CN107919986B - Optimization method for VM migration between MEC nodes in ultra-dense network - Google Patents

Abstract

The invention belongs to the technical field of wireless communication, and in particular relates to a VM migration optimization method between mobile edge computing nodes in an ultra-dense network. The applied ultra-dense network includes gateway nodes, aggregation nodes and edge nodes; when the VM of the MEC node needs to be migrated, the initialization feature is first calculated, and according to the predicted migration time of the user, the time when the user is active in the same gateway node is calculated. The energy consumption generated by interacting with the VM includes the energy consumption W ^mig of data transmission to the destination node, the energy consumption W ^pre of data transmission to the source node, and the connection with the VM in the coverage area of three types of nodes when the user's location changes. The energy consumption of data transmission is W ^after ; secondly, the optimal revenue model is established; finally, the optimal revenue model is solved, and the optimal VM migration strategy is selected. The method of the invention can realize the flexible migration function of the VM in special scenarios, effectively reduce the energy consumption of the system, and improve the migration efficiency; the resource collocation is reasonably carried out to meet the service requirements of users.

Description

Optimization method for VM migration between MEC nodes in ultra-dense network

technical field

The invention belongs to the technical field of wireless communication, and in particular relates to an optimization method for VM (virtual machine) migration between mobile edge computing (Mobile Edge Computing, MEC) nodes in an ultra-dense network (ultra-dense network, UDN).

Background technique

In recent years, the demand for wireless services has grown rapidly. According to industry forecasts, the demand for wireless services will increase by a thousand times in the near future. The existing access network service equipment is difficult to meet the increasing high-speed demand. Heterogeneous cellular (small cell) The intensive deployment of this problem has become the most critical solution. At the same time, in order to adapt to the flexibility of dense cells and reduce costs, wireless transmission solutions have become an important choice for cells to connect to the core network through gateway nodes. In addition, wireless backhaul in different bands, such as non-line-of-sight propagation (below 6 GHz) and point-to-point straight-line millimeter wave propagation, can satisfy wireless transmission in specific situations. Wireless backhaul technology has emerged as one of the backhaul solutions for cellular deployments.

New applications and cloud services have strict requirements on the transmission delay and computing delay of resources, and also have strict requirements on the placement and transmission conditions of cloud and edge computing servers. Deploying edge computing servers on cellular nodes, closer to users, can provide sufficient transmission and computing resources. However, deploying servers on cellular nodes inevitably reduces the coverage of the servers, the mobility of users is more frequent and complex, and the migration of VMs will be more frequent, especially between heterogeneous cells. Heterogeneous backhaul links are more complex, and VM migration technology needs to consider these new factors to reduce system energy consumption, delay and optimize throughput, and ensure overall system service efficiency.

The choice of the migration location and the migration method are the main research work of the current VM migration technology. In the case of multiple users, the resource occupancy of users is a balancing issue. During VM migration, not only the workload status of the relevant server, that is, the working status of other users on the server, but also the transmission status and user mobility are considered. Due to the consideration of VM components and memory updates, VM migration requires memory transfer and computation costs during the migration process, and migration benefits become the main determinant of VM migration. To improve the benefits of migration, mobility forecast and cost forecast become new influencing factors. In addition, the migration time can be reduced by changing the migration method, such as transferring a smaller amount of data through a compression algorithm, and reducing the VM interruption time through the memory pre-migration mode to reduce the migration cost, which can not only reduce the execution delay, but also reduce the migration cost. At the same time, it can increase the throughput of the system and reduce the pressure of the system load.

Literature [1] (L.Tong, Y.Li and W.Gao, "A hierarchical edge cloud architecture for mobile computing," IEEE INFOCOM 2016-The 35th Annual IEEE International Conference on Computer Communications, San Francisco, CA, 2016, pp.1 -9.) has established a heterogeneous MEC server framework, a layered framework, which can provide more flexible services, ensure the rational use of resources and the efficiency of user service unloading, and solve the task load of edge servers that change at different times. The deployment issues under the section, but there is no specific discussion of the backhaul link and VM migration issues in this case.

