SFC deployment and migration method based on VNF dependent component

Abstract

The invention discloses an SFC deployment and migration method based on a VNF dependent component, belongs to the technical field of communication, and is used for solving the problem of providing mobile user services by jointly optimizing and considering an SFC deployment and migration strategy of VNF dependent component migration. According to the method, service request information is put forward to an SDN controller based on a mobile terminal user, the SDN controller analyzes user demand types and resource demands including bandwidth resources, computing resources, storage resources, time delay demands and the like according to the service request information, and a network topology structure, the resource state of an edge server and the network operation condition are combined at the same time. And the deployment and migration of the SFC are optimized by adopting a deep reinforcement learning technology. According to the method, the influence among VNF dependent component migration, SFC deployment and SFC migration decision is comprehensively considered, and the mobile service is efficiently provided.

Description

Technical Field

[0001] The present invention relates to advanced network and communication technologies, and in particular to a SFC deployment and migration method based on VNF dependent components. Background Art

[0002] With the popularity of mobile devices and applications, the provision of mobile services has attracted widespread attention from academia and industry. The delivery of mobile services not only involves solving the problem of service offloading, but also includes service migration to maintain the quality of service (QoS). Some studies have proposed service migration strategies to mitigate server failures, alleviate overload conditions, or achieve cost-effectiveness, aiming to maintain the best performance of service delivery. However, these methods are not suitable for the scenario of mobile service delivery. Solving the migration problem caused by user mobility is crucial to ensure continuous service delivery and optimize overall performance.

[0003] Some studies have considered the migration problem caused by user mobility, but have not considered the significant impact of service state migration caused by user mobility on performance. Later, some studies considered the migration of service state to reduce the impact of user mobility. Researchers have extensively explored the complexity of SFC deployment and migration, including SFC deployment and migration under random user mobility and SFC deployment and migration assuming known user mobility trajectories. However, in most cases, accurately predicting the trajectory of mobile users remains a major challenge.

[0004] While previous research contributions have provided effective solutions for mobile services in multiple environments (such as MEC and cloud computing) and considered user mobility, these strategies are difficult to apply to new scenarios due to the issues of SFC deployment and migration with VNF dependent component migration. Zhang et al. addressed the migration of VNF dependent components during SFC deployment; however, their research focused on the static deployment of SFC. When considering user mobility, the SFC itself must be migrated at the same time. This paper proposes an innovative method for online deployment of mobility-aware SFCs in dense MEC networks integrated with 3GPP cellular networks, and proposes an SFC migration control strategy to redeploy these chains taking into account the migration of VNF dependent components. Summary of the invention

[0005] Purpose of the invention: In order to optimize the end-to-end communication delay and resource usage cost of mobile users, reduce interruption delay, improve service reception rate, efficiently manage limited resources, and provide high-quality mobile services, the present invention provides an SFC deployment and migration method based on VNF dependent components.

[0006] Technical solution: A SFC deployment and migration method based on VNF dependent components. The method is based on the service request information submitted by the mobile terminal user to the SDN controller. The SDN controller analyzes the user demand type and its resource demand according to the service request information, including bandwidth resources, computing resources, storage resources and latency requirements, and optimizes the deployment and migration of SFC by using deep reinforcement learning technology in combination with the network topology, resource status of the edge server and network operation status. The method includes the following steps:

[0007] (1) Build a communication architecture including mobile users, edge servers, and SDN controllers;

[0008] In this communication architecture, mobile users are used to generate dynamic business needs, including computing tasks, data storage and communication services. Edge servers are deployed close to users to provide computing and communication services to users. SDN controllers are used to monitor the business needs of mobile users and the resource status of edge servers in real time, and dynamically adjust resource allocation and scheduling strategies based on network topology information.

[0009] (2) Establish user request models, communication models, and computational models to describe the system’s operating mechanism and performance indicators;

[0010] The user request model includes SFC deployment request and migration request. The deployment request describes the user's demand for bandwidth resources, computing resources, memory resources, and storage resources. The migration request is used to transfer the currently deployed SFC to the target edge server when the user moves.

[0011] The communication model defines the bandwidth allocation strategy of the edge server and calculates the transmission rate and data transmission time based on the Shannon formula and channel interference index, thereby achieving reasonable allocation of communication resources;

[0012] The calculation model includes processing and calculating end-to-end communication delay, resource usage cost, and interruption delay during SFC migration, which is used to evaluate service performance and provide theoretical support for system resource optimization and scheduling;

[0013] (3) formulating resource scheduling optimization problems and clarifying optimization goals based on the network topology information, user request model, communication model, and computing model of the communication system;

[0014] The optimization problem includes maximizing the service access rate requested by users, minimizing the end-to-end communication delay and SFC migration delay, and maximizing resource utilization; during the SFC migration process, optimizing the migration path of VNF and the resource usage cost of dependent components, and reducing the service interruption time caused by migration;

[0015] (4) Deep reinforcement learning-based SFC deployment and migration considering VNF dependent components;

[0016] The SDN controller collects user requests, edge server resource status, and network topology information in real time, and uses a deep reinforcement learning algorithm to dynamically decide the following:

[0017] User access strategy: intelligently select the best edge server for access based on the user's geographic location and business needs;

[0018] SFC deployment and migration strategy: Dynamically adjust SFC deployment and migration paths by comprehensively considering resource utilization, communication latency, and migration latency;

[0019] VNF dependency component migration strategy: Based on the dependencies of VNFs in SFC, key VNF components are migrated first to ensure the integrity and continuity of the service chain.

[0020] Furthermore, the user request model, communication model, and calculation model described in step (2) are as follows:

[0021] (21) User Request Model

[0022] The user request r is denoted as SFCr, is a quintuple of SFCr, where b r Defined as the bandwidth requirement, Indicates the communication delay requirement, TTL r is the life cycle of SFCr, SFC is described as a directed graph G r =(F r ,E r ),in represents the VNF set in SFCr. Any VNF f rh Both require CPU resources and memory resources E r represents the virtual link of SFCr, and the arrival time of SFCr is expressed as m represents the number of time slots, Δ represents the length of the time slot;

[0023] For SFC deployment, use decision variables Indicates whether it passes through edge server node n i Access the network, Indicates whether to use VNF f rh Placed at edge server node n i superior, is a binary variable used to represent the logical link Is it mapped to link e? ij The mathematical expression is:

[0024]

[0025] For SFC migration, when VNF is deployed to edge server node n i Or migrate to edge server node n i If n i If the component resources required to support VNF ​​operation are missing, VNF dependent components must be migrated from other nodes to the edge server node. i , to ensure the normal operation of VNF;

[0026] (22)Communication model

[0027] Edge service node n i The bandwidth resources are evenly distributed to SFCr to ensure fair distribution. The mathematical expression is as follows:

[0028]

[0029] Among them, N ni Indicates that from edge service node n i Access the SFC collection, Indicates server node n i bandwidth resources on Represents r j Whether to use the server node n i Access the network, if from n i 1 if connected to the network, otherwise 0;

[0030] Use parameters Indicates SFCr access edge server node n i The channel interference is caused by the channel contention caused by other SFCs accessing the same node at the same time, where SFCr j The transmission capacity, Represents the channel gain, which is affected by SFCr j With edge service node n i The distance between them and the power gain effect caused by the small-scale fading effect in the channel can be expressed as:

[0031]

[0032] On this basis, \ in the formula represents the impact on SFCr and service node n in addition to SFCr i The transmission rate between is obtained by the following formula:

[0033]

[0034] Among them, ρ 2 is the signal-to-noise ratio, data transmission delay It refers to the data to be transmitted by SFCr to n i The time required is expressed as:

[0035]

[0036] (23) Computational model

[0037] End-to-end communication delay D of SFCr r It is defined as the sum of data transmission delay, VNF processing delay and link propagation delay as follows:

[0038]

[0039] The deployment cost of SFC is defined as follows:

[0040]

[0041] in, is the migration component ψ q The path used, is a binary variable, indicating the path on is mapped to link e ij superior, Indicates n i The unit storage cost.

