SFC deployment method based on VNF dependent component migration in multi-domain network

Abstract

The invention discloses an SFC deployment method based on VNF dependent component migration in a multi-domain network, and belongs to the technical field of communication. According to the method, the SFC deployment and the migration of the dependent component of the VNF are jointly optimized, and the service providing problem in the multi-domain network is solved. According to the method, firstly, information related to a multi-domain network is initialized, and then a multi-layer network architecture is constructed to support SFC deployment. Then, a multi-layer weighted network is established by using an analytic hierarchy process (AHP), and SFC deployment and a migration strategy of a dependent component of a VNF are executed on the basis of the multi-layer weighted network. According to the method, the influence between the SFC deployment and the VNF dependent component migration decision is comprehensively considered, and efficient SFC deployment and VNF dependent component migration in the multi-domain network are realized.

Description

Technical Field

[0001] The present invention belongs to network communication technology, and in particular to an SFC deployment method based on VNF dependent component migration in a multi-domain network. Background Art

[0002] SFC deployment is an emerging research topic in the service delivery field. Related research work can be divided into two categories: SFC deployment in a single domain and SFC deployment in multiple domains.

[0003] When deploying SFC in a single-domain network, some works effectively reduce service costs by studying VNF placement issues, optimizing resource allocation, and improving network throughput, but ignore the key performance indicator of end-to-end latency, which is crucial to ensuring the overall performance of the network. Some researchers are committed to ensuring service quality by reducing SFC end-to-end latency. At the same time, some studies jointly optimize cost and latency and propose SFC deployment methods based on ILP, DRL, and heuristics. However, the SFC deployment strategy in a single-domain network is difficult to support SFC deployment in a cross-domain network. In addition, 6G's support for vertical services and applications emphasizes the key need for cross-domain service provision. Since 2017, researchers in the field of network communications have discussed service provision in multi-domain networks, including new architectures and new methods for SFC deployment in multi-domain networks, aiming to minimize network resource consumption costs, reduce SFC communication latency, and achieve multi-domain network load balancing.

[0004] However, solving the problem of optimal SFC deployment in multi-domain networks to achieve cost-effective and low-latency service delivery is not an easy task. In addition, previous studies have ignored the dependent components required for VNF execution, assuming that physical nodes can independently execute all VNFs. However, network function instantiation requires corresponding component resources on physical nodes to support its operation, including system-level component resources (e.g., Ubuntu, VEKET, and CentOS), configuration files (e.g., network interface files and routing rule configuration files), and application-level component resources (e.g., IPFilter and Hash Switch). These resource components are called VNF dependent components, which support the operation of VNFs and are critical to the operation of VNFs.

[0005] In large-scale, multi-domain 6G networks, network heterogeneity between domains is becoming increasingly significant, and different domains have differentiated VNF placement platforms. In order to meet the time-varying extreme business needs and improve the competitiveness of the platform, the VNF platform needs to be frequently upgraded and additional configurations, such as network interface settings and routing configuration adjustments, are performed to ensure seamless operation of VNFs and reliable network communication. Although virtual machine and container technologies reduce the operation and management costs of service providers, they also bring higher latency. Compared with virtual machines, containers reduce communication latency by sharing operating systems, but they usually start slowly. Martines et al. proposed a lightweight virtual machine platform based on Xen Hypervisor, which uses ClickModular Router to define and execute VNFs. This approach enables VNFs to run as modular Click configurations, significantly reducing the overhead of network function virtualization. Similar to ClickOS, Click-on-OSv provides a lightweight, low-latency solution suitable for network functions with high performance requirements. In addition, considering VNF dependent components can speed up the update of VNFs, promote their flexible placement in large-scale, multi-domain heterogeneous 6G networks, and efficiently adapt to changing business needs. This approach is also closely aligned with the 6G vision and supports ultra-low latency applications. Summary of the invention

[0006] Purpose of the invention: In order to minimize communication delays and service provision costs, while ensuring load balancing in multi-domain networks, meeting extreme business requirements, and providing high-performance communication services, the present invention provides an SFC deployment method based on VNF dependent component migration in a multi-domain network.

[0007] Technical solution: The present invention provides a SFC deployment method based on VNF dependent component migration in a multi-domain network, the method comprising:

[0008] (1) Based on the SDN controller, a multi-domain network communication architecture is constructed, including multi-domain physical networks, end users, domain SDN controllers and master SDN controllers. End users submit service requests to domain controllers, and domain SDN controllers send information to the master SDN controller. The master SDN controller, in conjunction with the VNF dependent component migration, sends SFC deployment instructions and VNF dependent component migration instructions to the domain SDN controllers. The multi-domain network provides services to end users based on the decisions of the domain SDN controllers.

[0009] (2) Establishing network topology and service request model, including:

[0010] (21) Based on multi-domain network topology, business function chains are mapped through heterogeneous multi-domain physical networks;

[0011] The domain set of a heterogeneous multi-domain physical network is represented as in represents the τth network domain, |G| represents the number of domains, and the set of multi-domain physical network nodes is expressed as n τi represents the i-th node in the network domain τ, and the set of physical network links is expressed as Multi-domain network support types of VNFs and over a long period of time A collection of nodes that work within to support the service provider requested by the user Used to deploy virtual machines to run VNFs and forward data, using the maximum computing power of each server node Maximum memory capacity and maximum storage capacity Respectively represent the computing, memory and storage resources of the node. Each server node belongs to a subnetwork. Define the variables Represents server node n τi Domain, link Connect two server nodes n τi and represents the physical link between them; when τ is equal to When , the connection represents an intra-domain link; otherwise, it represents an inter-domain link; and Respectively represent links The starting and ending nodes of To represent the link bandwidth capacity; Used to represent service node n τi The remaining computing resources, Indicates service node n τi Remaining memory resources; use Indicates service node n τi Remaining storage resources; Indicates link Remaining bandwidth resources;

[0012] (22) Build a business request model;

[0013] All user requests with actual needs entering the NFV-based multi-domain network are represented as SFC requests. The SFC request set is represented as Γ={Γ1,Γ2,…,Γ r ,...,Γ |Γ|}; Define quintuple Used to represent an SFC request r, where s r ,d r denote the source node and the destination node respectively; each SFC request r has a known bandwidth requirement br and a maximum tolerable end-to-end delay requirement

[0014] For an SFC request r, it consists of a set of predefined VNFs and links, which is represented as a directed weighted graph G r ={F r ,E r},in Represents a collection of VNFs, Represents a collection of virtual network links; each VNFf rh ∈F r A certain amount of server computing resources are required and memory resources Similarly, mapping a virtual link requires allocating bandwidth resources b r ;

[0015] (23) Building a VNF dependent component migration model to migrate components from one server node to another;

[0016] When a virtual network function rh It is decided to be deployed on a server node with insufficient component resources n τi When on, f rh The related dependent components will be migrated to node n τi ; Ψ={ψ1,ψ2,...,ψ q ,...,ψ |Ψ|} represents the collection of all component resources that support VNF ​​operation;

[0017] For VNF placement and logical link mapping, two binary decision variables are used to define this relationship:

[0018]

[0019]

[0020] Logical Link Can be mapped to multiple physical links, indicating that communication between VNFs can span multiple physical connections;

[0021] (24) VNF dependent component migration decision

[0022] The working time period of the NFV-based multi-domain network is divided into several time slots. Assuming that new SFC request information is collected in time slot t-1, SFC will be deployed in time slot t. At the same time, the component resources ψ q The migration will be executed and completed within time slot t, defining a binary variable Represents server node n τiWhether VNFf is supported in time slot t rh , is a binary migration decision variable, indicating whether node n is τi Migration component ψ q :

[0023]

[0024] in, is a binary variable indicating that at time slot t, the component ψ q Is it at node n τi Up, rh Supports virtual network functions rh The set of dependent components to be run. ResourceConstraint is an indicator variable, indicating the service node n τi Are other resources on the network sufficient to support the virtual network function? rh When the service node n τi When the resources on the network are sufficient, its value is 1, and the corresponding mathematical expression is:

