cache on SBS is fully occupied by non-real-time tasks.

[0104] S24, average latency of non-real-time tasks processed on edge layer MBS and average battery level for:

[0105]

[0106]

[0107] in, It represents the probability of non-real-time tasks being offloaded to the edge layer SBS through the local output port for processing; It represents the flow intensity of non-real-time tasks in cluster I on MBS.

[0108] S3. Use the full probability formula to calculate the average battery power of the delayed task in the MEC system, and then obtain the average battery power E of the MEC system mec .

[0109] S31. Use the total probability formula to obtain the average battery power of real-time tasks and non-real-time tasks in the MEC system.

[0110]

[0111]

[0112] S32: Obtain the average battery power E of the MEC system by weighting mec :

[0113]

[0114] S4, weighing the average delay of delayed tasks and the average battery power of the MEC system, including the average delay of real-time tasks and non-real-time tasks in the MEC system With the help of the total probability formula, we can get:

[0115]

[0116]

[0117] Constructing MEC system evaluation function with multiple inequality constraints

[0118]

[0119] Among them, f 1 represents the impact factor of the average delay of real-time tasks on the system evaluation function; f 2 represents the influence factor of the average battery power of the MEC system on the system evaluation function; f 3 represents the impact factor of the average delay of non-real-time tasks on the system evaluation function, and satisfies f 1 +f 2 +f 3 =1.

[0120] S5, using the general constraint multiplier method PHR based on Lagrange multipliers, using MATLAB to solve the MEC system evaluation function in S4 The strategy optimization result that reaches the minimum value obtains the optimal task offloading probability combination

[0121] For the sake of convenience, the above mentioned: represents the probability of real-time tasks being offloaded to the edge layer MBS cluster II through the local output port for processing; represents the probability of offloading non-real-time tasks to the edge layer SBS through the local output port. In the following, they are respectively abbreviated as: the offloading probability of real-time tasks Offloading probability of non-real-time tasks

[0122] In order to evaluate the performance of the task offloading strategy proposed in this invention, numerical experiments and simulation experiments are given. The platform for numerical experiments is MATLAB 2017a, and the platform for simulation experiments is IntelliJ IDEA 2019.3.x64.

[0123] The settings of specific system experimental parameters are shown in Table 1.

[0124]

[0125] Table 1

[0126] like Figure 3 As shown, by setting the arrival rate of non-real-time tasks to λ 1 =1.0, depicting the average delay of real-time tasks in the MEC system under different real-time task arrival rates The offloading probability of real-time tasks The relationship between.

[0127] As Figure 3As shown in the figure, under the same arrival rate of real-time tasks, as the probability of unloading real-time tasks increases, the average delay of real-time tasks in the MEC system first decreases and then increases. This is because when the probability of unloading real-time tasks is small, most of the real-time tasks are processed on the local server CPU, and some real-time tasks are processed on the edge layer MBS cluster II, so the average delay of real-time tasks in the MEC system will decrease; however, as the probability of unloading real-time tasks increases, the average delay of real-time tasks transmitted at the local output port will increase, which will increase the average delay of real-time tasks in the MEC system. Therefore, the average delay of real-time tasks in the MEC system first decreases and then increases.

[0128] exist Figure 3 It can be seen that under the same probability of unloading real-time tasks, as the arrival rate of real-time tasks increases, the average delay of real-time tasks in the MEC system will increase. This is because as the arrival rate of real-time tasks increases, the load of the entire MEC system will increase, and the waiting time of the corresponding real-time tasks in the MEC system will increase, so the average delay of real-time tasks in the MEC system will increase.

[0129] exist Figure 3 It can also be seen that under different arrival rates of real-time tasks, the local and completely offload to the edge tier However, in the case of optimal unloading probability, there will be a minimum average delay. It is not difficult to find that as the arrival rate of real-time tasks increases, the optimal offloading probability also increases, which means that when the local layer load is large, offloading more real-time tasks to the edge layer service can better meet the user's QoS.