[0042] Furthermore, the method minimizes the interruption time and enhances the success rate through pre-copy migration. The interruption time is defined as the duration of the last stage of the migration process, during which the service is temporarily interrupted to complete the final synchronization of the data. Indicates the time when the blackout period starts. Indicates the time when the blackout period ends. The expressions of the two are as follows:

[0043]

[0044] in, represents the migration time of the qth iteration, is the actual number of iterations, Indicates the VNF to be migrated f rh The memory page size, Represents VNF f rh The migration rate is, therefore, the interruption time is defined as follows:

[0045]

[0046] M in the formula r Represents the set of VNFs that need to be migrated in SFCr.

[0047] Furthermore, the optimization problem and optimization objective described in step (3) are as follows:

[0048] Constraint 1: Each f rh A VNF instance can only be successfully deployed on one server node at most. It is indivisible, as shown in the following table:

[0049]

[0050] Constraint 2: The communication quality between users connected to the network through the same node affects each other. Each physical node can accommodate at most n max users to maintain the service quality of all users, which can be expressed mathematically as:

[0051]

[0052] Constraint 3: On any server node n i The total CPU and memory resource requirements for SFC deployment and migration must not exceed the maximum CPU and memory resource capacity, which can be expressed mathematically as:

[0053]

[0054] Constraint 4: The used storage resources will not exceed the maximum storage resource capacity, which can be expressed mathematically as:

[0055]

[0056] Among them, ni is node n i The collection of components on ;

[0057] Constraint 5: Any link e ij The total bandwidth consumption on the network must be less than the maximum bandwidth capacity, which is as follows:

[0058]

[0059] The first item indicates the link bandwidth occupied by SFC deployment, and the second item indicates the bandwidth used by SFC migration;

[0060] Constraint 6: The end-to-end delay of any SFCr must satisfy the following constraints:

[0061]

[0062] Based on the above constraints, the JSDM-VDSM problem is defined. The JSDM-VDSM problem includes SFC deployment strategy, SFC migration strategy, and VNF dependent component migration strategy Z. The goal of this method is to maximize QoS by improving service reception rate, reducing resource usage cost, and minimizing communication delay and interruption delay. The mathematical definition of the optimization target is as follows:

[0063]

[0064] Among them, Z r (t) is a decision variable used to determine whether the SFC request r is accepted. If accepted, Z r (t) is set to 1, otherwise it is set to 0; where w1, w2, w3 and w4 are the weights of different indicators.

[0065] Furthermore, step (4) further includes:

[0066] The network architecture identifies the SFCs that need to be orchestrated, including those that need to be deployed and those that are planned to be migrated, and considers SFC migration as a joint optimization of VNF redeployment and SFC rerouting;

[0067] Get the state of the environment and the current SFC to be orchestrated, which includes the available CPU, memory, storage, bandwidth, and component resources of the physical node; the CPU, memory, and component resource requirements of the VNF to be deployed; the number of VNFs that have not yet been placed in the deployed SFC; the partial communication delay of the current SFC, and the lifetime of the SFC;

[0068] The state s will be obtained ι Input to Actor old Network, obtain VNF deployment decision a ι ,Afterwards, iteratively interacts with the environment to collect a set of experimental data[s ι ,a ι ,r ι ,s ι+1 ] and store it in memory;

[0069] Calculation of advantage function based on experimental data Used to update the Critic network; using truncated objective function Update the Actor network. After multiple iterations, the parameters of the Actor network are passed to the Actor. old Network,iteration process continues until convergence;

[0070] The solution process of JSDM-VDSM problem includes obtaining SFC request r, determining user access node and SFC deployment or migration strategy Strategy-based and Determine the migration strategy for VNF dependent components Before learning the service provisioning policy, the SDN controller collects global information and releases SFCs that have reached their lifecycle, and then identifies SFCs that need to be deployed or migrated;

[0071] The SFC requested for deployment is the SFC request that arrives randomly in the current time slot t, while the SFC requested for migration is the SFC request that meets the migration criteria in the same time slot; if there is no request to be processed, the SDN controller will enter the next time slot and continue to obtain requests until r batch is not empty; then, r batch The SFC requests in will be deployed or migrated in sequence;

[0072] For deployment and migration requests, the user area associated with the SFC request is first determined, and then the SDN controller evaluates whether the base station in the current area has sufficient capacity. If there is sufficient capacity, access is made through the base station; if there is insufficient capacity, the SDN controller plans to connect through neighboring areas, first identifying neighboring areas and then calculating the access capacity of these areas; then, the number of accessible nodes with the maximum access capacity is determined, and the area that can accommodate the maximum capacity is selected, and further an access node is selected from the max_num_node with 1-∈, and an access node is selected from the n_max_num_node set with probability ∈;

[0073] After selecting the access point, determine the deployment or migration strategy of the SFC and the migration strategy of the VNF dependent components. First, identify the first VNF ​​of the SFCr, then obtain the current status, and according to the strategy π θold (a ι ,s ι ), determine VNF f rh The placement position, if f rh Can be placed at node a ι On, but a ι Currently not available rh To run the required component resources, you must add the required f rh Migrate the dependent components to a ι , then place the next VNF ​​and record the updated state s ι+1 ;

[0074] The above process continues until the SFC is successfully deployed or migrated; the reward is calculated based on the deployment result. If the deployment is successful, the reward for successful deployment is calculated; if the migration is successful, the reward for successful migration is calculated; if the SFC deployment or migration fails, the system will be restored to the last successfully achieved state sι-u+1 .

[0075] Beneficial effects: The method of the present invention comprehensively considers the mobility of terminal users and the service requests of mobile terminal users. First, an SFC deployment and migration network architecture considering the migration of VNF dependent components is constructed. Then, on the basis of the architecture, an SFC deployment and migration method for joint VNF ​​dependent component migration is proposed, including: obtaining SFC request information, determining access nodes, and designing SFC deployment and migration decisions and VNF dependent component migration strategies. Finally, an experimental evaluation was conducted, and the results showed that since the present invention comprehensively considers the impact between VNF dependent component migration, SFC deployment and SFC migration decisions, efficient and high-performance mobile services can be provided. BRIEF DESCRIPTION OF THE DRAWINGS

[0076] Figure 1 It is a schematic diagram of the process of the method of the present invention;

[0077] Figure 2 Provides an example diagram for mobile services under the MEC system architecture in an embodiment of the present invention;

[0078] Figure 3 A network architecture diagram of SFC deployment and migration considering VNF dependent component migration in an embodiment of the present invention;

[0079] FIG4 is a graph showing the convergence of the algorithm in an embodiment of the present invention, wherein: FIG4(a) is a graph showing the average acceptance rate of SFC and the convergence of the strategy gradient loss. FIG4(b) is a graph showing the average acceptance rate of SFC and the convergence of the strategy gradient loss.