[0025]

[0026] when , indicating that in time slot t, component ψ q Has been migrated to server node n τi Otherwise, it means component ψ q Not migrated, using Indicates that VNFf rh The component resource collection to be migrated, after determining the components to be migrated, updates the component resources on the corresponding nodes;

[0027] (3) Formulate a problem to be solved and determine the optimization goal based on the network topology and service request model constructed in step (2);

[0028] The problem to be solved is defined as follows:

[0029] SFC deployment problem: Given a network domain set and the SFC request set Γ, find the routing path of SFC and its deployment strategy, considering the quality of the selected path to minimize the SFC deployment cost and ensure the end-to-end delay of SFC while satisfying the physical network resource constraints;

[0030] VNF dependency component migration problem: given a set of network domains SD deployment results, current component resource distribution matrix Find component migration strategies and their migration paths, taking into account VNF’s requirements for dependent components and minimizing migration costs when the selected server nodes cannot provide sufficient component resources;

[0031] The overall optimization goal is determined based on the multi-domain network information, user terminal service request information, and VNF dependent component migration model. The overall optimization goal is to minimize the end-to-end delay of all SFCs and the costs associated with SFC deployment and VNF dependent components, namely, VNF dependent component migration cost and VNF dependent component deployment cost, where the end-to-end delay of the SFC request is defined as the sum of the link delays D r , mathematical calculation is expressed as:

[0032]

[0033] Yes Link The delay on is the service node n τi Processing delays on

[0034] The deployment cost of an SFC request r is defined as The total cost of SFC request r deployment is express, It is composed of the node resource usage cost, and its expression is as follows:

[0035]

[0036] It consists of the VNF dependent component migration cost and the VNF dependent component deployment cost, which is defined as follows:

[0037]

[0038] The service delivery cost C of SFC request r r The definition is as follows:

[0039]

[0040] The joint optimization objective is expressed as follows:

[0041]

[0042] Among them, ω1 and ω2 are adjustable weight factors used to balance various objectives. By adjusting the weight factor of each objective, the preference of a specific objective is highlighted, and the constraints of physical network resources and SFC service quality are met.

[0043] Constraints (10) and (11) ensure that at any server node n τiThe total CPU and memory resource requirements required for an SFC deployment must not exceed the respective maximum CPU and memory resource capacities:

[0044]

[0045] Constraint (12) ensures that the used storage resources do not exceed the maximum storage resource capacity:

[0046]

[0047] in, is node n τi Similarly, any link The total bandwidth consumption on must be less than the maximum bandwidth capacity, subject to the following constraints:

[0048]

[0049] To meet the quality of service, the delay constraint (14) ensures that the end-to-end delay of any SFCr must meet the following constraints:

[0050]

[0051] Constraint (15) ensures that every f rh It can only be successfully deployed on one server node at most, which means that the VNF instance is inseparable, corresponding to:

[0052]

[0053] To simplify the deployment of SFC, it is assumed that the entry node of SFC request r has only one outgoing link and the exit node has only one incoming link to facilitate the realization of user requests; at any intermediate node on the path selected by SFC request r, SFCr only uses one incoming link and one outgoing link if and only if SFCr uses physical link When the variable is equal to 1:

[0054]

[0055] (4) Design layered architecture and deployment and migration mechanisms;

[0056] A multi-layer network architecture is adopted, and each VNF corresponds to an independent layer of network. The matching relationship between each VNF and all nodes is calculated based on the hierarchical analysis method, and a multi-layer network weighted graph is constructed. In the multi-layer weighted graph, the shortest path for deploying SFC and the placement node of VNF are found based on the dynamic programming algorithm. After obtaining the node where VNF is placed, the migration strategy of VNF dependent components is determined.

[0057] Furthermore, the method is based on the terminal user submitting extreme service request information to the SDN controller within the domain, and the SDN controller within the domain transmits the information to the master control SDN controller. The master control SDN controller decides the multi-domain network to provide deployment location and migration strategy of VNF dependent components for the service function chain through the SFC deployment method of joint VNF ​​dependent component migration, thereby reducing the end-to-end communication delay and resource utilization cost of the service, the resource utilization cost of VNF dependent components and the migration cost of VNF dependent components, and ensuring multi-domain network load balancing.

[0058] In the above scheme, the network topology and service request model constructed in step (2) in implementing SFC deployment, VNF instantiation not only relies on virtualization technology, but also needs to fully consider the adaptability of component resources of physical nodes, including the network attributes such as the capacity of computing nodes, link bandwidth and latency of the multi-domain network implemented based on the network topology, and use a graph model to describe the communication topology within and between domains; define multiple service request types, extract key indicators of their network attributes, and use a random process to simulate the dynamics of request arrival; determine the VNF dependent component migration model based on the two established models.

[0059] Furthermore, the layered architecture design, deployment and migration mechanism described in step (4) are specifically:

[0060] (41) Algorithm initialization;

[0061] Build a multi-layer network structure, where each VNF corresponds to an independent layer to calculate the matching relationship between each VNF and all nodes;

[0062] Multi-dimensional resource indicators are introduced. In the initialization phase, a network graph MG is generated for each VNF in the SFC. rh ; Independent network graph MG rh By connecting the corresponding nodes to each other, a multi-layer network graph MG is constructed. r , whose number of layers is |F r |;

[0063] (42) Performance measurement based on AHP;

[0064] First, from the multi-layer network graph MG r The data of each layer of physical nodes are extracted from the network, including the remaining computing, memory and storage resources, component resources, and processing delays; then, a pairwise comparison matrix NCM is constructed to evaluate the criterion layer for the target; then, to compare the multi-layer network graph MG rh The pairwise relationship of all nodes of the same type of elements in the construct eight Matrices, named computing resource utilization Memory resource utilization Storage resource utilization Storage resource usage cost Countdown of remaining computing resources Remaining memory resources countdown Remaining storage resources countdown and processing delays Then the matrix is ​​tested for consistency;

[0065] The consistency ratio is used to evaluate the consistency of the matrix, and the consistency index CI is defined as Used to quantify the consistency within the judgment matrix, where λ max Represents the largest eigenvalue of a matrix;

[0066] The random consistency index RI depends on the dimension of the matrix and is obtained from a table of known RI values;

[0067] If the matrix satisfies the consistency criteria, the eigenvector corresponding to the largest eigenvalue is extracted; otherwise, the matrix is ​​reconstructed; then, h is calculated. τi The indicator value of

[0068] For the link index, a similar method is used to calculate, where the index value of the inter-layer link is always set to 0. Through the above iterative process, a multi-layer weighted network is finally generated, denoted as MWG r ;

[0069] (43) SFC deployment and VNF dependent component migration strategy;

[0070] Getting the multi-layer weighted graph MWG r After that, the next task is to determine the communication path from the source node to the target node and the node to place the VNF, specifically:

[0071] First, map the starting node to MWG r Node labels in, for each layer, select a placement node for the VNF and calculate the metrics of intra-layer links and inter-layer communication links, then calculate the comprehensive metrics of the selected placement service nodes and links, and finally, return the selected deployment service nodes and communication links;

[0072] The metric calculation of the service node takes into account potential component migration scenarios, thereby selecting the optimal migration strategy based on the VNF placement.