[0130] like Figure 4 As shown, by setting the arrival rate of non-real-time tasks to λ 2 =1.0, under different non-real-time task arrival rates, it describes the average delay of non-real-time tasks in the MEC system Offloading probability of non-real-time tasks The relationship between.

[0131] As Figure 4As shown in the figure, under the same arrival rate of non-real-time tasks, as the probability of unloading non-real-time tasks increases, the average delay of non-real-time tasks in the MEC system first decreases and then increases. This is because when the probability of unloading non-real-time tasks is small, most of the non-real-time tasks are processed on the local server CPU, and some of the non-real-time tasks are processed on the edge layer, so the average delay of non-real-time tasks in the MEC system will decrease; however, as the probability of unloading non-real-time tasks increases, and they are preempted by real-time tasks at the same time, the average delay of real-time tasks in transmitting at the local output port will increase, which will increase the average delay of non-real-time tasks in the MEC system. Therefore, the average delay of non-real-time tasks in the MEC system first decreases and then increases.

[0132] exist Figure 4 It can be seen that under the same probability of unloading non-real-time tasks, as the arrival rate of non-real-time tasks increases, the average delay of non-real-time tasks in the MEC system will increase. This is because as the arrival rate of non-real-time tasks increases, the load of the entire MEC system will increase, and the probability of being preempted by real-time tasks will increase. The corresponding waiting time of non-real-time tasks in the MEC system will increase, and the average delay of non-real-time tasks in the MEC system will increase.

[0133] By comparison Figure 3 and Figure 4 , noting Figure 3 The average latency of real-time tasks in the MEC system is significantly lower than Figure 4 The average delay of non-real-time tasks in the MEC system clearly reflects the effectiveness and superiority of the task offloading strategy proposed in this invention, which can effectively meet the user's QoS.

[0134] Figure 5 vividly depicts the battery power E of the MEC system under different arrival rates of real-time tasks and non-real-time tasks. mec As the probability of offloading real-time tasks increases and the offloading probability of non-real-time tasks The changing trend of Figure 5a , 5bIn 5c and 5d, when the probability of offloading real-time tasks and non-real-time tasks is small, the battery power of the MEC system is quite high. This is because most of the tasks in the system are processed on the local server CPU, and most of the virtual machines on the edge layer SBS and MBS are in an idling state, which generates a large amount of idling battery power. As the probability of offloading real-time tasks and non-real-time tasks increases, the average battery power of the MEC system decreases first and then increases. This is because part of the tasks are processed on the local server CPU, and part of the tasks are offloaded to the edge layer for processing, which reduces the waiting time of the tasks and reduces the idling battery power of the local server CPU, local output port and edge layer virtual machine. Therefore, the average battery power of the MEC system decreases first. However, as the probability of offloading real-time tasks and non-real-time tasks further increases, the average battery power of the MEC system increases accordingly. This is because a large number of tasks are processed on the edge layer, generating a large amount of execution battery power, and at the same time, the local layer generates a large amount of idling battery power. Therefore, the average battery power of the MEC system will increase.

[0135] exist Figure 5a , 5b In 5c and 5d, it is not difficult to find that when all real-time tasks or non-real-time tasks are in the local layer or all in the edge layer, the average battery power of the MEC system will be higher. Under a reasonable combination of offloading probabilities, the MEC system has the lowest average battery power. Therefore, the superiority and effectiveness of the proposed task offloading strategy can be effectively demonstrated.

[0136] By setting the arrival rate of real-time tasks and non-real-time tasks to 1.0, Figure 6 The MEC system evaluation function is vividly depicted As the probability of offloading real-time tasks increases and the offloading probability of non-real-time tasks Similar to the trend of change in Figure 3 , Figure 4 and Figures 5a to 5d From the analysis, it is not difficult to find that, under the optimal combination of offloading probabilities of real-time tasks and non-real-time tasks, there exists a minimum MEC system evaluation, which further proves the superiority and effectiveness of the proposed task offloading strategy.