[0080] Figure 5 The following is a schematic diagram showing a comparison of algorithm performance evaluation under different numbers of service function chains and different migration conditions in an embodiment of the present invention, wherein: Figure 5 (a) is the average reception rate when speed = high, |F| = 3, Figure 5 (b) is the average end-to-end communication delay when speed = high and |F| = 3. Figure 5 (c) is the average number of migrations of VNF dependent components when speed = high and |F| = 3. Figure 5 (d) is the average migration cost of VNF dependent components when speed = high, |F| = 3, Figure 5 (e) is the average placement cost of VNF dependent components when speed = high, |F| = 3, Figure 5 (f) is the average SFC interrupt delay when speed = high, |F| = 3, Figure 5 (g) is the total number of migrations of VNF dependent components when speed=high,|F|=3, Figure 5 (h) is the total migration cost of VNF dependent components when speed = high, |F| = 3, Figure 5 (i) is the total placement cost of VNF dependent components under speed = high, |F| = 3;

[0081] Figure 6 This is a schematic diagram of algorithm performance evaluation comparison under different migration conditions and different user moving speeds in an embodiment of the present invention, where: Figure 6 (a) is Γ d =240, average acceptance rate under |F|=3, Figure 6 (b) is Γ d =240, average end-to-end communication delay under |F|=3;

[0082] Figure 7 This is a schematic diagram of algorithm performance evaluation comparison under different user moving speeds and different service function chain lengths in an embodiment of the present invention, where: Figure 7 (a) is dis = 2, Γ d =240 average reception rate, Figure 7 (b) is dis = 2, Γ d = Average end-to-end communication delay under 240. DETAILED DESCRIPTION

[0083] In order to explain the technical solution disclosed by the present invention in detail, the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0084] The present invention provides a SFC deployment and migration method based on VNF dependent components, which maximizes QoS mainly by improving service reception rate, reducing resource usage cost, and minimizing communication delay and interruption delay.

[0085] Although new smart mobile devices, such as wearable devices, smart vehicles, robots, and drones, have significantly improved the quality of our lives by providing greater convenience and improving efficiency, they have also led to an increase in computationally intensive tasks that require high computing power and performance, especially in terms of low latency and continuous communication. Given the limited resources of mobile devices, such as computing power and battery life, it is usually not feasible to process these tasks locally. The exponential growth in the number of mobile devices and the increase in data traffic generated by computationally intensive tasks may increase the congestion of core networks in the future. To address these challenges, offloading mobile services to edge servers close to mobile terminals is a promising solution. This approach not only improves the quality of service by providing additional resources and reducing communication delays, but also helps alleviate network congestion.

[0086] By integrating network function virtualization (NFV) technology, mobile services are represented by ordered virtual network functions (VNFs) that form service function chains (SFCs). Deploying these SFCs helps alleviate the problem that resource-constrained edge nodes have difficulty accommodating large-scale, resource-intensive services, such as high-definition video streaming, real-time gaming, and complex data analysis services, thereby improving the success rate of mobile service offloading. However, the high mobility and uneven distribution of mobile devices lead to significant spatial and temporal network load imbalance. In addition, all mobile devices compete for limited edge resources. Therefore, meeting quality of service (QoS) requirements by deploying SFCs is a challenging task that needs to be solved urgently. In order to maintain service continuity when mobile devices are moving, seamless migration of SFCs is crucial, which makes service delivery more complicated.

[0087] In previous studies, VNFs have been implemented in the form of packages, and functional instantiation is achieved by placing these VNF packages on servers. However, in large-scale heterogeneous edge networks, the diversity of virtualization platforms and the frequent iterative updates of VNFs require that VNF ​​packages have decoupling characteristics. The decoupling of VNF packages is essential for flexible and efficient VNF ​​placement, ensuring rapid adaptation to changing business needs. The deployment or migration of SFC requires that VNFs be deployed on servers with the necessary component resources (e.g., Ubuntu, VEKET, IPFilter, and Hash Switch) - these components are collectively referred to as VNF dependent components - to support the operation of VNFs. In previous SFC deployment or migration processes, it is unrealistic to assume that each server node can support the operation of all VNFs, especially considering the limited resources of edge server nodes.

[0088] When the node where the VNF is located lacks the necessary components to support its operation, the dependent components of the VNF can be migrated to the corresponding server node to enable the VNF to work properly. The deployment and migration of SFC is a complex process in itself. Including the migration of VNF dependent components, while more accurately reflecting the real-world scenario, further exacerbates the challenges of mobile service provision.

[0089] When considering VNF dependent components, component reuse and migration strategies must be evaluated during the SFC deployment process to ensure optimal operation of the VNF. Reusing component resources can reduce storage costs, but may result in increased communication latency. On the other hand, migrating components incurs migration and storage costs, but has the potential to reduce communication latency. During the SFC migration process, it is important to consider the VNF to be migrated and the target migration node. VNFs can be migrated to nodes that support VNF ​​operation or do not support VNF ​​operation. If a node does not support VNF ​​operation, the VNF dependent components must also be migrated to that node. Therefore, how to efficiently manage limited resources while providing high-quality mobile services is a pressing issue that needs to be addressed.

[0090] The present invention provides a method for deploying and migrating SFCs based on VNF dependent components. The method is based on mobile users sending service request information to SDN controllers, and the SDN controllers perceive user requests, track the locations of mobile users, coordinate the deployment and migration of SFCs (including migration of VNF dependent components), and manage network resource allocation. The user's mobility, the requirements for high computing power and low latency, the resource limitations of edge nodes, and the challenges brought by the migration of VNF dependent components are taken into consideration. Due to the complexity of the problem, we propose a JSDM-VDSM architecture based on deep reinforcement learning, which lays the foundation for the design of the JSDM-VDSM algorithm. The algorithm includes a mechanism for determining the SFC to be migrated, a user access strategy, an SFC deployment and migration strategy, and a VNF dependent component migration strategy. Compared with traditional methods, the proposed mechanism provides a more flexible, efficient and adaptable solution because it can intelligently adapt to dynamic and complex network environments. The performance of JSDM-VDSMA is evaluated through simulation experiments. The results show that the proposed algorithm shows superior performance compared to the benchmark algorithm, especially in terms of reception rate and end-to-end delay.

[0091] The implementation process of the technical solution provided by the present invention is described in detail below.

[0092] The method described in the present invention is to realize SFC deployment and migration considering the migration of VNF dependent components. It mainly includes mobile terminal users, 3GPP-based MEC network and SDN controller. Mobile terminal users make service requests and move over time. MEC network servers provide communication and computing resources for mobile terminal users, and SDN controllers are responsible for managing terminal users and network resources.

[0093] The main implementation process of the method of the present invention is as follows Figure 1 As shown, based on the above technical solution, further detailed description is made in the embodiment, which specifically includes the following steps:

[0094] (1) Establishing the communication system architecture

[0095] This embodiment constructs a communication system architecture consisting of mobile users, edge servers, and SDN controllers. Mobile users are mainly responsible for dynamically generating business needs, edge servers provide computing and communication services for mobile users, and SDN controllers play a coordination and management role in the entire system, responsible for real-time allocation of user business needs and edge server resources. This architecture ensures that the network can efficiently handle complex resource scheduling and dynamic business requests through the coordinated work of mobile users, edge servers, and SDN controllers.

[0096] (2) Establishment of user request model, communication model and calculation model

[0097] This embodiment establishes a user request model, a communication model, and a computing model to provide a comprehensive description of the communication system. The user request model defines SFC deployment requests and migration requests, and covers the requirements for bandwidth resources, computing resources, memory resources, and storage resources, as well as the requirements for life cycle and communication delay; the communication model designs the bandwidth allocation strategy of the edge server in detail, quantifies the communication channel interference, data transmission rate, and transmission time; the computing model starts from the end-to-end communication delay, resource usage cost, and interruption delay of SFC migration, providing a theoretical basis for system performance analysis and optimization.