[0073] In terms of implementation and principle, the SFC deployment method based on VNF dependent component migration in a multi-domain network described in the present invention is mainly implemented from the following four perspectives:

[0074] (1) Construct a multi-domain network communication architecture based on SDN controller. The multi-domain network communication architecture based on SDN controller separates the control plane from the data plane, aggregates the global information (topology, traffic, etc.) of the multi-domain network, reduces the complexity of operation and maintenance, realizes cross-domain resource optimization, supports the rapid deployment of new functions and services, and makes network management more flexible. At the same time, the constructed architecture fully considers the heterogeneity of multi-domain networks, the differences in service node resources, and the diversity of extreme business needs, laying a technical foundation for the intelligent, agile and efficient operation and maintenance of multi-domain networks, while supporting higher-level business innovation and service differentiation. The architecture includes the master SDN controller, the sub-domain SDN controller, the communication link between the master SDN controller and the sub-domain SDN controller, the communication link between the users in the domain and the sub-domain SDN controller, and the definition and design of communication instruction information.

[0075] (2) Establishing a network topology and service request model. In SFC deployment, VNF instantiation not only relies on virtualization technology, but also needs to fully consider the adaptability of component resources of physical nodes. VNF instances are usually provided in the form of virtual machines or container images. Although the deployment process is simplified, its operation depends on specific component resources. These resources include system-level components (such as Ubuntu, VEKET, CentOS), configuration files (such as network interface files, routing rule configuration files), and application-level components (such as IPFilter, Hash Switch). These are called VNF dependent component resources and are an important part of the construction of network topology and service request models. They are crucial to ensuring the normal operation and service quality of VNFs. By establishing a network topology and service request model that takes into account VNF ​​dependent component resources, VNF resources and requirements can be managed in a refined manner, promoting the smooth implementation of diversified business scenarios.

[0076] (3) Formulate the problem to be solved and determine the optimization objectives. The SFC deployment and VNF dependency component migration problem in a multi-domain network is formulated, where multiple domains are interconnected through an SDN controller. The goal is to reduce communication delays and costs associated with service provision while ensuring load balancing in the multi-domain network and satisfying constraints related to link and physical node resources. Solving this problem requires a comprehensive decision-making process, including strategic placement of VNFs, selection of optimal communication links, and reasonable VNF dependency component migration strategies. Considering the complex coupling between SFC deployment decisions and VNF dependency component migration, these two sub-problems are integrated into one problem, and the problem of minimizing end-to-end communication delays and costs associated with service provision while ensuring load balancing in the multi-domain network is proposed.

[0077] (4) Layered architecture design and deployment and migration mechanisms. VNFs in SFC have different requirements for computing, memory, and component resources. The matching degree of each physical node for VNF placement is different, which increases the complexity of service delivery. A multi-layer network architecture is proposed. Each VNF corresponds to an independent layer of network, which is convenient for calculating the matching relationship between each VNF and all nodes based on the hierarchical analysis method, and further constructing a multi-layer network weighted graph. In the multi-layer weighted graph, the shortest path for deploying SFC and the placement node of VNF are found based on the dynamic programming algorithm. After obtaining the node where VNF is placed, the migration strategy of VNF dependent components is determined. The designed layered architecture and the proposed deployment and migration mechanism enable VNFs to be placed on appropriate physical nodes, effectively solving the challenges brought by VNF dependent component resource requests.

[0078] Beneficial effects: The present invention discloses a SFC deployment method based on VNF dependent component migration in a multi-domain network. The method comprehensively considers the information of the multi-domain network topology and the diversified extreme service request information. First, the information related to the multi-domain network is initialized, and then a multi-layer network framework is constructed to support the deployment of SFC. Then, using the hierarchical analysis method, we established a multi-layer weighted network, on which the SFC deployment and VNF dependent component migration strategy are implemented. Finally, a comprehensive experimental evaluation was conducted, and the results showed that the JSD-VDSM algorithm proposed in the present invention outperforms other comparative algorithms in performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 It is a schematic diagram of the process of the method of the present invention;

[0080] Figure 2 This is a diagram of an SFC deployment instance considering the migration of VNF dependent components across domains in the present invention;

[0081] Figure 3 This is an example diagram of the implementation of the JSD-VDSM algorithm in the embodiment of the present invention;

[0082] Figure 4 AHP-based node link performance measurement diagram in an embodiment of the present invention;

[0083] Figure 5 is the average receiving rate at different arrival rates of SFC requests in the embodiment of the present invention, N τ =8;

[0084] Figure 6 is the average SFC receiving rate that changes with time under a specific arrival rate in the embodiment of the present invention, N τ =8;

[0085] Figure 7is the average SFC reception rate under different QoS requests and network sizes in the embodiment of the present invention, where Figure 7 (a) corresponds to λ=0.1,N τ =8, Figure 7 (b) corresponds to λ = 0.1;

[0086] Figure 8 is the average end-to-end communication delay at different arrival rates of SFC requests in the embodiment of the present invention, N τ =8;

[0087] Fig. 9 is the average SFC end-to-end communication delay that varies with time at a specific arrival rate in the embodiment of the present invention, N τ =8;

[0088] Fig.10 is the average SFC end-to-end communication delay under different QoS requests and network sizes in the embodiment of the present invention, where Fig.10 (a) corresponds to λ=0.1,N τ =8, Fig.10 (b) corresponds to λ = 0.1;

[0089] Fig.11 is the average resource usage cost at different arrival rates of SFC requests in the embodiment of the present invention, N τ =8;

[0090] Fig.12 is the average SFC resource usage cost under different QoS requests and network sizes in the embodiment of the present invention; Fig.12 (a) corresponds to λ = 0.1, Fig.12 (b) corresponds to λ = 0.1;

[0091] Fig.13 is the relative storage resource usage cost under different arrival rates of SFC requests in the embodiment of the present invention, N τ =8;

[0092] Fig.14 is the average SFC relative storage resource usage cost under different QoS requests and network sizes in the embodiment of the present invention;

[0093] Fig.15 is the average component migration cost at different arrival rates of SFC requests in the embodiment of the present invention, N τ =8;

[0094] Fig.16 is the average SFC component migration cost under different QoS requests and network sizes in the embodiment of the present invention;

[0095] Fig.17 is the load balancing under different arrival rates of SFC requests in the embodiment of the present invention, N τ =8;

[0096] Fig.18 Load balancing under different QoS requests and network sizes in an embodiment of the present invention;

[0097] FIG. 19 shows the effect of the length change of SFC on the algorithm performance in an embodiment of the present invention, N τ =8. DETAILED DESCRIPTION

[0098] 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.

[0099] The present invention provides an SFC deployment method based on VNF dependent component migration in a multi-domain network, which aims to minimize end-to-end communication delay and improve service quality (QoS); balance multi-domain network load and ensure resource utilization efficiency; reduce deployment overhead and optimize resource utilization and service cost.

[0100] From the application perspective of the present invention, as the application and widespread deployment of 5G technology have made significant progress around the world, we have witnessed a significant improvement in public communication capabilities and service quality. This progress has significantly accelerated the digital transformation of various vertical industries and promoted social and economic development. Entering the 6G era, the demand for extreme services is growing, including immersive extended reality (XR), holographic communication, perceptual interconnection and intelligent interaction, which have the characteristics of low latency and cost-effectiveness. Multi-domain networks are built and managed by different network operators and service providers. With their large capacity and wide coverage, they play a key role in meeting extreme business needs and providing high-performance communication services. In view of the advantages and extreme business needs of multi-domain networks, the present invention provides an SFC deployment strategy for VNF-based dependent component migration in multi-domain networks.

[0101] The present invention provides a SFC deployment method based on VNF dependent component migration in a multi-domain network, which is used for the integrated deployment of SFC and VNF dependent component migration. Using the hierarchical analysis method and dynamic programming method, our method aims to minimize the end-to-end communication delay and service delivery cost of SFC.

[0102] The implementation process of the technical solution provided by the present invention is described in detail below.

[0103] The method described in the present invention is to realize the deployment of the service function chain and the migration decision of the VNF dependent components. It mainly includes the master SDN controller, the domain SDN controller, the extreme service requirements and the multi-domain network topology. The domain SDN controller collects the network topology and the extreme service request information in the current domain, and transmits the information to the master SDN controller, and is responsible for the deployment of SFC and the migration of VNF dependent components; the terminal user makes a deployment request to the controller in the domain, and waits for the network node to provide services for it; the nodes and links in the multi-domain network topology are scheduled by the master SDN controller to provide services for the terminal users.