[0137] Using the general constraint multiplier method PHR based on Lagrange multipliers, MATLAB is used to solve the strategy optimization result that makes the MEC system evaluation function reach the minimum value, and the optimal task offloading probability combination is obtained. The system optimization results are shown in the following table:

[0138]

[0139] Table 2

[0140] The present invention designs a performance evaluation method for a MEC task offloading strategy based on a double-layer edge structure. Aiming at the scenario where a large number of tasks are offloaded to the edge layer and taking into account the pressure of the edge layer, a MEC system architecture based on a double-layer edge structure is built; based on the MEC system architecture, a flexible task offloading strategy is proposed for heterogeneous task requirements, and the strategy has the characteristics of low latency, low battery power and efficient offloading performance; to ensure the real-time performance of real-time tasks, a preemption priority strategy is introduced; to ensure the throughput of non-real-time tasks, an edge secondary offloading strategy is introduced; using queuing theory to establish a computational model of tasks in the local layer and the edge layer, a steady-state analysis of the system is performed, and performance indicators such as the average latency of tasks in the steady state and the average battery power of the MEC system are solved; a MEC system evaluation function with multiple inequality constraints is constructed, and the PHR algorithm based on Lagrange multipliers is used to provide an optimization scheme for the offloading probability of two types of tasks.

[0141] 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 performance evaluation method for MEC task offloading strategy based on a double-layer edge structure, characterized in that: It includes the following steps: S1. Calculate the average delay and average battery power of the delayed tasks processed locally, where the delayed tasks include real-time tasks and non-real-time tasks; S2, calculate the average delay and average battery power of the delay task processed in the edge layer; S3. Use the full probability formula to calculate the average battery power of the delayed task in the MEC system, and then obtain the average battery power E of the MEC system mec ; S4. Weigh the average latency of the delayed tasks and the average battery power of the MEC system, and construct a MEC system evaluation function with multiple inequality constraints in, It represents the probability of non-real-time tasks being offloaded to the edge layer SBS through the local output port for processing; represents the probability of real-time tasks being offloaded to the edge layer MBS cluster II through the local output port for processing; f1 represents the impact factor of the average delay of real-time tasks on the system evaluation function; f2 represents the impact factor of the average battery power of the MEC system on the system evaluation function; f3 represents the impact factor of the average delay of non-real-time tasks on the system evaluation function, and satisfies f1+f2+f3=1; They represent the average delay of real-time tasks and non-real-time tasks in the MEC system respectively; E mec represents the average battery power of the MEC system; λ1 and λ2 represent the arrival rates of non-real-time tasks and real-time tasks in the system, respectively, and satisfy λ1≥0 and λ2≥0; μ lc represents the service rate of the local server CPU and satisfies μ lc ≥0;μ lo represents the service rate of the local output port and satisfies μ lo ≥0; μ1 represents the service rate of a virtual machine on SBS, and satisfies μ1≥0; μ2 represents the service rate of a virtual machine on MBS cluster I, and satisfies μ2≥μ1; μ3 represents the service rate of a virtual machine on MBS cluster II, and satisfies μ3≥μ2; n1, n2, n3 are all constants; S5, using the general constraint multiplier method PHR based on Lagrange multipliers, using MATLAB to solve the MEC system evaluation function in S4 The strategy optimization result that reaches the minimum value obtains the optimal task offloading probability combination 2. According to claim 1, the performance evaluation method of the MEC task offloading strategy based on the double-layer edge structure is characterized in that: The step S1 specifically includes the following steps: S11, average latency of real-time tasks processed by the local server CPU and average battery level They are: in, Indicates the probability of real-time tasks being processed by the local server CPU; and They represent the execution and idling power of the local server CPU respectively; Indicates the circulation intensity of real-time tasks in the local server CPU; S12, average transmission delay of local output port of real-time task and average battery level They are: in, It represents the probability of real-time tasks being transmitted to the edge layer MBS for processing via the local output port; and denote the execution and idling power of the local output port, respectively; is the flow intensity of the real-time task at the local output port; S13, average latency of non-real-time tasks processed by the local server CPU and average battery level for: in, Indicates the probability of a non-real-time task being accepted by the local server CPU; Indicates the circulation intensity of non-real-time tasks in the local server CPU; S14, average transmission delay of local output port of non-real-time task and average battery level for: in, It represents the probability of non-real-time tasks being transmitted to the edge layer SBS for processing via the local output port; Indicates the traffic intensity of non-real-time tasks at the local output port.

