An Active Migration Method for Edge Services Based on Mobility Prediction

Abstract

The present invention discloses an active migration method for edge services based on mobility prediction, belonging to the field of mobile edge computing. Specifically, it is based on a hierarchical mobile edge computing network composed of mobile users, edge servers, and central servers. For a mobile user U, the target server for service migration is obtained through mobility prediction. Then, it is judged whether performance indicators such as the end-to-end delay and received signal strength indication received by the mobile device meet the service quality requirements. If so, the prediction of the next target server is carried out again; otherwise, the mobile user U notifies the target edge server s′ to reserve a container, and copies the disk data from the source edge cloud server s to the target edge server s′. Finally, the checkpoint / restore technology is used to achieve the data migration of the memory state, so that the edge server s′ provides services for the mobile user U. The present invention reduces the service migration time and end-to-end delay based on containers.

Description

Technical Field

[0001] The present invention belongs to the field of mobile edge computing, and particularly relates to an active migration method for edge services based on mobility prediction. Background Art

[0002] In a mobile edge computing environment, the coverage range of a single edge cloud server is limited. When a user leaves the coverage range of the source edge cloud, the quality of service of the services running on the source edge cloud (such as the end-to-end delay between the mobile device and the edge cloud being too long) will seriously deteriorate, and may even be interrupted. Therefore, it is particularly important to ensure the quality of service and continuity of the edge cloud.

[0003] Service migration is an important method to ensure the continuity of edge cloud services. Service migration refers to migrating the services running on the source edge server to a new target edge server. Service migration technology includes two aspects: migration decision-making and migration implementation.

[0004] Migration decision-making includes migration time decision-making and migration target decision-making, mainly determining whether to migrate the running services, where to migrate, and when to perform the migration; the implementation of migration needs to determine how to implement the service migration process, and the roles involved include the source edge server and the target edge server (migration target). And the migration target is one of the issues studied in migration decision-making. Therefore, migration target decision-making is the premise and foundation of migration implementation, and migration decision-making and migration implementation are closely related.

[0005] The implementation process of service migration involves aspects such as virtualization technology, communication handover, and computing migration.

[0006] According to the sequence relationship between computing migration and communication handover, service migration technology is divided into passive migration and active migration:

[0007] Passive migration means that computing migration is performed after communication handover, that is, computing migration is triggered by communication handover. The migration of a large number of files will cause the end-to-end delay between the mobile user and the edge server and the service migration delay to be too long; active migration pre-obtains the service migration target server based on mobility prediction and starts performing computing migration before communication handover;

[0008] According to the virtualization technology adopted, service migration can be divided into virtual machine-based migration and container-based migration;

[0009] Regardless of the virtualization technology adopted, the implementation of migration imposes a load on the roles involved in migration, such as the source edge cloud server, the target edge cloud server, and network bandwidth resources. In terms of the performance metrics of live migration, the above-related research mainly uses service migration time and service downtime as metrics. With the development of edge computing and new mobile applications, the end-to-end delay (End to End, E2E) between mobile users and edge servers has become an important indicator for measuring service quality and thus also a metric for measuring service migration performance.

[0010] Generally speaking, the implementation of service migration involves the following issues: First, the selection of virtualization technology. The main difference between virtual machines and containers is that virtual machines are installed on system hardware and rely on a hypervisor to manage system resources. Containers are implemented through operating system virtualization and share the underlying operating system kernel. From the basic principles of these two virtualization technologies, containers reduce the level of the guest operating system, so they are more lightweight and have higher performance, making them more suitable for the migration of latency-sensitive application services. Second, the process design of the migration method. The migration of stored data (hard disk data) and memory data are important issues that need to be addressed in service migration. Among them, the migration of stored data includes the base image files of containers or virtual machines, and the migration of memory data requires migrating all the running state data of the service (CPU state, memory data, devices, etc.) to the target server. However, since virtual machine image files are larger than container image files, and the iterative copying of memory data files also generates a large amount of file transfer, it is easy to cause excessive service migration time and downtime.