[0104] 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:

[0105] (1) Constructing a multi-domain network communication architecture based on an SDN controller. The construction of a multi-domain network communication architecture based on an SDN controller in the present invention aims to achieve high flexibility, scalability and automated management of the network. A software-defined control plane is adopted, and a centralized SDN controller is used to uniformly schedule and manage resources, traffic and service functions between different network domains, thereby achieving global optimization of the network, improving performance, ensuring service quality, and significantly reducing the complexity of network operation and maintenance. This architecture not only supports cross-domain collaboration and resource sharing, but also can dynamically adapt to changes in demand for different business scenarios, providing flexible, reliable and efficient network services to meet today's diverse and complex business needs.

[0106] It is further pointed out that in the present invention, the domain SDN controller sends a request to the master SDN controller according to the service needs of the terminal user. Under the assumption that each terminal user has an extreme service request, the multi-domain network must provide it with the resources required for this service request. The purpose of the service is to reduce the end-to-end delay of the service function chain, the resource usage cost, the VNF dependent component migration cost and the resource usage cost during placement, and ensure the load balancing of the multi-domain network. The reduction of end-to-end delay may increase the resource usage cost and the cost related to the VNF dependent components, and may cause an unbalanced network load. The cost reduction may cause excessive communication delay. How to balance these three goals and how to deploy SFC while considering VNF dependent components is the difficulty of this problem.

[0107] (2) Establishing a network topology and service request model. The present invention ensures accurate adaptation and efficient management of component resources during VNF instantiation by establishing a network topology and service request model. By introducing a VNF dependent component resource identification and matching mechanism and utilizing dynamic adaptation of system-level and application-level resources (such as Ubuntu, VEKET, CentOS, IPFilter, Hash Switch, etc.), it effectively ensures that each VNF instance can run on the most suitable physical node and meet its performance and reliability requirements. And through intelligent optimization of resource scheduling, configuration management and cross-domain collaboration, efficient instantiation, stable operation and excellent service quality assurance of VNF are achieved, thereby promoting the smooth deployment and implementation of diversified business scenarios.

[0108] It is further pointed out that the present invention first models the multi-domain network, including defining the resource attributes of each domain, including the capabilities of the computing nodes, the bandwidth and latency of the links, etc., describing the communication topology within and between domains, and using a graph model to represent it; then, modeling the extreme service request, defining multiple service request types, and extracting their key indicators, such as the maximum allowed latency, using a random process to simulate the dynamics of request arrival; and determining the VNF dependent component migration model based on the two established models. A systematic framework is provided through model establishment, which can clearly describe the characteristics of the multi-domain network, the requirements of service requests, and the dependencies of VNF migration, avoiding vague or unclear problems in the design process.

[0109] (3) Formulate the problem to be solved and determine the optimization goal. The present invention takes into account the complex coupling relationship between SFC deployment decisions and VNF dependent component migration. Traditional methods are difficult to effectively handle these interdependent decisions. To this end, an integrated solution is proposed to integrate these two sub-problems into an overall optimization problem, with the goal of minimizing the end-to-end communication delay and the cost associated with service provision, and ensuring load balancing in multi-domain networks. Through this integrated approach, the limitations of traditional single optimization strategies have been successfully broken through, providing a new solution to the complex multi-domain network SFC deployment and VNF dependent component migration problems.

[0110] It is further pointed out that the present invention studies and designs an SFC deployment strategy based on VNF dependent component migration suitable for a multi-domain network environment. SFC deployment in a multi-domain network faces the dual challenges of complexity and resource constraints, especially the adaptation and migration of VNF dependent component resources, which directly affects the performance and deployment efficiency of the service. By deeply analyzing the resource distribution characteristics and VNF instantiation requirements in a multi-domain network environment, the following goals are determined: minimize end-to-end communication delays and improve service quality; balance multi-domain network loads and ensure resource utilization efficiency; reduce deployment overhead and optimize resource usage and service costs. Achieve deep coupling optimization of VNF dependent component migration and SFC deployment, and effectively solve the problems of resource adaptability and migration complexity.

[0111] (4) Layered architecture design, deployment and migration mechanism. The present invention ensures that VNFs can be accurately placed according to the matching degree between their computing, memory and component resource requirements and physical nodes by designing a layered architecture design, deployment and migration mechanism. The matching relationship between each VNF and all physical nodes is calculated by the hierarchical analysis method, and the adaptability of each node is effectively quantified using a multi-layer network weighted graph, providing a systematic decision-making basis for the optimal placement of VNFs. At the same time, the dynamic programming algorithm is used to achieve the shortest path selection for SFC deployment and the precise determination of VNF placement nodes, thereby ensuring the efficient migration and reasonable scheduling of VNF dependent component resources.

[0112] It is further pointed out that the present invention introduces a multi-layer network architecture, calculates the matching degree between VNF and physical nodes through hierarchical analysis method, and optimizes resource placement decisions. The dynamic programming algorithm is used to select the optimal SFC deployment path and node in the multi-layer weighted network to improve deployment efficiency and reduce communication delay. A dependent component migration optimization mechanism is designed to achieve efficient migration of component resources in a cross-domain environment and reduce service delivery overhead.

[0113] Further, the problem described in step (1) is described through an example. Figure 2 This paper presents an SFC deployment example considering the migration of VNF dependent components in a large network. The network consists of three domains with different characteristics. Each domain is equipped with different component resources. Each domain has an SDN controller, which is controlled by the overall SDN controller. To simplify the problem, it is assumed that each physical node has only one component resource or no component resource. For example, in domain1, physical node n 11 Assigned software2, management node n 12 Configured software3, physical node n 13In domain2, each node has software1, 2, 3, and 3. Similarly, in domain3, each node has software0, 1, and 2. software0 represents the physical node n. 31 There are no component resources on the physical node. The diversity of physical node resources and their associated resource usage costs will be discussed in detail in the subsequent modeling and methods. In addition, it is assumed that each physical node can only host one VNF instance.

[0114] For a given SFC1:start→VNF2→VNF3→end, the source node and the destination node are used as communication endpoints. Under the goal of minimizing link and node processing delays and service delivery costs, the routing path between the communication endpoints is determined and VNFs are placed on the physical nodes on the path. For example, Figure 1 The communication path selected by SFC1 is shown as a green dashed line. The deployment process of SFC1 involves two key issues: determining the optimal node for VNF placement and evaluating whether component migration is required when deploying VNFs on physical nodes, including selecting which component to migrate. VNF1, VNF2, and VNF3 are deployed on node n, respectively. 13 、n 22 and n 31 The node must have the component resources to support the operation of each VNF. 31 There is a lack of component resources required by VNF3. Therefore, software3 needs to be removed from node n. 24 Migrate to n 31 Therefore, the end-to-end communication delay and service delivery cost of SFC1 are calculated as (50+60+40)+(15+2+4+14+3)+14=202, where the first term represents the node processing delay, the second term represents the link delay, and the third term represents the cost of moving software3 from n 24 Migrate to n 31 cost.