[0098] (3) Formulation of optimization problems and determination of optimization goals

[0099] This embodiment formulates optimization problems based on the system architecture and related model information, and clarifies the overall optimization goals. The optimization problems mainly revolve around the user request model, communication model, computing model, and topology information of the edge server, aiming to cope with extreme user needs and edge resource constraints. The optimization goals must not only meet the user's mobility, high computing power, and low latency requirements, but also maximize the use of limited resources at the edge nodes, and effectively cope with the complexity and challenges of VNF dependent component migration, ensuring the dual improvement of network performance and resource efficiency.

[0100] (4) SFC deployment and migration method based on VNF dependent component migration

[0101] This embodiment proposes a SFC deployment and migration method based on VNF dependent components, and builds a network architecture on this basis. By introducing deep reinforcement learning technology, this method intelligently decides the SFC migration mechanism, user access strategy, and SFC deployment and migration strategy, while optimizing the migration of VNF dependent components. Compared with traditional methods, this method can not only better adapt to dynamic and complex network environments, but also significantly reduce migration delays, improve the flexibility and efficiency of resource allocation, and provide important support for efficient management of future network communications.

[0102] The implementation process is described in detail below.

[0103] 1) Construction of communication system architecture

[0104] The architecture consists of three different layers: the mobile user (MU) layer, the MEC layer, and the software defined network (SDN) control layer. That is, it consists of mobile users, edge servers, and SDN controllers to meet the communication and computing needs in a dynamic network environment. In this architecture, mobile users interact with the system by generating business needs (such as computing tasks or data transmission requests); edge servers are deployed close to users and provide users with low-latency, high-performance services by allocating computing resources, storage resources, and bandwidth resources; the SDN controller, as the core unit of global resource scheduling, monitors users' business needs, network topology information, and resource status of edge servers in real time, and intelligently optimizes resource allocation and scheduling strategies. In the communication process, the mobile user's request is first parsed by the SDN controller, which schedules the request to the optimal edge server based on the user's demand type and the current network status, thereby achieving efficient use of computing and communication resources. By simulating different user needs and network scenarios, the architecture verifies its flexibility in resource management and its ability to respond quickly to user needs in a dynamic environment.

[0105] Specifically, the running time is divided into discrete time slots The time interval is Δ.

[0106] MU (mobile user): Let MU = {u1,u2,…,u r ,…,u |MU|} is a set of mobile users. Each mobile user generates a computationally intensive SFC request within time T. Mobile user u r The position in time slot t is given by In each time slot, the communication system has to handle both newly arrived requests and ongoing user requests, which are caused by the random arrival and movement of users. Therefore, for these newly arrived requests, corresponding SFCs need to be deployed, while for existing users, migration may be required to ensure continuous service.

[0107] Edge Server: is a collection of edge servers. Each edge server is located at the base station and is equipped with specific computing, memory, storage, and bandwidth resources. and These resources enable the server to provide communication and computing services to mobile users. The location of the edge server is fixed, with a height of 0. The location of the edge server is represented by Indicates that Yes i The horizontal coordinates of the network, is its network vertical coordinate. The definition of edge server coordinates helps provide precise geographic location information, which is crucial for optimizing service selection for mobile users. Due to the limited resources of edge servers and the changing locations of mobile users, the VNFs of SFC may be hosted on different edge servers, and each server may serve different mobile users at different time slots.

[0108] Controller: The SDN controller plays a key role in sensing user requests, tracking the location of mobile users, coordinating the deployment and migration of SFCs (including the migration of VNF dependent components), and managing network resource allocation. Figure 2 As shown in the figure, in time slot t, the SDN controller receives newly arrived service requests from users 1, 2, and 3, each of which accesses the network through their respective base stations. The SDN controller is responsible for coordinating the deployment of SFCs to edge servers to meet user requests. VNF1 and VNF2 of SFC1 are placed on MEC1, VNF1 of SFC2 is placed on MEC2, and VNF2 is placed on MEC3. For SFC3, VNF2 is placed on MEC1 and VNF4 is placed on MEC2. However, since MEC1 lacks the necessary component resources to support the operation of VNF1, software1 needs to be migrated from MEC2 to MEC1 to ensure the normal operation of VNF1. In the next time slot t+1, users 1, 2, and 3 have moved to new locations. User 1 moves slowly and continues to access services through the previous base station and edge server. User 2 moves at a moderate speed and modifies its access method with a rerouting strategy. At the same time, user 3 moves quickly and ensures service quality and continuity by strategically migrating VNFs and switching to new access points. SFC3's VNF2 is migrated from MEC1 to MEC3, while SFC3's VNF4 is migrated from MEC2 to MEC3. However, since MEC3 does not have the necessary components to support VNF4 execution, software4 needs to be migrated from MEC2 to MEC3. During the SFC deployment and migration process, the SDN controller manages the allocation and release of network resources to adapt to changing needs.

[0109] 2) Establishment of user request model, communication model and calculation model

[0110] According to the needs of the communication system, user request model, communication model and calculation model are established to describe the operation mechanism and performance indicators of each part of the system.

[0111] User request model: User requests are divided into two categories, namely SFC deployment requests and migration requests. SFC deployment requests include requirements for bandwidth resources, computing resources, memory resources, and storage resources, and define the request lifecycle and communication latency requirements; migration requests are mainly for user mobility scenarios, requiring the currently deployed SFC to be migrated to a new edge server while ensuring service continuity and low latency during the migration process.

[0112] Communication model: The communication model defines the bandwidth allocation strategy of the edge server and establishes a calculation method for the interference, transmission rate, and data transmission time of the communication channel. Specifically, the communication model estimates the channel capacity based on Shannon's law, and calculates the transmission rate and communication delay of each request based on the interference and bandwidth resource status.

[0113] Computational model: The computational model defines the calculation method of end-to-end communication delay by analyzing the computing resource allocation and usage cost of edge servers. At the same time, the model also describes the interruption delay of the business function chain during the migration process, which is used to evaluate the impact of the migration strategy on service performance. Together, these models provide a clear theoretical basis for the design of optimization algorithms.

[0114] Specifically, the user request is represented as an SFC request. In order to more clearly explain the process of mathematical processing, the present invention represents any user request as SFCr. The SFC request is all user requests, and SFCr is a specific one of the SFC requests. is a quintuple of SFCr, where b r Defined as the bandwidth requirement, Indicates the communication delay requirement, TTL r = l·Δ is the life cycle of SFCr. SFC is described as a directed graph G r =(F r ,E r ),in represents the VNF set in SFCr. Each VNF f rh Requires CPU resources and memory resources E r represents the virtual link of SFCr. The arrival time of SFCr is expressed as

[0115] Γ d represents the set of SFC deployment requests, Γ m Represents a collection of migration requests. In the present invention, "base station", "edge server" and "physical node" are synonymous and can be used interchangeably. For the deployment of SFC, it is necessary to determine which base station the user uses to access the network, the deployment location of VNFs, the migration of VNF dependent components, and the mapping path of the virtual link, using the following representation:

[0116] The decision variables in (1) Determine whether to pass through the edge server n i Access the network.

[0117] (2) Decide whether to use VNF f rh Place in n i superior.

[0118] (3) is a binary variable representing Is it mapped to link e? ij superior.

[0119]

[0120]

[0121] For SFC migration, it is necessary to determine the appropriate time for migration, confirm whether the access point needs to be switched, identify the VNFs that need to be migrated, and determine whether the VNF dependent components need to be migrated. r Indicates the set of VNFs waiting to be migrated. Records the source node where each VNF to be migrated is placed. Memory Dirty Rate and the size of the memory pages used In addition, the target node to which the VNF is to be migrated needs to be carefully selected. Ψ={ψ1,ψ2,...,ψ q ,...,ψ |Ψ|} is a collection of components that support VNF ​​operation. Represents node n i The VNF dependency components on the node n. i Or migrate to node n i If n i If the component resources required to support VNF ​​operation are missing, VNF dependent components must be migrated from other nodes to n i , to ensure the normal operation of VNF. Represents VNF dependent components ψ q The storage resources occupied by binary variables In (4), we denote the component ψ q Is it present at node n? i We use Represented as f rh The collection of all components that were migrated.