Literature [2] (M.N.Islam, A.Sampath, A.Maharshi, O.Koymen and N.B.Mandayam, "Wireless backhaul node placement for small cell networks," 2014 48th Annual Conference on Information Sciences and Systems (CISS), Princeton, NJ, 2014, pp.1-6.) discussed a wireless backhaul node deployed in a dense network. This backhaul method utilizes different frequency bands for communication between nodes, which enables flexible deployment of cells and solves the problem in urban areas. The deployment of vertical nodes is not combined with other application scenarios, such as the deployment of MEC computing resources, and the parts related to VM migration.

In an ultra-dense network scenario, access nodes are densely deployed, edge computing servers are getting closer and closer to users, and their service scope is also becoming smaller. As users move, virtual machines on edge servers need to be migrated to appropriate nodes. To ensure user service efficiency and energy consumption in the network, the existing VM migration does not take into account the deployment characteristics of heterogeneous networks; in addition, the deployment of many nodes requires high flexibility, and wireless backhaul has become the backhaul technology choice in the scenario , the transmission power and transmission rate of different bands are also different, which also need to be considered in VM migration.

SUMMARY OF THE INVENTION

In order to reduce the energy consumption in the ultra-dense network and improve the VM migration efficiency, the present invention does not consider the characteristics of the heterogeneous network for the VM migration in the current ultra-dense network scenario, and introduces the heterogeneous scenario of the cell to provide a MEC in the ultra-dense network. An optimization method for VM migration between nodes.

In the method for optimizing VM migration between MEC nodes in an ultra-dense network provided by the present invention, the applied ultra-dense network includes gateway nodes, aggregation nodes and edge nodes. The coverage of the three types of nodes is from large to small. The entry point has an MEC server. The MEC node refers to the node where the MEC server is deployed. When the VM of the MEC node needs to be migrated, perform the following steps:

计算用户在同一个网关节点范围内活动时与VM进行交互所产生的能耗，能耗包括数据传输到目的节点的能耗W^mig，数据传输到源节点的能耗W^pre，以及用户位置变化时在三种节点覆盖区域与VM连接进行数据传输的能耗W^after；First, calculate the initialization features, including: according to the predicted user's migration time

Calculate the energy consumption generated by the interaction between the user and the VM when the user is active within the same gateway node. The energy consumption includes the energy consumption W ^mig of data transmission to the destination node, the energy consumption W ^pre of data transmission to the source node, and the change of user location. The energy consumption W ^after of connecting with the VM for data transmission in the three types of node coverage areas;

第二阶段，用户在边缘节点B的覆盖范围下，边缘节点A和B在同一个聚合节点下，该阶段用户与VM进行数据传输的能耗为

第三阶段，用户移动到边缘节点C，边缘节点A和C在同一个网关节点下，该阶段用户与VM进行数据传输的能耗为

The energy consumption W ^after is divided into three stages: in the first stage, the user is under the coverage of the edge node A where the VM is located, and the energy consumption of data transmission between the user and the VM in this stage is

In the second stage, the user is under the coverage of edge node B, and edge nodes A and B are under the same aggregation node. In this stage, the energy consumption of data transmission between the user and the VM is:

In the third stage, the user moves to the edge node C, and the edge nodes A and C are under the same gateway node. In this stage, the energy consumption of data transmission between the user and the VM is:

Second, establish the optimal revenue model:

表示VM的数量，ψ表示网络中节点的集合；g(l)取值为1时，表示VM迁移到节点l上，g(l)取值为0时，表示VM未迁移到节点l上；D_k表示VM k需要占用的资源；M_l表示节点l的资源总量。in,

Represents the number of VMs, ψ represents the set of nodes in the network; when g(l) is 1, it means that the VM has migrated to node l, and when g(l) is 0, it means that the VM has not been migrated to node l; D _k represents the resources that VM k needs to occupy; M _l represents the total resources of node l.

Finally, the optimal revenue model is solved, and the optimal VM migration strategy is selected.

Compared with the prior art, the advantages and positive effects of the present invention are:

(1) The method of the present invention can realize the flexible migration function of VMs in special scenarios, and combines technologies such as wireless backhaul and user mobility prediction. According to the simulation results, it can be seen that the method of the present invention can effectively reduce the energy consumption of the system, Improve migration efficiency;

(2) The method of the present invention considers the coverage of different nodes in the ultra-dense network scenario, and comprehensively considers various factors such as resource deployment, user mobility characteristics and needs, channel conditions and other factors according to the size of the node coverage. The VM selects appropriate nodes and reasonably allocates resources to ensure computing requirements and meet user service requirements.