[0115] In the above method, in step (2) represents the set of domains of a heterogeneous multi-domain physical network, and in represents the τth network domain, responsible for mapping SFCs. denote the set of physical network nodes and the set of physical network links in the τth domain respectively. Assuming that the multi-domain network supports There are three types of VNFs. VFNs represent more than one VFN, such as Firewall (FW), Deep Packet Inspection (DPI), Network Address Translation (NAT), etc., and they are used over a long period of time. Works within to support the provision of services requested by users. Virtual Machines (VMs) can be deployed to run VNFs and forward data. These virtual machines are called VNF instances. With maximum computing power (Superscript c indicates CPU), maximum memory capacity and maximum storage capacity Respectively represent the computing, memory and storage resources of the node. Each server node belongs to a sub-network. Is a variable representing the server node n τi The domain to which it belongs. Link Connect two server nodes n τi and represents the physical link between them. When τ is equal to When , the connection represents an intra-domain link; otherwise, it represents an inter-domain link, that is, the service nodes on the link belong to different domains, and τ and To represent different physical network domains. and Respectively represent links The starting and ending nodes of the link are represented by the subscripts 1 and 2, respectively. Each link is composed of two nodes. To represent the link bandwidth capacity. Used to represent node n τi The remaining computing resources, Represents node n τi Remaining memory resources. Represents node n τi Remaining storage resources. Indicates link Remaining bandwidth resources.

[0116] All user requests entering the NFV-based multi-domain network with actual requirements (such as bandwidth requirements and delay requirements) are represented as SFC requests. r ,...,Γ |Γ|} represents an SFC request set. A five-tuple Used to define SFC request r, where denote the source node and the destination node respectively. Each SFC request r has a known bandwidth requirement b r and a maximum tolerable end-to-end delay requirement An SFC request r consists of a set of predefined VNFs and links and is represented as a directed weighted graph G r ={F r ,E r},in Represents a collection of VNFs, Represents a collection of virtual network links. Each VNFf rh ∈F r A certain amount of server computing resources are required and a certain amount of memory resources Similarly, mapping a virtual link requires a certain amount of bandwidth resources b r .

[0117] NFV-based multi-domain network is enabled to provide There are several types of VNFs. A VNF with a special type running on a virtual machine needs to support some dependent components to run it, and the dependent components required for different types of VNFs are different. However, due to the heterogeneity of multi-domain networks and the limitation of storage resources, it is not realistic to accommodate all the dependent components that support the operation of VNFs on each server, which means that each server node only supports a few specific types of VNFs. Usually, a server node does not need to support all types of VNFs because components are a kind of resource that can be migrated in a NFV-based network.

[0118] Migration of VNF dependent components refers to the process of migrating components from one server node to another due to insufficient component resources. rh It is decided to be deployed on a server node with insufficient component resources n τi When on, f rh The related dependent components will be migrated to node n τi Ψ={ψ1,ψ2,...,ψ q ,…,ψ |Ψ|} represents the set of all component resources that support VNF ​​operation. Before describing the component migration model, we first introduce the mapping relationship between SFC requests and physical network topology, namely VNF placement and logical link mapping. Two binary decision variables are used to define this relationship:

[0119]

[0120] It is worth noting that the logical link It can be mapped to multiple physical links because communication between logically adjacent VNFs may span multiple physical connections.

[0121] Next, we describe the process of VNF dependent component migration. Assume that the working time period of the NFV-based multi-domain network is divided into multiple time slots. Assume that new SFC request information is collected at time slot t-1, SFC will be deployed at time slot t, and component ψ q The migration of will be executed and completed in time slot t. A binary variable Represents server node n τi Whether VNFf is supported in time slot t rh ,and is a binary migration decision variable, indicating whether node n is τi Migration component ψ q .

[0122]

[0123] in, is a binary variable indicating that at time slot t, the component ψ q Is it at node n τi Up. rh Yes VNFf rh The set of dependent components to be run. ResourceConstraint is an indicator variable, indicating that node n τi Are other resources on the system (e.g., CPU, memory, and storage) sufficient to support the rh When node n τi When sufficient resources are available, its value is 1.

[0124]

[0125] when , indicating that in time slot t, component ψ q Has been migrated to server node n τi Otherwise, it means component ψ q Next, use Indicates that VNFf rh A collection of component resources to be migrated. After determining the components that need to be migrated, you need to update the component resources on the corresponding nodes.

[0126] Furthermore, in step (3), the problem to be solved is first defined, and then a mathematical model of the SFC deployment problem considering the migration of VNF dependent components in a multi-domain network is given.

[0127] The problem to be solved is defined as follows:

[0128] Definition 1: SFC Deployment (SD). Given network information and the SFC request set Γ, find the routing path of SFC and its deployment strategy, considering the quality of the selected path to minimize the SFC deployment cost and ensure the end-to-end delay of SFC while satisfying the physical network resource constraints.

[0129] Definition 2: VNF-dependent Software Migration (VDSM). Given network information SD deployment results, current component resource distribution matrix Find component migration strategies and their migration paths, consider VNF’s requirements for dependent components, and minimize migration costs when the selected server nodes cannot provide sufficient component resources.

[0130] According to the above two definitions, it is obvious that the deployment strategy of SFC will affect the component migration strategy. In turn, the component migration strategy will also affect the deployment strategy of SFC in subsequent time slots. The strategies of these two problems are strongly coupled. We integrate these two problems into one problem, called SD-VDSM.

[0131] Objective: The objective of the SD-VDSM problem is to minimize the end-to-end latency of all SFCs and the costs associated with SFC deployment and VNF dependent components (Dependent Software Cost, DSC), namely, VNF dependent component migration cost (Dependent Software Migration Cost, DSMC) and VNF dependent component deployment cost (Dependent Software Placement Cost, DSPC).

[0132] The end-to-end delay of an SFC request is defined as the sum of the link delay and the sum of the VNF processing delay on the server node, and its formula is as follows:

[0133]

[0134] in, Yes Link The delay on is node n τi processing delay on .

[0135] Subsequently, the costs associated with SFC deployment and DSC are defined. τi , expressing the unit CPU cost as The unit memory cost and unit storage cost are defined as and These three indicators are related to the load status of the node. The symbol * indicates the resource type of the service node, which may include CPU, memory or other resources. In the case of The node resource utilization rate is arrive In the range of , the unit cost is linearly distributed. In addition, for nodes with resource utilization in the interval In the case between Afterwards, the deployment cost of SFCr is defined as The DSCs deployed by SFC r are used express. It is composed of the node resource usage cost, and its expression is as follows:

[0136]

[0137] As VNF dependent components migrate, additional migration costs will be incurred to support the deployment of SFCr. τi When it is deployed, it will occupy the storage resources on the server node, thus incurring the placement cost of VNF dependent components. As for DSCs, it consists of DSMCs and DSPCs, which are defined as follows:

[0138]

[0139] Among them, DSMCs is the first part, and DSPCs is defined as the second part. In formula (7), is used to migrate the component ψ q The path, Is a binary variable used to determine the path of Is it mapped to a link? superior, Represents ψ q The amount of storage resources used. Map to On time, SFCr Service Delivery Cost C r The definition is as follows:

[0140]

[0141] It is the sum of SFC deployment cost and DSCs.

[0142] The SD-VDSM problem involves selecting the optimal deployment and migration strategy of VNF dependent components to improve the service quality of SFC and minimize the costs associated with service delivery. Given that the problem involves multiple decisions, the joint optimization objective is expressed as follows:

[0143]

[0144] Among them, ω1 and ω2 are adjustable weight factors used to balance various objectives. By adjusting the weight factor of each objective, the preference of a specific objective can be highlighted. In addition, the constraints of physical network resources and SFC service quality must be met.

[0145] Constraints (10) and (11) ensure that at any server node n τi The total CPU and memory resource requirements required for an SFC deployment must not exceed the respective maximum CPU and memory resource capacities:

[0146]

[0147] Constraint (12) ensures that the used storage resources do not exceed the maximum storage resource capacity:

[0148]

[0149] in, is node n τi Similarly, any link The total bandwidth consumption on the network must be less than the maximum bandwidth capacity. The constraints are as follows:

[0150]

[0151] To meet the quality of service, the delay constraint (14) ensures that the end-to-end delay of any SFCr must meet the following constraints:

[0152]

[0153] Constraint (15) ensures that every f rh It can only be successfully deployed on one server node at most, which means that the VNF instance is indivisible:

[0154]

[0155] To simplify the deployment of SFC, it is assumed that the ingress node of SFCr has only one outgoing link and the egress node has only one incoming link to facilitate user requests. At any intermediate node on the path selected by SFCr, SFCr only uses one incoming link and one outgoing link. When the variable is equal to 1:

[0156]

[0157] Furthermore, in step (4), the SFC deployment method combined with the migration of VNF dependent components is constructed around three main steps: algorithm initialization, calculation of service node and link performance using the hierarchical analysis method, and finally SFC deployment and VNF dependent component migration. The flowchart of the whole process is shown as follows Figure 1 Then, the proposed algorithm is elaborated in detail. Figure 3 The example shown explains the algorithm flow step by step.