[0011] Based on the above analysis, containers are more suitable for the migration of latency-sensitive application services. In the aspect of the live migration of containers, checkpoint / restore is a commonly used technology for the migration of container memory data. Its main process is to save the memory state of the process to a file and then restore the process from the checkpoint on the destination host. On the other hand, containers have a layered storage structure, and this layered structure involves disk files such as a read-only base image layer and a writable container top layer (application image layer related to the application), and this layered structure can achieve the rapid packaging and migration of containers.

[0012] In the prior art, Document 1: A method for migrating edge computing services based on trajectory prediction with the application number 202210440772 selects the optimal transfer server according to the trajectory prediction result, reduces the user waiting time for service migration caused by user transfer, and can dynamically update service remedial migration according to the comprehensive migration time. This method obtains the migration target based on mobile trajectory prediction and determines the optimal service migration path through the path score of the optimal migration base station to reduce the user waiting time for service migration;

[0013] Document 2: A joint mobility management method for communication handover and service migration with the application number 201911417106. In a network with mobile edge computing nodes, logical function entities related to mobility management, namely, a handover management entity and a service migration management entity, are set up. The handover management entity predicts the communication handover moment and the target access point, and makes decisions on the time sequence during the communication handover and service migration processes, reducing the total service interruption time during user movement and improving the quality of mobile communication services. The research on whether to perform service migration and how to perform the service migration process in this method does not specify a particular virtualization technology, and both containers and virtual machines are applicable. The pre-copy method is used to make decisions on the service migration process;

[0014] Document 3: A container migration method with the application number 201980082561 discloses a technology for migrating containerized software packages between a source computing device and a destination computing device. This method does not address the issues related to the MEC environment and how the destination computing device is selected;

[0015] The above three comparative documents are respectively a service migration method based on mobility prediction, a method jointly considering communication handover and computing migration, and a container-based service migration method; each has the following disadvantages:

[0016] Document 1 predicts the target trajectory of a user's future movement based on the user's historical movement trajectory, then determines a list of alternative base stations according to the target prediction trajectory and obtains the optimal migration base station, and determines the optimal service migration path. The differences from the present invention are as follows:

[0017] (1) Different considerations for mobility: This invention uses the user's trajectory information and location information to determine the user's target prediction trajectory; the present invention predicts the next position of the user's movement based on the user's historical movement trajectory and uses the edge server where this position is located as the target edge server for migration.

[0018] (2) Different service migration optimization methods are adopted: This invention optimizes the time of service migration by optimizing the service migration path; the main idea of the present invention is to use the hierarchical storage characteristics of containers to optimize the implementation process of service migration, thereby optimizing the service migration time and end-to-end delay.

[0019] Document 2 first makes a decision on whether to perform service migration. If necessary, it then predicts the start moment when the service migration enters the stop-copy phase, and then plans the time of the service migration process in combination with the moments of communication handover, pre-copy, etc., to reduce the total service interruption time during user movement. The differences from the present invention are as follows:

[0020] (1) Different virtualization technologies are adopted: The present invention does not limit to a specific virtualization technology, and both virtual machines and containers are applicable; the present invention specifically refers to container technology and adopts the checkpoint recovery technology in containers for the design of the service migration process.

[0021] (2) Different optimization objectives: The present invention takes the service interruption time as the optimization objective; the present invention aims to optimize the end-to-end delay between the user and the edge server and the service migration time.

[0022] Document 3 discloses a technology for migrating containerized software packages between a source computing device and a destination computing device. The method includes receiving, at the destination device, a request to migrate a source container currently executing on the source device to the destination device; it also includes synchronizing a handle list utilized by the source container on the source device between the destination device and the source device, and instantiating a destination container in the destination device using a copy of the image of the source container on the source device, a memory snapshot, and the synchronized handle list. The differences from the present invention are as follows:

[0023] (1) Different scenarios and software management considered: The present invention considers the use of sandboxes as a software management strategy in a computer network; the present invention considers the mobile edge computing environment and jointly considers communication handover and the hierarchical storage characteristics of containers for the design of the service migration process.