Description of drawings

Figure 1 is a schematic diagram of a system model of an ultra-dense network;

Fig. 2 is a schematic flow diagram of the main flow of the VM migration optimization method provided by the present invention;

Figure 3 is a graph showing the change in energy consumption of different schemes when the number of users increases linearly;

Fig. 4 is the change diagram of the energy consumption benefit that different types of nodes increase with the user when the method of the present invention is applied;

Fig. 5 is the change diagram of the energy consumption benefit of the average user under different types of situations when the method of the present invention is applied;

Fig. 6 is a graph showing the change of energy consumption benefit of different types of nodes with the average memory occupancy of the VM when the method of the present invention is applied.

Detailed ways

The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.

The VM migration optimization problem between MEC nodes in an ultra-dense network is a multi-server and multi-user problem. During the VM migration process, since the VM has a variety of locations to choose from, and the multi-user changes are complex and random, it is completely optimized. , it becomes a Nash equilibrium problem, its complexity will increase greatly, and it will be difficult to calculate in a very short time. The energy consumption optimization strategy based on VM dynamic migration is a simplified algorithm, and the method of the present invention will maximize the profit as optimize the target. This kind of objective equation has a certain amount of resources and also produces corresponding value. This selection problem can be transformed into a knapsack problem and solved by dynamic programming. Due to the heterogeneity of MEC servers, the improved distributed dynamic planning can ensure that the service revenue is maximized, and it can also ensure the problem of overloading a single MEC and ensure the service quality of users.

As shown in Figure 1, it is an ultra-dense network model. Gateway Nodes (GN) are located at the top of the building, and its special function is to access the core network through optical fiber communication. Aggregator Nodes (AN) are located at the top of the building where the cells are deployed. On the one hand, they can be connected to the gateway node and enter the core network through the gateway node. Edge Nodes (EN) are located inside the building, and this node directly provides network services to users. The three types of nodes are responsible for different tasks, and their different coverages determine their server size deployment. In the ultra-dense network to which the method of the present invention is applied, each access point is arranged with an MEC server. The MEC node in the present invention refers to a node on which an MEC server is deployed, such as shown in FIG. 1 , including heterogeneous gateway nodes, Aggregation nodes and edge nodes, the signal coverage of the three decreases in turn.

The method for optimizing VM migration between MEC nodes in an ultra-dense network of the present invention, for a single user, selects an appropriate node for VM migration according to the user's mobility characteristics, which can improve the work efficiency of the VM and reduce the interruption caused by the VM migration. , and can reduce the overall energy consumption. In the case of multiple users, it is necessary to consider the resource occupation and other situations, and adjust the location of VM migration to ensure the overall benefits.

During the migration process, in order to ensure the user's QoS, the user equipment still needs to maintain a normal working state with the VM at the initial location. The longer the VM migration time is, the more energy consumption will be, and the frequency of migration will decrease relatively. The method of the invention firstly predicts the user's moving time reasonably based on the user's moving characteristics, then calculates the energy consumption of the migration, allocates according to the calculation result, and then optimizes the selection of each node to obtain the best result.

The power p _i, j between node i and node j can be obtained by the Shannon formula, as follows:

Among them, r is the transmission rate, N ₀ is the channel noise, w is the channel bandwidth, and g _i,j is the link gain.

为VM迁移的总时间，在传输速率固定的情况下，迁移时间满足以下公式：

is the total time of VM migration. In the case of a fixed transmission rate, the migration time satisfies the following formula:

Among them, M _k,0 represents the initial memory size of the VM with the label k, and the labels of the VM are numbered in the order of natural numbers; R is the data transfer rate of VM migration, which is set to a fixed value here; D _k is the dirty page rate, Represents the generation rate of memory, which must satisfy R>D _k , otherwise the migration process cannot converge, VM pre-migration deteriorates, and server resources are extremely occupied; M ^th is the dirty page threshold, and when the threshold is reached, the switching stage is entered.

An implementation process of the VM migration optimization method of the present invention will be described below with reference to FIG. 2 .

Step 1: First determine whether the VM needs to be migrated. If the amount of computing data of the user is not large enough, or the resources of the destination server are insufficient, the migration operation is not required. If migration is required, the user's mobility characteristics are retrieved.