[0158] For each VNF in the SFC, there are different requirements for computing, memory, and component resources. Therefore, each physical node has a different matching degree for VNF placement, which increases the complexity of service delivery. Therefore, a multi-layer network structure is proposed, in which each VNF corresponds to an independent layer. This structure facilitates the calculation of the matching relationship between each VNF and all nodes. Different from previous work, a multi-dimensional resource indicator is introduced and a novel algorithm is proposed. In the initialization phase, a network graph MG is generated for each VNF in the SFC. rh These independent network graphs are connected to each other through corresponding nodes to construct a multi-layer network graph MG. r , whose number of layers is |F r |.

[0159] For example, SFC1 contains 3 VNFs, so it is necessary to build a three-layer network diagram consisting of the first layer, the second layer, and the third layer, as shown in Figure 3 As shown. Inter-layer communication is achieved by connecting corresponding nodes, for example, 10011 Connect to n 20011 , and n 20011 Connect to n 30011 The connection between corresponding nodes in different layers is defined as inter-layer link. r After that, the performance indicators of the corresponding VNFs deployed on the physical nodes of each layer need to be calculated. The multi-dimensional characteristics of node indicators bring challenges to the evaluation of physical node performance. When evaluating the performance of VNFs placed on physical nodes, potential component migration further increases the complexity of the measurement process. AHP decomposes the problem into a hierarchical structure problem. Using AHP (such as Figure 4As shown in the figure, the performance of physical nodes is quantified, taking into account factors such as computing, memory and storage resources, component resources, and node processing delay. These factors are specifically reflected in indicators such as resource utilization, storage resource usage cost, remaining computing resources, the inverse of memory and storage resources, and node processing delay. Similarly, the performance of physical links is also evaluated through AHP, taking into account factors such as bandwidth resources and link delay, which is specifically reflected in indicators such as bandwidth utilization, the inverse of remaining bandwidth resources, and the communication delay of the link. Throughout the process of node and link performance evaluation, smaller indicator values ​​are considered to represent better performance. Therefore, smaller indicator values ​​correspond to higher performance of nodes and links. When obtaining the multi-layer weighted graph MWG r After that, the next task is to determine the communication path from the source node to the target node and the node to place the VNF.

[0160] 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, SFC-MAP, algorithms based on BestFit, Greedy and Random were used for comparison. Our simulation experiment was implemented using Python 3.7. In order to simulate a multi-domain network, we chose USANet as the network topology. The topology consists of 24 physical nodes and 43 physical links. We divide the topology into 3 domains, each containing 8 physical nodes. In the experiment, all physical nodes are regarded as server nodes. On this basis, we also explored the scenarios where the number of nodes in each domain is reduced to 7 or increased to 9. Its effects mainly include the following points:

[0161] (1) Acceptance rate

[0162] Figure 5 The comparative analysis of the average SFC acceptance rate of five different algorithms at different arrival rates is shown. The figure shows that the SFC acceptance rate of JSD-VDSMA, BestFit, SFC-MAP and Greedy algorithms decreases with the increase of SFC arrival rate. This phenomenon can be attributed to the proportional increase in the number of received SFC requests with the increase of arrival rate. However, the increase rate of the number of received SFC requests is relatively slow, resulting in a decrease in the SFC acceptance rate.

[0163] The figure also shows that JSD-VDSMA outperforms other algorithms in terms of reception rate, followed by BestFit algorithm. Greedy algorithm shows higher reception rate compared to SFC-MAP algorithm. This difference can be attributed to the JSD-VDSMA algorithm, which carefully selects the best match of physical nodes and links for SFC deployment. BestFit algorithm focuses on finding the best match for VNFs and their connected logical links in each SFC. On the other hand, Greedy algorithm only considers the deployment performance of VNFs on physical nodes. In contrast, SFC-MAP algorithm prioritizes the low latency requirement of SFC, although this sacrifices reception rate.

[0164] In addition, the fluctuation of SFC reception rate over time under a fixed arrival rate is analyzed, e.g. Figure 6 As shown, in particular, from Figure 6 It can be seen from (b), (c) and (d) that under a specific service arrival rate, the reception rate of SFC shows a downward trend as the time slot increases. Figure 6 (a) It can be clearly seen that when λ = 0.05, the SFC reception rate under the SFC-MAP and Greedy algorithms does not continue to decrease with the increase of time slots. The reason for this phenomenon is that although the number of received SFCs increases with the increase of time slots, the growth rate of the number of SFC requests received in each time slot fluctuates.

[0165] We also analyze the average reception rate of SFC under different QoS requirements and network scales. Figure 7 As shown in (a), Figure 7 (a), λ = 0.1, Figure 7 (b) corresponds to λ = 0.1. No matter the delay constraint is high or low, JSD-VDSMA always maintains a high reception rate. Similarly, Figure 7 (b) shows that JSD-VDSMA also maintains a similar high reception rate under different network scales.

[0166] (2) Communication delay

[0167] Figure 8represents the end-to-end communication delay of SFC at different arrival rates. As the SFC arrival rate increases, the average end-to-end delay of SFC under JSD-VDSMA, BestFit, SFC-MAP and Greedy algorithms shows an upward trend. The reason for this trend is that as the SFC arrival rate increases, the number of received SFCs also increases, resulting in more physical nodes with lower processing delay characteristics being utilized to accommodate additional SFC deployments to provide services, which further prompts physical nodes with higher processing delay characteristics to be used, thereby increasing the average processing delay. In addition, as the number of successfully deployed SFCs increases, longer physical links are required to support end-to-end communication, which leads to larger communication delays.

[0168] Since JSD-VDSMA, BestFit, and SFC-MAP algorithms perform well in terms of average end-to-end communication delay, we Fig. 9 The figure shows how the average end-to-end communication delay of the SFC of the three algorithms changes over time under a specific arrival rate. Fig. 9 (a), (b), (c), and (d) all show an upward trend, which can be attributed to the increasing number of received SFCs as the number of time slots increases. This leads to the occupation of nodes with high processing delays and the need to use longer physical links to ensure end-to-end communication. Fig.10 The end-to-end communication delay of SFCs under different QoS requests and network sizes is shown. Under different QoS and network sizes, JSD-VDSMA always maintains a low communication delay. Fig.10 (a) Corresponding to λ=0.1,N τ =8; Fig.10 (b) corresponds to λ=0.1.

[0169] (3) Resource usage cost

[0170] Fig.11The average resource usage cost of SFC at different arrival rates is shown. Given the poor performance of the Random algorithm in terms of SFC reception rate and end-to-end communication delay, we only focus on showing the performance of the JSD-VDSMA, BestFit, SFC-MAP, and Greedy algorithms in terms of resource usage cost. As λ increases from 0.05 to 0.1, the average resource cost increases significantly. This is because, at λ = 0.05, fewer physical node resources are used, and the unit cost of resources is proportional to the resource usage. At λ = 0.1, as the number of deployed SFCs increases, SFCs tend to be deployed on physical nodes with lower latency and cost. Although the latency of these nodes is relatively low, their higher utilization leads to relatively high resource usage costs. While at λ = 0.15 and λ = 0.2, despite the increase in the number of deployed SFCs, the resource cost remains relatively stable compared to λ = 0.1. This is because the optimization objective prioritizes the communication latency and service provision cost of SFCs, so that SFCs are deployed on physical nodes with both low latency and low cost. Fig.12 show, Fig.12 (a) corresponds to λ=0.1,N τ =8, Fig.12 (b) corresponds to λ = 0.1. Under different QoS constraints and network scales, the proposed algorithm achieves relatively low resource usage cost.