[0024] (2) Different purposes of container migration: In the present invention, container migration is to effectively switch between local resources and remote resources by tracking the memory state of the migrated container and migrating the container from a local device to a remote server; the present invention realizes the migration of computing tasks of edge services through container migration between edge servers.

[0025] From the perspective of the virtualization technology adopted, most studies use virtual machines as the virtualization environment, and the iterative copying of virtual machine image files during the migration process generates a large amount of file transmission, resulting in excessive time overhead. From the perspective of the timing planning of the migration process, the passive service migration triggered by communication handover advocates that the computing migration occurs after the communication handover, which is also one of the factors leading to excessive service migration time. In the passive method, the service migration process can only be executed after the user moves to the coverage area of the new edge cloud. A small number of studies jointly consider communication handover and the migration process. Even considering the active method based on mobility prediction, these works assume that the service on the target edge server is already available. For example, in a predefined movement pattern, it is assumed that the target edge server already has a container image or an application image, or it is assumed that after the mobile user moves, the basic container image file and the application image are downloaded from the remote cloud server.

[0026] In terms of mobility, existing research mainly uses random mobility models to model user mobility and designs corresponding service migration decision optimization schemes. These studies only consider the randomness of mobile behavior, ignoring characteristics such as the periodicity and predictability of user movement behavior, lacking the characterization and representation of users' true mobility, which affects the improvement of service migration performance. Summary of the Invention

[0027] To address the above problems, the present invention uses mobility prediction and container migration technologies to propose an active edge service migration method based on mobility prediction. The target server for migration is obtained through mobility prediction. Before migrating the memory state data, the container base image and application program image are pre - copied from the source edge server to avoid transferring disk files during the actual migration process, thereby reducing the service migration time and end - to - end delay based on containers and providing a solution for the migration of delay - sensitive application services.

[0028] The specific steps of the active edge service migration method based on mobility prediction are as follows:

[0029] Step 1: Build a hierarchical mobile edge computing network communication scenario consisting of mobile users, edge servers, and central servers;

[0030] Mobile users carry mobile devices and move. Edge servers receive offloading requests from mobile users and provide edge cloud services. Edge servers are located at wireless access points in the access network and have a limited coverage area.

[0031] All service replicas of the central server are hosted on the central server in a containerized manner, and at the same time, the trajectory data of mobile users, including specific time and location information, is collected and stored in the database;

[0032] Step 2: For mobile user U, predict the next moving location through the historical trajectory data of the user's movement, and set the edge server where the location is located as the target server s′ for service migration.

[0033] Step 3: Determine whether performance indicators such as the end - to - end delay and received signal strength indication received by the mobile device meet the service quality requirements. If so, re - predict the next moving location and re - obtain the target server; otherwise, proceed to Step 4;

[0034] Step 4: When mobile user U is within the coverage range of the source wireless access point b and edge server s at time slice t, notify the target edge server s′ to reserve a container, and copy the disk data from the source edge cloud server s to the target edge server s′;

[0035] Replication is preferably implemented using the rsync command; the disk data includes the container base image and the application data image.

[0036] Step Five: Use checkpoint / restore technology to implement data migration of the memory state, so that the edge server s' provides services for the mobile user U.

[0037] The specific process is as follows:

[0038] a. Send a migration instruction to the source edge server s and the target server s'. Store the memory data of the current container on the source edge server s and copy it to the target server s'. Meanwhile, the application in the container on the source edge server s continues to run.

[0039] This period of time T mem depends on the size S of the memory data f and the bandwidth B between the source edge server s and the target edge server s', that is: T ss ′, namely: T mem = S f / B ss ′.