有直接的关系。Step 2: Initialize feature calculation. Heterogeneous nodes result in multiple locations for VMs to be placed. When a user moves to a location, the location will be covered by three different nodes. In theory, VMs can be migrated to any of these three nodes, but it needs to be Choose the one with the best yield. In the present invention, for each task (ie VM) covering all different types of nodes of the user, the energy consumption rate at different nodes should be calculated, and the migration energy consumption should be estimated according to the user's mobility characteristics, and then the relative energy consumption should be calculated. Revenue, allocate VMs to the optimal MEC server nodes according to the size of the revenue. During the migration process of the VM, it is necessary to ensure the connection of the link and a certain transmission rate. The consumption of this time is the same as the above-mentioned migration time.

have a direct relationship.

W ^mig represents the energy consumed by the configuration files contained in the VM and the necessary data cache transferred to the destination node during the VM migration process.

f(x) is a normalization function, when x=0, f(x)=0, and when x≠0, f(x)=1.

为不同类型节点的代号，上标t+1和t代表两个不同的时间段，t表示上一个时段，一般代表VM的源节点位置，t+1代表用户当前所在位置。

分别代表VM在上一时刻和当前时刻所在的边缘节点，

分别代表VM在上一时刻和当前时刻所在的聚合节点，

分别代表VM在上一时刻和当前时刻所在的网关节点。以

为例，若

为1，表明VM k在t时刻在代号为

的边缘节点上运行。φ _k = (i, j, z) represents the location of virtual machine k, i represents the gateway node number of the node where the VM is located, j is the aggregation node number of the node where the VM is located, and z represents the edge node number of the node where the VM is located. On the edge node, z is 0, if it is not on the aggregation node, then j is 0, i is the gateway node code, it will not be 0, the VM is not necessarily on the edge node or the aggregation node server, but must be on a certain one covered by the gateway node, where

For the codes of different types of nodes, the superscripts t+1 and t represent two different time periods, t represents the previous period, which generally represents the location of the source node of the VM, and t+1 represents the current location of the user.

represent the edge nodes where the VM is located at the previous moment and the current moment, respectively,

Represents the aggregation node where the VM is located at the previous moment and the current moment, respectively.

Represents the gateway node where the VM is located at the previous moment and the current moment, respectively. by

For example, if

is 1, indicating that VM k is at time t at codename

run on edge nodes.

表示任务k从聚合节点向边缘节点传输时的能耗功率，

表示任务k从网关节点向聚合节点传输时的能耗功率，

表示任务k从聚合节点向网关节点传输时的能耗功率，

表示任务k边缘节点向聚合节点传输时的能耗功率。

表示任务k光纤传输的能耗功率。The migration of VM involves four kinds of inter-node migration, including migration between edge nodes and aggregation nodes, and migration between aggregation nodes and gateway nodes. Let en represent the edge node, an represent the aggregation node, D represent the destination, and R represent the source.

represents the energy consumption of task k when it is transmitted from the aggregation node to the edge node,

represents the energy consumption of task k when it is transmitted from the gateway node to the aggregation node,

represents the energy consumption of task k when it is transmitted from the aggregation node to the gateway node,

Represents the energy consumption when the task k edge node transmits to the aggregation node.

Represents the energy consumption power of task k fiber transmission.

The energy W ^pre of calculating data transmission consumption during VM migration means that the calculated data still needs to be transmitted to the source server during the migration time, and the transmission energy consumption is:

After the VM migration is completed, the user's location changes, and its location can be divided into three types of areas, so the consumption can also be divided into three stages, and the consumption of data transfer W ^after is calculated:

这一阶段是用户在初始边缘节点的能耗时间，用户的能量消耗：1) The first stage, the time is

This stage is the energy consumption time of the user at the initial edge node, and the energy consumption of the user:

表示VM迁移所需的时间。由公式(5)可知，若VM迁移到边缘节点，此时

为0，表示VM在边缘节点就可以为用户服务，节点之间没有能量消耗；若迁移到聚合节点或者网关节点，此时节点之间需要数据传输，有能量消耗。Among them, T represents time, and let T _{tier_3} , T _{tier_2} , and T _{tier_1} represent the predicted stay time of the user under the same edge node, the same aggregation node, and the same gateway node as the VM, respectively.