[0122]

[0123] Edge Server i The bandwidth resources are evenly allocated to SFCr(MUr) to ensure fair distribution, and its expression is as follows:

[0124]

[0125] in, Indicates that from n i The parameters defined in formula (6) are Represents MUr accessing edge service node n i The channel interference is caused by the channel contention caused by other MUs accessing the same node at the same time. MUr j The transmission capacity, Represents the channel gain, which is affected by MUr j With node n i The distance between them and the power gain caused by the small-scale fading effect in the channel,

[0126]

[0127] On this basis, SFCr and service node n i The transmission rate between can be obtained by the following formula:

[0128]

[0129] Among them, ρ 2 is the signal-to-noise ratio. The transmission delay expressed in formula (8) refers to the transmission delay of the data to be transmitted from SFCr to n i The time required.

[0130]

[0131] The end-to-end communication delay of SFCr is defined as the sum of data transmission delay, VNF processing delay and link propagation delay as follows:

[0132]

[0133] During the SFC deployment process, certain costs are incurred due to the use of CPU and memory resources and the migration of VNF dependent components. Because CPU and memory usage costs are an inherent part of the deployment process, this study focuses on the costs associated with the migration of VNF dependent components. Therefore, the deployment cost of SFC is defined as follows:

[0134]

[0135] in, is the migration component ψq The path used, is a binary variable, indicating the path on is mapped to link e ij superior, Indicates n i The unit storage cost.

[0136] In the present invention, a pre-copy migration technology is adopted, which has the characteristics of flexible migration characteristics, minimizing interruption time and enhancing success rate. In the pre-copy migration technology, the interruption time is defined as the duration of the last stage of the migration process, during which the service is temporarily interrupted to complete the final synchronization of data. Indicates the time when the blackout period starts. Indicates the time when the blackout period ends. They are described in detail in (11) and (12) respectively:

[0137]

[0138] Among them, (11) and (12) represents the migration time of the qth iteration, is the actual number of iterations, and Represents VNF f rh Therefore, the definition of the interruption time is as shown in (13):

[0139]

[0140] 3) Formulation of optimization problems and optimization goals

[0141] Based on the user request model, communication model, and computing model, the optimization problem aims to solve the following difficulties: how to reasonably allocate the limited resources of edge servers when user requests change dynamically, while meeting users' needs for low latency and high computing power. The optimization goals include:

[0142] Improve business acceptance rate: maximize the number of user requests received and ensure the system's service capabilities under high-load scenarios.

[0143] Minimize communication and migration delays: Optimize SFC deployment and migration strategies, reduce end-to-end communication delays, and reduce the impact of SFC migration on user experience.

[0144] Improve resource utilization: Under the premise of ensuring service quality, make full use of the computing, bandwidth and storage resources of edge servers to avoid resource waste.

[0145] Addressing the challenges of VNF dependency migration: For the dependencies between VNFs in the business function chain and the dependencies between VNFs and components, the migration complexity and performance overhead caused by dependencies can be reduced by optimizing the migration path and strategy.

[0146] Specifically, this step analyzes the constraints that arise when deploying and migrating SFCs, especially those related to the migration of VNF dependent components, and then formulates this problem. Constraint (14) ensures that each f rh It can only be successfully deployed on one server node at most, which means that the VNF instance is indivisible:

[0147]

[0148] The communication quality between users connected to the network through the same node affects each other. It is proposed that each physical node can accommodate up to n max users to maintain quality of service for all users:

[0149]

[0150] Constraints (16) and (17) ensure that at any server node n i The total CPU and memory resource requirements for SFC deployment and migration must not exceed the maximum CPU and memory resource capacity:

[0151]

[0152] Constraint (18) ensures that the used storage resources do not exceed the maximum storage resource capacity:

[0153]

[0154] in, is node n i A collection of components on .

[0155] Any link ij The total bandwidth consumption on the network must be less than the maximum bandwidth capacity. The condition is as follows:

[0156]

[0157] The first term represents the link bandwidth occupied by SFC deployment, and the second term represents the bandwidth used by SFC migration. In order to ensure the quality of service delivered, the delay constraint (20) ensures that the end-to-end delay of any SFCr must meet the following specific constraints:

[0158]

[0159] Defining JSDM-VDSM issues: SFC deployment strategy and SFC migration strategy (i.e. and ), and VNF dependent component migration strategy This will significantly affect the quality of service provided. and Under the decision of , the goal of the present invention is to maximize QoS by improving service reception rate, reducing resource usage cost, and minimizing communication delay and interruption delay. The optimization goal is defined as follows:

[0160]

[0161] Among them, Z r (t) is a decision variable used to determine whether the SFC request r is accepted. If accepted, Z r (t) is set to 1, otherwise it is set to 0. w1, w2, w3 and w4 are the weights of different indicators.

[0162] 4) SFC deployment and migration method based on VNF dependent component migration

[0163] This step is also a SFC deployment and migration method based on VNF dependent components proposed by the method described in the present invention, and a deep reinforcement learning-driven intelligent network architecture is constructed.

[0164] A network architecture for joint VNF ​​dependency component migration (JSDM-VDSM) is constructed, including a mobile user access module, an edge server resource management module, and an SDN control module. The architecture supports dynamic user access, resource scheduling, and service migration functions, and models the dependencies between different VNFs in the service function chain.

[0165] Deep reinforcement learning algorithm design: Deep reinforcement learning technology is used to design intelligent decision-making modules, including SFC migration mechanism, user access strategy, SFC deployment and migration strategy, and VNF dependent component migration strategy. By defining a reward function, the algorithm is encouraged to maximize resource utilization, reduce migration latency, and ensure the integrity of the service function chain during the migration process.

[0166] Policy optimization and training: In a simulation environment, user requests and network resource status are generated as training samples, and resource allocation strategies are continuously optimized through deep reinforcement learning models. The VNF migration path is optimized, and VNF components with key dependencies are migrated first to ensure that service quality is not affected by the migration process.

[0167] The specific implementation process includes the following steps:

[0168] Firstly, the network architecture of JSDM-VDSM is introduced. JSDM-VDSMA is also proposed to obtain SFC deployment and migration strategies with VNF dependent component migration strategies.

[0169] Figure 3 The network architecture of JSDM-VDSM is presented, which mainly consists of two key steps: environment and training network. Since users arrive and move randomly, JSDM-VDSM uses online learning methods in a dynamic network environment to dynamically determine A, X, Y, and Z.

[0170] Initially, the network architecture identifies the SFCs that need to be orchestrated, including the SFCs that need to be deployed and the SFCs that are planned to be migrated. SFC deployment involves placing VNFs on physical nodes and mapping virtual links between VNFs to corresponding physical links. The migration process migrates VNFs from the original physical nodes to the new physical nodes while rerouting SFCs to improve service quality. Therefore, SFC migration can be viewed as a joint optimization of VNF redeployment and SFC rerouting.

[0171] Next, the system obtains the state of the environment and the current SFC to be orchestrated. This state includes the available CPU, memory, storage, bandwidth, and component resources of the physical node; the CPU, memory, and component resource requirements of the VNF to be deployed; the number of VNFs that have not yet been placed in the deployed SFC; the partial communication delay of the current SFC; and the TTL (time to live) of the SFC.