[0171] (4) Relative storage resource usage cost

[0172] Fig.13 The relative average storage resource usage cost of SFC under different arrival rates is shown. The relative average resource usage cost is obtained by dividing the average storage resource usage cost by the average storage resource utilization rate. Under the JSD-VDSMA and BestFit algorithms, the relative average storage resource usage cost gradually decreases with the increase of the SFC arrival rate. This trend is attributed to the increase in the SFC arrival rate, which leads to an increase in the number of received SFC requests and requires more components to support the operation of VNFs, resulting in an increase in the use of physical node storage resources. It can be clearly seen from the figure that JSD-VDSMA is able to make full use of storage resources and receive more SFC requests with lower latency. Under the Greedy and SFC-MAP algorithms, when λ=0.05, the relative average storage resource usage cost is almost zero, because the received SFC gives priority to using existing components, and thereafter, the relative average storage resource usage cost remains at a relatively high level. As Fig.14 As shown, Fig.14 (a) corresponds to λ=0.1,N τ =8, Fig.14 (b) is λ=0.1.

[0173] Under different QoS requirements and network scales, the SFC under JSD-VDSMA maintains a relatively low average storage resource usage cost.

[0174] (5) Component Migration Cost and Load Balancing

[0175] Through the analysis of the above five algorithms, it can be observed that when solving the SD-VDSM problem, the JSD-VDSMA and BestFit algorithms perform well in terms of reception rate, communication delay, resource usage cost, and relative storage resource usage cost. Next, we analyzed the impact of the JSD-VDSMA and BestFit algorithms on the average component migration cost and network load balancing.

[0176] The component migration cost is defined in formula (7). For example, when a component needs to be moved from physical node n τ1 Migrate to n τ2 When n τ1 to n τ2 The shortest latency path is determined and the component migration is performed using the shortest path. Fig.15 The average component migration cost of SFC at different arrival rates is shown. According to the JSD-VDSMA algorithm, the average migration cost rises slowly as the SFC arrival rate increases. This is because as the SFC arrival rate increases, the number of component migrations increases, resulting in a more uniform distribution of component resources in the multi-domain network, which leads to a gradual increase in the migration cost. Under the BestFit algorithm, the average component migration cost is highest when λ=0.15, and the migration cost gradually decreases as λ decreases from 0.15 to 0.2. This change is because when λ=0.2, the distribution of components is more uniform, which reduces the time required for components to migrate from one location to another, thereby reducing the average component migration cost.

[0177] Fig.17 The network load balancing under the JSD-VDSMA and BestFit algorithms is shown. As can be seen from the figure, when λ = 0.1, the load balancing degree under both algorithms is high. The experimental results show that when λ = 0.1, SFC occupies a large amount of CPU resources in the third domain, resulting in a more significant load balancing effect. From λ = 0.1 to λ = 0.2, the load balancing degree shows a downward trend, indicating that SFC is more evenly distributed among the three network domains. Fig.16 and 18 It is demonstrated that JSD-VDSMA performs better in average component migration cost and load balancing performance in multi-domain networks of SFC under different QoS requirements and network scales.

[0178] (6) Performance of JSD-VDSMA at different SFC lengths

[0179] To further illustrate the scalability of JSD-VDSMA, we analyze the impact of different SFC lengths on the algorithm performance, as shown in Figure 19. Figure 19(a) shows the average reception rates of JSD-VDSMA and BestFit algorithms when SFC arrival rates are 0.15 and 0.2, and SFC lengths are 2 and 3. It can be seen from the figure that when λ is 0.15 and SFC length is 2, both JSD-VDSMA and BestFit algorithms achieve 100% reception rate. When λ is 0.2 and SFC length is 2, the reception rate of JSD-VDSMA is slightly higher than that of BestFit algorithm. In addition, for λ of 0.15 and 0.2, when SFC length is 3, the reception rate of JSD-VDSMA is always better than that of BestFit algorithm. Figure 19(b) shows the average end-to-end delay of JSD-VDSMA and BestFit algorithms, considering the cases of SFC arrival rates of 0.15 and 0.2, and SFC lengths of 2 and 3. When λ is 0.15 and 0.2, and SFC length is 2 and 3, the end-to-end communication delay of JSD-VDSMA is always lower than that of BestFit algorithm. Figure 19(c) shows the network load of JSD-VDSMA and BestFit algorithm when SFC arrival rate is 0.15 and 0.2, and SFC length is 2 and 3. In terms of load balancing, JSD-VDSMA outperforms BestFit algorithm, especially when λ is 0.15 and 0.2, and SFC length is 2 and 3.

[0180] In summary, in the experimental results, the algorithm has high performance in terms of service reception rate, end-to-end communication delay, resource utilization cost, etc. Therefore, the method proposed in the present invention is better overall.