Indicates the time required for VM migration. It can be seen from formula (5) that if the VM migrates to the edge node, then

If it is 0, it means that the VM can serve users at the edge node, and there is no energy consumption between nodes; if it is migrated to an aggregation node or a gateway node, data transmission is required between nodes at this time, and there is energy consumption.

2) The second stage, the time is T _{tier_2} -T _{tier_3} , this stage is the energy consumption time of the user in the same aggregation node but different edge nodes, the user leaves the original edge node, but is still within the coverage of the same aggregation node, Energy consumption at this stage:

表示任务k从聚合节点向另外一个边缘节点en’传输时的能耗功率。Among them, en' represents another edge node located under the same aggregation node an as the edge node en.

Represents the power consumption when task k is transmitted from the aggregation node to another edge node en'.

3) In the third stage, the time is T _{tier_1} -T _{tier_2} , which represents the energy consumption time of the user at the same gateway node but different aggregation nodes. This stage is under the coverage of the gateway node, so there is no need to consider the situation of VM migration again. Its energy consumption is:

Among them, en" indicates that the user's location is under the edge node of another aggregation node in the same gateway node range, and an' indicates another aggregation node in the same gateway node range.

Through the initialization feature calculation in this step, according to the predicted time of the user, determine the energy consumption generated by the interaction between the user and the VM when the user is active within the same gateway node range.

Step 3: Calculation of distribution optimal income. After the initialization calculation is completed, the tasks are pre-assigned to the corresponding stages, and the node calculates the total revenue according to its own resources, user space and value as follows:

表示用户或者VM的数量，用户和VM是一一对应的。ψ表示所有节点的集合。g(l)表示VM是否迁移到节点l上，M_l表示节点l的资源总量。D_k表示用户k或VM k需要占用的资源。公式(8)中的限制条件是，保证一个节点上的VM占用的资源不超该节点的资源最大值。in,

Indicates the number of users or VMs. Users and VMs are in one-to-one correspondence. ψ represents the set of all nodes. g( _l ) represents whether the VM is migrated to node 1, and M1 represents the total resources of node 1. D _k represents the resources required to be occupied by user k or VM k. The restriction in formula (8) is to ensure that the resource occupied by the VM on a node does not exceed the maximum resource value of the node.

可以通过更改W^pre条件求出，公式如下：The present invention takes the energy consumption W ^static of the VM without migration as the baseline,

It can be obtained by changing the W ^pre condition, the formula is as follows:

根据公式(9)获得，设

则将

的差值作为收益目标，计算最优收益的目标表达式可以转化为：The baseline energy consumption of VM k is

According to formula (9), we can get

will

The difference is used as the income target, and the target expression for calculating the optimal income can be transformed into:

Due to the existence of server distribution, to find the overall optimal solution, the computational complexity and time consumption are very high, so the goal of the present invention is to find the sub-optimal solution, that is, each server recursively finds the local optimal solution through dynamic programming , the core formula of its iteration is as follows:

W(n,m)=max{W(n-1,m),W(n-1,mD _n )+w _k } (10)

通过这种方式可以求出最大收益。Among them, n is the number of iterations, m is the resource occupancy of the MEC server on the node, and the maximum value of the resource that can be occupied by the server is M. w _k is the expected return of a single VM k,

In this way, the maximum profit can be obtained.

Step 4: Circular filling calculation. In step 3, the VM without resource storage will select a higher-level server node, perform a solution on the original basis, and perform an iterative solution again.

The present invention proposes a VM migration scheme based on heterogeneous nodes. First, the network system model is introduced, then the VM migration scenario and specific amplification are described, and finally the performance simulation is given. The proposed method can effectively reduce the energy consumption of the system and improve the service quality of users.

Example:

Consider the following scenario: two gateway nodes, four aggregation nodes, and twelve edge nodes. As shown in Table 1, in this embodiment, a bandwidth of 5.8 GHz is selected as the over-the-horizon transmission, and 60 GHz is selected as the line-of-sight. The maximum number of users is set to 10,000, the transmission rate requirements meet the normal distribution N (300Mbps, 100Mbps), the resource occupancy is randomly distributed, and its range is [5, 20]; the dirty page rate is [10, 20] k/s, the transfer rate of the migrated data is set to 200Mbps, the Gaussian noise is 10 ⁴ , the time of users under three nodes is randomly generated, and the generation rule is T _{tier_1} ∈(T _{tier_2} ,3T _{tier_2} ), T _{tier_2} ∈(T _{tier_1} , 3T _{tier_1} ), T _{tier_3} ∈ (0,7200)s.