[0172] The status obtained ι Input to Actor old Network, obtain VNF deployment decision a ι . Then, we iteratively interact with the environment to collect a set of experimental data [s ι ,a ι ,r ι ,s ι+1 ] and store it in memory. Calculate the advantage function based on experimental data Used to update the Critic network. Next, use the truncated objective function Update the Actor network. After several iterations, the parameters of the Actor network are passed to the Actor old The iteration process continues until convergence.

[0173] JSDM-VDSMA includes algorithms for obtaining SFC requests, determining user access nodes, methods, and SFC deployment or migration strategies approach, based on strategy and VNF dependent component migration strategy Before learning the service provisioning policy, the SDN controller first collects global information and releases the SFCs that have reached their lifecycle. Then, it identifies the SFCs that need to be deployed or migrated. The SFCs requested for deployment are SFC requests that arrive randomly within the current time slot t, while the SFCs requested for migration are SFC requests that meet the migration criteria within the same time slot. If there are no requests to be processed, the controller will enter the next time slot and continue to obtain requests until r batch is not empty. Then, r batch The SFC requests in the will be deployed or migrated sequentially.

[0174] For deployment and migration requests, it is crucial to select an access node for network communication. This selection plays a key role in determining the access delay, which is affected by the remaining capacity of the node and the communication distance between the node and the user. Initially, the user area associated with the SFC request is determined. The controller then evaluates whether the base station in the current area has sufficient capacity. If there is sufficient capacity, access is made through this base station. If there is not enough capacity, the controller plans to connect through neighboring areas. First, neighboring areas are identified, and then the access capacity of these areas is calculated. Next, the number of accessible nodes with the maximum access capacity is determined, and the area that can accommodate the maximum capacity is selected. Next, an access node is selected from max_num_node with 1-∈, and an access node is selected from the set of n_max_num_node with probability ∈.

[0175] After selecting the access point, determine the deployment or migration strategy of the SFC and the migration strategy of the VNF dependent components. The process first identifies the first VNF ​​of the SFCr and then obtains the current status. According to the strategy π θold (a ι ,s ι ), determine f rh If f rh Can be placed at node a ι On, but a ι Currently not available rh To run the required components, you must install the required f rh Migrate the dependent components to a ι Then place the next VNF ​​and record the updated state s ι+1 This process continues until the SFC is successfully deployed or migrated. The rewards are calculated based on the deployment results. If the deployment is successful, the rewards for successful deployment are calculated. If the migration is successful, the rewards for successful migration are calculated. If the SFC deployment or migration fails, the system will be restored to the last successfully achieved state. ι-u+1 .

[0176] In this embodiment, in order to verify the actual effect of the present invention, a simulation experiment was carried out. In order to better illustrate the effect of the present invention, a rerouting-based algorithm was used for comparison. Figure 4 shows the convergence behavior of JSDM-VDSMA and JSDR-VDSMA, highlighting the changing trends of the reception rate and policy gradient loss. As shown in Figure 4(a), the average reception rate effect of SFC shows a continuous upward trend with the increase of learning steps, and finally stabilizes at a stable value. At the same time, the policy gradient loss in Figure 4(b) gradually decreases with the increase of learning steps, and finally converges to a stable state. This shows that the learning process of the two algorithms is effective and stable. The experimental results show that compared with traditional methods, the JSDM-VDSM method has significant advantages in communication delay, business continuity and the ability to adapt to dynamic environments. Its effects mainly include the following points:

[0177] (1) Performance analysis as the number of SFC requests changes

[0178] Figure 5 The performance evaluation of JSDM-VDSMA and JSDR-VDSMA algorithms under different migration conditions (i.e., dis, where dis represents the SFC migration threshold) and different numbers of SFC requests is shown. In this evaluation, the user's moving speed is set to a faster scenario (e.g., a car with a speed range of [135km / h, 150km / h]), and the length of the SFC is fixed to 3. Figure 5 As shown in (a), the acceptance rate decreases as the number of SFC requests increases. The reason for this decrease is that the number of received SFCs increases slower than the number of SFC requests. However, the total number of received SFCs is still increasing. When the shortest hop distance between the area where the SFC was last deployed and the current area is greater than or equal to the migration threshold dis, and the SFC is within its life cycle, the SFC will be migrated. In addition, Figure 5 (a) also shows that the change of the migration threshold dis leads to the performance difference between JSDM-VDSMA and JSDR-VDSMA algorithms.

[0179] For JSDM-VDSMA, the SFC reception rate decreases as the migration threshold dis increases. This is because it is assumed that all users generating SFC requests move randomly from the center of the MEC network. As dis increases, the SFC stays at its original access and deployment location for a longer time, causing congestion in nodes close to the center of the MEC network. When a new SFC request arrives at the MEC center, there may not be enough resources to support its deployment, resulting in a decrease in the reception rate. In contrast, when dis is smaller, SFC migrations are more frequent. In this case, previously deployed SFCs migrate to servers farther away from the MEC center, freeing up resources for new requests, which in turn increases the SFC reception rate. For JSDR-VDSMA, the SFC reception rate also decreases as the migration condition dis increases. In an MEC network, users' QoS are interdependent, so the maximum number of users that each base station can accommodate is limited. As dis increases, access resources close to the MEC network center are rapidly consumed. When a new SFC request arrives at the MEC center, there are not enough resources to support its deployment, resulting in a decrease in the SFC reception rate. An interesting observation is that from Figure 5 As can be seen in (a), JSDM-VDSMA consistently outperforms JSDR-VDSMA in terms of SFC reception rate under all migration conditions. This superior performance is attributed to the fact that JSDM-VDSMA not only switches access nodes but also migrates VNFs when the migration threshold is reached, thus ensuring the best performance of SFC. In contrast, JSDR-VDSMA only switches access nodes without migrating VNFs, so its performance is relatively low.

[0180] Figure 5 (b) shows the end-to-end communication delay, from which it can be observed that as the number of SFC requests increases, the delay also increases. This is because the increase in the number of received SFCs leads to an increase in access delay, processing delay, and link delay. In addition, Figure 5 (b) also shows that as the migration condition dis increases, the latency of both JSDM-VDSM and JSDR-VDSM algorithms increases. This is because higher dis reduces the frequency of SFC migration, which means that when users move to farther locations, they still communicate with the previously deployed SFCs through the original access nodes, resulting in higher access latency. Therefore, an increase in dis leads to an increase in overall latency.

[0181] Under different migration conditions, JSDM-VDSMA is always better than JSDR-VDSMA in terms of latency. This is because JSDM-VDSMA not only switches access nodes during SFC migration, but also migrates VNFs, providing more high-quality resources for the newly arrived SFCs that need to be deployed, thereby reducing the overall latency and performing better than JSDR-VDSMA.

[0182] Figure 5 (c), (d), (e) and (f) respectively show the average number of migrations of SFC, the average migration cost, the average placement cost of VNF dependent components and the average interruption delay of SFC, where the formula for interruption delay is defined in (13). Figure 5 As shown in (c), under different migration conditions, the average migration times of VNF dependent components in JSDM-VDSM and JSDR-VDSM algorithms are similar, which is about half of the SFC length. Figure 5 (d) shows that under different migration conditions, the average migration costs of VNF dependent components of JSDM-VDSM and JSDR-VDSM algorithms are also similar. The migration cost of VNF dependent components is defined as the shortest path to migrate VNF dependent components. Figure 5 In (e), it can be clearly seen that under different migration conditions, JSDM-VDSMA always outperforms JSDR-VDSMA in terms of VNF dependent component placement cost. Figure 5 (f) shows that the interruption delay of JSDM-VDSMA remains constant under different migration conditions. This is because, by definition, the calculation of the interruption delay is independent of the migration condition (i.e. and ).