3. The performance evaluation method of the MEC task offloading strategy based on the double-layer edge structure according to claim 2 is characterized in that: The step S2 specifically includes the following steps: S21, average latency of real-time tasks processed on edge layer MBS and average battery level for: in, represents the probability of real-time tasks being offloaded to the edge layer MBS cluster II through the local output port for processing; and They represent the execution and idling power of a virtual machine on the edge layer MBS, respectively; represents the flow intensity of real-time tasks in cluster II on the MBS; k represents the number of real-time tasks on the edge layer MBS cluster II; L R represents the loss rate of delay-sensitive tasks on edge layer MBS cluster II, and represents the probability that the number of real-time tasks on MBS cluster II is n3; S22, average latency of non-real-time tasks processed on edge layer SBS and average battery level for: in, It represents the probability of non-real-time tasks being offloaded to the edge layer SBS through the local output port for processing; and They represent the execution and idling power of a virtual machine on the edge layer SBS, respectively; represents the flow intensity of real-time tasks in MEC servers on SBS; π i,j represents the steady-state probability when the number of non-real-time tasks on the edge layer MBS cluster I is i and the number of non-real-time tasks on the edge layer SBS is j; K represents the system capacity of the edge layer SBS, which includes the total number of virtual machines and cache capacity on the edge layer SBS, It represents the probability that the non-real-time task is offloaded to the edge layer MBS cluster I for the second time when the edge layer SBS is overloaded; S23, the probability of non-real-time tasks being offloaded to edge layer MBS cluster I for the second time when edge layer SBS is overloaded for: Among them, π i,K represents the steady-state probability when the number of non-real-time tasks on edge layer MBS cluster I is i and the number of non-real-time tasks on edge layer SBS is K, that is, the steady-state probability that all virtual machines on edge layer SBS are fully running and the cache on SBS is also fully occupied by non-real-time tasks; S24, average latency of non-real-time tasks processed on edge layer MBS and average battery level for: in, It represents the probability of non-real-time tasks being offloaded to the edge layer SBS through the local output port for processing; It represents the flow intensity of non-real-time tasks in cluster I on MBS.

4. The performance evaluation method of the MEC task offloading strategy based on the double-layer edge structure according to claim 3 is characterized in that: The step S3 specifically comprises the following steps: S31. Use the total probability formula to obtain the average battery power of real-time tasks and non-real-time tasks in the MEC system. S32: Obtain the average battery power E of the MEC system by weighting mec :

5. The performance evaluation method of the MEC task offloading strategy based on the double-layer edge structure according to claim 4 is characterized in that: The average delay of the delay task in step S4 includes the average delay of real-time tasks and non-real-time tasks in the MEC system. With the help of the total probability formula, we can get:

6. The performance evaluation method of the MEC task offloading strategy based on the double-layer edge structure according to claim 5 is characterized in that: The MEC system includes a local layer and an edge layer. A task scheduler, a local server CPU and a local output port are deployed in the local layer, and two types of base stations, SBS and MBS, are deployed in the edge layer. A preemption priority strategy is introduced in the local server CPU and the local output port, and an edge secondary unloading strategy is introduced in the edge layer.

Images