Table 1 5.8GHz and 60GHz wireless channel characteristics table

feature 5.8GHz 60GHz Rain Fading (dB) 0 10 Oxygen absorption (dB) 0 15 Channel Gain (dB) 17 38 Maximum transmit power (dBm) 19 25 Fading Margin (dB) 15 25 Channel bandwidth (MHz) 40 160 number of channels 6 6

For the above scenario, the method of the present invention, the migration scheme with the user and the non-migration scheme are used to conduct simulation experiments on the VM. Fig. 3 shows the change diagram of energy consumption when the number of users increases linearly for three different schemes, wherein the curve of the method of the present invention adopts the selection migration strategy. This figure is mainly reflected in the situation of different users, and the general trend of its total energy consumption is gradually increasing. The state of no migration scheme is almost in a linear state, and the choice of migration is better than no migration. Convergence is better, but when the number of users increases To a certain extent, it will perform better. In general, under normal circumstances, the method of the present invention has a very good effect of saving energy.

Figure 4 is a graph showing the change of energy consumption with the total energy return of each type of MEC server in the optimization strategy. Compared with the non-migration state, the most profitable is the edge node, because the edge node is the node closest to the user, if the user's characteristics belong to a long time in a certain place, if the VM migrates to the edge node, the energy consumption rate at this time is is the lowest, so the benefits of VMs migrated to edge nodes are the largest and become the main part of the total revenue, so the deployment of edge nodes becomes the key.

Figure 5 shows the average energy consumption change of each user. With the increase of users, the revenue on different server nodes continues to decrease. This is due to the increase of users and the conflict of their resources, causing some VMs to select servers for the second time. Earnings are bound to decline. Both of these points indicate that in order to achieve a better effect of reducing energy consumption, more suitable edge nodes need to be deployed, and more resources need to be available for flexible VM migration.

In Figure 6, a graph of the change in energy consumption benefit with the average memory footprint of the VM is shown. As the resource volume of the VM increases, the energy benefit of different types of servers decreases. On different types of servers, the energy consumption of the gateway node is the largest, which reflects the relatively high mobility of users. The larger the coverage area of the VM migration location, the higher the income. Therefore, when the server resources are deployed, the coverage area is large. Nodes, such as gateway nodes, can deploy more resources, but their transmission distance is long, which may sacrifice part of the transmission delay, thereby reducing the calculation rate of the task.

Claims (7)

1. VM migration optimization method between MEC nodes in an ultra-dense network, is characterized in that, in the applied ultra-dense network, gateway node, aggregation node and edge node are included, and each node is deployed with MEC server, wherein, MEC will be deployed with The node of the server is called MEC node, MEC means mobile edge computing, VM means virtual machine; In the described method, when the VM of the MEC node needs to be migrated, the following steps are performed: First, calculate the initialization features, including: according to the predicted user's migration time

Calculate the energy consumption generated by the interaction between the user and the VM when the user is active within the same gateway node. The energy consumption includes the energy consumption W ^mig of data transmission to the destination node, the energy consumption W ^pre of data transmission to the source node, and the change of user location. The energy consumption W ^after of connecting to the VM in the three types of node coverage areas; The energy consumption W ^after is divided into three stages: in the first stage, the user is located under the coverage of the edge node A where the VM is located, and the energy consumption of data transmission between the user and the VM in this stage is W ₁ ^after ; in the second stage, the user is on the edge Under the coverage of node B, edge nodes A and B are under the same aggregation node, and the energy consumption of data transmission between users and VMs at this stage is

In the third stage, the user moves to the edge node C, and the edge nodes A and C are under the same gateway node. In this stage, the energy consumption of data transmission between the user and the VM is:

Secondly, establish the optimal income model, as follows:

g(x)∈(0,1) in,

Represents the number of VMs, ψ represents the set of nodes in the network; g(l) represents whether the VM is migrated to node l, a value of 1 means migrated to, and a value of 0 means not migrated to; D _k means that VM k needs to occupy resources; M _l represents the total resources of node l; Finally, the optimal revenue model is solved, and the optimal VM migration strategy is selected. 2. The method according to claim 1, wherein the migration time of the user is