[0183] Figure 5 (g), (h), and (i) show the total number of migrations, total migration cost, and total placement cost of VNF dependent components, respectively. These results clearly show that the JSDM-VDSM algorithm consistently outperforms the JSDR-VDSM algorithm under different migration conditions. In addition, as the number of SFC requests increases, the total number of migrations, total migration cost, and total placement cost of VNF dependent components increase significantly.

[0184] (2) Performance analysis at different moving speeds

[0185] Figure 6 The performance evaluation of JSDM-VDSM and JSDR-VDSM algorithms under different migration conditions and user moving speeds is shown. In this evaluation, the number of SFCs is fixed to 240 and the length of each SFC is 3. Figure 6 (a) shows that JSDM-VDSMA consistently outperforms JSDR-VDSMA in terms of SFC acceptance rate under various migration conditions.

[0186] For JSDM-VDSMA, when dis is set to 1 or 2, the SFC reception rate increases as the user's mobile speed increases. This is because higher speeds increase the possibility of meeting the migration conditions (dis=1 or 2), thereby increasing the chance of SFC migration. The migration of SFC releases resources, which can be allocated to newly arriving SFC requests, thereby improving the reception rate. In contrast, when dis=3, the reception rate remains almost unchanged. This is because even at higher mobile speeds, it is difficult for SFC to meet the dis=3 condition within its limited life cycle, so the probability of SFC migration is low at all mobile speeds, and the reception rate remains stable. Similarly, for JSDR-VDSMA, when dis=1, the reception rate increases as the user's mobile speed increases. However, when dis=2 or 3, the reception rate remains almost unchanged, for reasons similar to JSDM-VDSMA.

[0187] Figure 6 (b) shows the end-to-end delay, revealing that JSDM-VDSMA consistently outperforms JSDR-VDSMA in reducing delay under various constraints. For both JSDM-VDSM and JSDR-VDSMA algorithms, when dis=1, the delay decreases as the user's mobility speed increases. This is because higher speeds make it easier to meet the migration condition of dis=1, allowing SFC migration. This allows the VNF to migrate to a more suitable node and obtain a suitable access node, thereby reducing the delay. However, when dis=3, the delay increases as the speed increases. This is because even under high-speed conditions, it is still difficult to meet the threshold of dis=3. Therefore, the user continues to connect through the original access node despite the long distance, resulting in an increase in access delay and causing an increase in end-to-end delay. When dis=2, the delay of both algorithms is relatively high under medium-speed conditions. This is because the probability of meeting the migration condition dis=2 is lower than that of high-speed movement. In addition, compared with low-speed movement, the distance between the user's current location and the access node is larger, resulting in higher access delay. Therefore, in this medium-speed scenario, the end-to-end delay increases significantly.

[0188] (3) Performance analysis as SFC length changes

[0189] Different services require different VNF ​​functions, so the length of the SFC will vary according to service requirements. For example, some services may require more complex VNFs, while other services may only need simple VNFs. An effective algorithm should be able to adapt to multiple services and ensure flexibility in VNF placement and SFC configuration to meet the specific needs of various services. Figure 7The performance evaluation of JSDM-VDSM and JSDR-VDSM algorithms under different user mobility speeds and different SFC lengths is shown. In this evaluation, the migration condition dis is fixed to 2, while the number of SFCs is kept as 240.

[0190] Figure 7 (a) shows that the reception rate of both algorithms decreases as the SFC length increases. This is because longer SFCs consume more resources, thereby reducing the number of SFCs that can be received. In addition, it can be observed that for different SFC lengths, JSDM-VDSMA achieves higher reception rates at higher mobility speeds. This is because higher speeds increase the likelihood of satisfying the migration condition (dis=2), thereby allowing occupied resources to be released through migration. These released resources can be used to deploy newly arrived SFC requests. For JSDR-VDSMA, the reception rate is relatively high at higher speeds when the SFC length is 2 or 3. However, when the SFC length increases to 4, the reception rate decreases under high-speed conditions. This is because, even if the migration condition (dis=2) is satisfied, JSDR-VDSMA only switches access points without migrating VNFs. Longer SFCs occupy more resources, resulting in a reduction in the number of SFCs received. Overall, JSDM-VDSMA consistently outperforms JSDR-VDSMA in terms of reception rate.

[0191] Figure 7 (b) shows that as the SFC length increases, the end-to-end delay of both algorithms also increases. This is because longer SFCs result in more VNF processing delays and link delays. When the SFC length is 2, the two algorithms perform similarly in terms of delay. For cases where the SFC length is 3 or 4, JSDM-VDSMA significantly outperforms JSDR-VDSMA in reducing delay under high-speed conditions. This is because, at higher speeds, it is easier to meet the migration condition (dis=2), and the migration-based approach of JSDM-VDSMA reduces delay more effectively than the rerouting approach of JSDR-VDSMA.

[0192] In summary, in the experimental results, compared with the algorithm based on rerouting, the method of the present invention always maintains better performance in terms of average service function link yield, average end-to-end delay, and average resource usage cost, especially when the user moves at high speed.

Claims

1. A method for deploying and migrating SFCs based on VNF dependent components. The method is based on the service request information submitted by a mobile terminal user to an SDN controller. The SDN controller analyzes the user demand type and its resource demand according to the service request information, including bandwidth resources, computing resources, storage resources, and latency requirements, and optimizes the deployment and migration of SFCs by using deep reinforcement learning technology in combination with the network topology, the resource status of the edge server, and the network operation status. The method is characterized by: The steps include: (1) Build a communication architecture including mobile users, edge servers, and SDN controllers; In this communication architecture, mobile users are used to generate dynamic business needs, including computing tasks, data storage and communication services. Edge servers are deployed close to users to provide computing and communication services to users. SDN controllers are used to monitor the business needs of mobile users and the resource status of edge servers in real time, and dynamically adjust resource allocation and scheduling strategies based on network topology information. (2) Establish user request models, communication models, and computational models to describe the operation mechanism and performance indicators of the communication system; The user request model includes SFC deployment request and migration request. The deployment request describes the user's demand for bandwidth resources, computing resources, memory resources, and storage resources. The migration request is used to transfer the currently deployed SFC to the target edge server when the user moves. The communication model defines the bandwidth allocation strategy of the edge server and calculates the transmission rate and data transmission time based on the Shannon formula and channel interference index, thereby achieving reasonable allocation of communication resources; The calculation model includes processing and calculating end-to-end communication delay, resource usage cost and interruption delay during SFC migration, which is used to evaluate service performance and provide theoretical support for communication system resource optimization and scheduling; (3) formulating resource scheduling optimization problems and clarifying optimization goals based on the network topology information, user request model, communication model, and computing model of the communication system; The optimization problem includes maximizing the service access rate of user requests, minimizing the end-to-end communication delay and SFC migration cost, and maximizing resource utilization; During the SFC migration process, the migration path of VNF and the resource cost of dependent components are optimized to reduce the service interruption time caused by the migration; (4) Deep reinforcement learning-based SFC deployment and migration considering VNF dependent components; The SDN controller collects user requests, edge server resource status, and network topology information in real time, and uses a deep reinforcement learning algorithm to dynamically decide the following: User access strategy: intelligently select the best edge server for access based on the user's geographic location and business needs; SFC deployment and migration strategy: Dynamically adjust SFC deployment and migration paths by comprehensively considering resource utilization, communication latency, and migration latency; VNF dependency component migration strategy: Based on the dependencies of VNFs in SFC, key VNF components are migrated first to ensure the integrity and continuity of the service chain.