[0040] b. When the fixed interval time is reached, store the memory data of the container again; save the data change between this storage and the previous storage as the incremental memory state data f d , and stop the service running on the source edge server s;

[0041] The fixed interval time is set manually according to the actual scenario.

[0042] c. Copy the incremental memory state data f d to the target edge server s' and restore the service on the target server s'. The edge server s' provides services for the mobile user U.

[0043] The time T for copying the incremental memory del depends on the size of the incremental memory state file f d and the bandwidth B between the source edge server and the target edge server, that is: The bandwidth B between the source edge server and the target edge server ss′ , that is:

[0044] d. The mobile user U enters the coverage range of the wireless access point b' of the target server s' and performs a communication handover to the wireless access point b'; the service migration time is calculated as:

[0045] T total = T mem + T del + T res

[0046] T resThe time for the migrated container to be restored on the target edge server.

[0047] e. Delete the stopped application on the source edge server s.

[0048] The advantages of the present invention are as follows:

[0049] An active migration method for edge services based on mobility prediction in the present invention, compared with the existing technologies, jointly considers the migration target decision and the implementation process of container migration. Based on the mobility prediction results, the target server can be obtained. Before migrating the memory state data, the container base image and application image can be pre - copied from the source edge server, avoiding transferring the container base image file and application image during the actual migration process, thereby reducing the service migration time and end - to - end latency based on containers. Description of the Drawings

[0050] Figure 1 It is a flowchart of an active migration method for edge services based on mobility prediction in the present invention;

[0051] Figure 2 It is a hierarchical mobile edge computing network diagram composed of mobile users, edge servers, and central servers in the present invention;

[0052] Figure 3 It is a scenario diagram of service migration by mobile users adopted in the embodiments of the present invention;

[0053] Figure 4 It is a schematic diagram of service migration by mobile users adopted in the embodiments of the present invention; Detailed Embodiments

[0054] The present invention will be further described below in conjunction with the embodiments and the drawings of the specification. It should be understood that these embodiments are only used to illustrate the technical solutions and implementation processes of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art without any creative modifications based on the embodiments of the present invention belong to the protection scope of the present invention.

[0055] Related research findings show that mobility is predictable, and mobility prediction can be used to predict the next location of a user's future movement through historical movement trajectory data. Therefore, the present invention can obtain the edge server where the location is located in the above manner; and the container has a hierarchical storage structure, including disk files such as container base image files and application image files, as well as memory status data, etc. The present invention jointly considers the decision-making of the migration target server and the implementation of the container migration process, and proposes an active service migration method for containers based on mobility prediction. First, based on the historical trajectory prediction of user movement, the next location of user movement is obtained, and the edge server where the location is located is used as the target edge server for migration; then, based on the hierarchical storage of the container (including container base image, application image, memory status data), some files of the container are pre-migrated before the communication handover occurs, and a container migration time planning method based on checkpoint recovery technology is proposed. This process includes pre-migration of container images and application images, memory status data migration, incremental memory migration, and communication handover, etc., so as to optimize the container migration process to reduce the end-to-end delay and the time of service migration.

[0056] The proactive edge service migration method based on mobility prediction jointly considers the timing planning of container migration and communication handover (Proactive Migration-Handover Method, PMHM). This method includes two aspects: First, the target selection of proactive migration, determining the target edge server for migration and the wireless access point for handover according to the prediction results; Second, the time planning for each stage during the service migration process.

[0057] As Figure 1 shown, the specific steps are as follows:

[0058] Step 1: Build a hierarchical mobile edge computing network communication scenario composed of mobile users, edge servers, and central servers;

[0059] As Figure 2 shown, the mobile user carries a mobile device and moves; the edge server receives the offloading request from the mobile user and provides edge cloud services; the edge server is located at the wireless access point in the access network and has a limited coverage area.