Calculated according to the following formula:

Among them, M _k,0 represents the initial memory size of the VM labeled k; R is the data transfer rate of VM migration; D _k is the dirty page rate; M ^th is the dirty page threshold. 3. The method according to claim 1, wherein the energy consumption W ^mig of the data transmission to the destination node is calculated and obtained according to the following formula:

Among them, f(x) is a normalization function, when x=0, f(x)=0, when x≠0, f(x)=1; f(x)=0 means that VM is on node x Not running, f(x)=1 means the VM is running on node x; Let φ _k =(i,j,z) represent the position of virtual machine k, i represents the gateway node serial number of the node where the VM is located, j represents the aggregation node serial number of the node where the VM is located, and z represents the edge node serial number of the node where the VM is located;

Indicates the edge node where VM k is located at the current moment;

Indicates the aggregation node where VM k is located at the current moment;

respectively represent the edge node and aggregation node where VM k is located at the last moment; where

are the code names of different types of nodes, the superscripts t+1 and t represent two different time periods, t represents the previous time period, representing the source node location of the VM, and t+1 represents the current location of the user;

represents the energy consumption of task k when it is transmitted from the aggregation node to the edge node,

represents the energy consumption of task k when it is transmitted from the gateway node to the aggregation node,

represents the energy consumption of task k when it is transmitted from the edge node to the aggregation node,

represents the energy consumption of task k when it is transmitted from the aggregation node to the gateway node,

represents the energy consumption power of optical fiber transmission of task k. 4. method according to claim 3, is characterized in that, described energy consumption W ^pre is obtained according to following formula:

in,

represents the energy consumption of task k when it is transmitted from the aggregation node to the edge node,

represents the energy consumption of task k when it is transmitted from the gateway node to the aggregation node,

represents the energy consumption of task k when it is transmitted from the edge node to the aggregation node,

represents the energy consumption of task k when it is transmitted from the aggregation node to the gateway node,

represents the energy consumption power of optical fiber transmission of task k. 5. The method according to claim 3, wherein the consumption W ^after is divided into three stages according to the position change of the user, and let T _{tier_3} , T _{tier_2} , T _{tier_1} respectively represent that the user is on the same edge as the VM Predicted residence time under the same node, the same aggregation node, and the same gateway node; (1) The first stage, the time is

The energy consumption W ₁ ^after is:

in,

represents the energy consumption of task k when it is transmitted from the gateway node to the aggregation node an,

Represents the energy consumption of task k when it is transmitted from the aggregation node an to the edge node en; (2) The second stage, the time is T _{tier_2} -T _{tier_3} , the energy consumption

for:

in,

Represents the energy consumption when task k is transmitted from the aggregation node an to another edge node en', and the nodes en' and en are under the aggregation node an; (3) The third stage, the time is T _{tier_1} -T _{tier_2} , the energy consumption

for:

in,

Represents the energy consumption of task k when it is transmitted from the aggregation node an to another edge node en”, and the nodes en” and en are under different aggregation nodes under the same gateway node;

Represents the energy consumption when task k is transmitted from the gateway node to the aggregation node an', and nodes an' and an are under the same gateway node. 6. The method according to claim 1 or 3, wherein, when the optimal revenue model is solved, each VM takes the energy consumption without migration as the reference energy consumption; Baseline energy consumption

T _{tier_1} represents the predicted stay time of the user under the same gateway node as the VM; For user k, let its baseline energy consumption be

Total energy consumption from interacting with the VM

Will

The difference of is the revenue target, then the target expression for calculating the optimal revenue is transformed into:

7. The method according to claim 6, wherein, when the optimal income model is solved, it is realized by dynamically recursively finding a local optimal solution for each node; Let the energy consumption of a node at the nth iteration W(n,m)=max{W(n-1,m),W( _n -1,mDn)+ _wk }; where n is the number of iterations, m is the resource occupancy of the MEC server on the node, D _n is the memory size of the VM, w _k is the expected revenue of VM k,

When the node has no resources to store VM k, a higher-level server node will be selected for VM k, and then recursively iteratively solve the node.

Images