2. The SFC deployment and migration method based on VNF dependent components according to claim 1 is characterized in that: The user request model, communication model, and calculation model described in step (2) are as follows: (21) User Request Model The user request r is denoted as SFCr, is a quintuple of SFCr, where b r Defined as the bandwidth requirement, Indicates the communication delay requirement, TTL r is the life cycle of SFCr, SFC is described as a directed graph G r =(F r ,E r ),in Represents the VNF set in SFCr. Any VNFf rh Both require CPU resources and memory resources E r represents the virtual link of SFCr, and the arrival time of SFCr is expressed as m represents the number of time slots, Δ represents the length of the time slot; For SFC deployment, use decision variables Indicates whether it passes through edge server node n i Access the network, Indicates whether VNFf rh Placed at edge server node n i superior, is a binary variable used to represent the logical link Is it mapped to link e? ij The mathematical expression is: For SFC migration, when VNF is deployed to edge server node n i Or migrate to edge server node n i If n i If the component resources required to support VNF ​​operation are missing, VNF dependent components must be migrated from other nodes to the edge server node. i , to ensure the normal operation of VNF; (22)Communication model Edge server node n i The bandwidth resources are evenly distributed to SFCr to ensure fair distribution. The mathematical expression is as follows: in, Indicates that from edge server node n i Access the SFC collection, Represents server node n i bandwidth resources on Represents r j Whether to use the server node n i Access the network, if from n i 1 if connected to the network, otherwise 0; Use parameters Indicates SFCr access edge server node n i The channel interference is caused by the channel contention caused by other SFCs accessing the same node at the same time, where SFCr j The transmission capacity, Represents the channel gain, which is affected by SFCr j With edge server node n i The distance between them and the power gain effect caused by the small-scale fading effect in the channel can be expressed as: On this basis, SFCr and server node n i The transmission rate between is obtained by the following formula: Among them, ρ 2 is the signal-to-noise ratio, data transmission delay It refers to the data to be transmitted by SFCr to the service node n i The time required is expressed as: (23) Computational model End-to-end communication delay D of SFCr r It is defined as the sum of data transmission delay, VNF processing delay and link propagation delay as follows: The deployment cost of SFC is defined as follows: in, is the migration component ψ q The path used, is a binary variable, indicating the path on is mapped to link e ij superior, Indicates n i The unit storage cost.

3. The SFC deployment and migration method based on VNF dependent components according to claim 2 is characterized in that: This method minimizes downtime and increases success rate through pre-copy migration. Downtime is defined as the duration of the last stage of the migration process, during which the service is temporarily interrupted to complete the final synchronization of data. Indicates the time when the blackout period starts. Indicates the time when the blackout period ends. The expressions of the two are as follows: in, represents the migration time of the qth iteration, is the actual number of iterations, Indicates the VNF to be migrated f rh The memory page size, VNFf rh The migration rate is, therefore, the interruption time is defined as follows: M in the formula r Represents the set of VNFs that need to be migrated in SFCr.

4. The SFC deployment and migration method based on VNF dependent components according to claim 1, characterized in that: The optimization problem and optimization objective described in step (3) are as follows: Constraint 1: Each f rh A VNF instance can only be successfully deployed on one server node at most. It is indivisible, as shown in the following table: Constraint 2: The communication quality between users connected to the network through the same node affects each other. Each physical node can accommodate at most n max users to maintain the service quality of all users, which can be expressed mathematically as: Constraint 3: On any server node n i The total CPU and memory resource requirements for SFC deployment and migration must not exceed the maximum CPU and memory resource capacity, which can be expressed mathematically as: Constraint 4: The used storage resources will not exceed the maximum storage resource capacity, which can be expressed mathematically as: in, is node n i A collection of components on ; Constraint 5: Any link e ij The total bandwidth consumption on the network must be less than the maximum bandwidth capacity, which is as follows: The first item indicates the link bandwidth occupied by SFC deployment, and the second item indicates the bandwidth used by SFC migration; Constraint 6: The end-to-end delay of any SFCr must satisfy the following constraints: Based on the above constraints, the JSDM-VDSM problem is defined. The JSDM-VDSM problem includes SFC deployment strategy, SFC migration strategy, and VNF dependent component migration strategy. The goal of this method is to maximize QoS by improving service reception rate, reducing resource usage cost, and minimizing communication delay and interruption delay. The mathematical definition of the optimization objective is as follows: Among them, Z r (t) is a decision variable used to determine whether the SFC request r is accepted. If accepted, Z r (t) is set to 1, otherwise it is set to 0; where w1, w2, w3 and w4 are the weights of different indicators.

5. The SFC deployment and migration method based on VNF dependent components according to claim 1, characterized in that: Step (4) further includes: The network architecture identifies the SFCs that need to be orchestrated, including those that need to be deployed and those that are planned to be migrated, and considers SFC migration as a joint optimization of VNF redeployment and SFC rerouting; Get the state of the environment and the current SFC to be orchestrated, which includes the available CPU, memory, storage, bandwidth, and component resources of the physical node; the CPU, memory, and component resource requirements of the VNF to be deployed; the number of VNFs that have not yet been placed in the deployed SFC; the partial communication delay of the current SFC, and the lifetime of the SFC; The state s will be obtained ι Input to Actor old Network, obtain VNF deployment decision a ι ,Afterwards, iteratively interacts with the environment to collect a set of experimental data[s ι ,a ι ,r ι ,s ι+1 ] and store it in memory; Calculation of advantage function A based on experimental data θold , used to update the Critic network; using the truncated objective function J θold (θ) Update the Actor network. After multiple iterations, the parameters of the Actor network are passed to the Actor old Network,iteration process continues until convergence; The solution process of the JSDM-VDSM problem includes obtaining SFC request r, determining user access node A, and SFC deployment or migration strategy X, Y; based on strategy X and Y, obtaining VNF dependent component migration strategy Z; before learning the service provision strategy, the SDN controller collects global information and releases SFCs that have reached their life cycle, and then identifies the SFCs that need to be deployed or migrated; The SFC requested for deployment is the SFC request that arrives randomly in the current time slot t, while the SFC requested for migration is the SFC request that meets the migration criteria in the same time slot; if there is no request to be processed, the SDN controller will enter the next time slot and continue to obtain requests until r batch is not empty; then, r batch The SFC requests in will be deployed or migrated in sequence; For deployment and migration requests, the user area related to the SFC request is first determined, and then the SDN controller evaluates whether the base station in the current area has sufficient capacity. If there is sufficient capacity, access is made through the base station; if there is insufficient capacity, the SDN controller plans to connect through neighboring areas, first identifying neighboring areas and then calculating the access capacity of these areas; then, the accessible node max_num_node with the maximum access capacity is determined, and the area that can accommodate the maximum capacity is selected, and further an access node is selected from max_num_node with 1-∈, and an access node is selected from the set n_max_num_node with probability ∈, where n_max_num_node is the set of accessible nodes; After selecting the access point, determine the deployment or migration strategy of the SFC and the migration strategy of the VNF dependent components. First, identify the first VNF ​​of the SFCr, then obtain the current status, and according to the strategy π θold (a ι ,s ι ), determine VNFf rh The placement position, if f rh Can be placed at node a ι On, but a ι Currently not available rh To run the required components, you must install the required f rh Migrate the dependent components to a ι , then place the next VNF ​​and record the updated state s ι+1 ; The above process continues until the SFC is successfully deployed or migrated; the reward is calculated based on the deployment result. If the deployment is successful, the reward for successful deployment is calculated; if the migration is successful, the reward for successful migration is calculated; if the SFC deployment or migration fails, the communication system will be restored to the last successfully achieved state s ι-u+1 .

Images