[0060] All service replicas of the central server are hosted on the central server in a containerized manner, and at the same time, the trajectory data of the mobile user is collected and stored in the database, including specific time and location information;

[0061] Step 2: For mobile user U, predict the next moving location through the historical trajectory data of the user's movement, and set the edge server where the location is located as the target server s′ for service migration.

[0062] Step 3: Determine whether performance metrics such as the end-to-end delay and received signal strength indication received by the mobile device meet the quality of service requirements. If so, re-predict the next mobile location and re-obtain the target server; otherwise, proceed to Step 4;

[0063] Step 4: When the mobile user U is within the coverage of the source wireless access point b and the edge server s at time slice t, notify the target edge server s′ to reserve a container, and copy the disk data from the source edge cloud server s to the target edge server s′;

[0064] The copying is implemented using the rsync command; the disk data includes the container base image and the application data image.

[0065] Step 5: Use the checkpoint / restore technology (Checkpoint / Restore In Userspace, CRIU) to implement the data migration of the memory state, so that the edge server s′ provides services for the mobile user U.

[0066] The specific process is as follows:

[0067] a. Send a migration instruction to the source edge server s and the target server s′. Execute the pre-dump command on the source edge server s to store the memory data of the current container, and execute the copy command to copy it to the target server s′. At the same time, the application in the container on the source edge server s continues to run.

[0068] This period of time T mem depends on the size S of the memory data f and the bandwidth B between the source edge server s and the target edge server s′ ss′ , that is: T mem = S f / B ss′ .

[0069] b. When the fixed interval time is reached, execute the dump command to store the memory data of the container again; save the data change between this storage and the previous storage as the incremental memory state data f d , and stop the service operation on the source edge server s;

[0070] The fixed interval time is set manually according to the actual scenario.

[0071] c. Execute the copy command to copy the incremental memory state data f d to the target edge server s′, and execute the restore command on the target server s′ to restore the service. The edge server s′ provides services for the mobile user U.

[0072] The time T for copying the incremental memory delDepending on the incremental memory status file f d with size S fd and the bandwidth B between the source edge server and the target edge server ss′ That is:

[0073] d. The mobile user U enters the coverage area of the wireless access point b' of the target server s', and performs a communication handover to the wireless access point b'; the service migration time is calculated as:

[0074] T total = T mem + T del + T res

[0075] T res is the time for the migrated container to resume on the target edge server.

[0076] e. Use the kill command to delete the stopped application on the source edge server s.

[0077] Example:

[0078] Consider a hierarchical mobile edge computing network, which consists of a mobile user layer, an edge server layer, and a central server layer. In the mobile user layer, mobile users carry mobile devices and can move from one location to another. In the edge server layer, edge servers receive offloading requests from mobile users and provide edge cloud services. These edge servers are located at wireless access points in the access network, such as base stations, WiFi, etc., and have a limited coverage area.

[0079] When the user leaves the coverage area of the source edge cloud server, the quality of service of the services running on the source edge cloud server may decline or even the service may be interrupted. Then, the running services may need to be migrated and a communication handover may be performed, or a new communication path may be found to ensure the data transmission between the computing tasks on the source edge cloud and the users. This example adopts the former processing method to ensure the continuity of the service. In the central server layer, all service replicas are hosted on the central server in a containerized manner. At the same time, the central server can collect and store mobile trajectory data in the database. These data contain specific time and location information, and the controller on it includes the following parts:

[0080] (1) Mobility prediction module: Predict the next moving location through the historical trajectory data of user movement, and set the edge server where the location is located as the target server for service migration.

[0081] (2) Service deployment module: Receive offloading requests from mobile users, deploy services on appropriate edge servers, and send the instruction to the corresponding edge server.

[0082] (3) Monitoring module: Based on performance metrics such as the end-to-end delay and Received Signal Strength Indication (RSSI) received from the mobile device, if the performance fails to meet the quality of service requirements, the monitoring module notifies the mobility prediction module to predict the next mobile location, and the mobility prediction module sends the predicted result of the next location to the planning module.