Claims

1. A SFC deployment method based on VNF dependent component migration in a multi-domain network, characterized in that: The method includes: (1) Based on the SDN controller, a multi-domain network communication architecture is constructed, including multi-domain physical networks, end users, domain SDN controllers and master SDN controllers. End users submit service requests to domain SDN controllers, and domain SDN controllers send information to the master SDN controller. The master SDN controller, in conjunction with the VNF dependent component migration, sends SFC deployment instructions and VNF dependent component migration instructions to the domain SDN controllers. The multi-domain network provides services to end users based on the decisions of the domain SDN controllers. (2) Establishing network topology and service request model, including: (21) Based on multi-domain network topology, business function chains are mapped through heterogeneous multi-domain physical networks; The domain set of a heterogeneous multi-domain physical network is represented as in represents the τth network domain, |G| represents the number of domains, and the set of multi-domain physical network nodes is expressed as n τi represents the i-th node in the network domain τ, and the set of physical network links is expressed as Multi-domain network support types of VNFs and over a long period of time A collection of nodes that work within to support the service provider requested by the user Used to deploy virtual machines to run VNFs and forward data, using the maximum computing power of each server node Maximum memory capacity and maximum storage capacity Respectively represent the computing, memory and storage resources of the node. Each server node belongs to a subnetwork. Define the variables Represents server node n τi Domain, link Connect two server nodes n τi and represents the physical link between them; when τ is equal to , the physical link connection represents an intra-domain link; otherwise, the link Indicates an inter-domain link; and Respectively represent links The starting and ending nodes of To represent the link bandwidth capacity; Used to represent service node n τi The remaining computing resources, Indicates service node n τi Remaining memory resources; use Indicates service node n τi Remaining storage resources; Indicates link Remaining bandwidth resources; (22) Build a business request model; All user requests with actual needs entering the NFV-based multi-domain network are represented as SFC requests. The SFC request set is represented as Γ={Γ1,Γ2,...,Γ r ,...,Γ |Γ| }; Define quintuple Used to represent an SFC request r, where s r ,d r denote the source node and the destination node respectively; each SFC request r has a known bandwidth requirement b r and a maximum tolerable end-to-end delay requirement For an SFC request r, it consists of a set of predefined VNFs and links, where the VNFs represent more than one virtual network function, represented as a directed weighted graph G r ={F r ,E r },in Represents a collection of VNFs, Represents a collection of virtual network links; each VNFf rh ∈F r A certain amount of server computing resources are required and memory resources Similarly, mapping a virtual link requires allocating bandwidth resources b r ; (23) Building a VNF dependent component migration model to migrate components from one server node to another; When a virtual network function rh It is decided to be deployed on a server node with insufficient component resources n τi When on, f rh The related dependent components will be migrated to node n τi ; Ψ={ψ1,ψ2,...,ψ q ,...,ψ |Ψ| } represents the collection of all component resources that support VNF ​​operation; For VNF placement and logical link mapping, two binary decision variables are used to define this relationship: Logical Link Can be mapped to multiple physical links, indicating that communication between VNFs can span multiple physical connections; (24) VNF dependent component migration decision The working time period of the NFV-based multi-domain network is divided into several time slots. Assuming that new SFC request information is collected in time slot t-1, SFC will be deployed in time slot t. At the same time, the component resources ψ q The migration will be executed and completed within time slot t, defining a binary variable Represents server node n τi Whether VNFf is supported in time slot t rh , is a binary migration decision variable, indicating whether node n is τi Migration component ψ q : in, is a binary variable indicating that at time slot t, the component ψ q Is it at node n τi Up, rh Supports virtual network functions rh The set of dependent components to be run. ResourceConstraint is an indicator variable, indicating the service node n τi Are other resources on the network sufficient to support the virtual network function? rh When the service node n τi When the resources on the network are sufficient, its value is 1, and the corresponding mathematical expression is: when , indicating that in time slot t, component ψ q Has been migrated to server node n τi Otherwise, it means component ψ q Not migrated, using Indicates that VNFf rh The component resource collection to be migrated, after determining the components to be migrated, updates the component resources on the corresponding nodes; (3) Formulate a problem to be solved and determine the optimization goal based on the network topology and service request model constructed in step (2); The problem to be solved is defined as follows: SFC deployment problem: Given a network domain set and the SFC request set Γ, find the routing path of SFC and its deployment strategy, considering the quality of the selected path to minimize the SFC deployment cost and ensure the end-to-end delay of SFC while satisfying the physical network resource constraints; VNF dependent component migration problem: Based on the given network domain set And based on the SFC deployment results, combined with the distribution matrix of current component resources Find component migration strategies and their migration paths, taking into account VNF’s requirements for dependent components and minimizing migration costs when the selected server nodes cannot provide sufficient component resources; The overall optimization goal is determined based on the multi-domain network information, user terminal service request information, and VNF dependent component migration model. The overall optimization goal is to minimize the end-to-end delay of all SFCs and the costs associated with SFC deployment and VNF dependent components, namely, VNF dependent component migration cost and VNF dependent component deployment cost, where the end-to-end delay of the SFC request is defined as the sum of the link delays D r , mathematical calculation is expressed as: Yes Link The delay on is the service node n τi Processing delays on The deployment cost of an SFC request r is defined as The total cost of SFC request r deployment is express, It is composed of the node resource usage cost, and its expression is as follows: It consists of the VNF dependent component migration cost and the VNF dependent component deployment cost, which is defined as follows: The service delivery cost C of SFC request r r The definition is as follows: The joint optimization objective is expressed as follows: Among them, ω1 and ω2 are adjustable weight factors used to balance various objectives. By adjusting the weight factor of each objective, the preference of a specific objective is highlighted, and the constraints of physical network resources and SFC service quality are met. Constraints (10) and (11) ensure that at any server node n τi The total CPU and memory resource requirements required for an SFC deployment must not exceed the respective maximum CPU and memory resource capacities: Constraint (12) ensures that the used storage resources do not exceed the maximum storage resource capacity: in, is node n τi Similarly, any link The total bandwidth consumption on must be less than the maximum bandwidth capacity, subject to the following constraints: To meet the quality of service, the delay constraint (14) ensures that the end-to-end delay of any SFC request r must meet the following constraints: Constraint (15) ensures that every f rh It can only be successfully deployed on one server node at most, which means that the VNF instance is inseparable, corresponding to: In order to simplify the deployment of SFC, it is assumed that the entry node of SFC request r has only one outgoing link and the exit node has only one incoming link to facilitate the realization of user requests; at any intermediate node on the path selected by SFC request r, SFC request r is mathematically expressed as SFCr, SFCr uses only one incoming link and one outgoing link if and only if SFCr uses physical link When the variable is equal to 1: (4) Design layered architecture and deployment and migration mechanisms; A multi-layer network architecture is adopted, and each VNF corresponds to an independent layer of network. The matching relationship between each VNF and all nodes is calculated based on the hierarchical analysis method, and a multi-layer network weighted graph is constructed. In the multi-layer weighted graph, the shortest path for deploying SFC and the placement node of VNF are found based on the dynamic programming algorithm. After obtaining the node where VNF is placed, the migration strategy of VNF dependent components is determined.

2. The SFC deployment method according to claim 1, characterized in that: The method is based on the terminal user submitting extreme service request information to the sub-domain SDN controller, and the sub-domain SDN controller transmits the information to the master control SDN controller. The master control SDN controller decides the multi-domain network to provide deployment location and migration strategy of VNF dependent components for the service function chain through the SFC deployment method of joint VNF ​​dependent component migration, thereby reducing the end-to-end communication delay and resource usage cost of the service, the resource usage cost of VNF dependent components and the migration cost of VNF dependent components, and ensuring multi-domain network load balancing.

3. The SFC deployment method according to claim 1, characterized in that: In the implementation of SFC deployment, the network topology and service request model constructed in step (2) not only rely on virtualization technology, but also need to fully consider the adaptability of component resources of physical nodes, including the network attributes such as the capacity of computing nodes, link bandwidth and latency of the multi-domain network implemented based on the network topology, and use a graph model to describe the communication topology within and between domains; define multiple service request types, extract key indicators of their network attributes, and use a random process to simulate the dynamics of request arrival; determine the VNF dependent component migration model based on the two established models.

4. The SFC deployment method according to claim 1, characterized in that: The layered architecture design, deployment, and migration mechanism described in step (4) are specifically: (41) Algorithm initialization; Build a multi-layer network structure, where each VNF corresponds to an independent layer to calculate the matching relationship between each VNF and all nodes; Multi-dimensional resource indicators are introduced. In the initialization phase, a network graph MG is generated for each VNF in the SFC. rh ; Independent network graph MG rh By connecting the corresponding nodes to each other, a multi-layer network graph MG is constructed. r , whose number of layers is |F r |; (42) Performance measurement based on AHP; First, from the multi-layer network graph MG r The data of each layer of physical nodes are extracted from the network, including the remaining computing, memory and storage resources, component resources, and processing delays; then, a pairwise comparison matrix NCM is constructed to evaluate the criterion layer for the target; then, to compare the multi-layer network graph MG rh The pairwise relationship of all nodes of the same type of elements in the construct eight Matrices, named computing resource utilization Memory resource utilization Storage resource utilization Storage resource usage cost Countdown of remaining computing resources Remaining memory resources countdown Remaining storage resources countdown and processing delays Then the matrix is ​​tested for consistency; The consistency ratio is used to evaluate the consistency of the matrix, and the consistency index CI is defined as Used to quantify the consistency within the judgment matrix, where λ max Represents the largest eigenvalue of a matrix; The random consistency index RI depends on the dimension of the matrix and is obtained from a table of known RI values; If the matrix satisfies the consistency criterion, the eigenvector corresponding to the largest eigenvalue is extracted; Otherwise, rebuild the matrix; then, calculate h τi The indicator value of For the link index, a similar method is used to calculate, where the index value of the inter-layer link is always set to 0. Through the above iterative process, a multi-layer weighted network is finally generated, denoted as MWG r ; (43) SFC deployment and VNF dependent component migration strategy; Getting the multi-layer weighted graph MWG r After that, the next task is to determine the communication path from the source node to the target node and the node to place the VNF, specifically: First, map the starting node to MWG r Node labels in, for each layer, select a placement node for the VNF and calculate the metrics of intra-layer links and inter-layer communication links, then calculate the comprehensive metrics of the selected placement service nodes and links, and finally, return the selected deployment service nodes and communication links; The metric calculation of the service node takes into account potential component migration scenarios, thereby selecting the optimal migration strategy based on the VNF placement.

Images