[0083] (4) Planning module: Uses PMHM to formulate a service migration plan, including when and where to migrate and switch.

[0084] Each MEC server consists of the following parts:

[0085] (1) Edge deployment module: Deploys multiple containers according to the instructions of the controller and executes the service requests of mobile users.

[0086] (2) Migration module: The migration modules on the source edge server and the target edge server cooperate with each other according to the migration instructions of the planning module of the controller, so as to execute each process of container migration.

[0087] For the mobile devices in the mobile user layer, whenever a mobile device connects to a wireless access point, the server in the central cloud layer collects the location information (such as GPS) of the mobile device at fixed time intervals. The mobile device uses the service discovery module to request to offload tasks to the edge server. The local monitoring module of the mobile device is responsible for monitoring information such as the end-to-end delay and RSSI related to the nearby wireless access points and edge servers, and sending the monitored information to the controller.

[0088] As Figure 3 shown, at a certain time slice t, the mobile users U1 and U4 are respectively within the coverage ranges of the edge cloud servers S1 and S4. Due to user movement, at a future time slice t′, the mobile users U1 and U4 may leave the coverage ranges of S1 and S4. The controller entity needs to make a service migration plan based on the monitoring module and the mobility prediction module to decide where (S2, S3 or S5) the corresponding services from the mobile users U1 and U4 should be migrated to and how to implement the entire migration process;

[0089] Based on the results of the mobility prediction module and container migration technology, this embodiment proposes a timing planning method PMHM for joint container migration and communication handover to minimize service migration time and end-to-end delay time. This method includes two aspects: First, during the migration process, some data will be pre-migrated to the target edge server, including disk data such as container base images and application images. Second, determine the optimal time for each stage to execute, which determines the time for triggering each stage of the migration process and handover process.

[0090] Take the edge server where the prediction result of the mobility prediction module is located as the target server, denoted as s′, and use the checkpoint / restore technology to migrate the memory state data of the container. The overall process is as Figure 4 shown:

[0091] Step 1: Disk migration. The mobile user is within the coverage of the source wireless access point b and the edge server s at time slice t. Notify the target edge server s′ to reserve the container, and copy the disk data from the source edge cloud server to the target edge server s′;

[0092] It mainly includes the container base image and the application data image.

[0093] Step 2: Memory state migration. First, dump the memory data of the container for the first time. Then the user performs a communication handover and sends migration instructions to the source and target servers. After a fixed set time interval, store the memory data of the container again, and save the data change between this storage and the previous storage as the incremental memory state data. Stop the service running on the source edge server s. Finally, copy the incremental memory state data to the target server s′ and restore the service on the target server s′.

[0094] Specifically as follows:

[0095] a. Dump memory data: In this stage, store the memory data (i.e., memory snapshot) of the current container on the source edge server s and copy it to the target server s′. At the same time, the application in the container on the source edge server continues to run.

[0096] b. Incremental memory state migration: In this stage, store the memory data of the container again, and save the data change between this storage and the previous storage as the incremental memory state data. The amount of this memory state data is much smaller than the memory state data in stage a, and copy it to the target edge server s′. If the mobile user enters the coverage of the wireless access point b′, perform the communication handover to b′.

[0097] Step 3: Restore the application. In this stage, the service on the target edge server s′ is restored. After that, the edge server s′ provides services for the mobile user.

[0098] Step 4: Delete the applications on the source edge server. If the previous steps are successful, the stopped applications on the source server can be deleted.

[0099] Through experimental simulations of PMHM in terms of performance such as end-to-end delay, downtime delay, and service migration time, the simulation results show that the MAPSM proposed in this embodiment is superior to the compared benchmark service migration method.

[0100] This embodiment uses Pycharm and Python 3.6 to simulate the proposed PMHM solution. The connection configuration between any two edge servers is 100 Mbps or 300 Mbps bandwidth, the connection between the mobile user and the edge server is configured as 100 Mbps or 200 Mbps, and the connection between each edge server and the cloud server is configured as 50 Mbps bandwidth.

[0101] Three applications are used and packaged into Docker containers. Among them, the file system size of the object recognition service based on Yolo is 792 MB, the file system size of the Busybox container is 290 KB, and the file system size of the service based on Openface is 1.86 GB. This embodiment compares PMHM with the benchmark method:

[0102] When the bandwidth between the source edge server and the target edge server is set to 100 Mbps, for Busybox, the end-to-end delay under PMHM is at least 16.7% lower than the benchmark method; for Yolo, the end-to-end delay under PMHM is at least 54.1% lower than the benchmark method; for Openface, the end-to-end delay under PMHM is at least 71.7% lower than the benchmark method. From the above results, it can be seen that the method proposed in this embodiment is more beneficial for applications with larger container base images and application files.

[0103] When the bandwidth between the source edge server and the target edge server is set to 200 Mbps, the service migration time under PMHM is significantly lower than the other three benchmark comparison schemes, and the performance of the two benchmark schemes is roughly the same. The simulation results show that PMHM is a more promising service migration solution in the case of low bandwidth and large file system applications because it realizes the replication of the base container image and application before the memory state migration and only needs to migrate the memory state file.

Claims

1. An active migration method for edge services based on mobility prediction, characterized in that, Specifically, it includes: First, build a hierarchical mobile edge computing network communication scenario composed of mobile users, edge servers, and central servers; For mobile user U, predict the next moving position through the historical trajectory data of the user's movement, and set the edge server where this position is located as the target server s' for service migration; Then, determine whether the indicated performance metrics received by the mobile device meet the quality of service requirements. If so, predict the next moving position again and obtain the target server again; Otherwise, when mobile user U is within the coverage of the source wireless access point b and edge server s in time slice t, notify the target edge server s' to reserve a container, and copy the disk data from the source edge cloud server s to the target edge server s'; Finally, use checkpoint / restore technology to implement the data migration of the memory state, so that the edge server s' provides services for mobile user U; The specific process is as follows: a. Send a migration instruction to the source edge server s and the target server s'. Store the memory data of the current container in the source edge server s and copy it to the target server s'. At the same time, the application in the container on the source edge server s continues to run; This period of time T mem depends on the size S of the memory data f , the bandwidth B between the source edge server s and the destination edge server s' ss′ , that is: T mem = S f / B ss′ ; b. When the fixed interval time is reached, store the memory data of the container again; save the data change between this storage and the previous storage as the incremental memory state data f d , stop the service running on the source edge server s; The fixed interval time is set artificially according to the actual scenario; c. Copy the incremental memory state data f d to the target edge server s′, and resume the service on the target server s′, where the edge server s′ provides services for the mobile user U; Time T for copying incremental memory del depends on the incremental memory status file f d size of bandwidth B between the source edge server and the target edge server ss′ , that is: d. Mobile user U enters the coverage of the wireless access point b' of the target server s', and performs the communication handover to the wireless access point b'. The service migration time is calculated as: T total = T mem + T del + T res T res is the recovery time of the container to be migrated at the target edge server; e. Delete the stopped application on the source edge server s.

2. The active migration method for edge services based on mobility prediction according to claim 1, characterized in that, The mobile user carries a mobile device and moves. The edge server receives the offloading request from the mobile user and provides edge cloud services. The edge server is located at the wireless access point in the access network and has a limited coverage area; All service replicas of the central server are hosted on the central server in a containerized manner, and at the same time, collect and store the trajectory data of mobile users in the database, including specific time and location information.

3. The active migration method for edge services based on mobility prediction according to claim 1, characterized in that, The disk data includes a container base image and an application data image.

Images