Cross-domain service function chain deployment method and system

Abstract

The invention relates to the technical field of cross-domain deployment, and discloses a cross-domain service function chain deployment method and system, and the method comprises the following steps: dividing a service function chain into at least one service function chain segment according to the delay demand of a virtual network function in the service function chain; evaluating the reliability of physical nodes in the domain, and calculating the overall reliability of the domain based on the reliability of the physical nodes; based on the segmentation result and the reliability evaluation result, selecting at least one domain for deploying the service function chain; and in the selected domain, utilizing deep learning to allocate and deploy resources for the virtual network functions in the service function chain. Through a delay segmentation strategy, the SFC is decomposed into smaller fragments, and the cross-domain delay characteristic is better adapted; through domain reliability evaluation, a domain with higher reliability is selected for deployment, and the overall reliability of the SFC is improved; in combination with deep reinforcement learning, efficient resource allocation and deployment in the domain are realized; and meanwhile, the system adopts a hierarchical control architecture, so that intra-domain privacy security is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of cross-domain deployment technology, and in particular to a cross-domain service function chain deployment method and system. Background Art

[0002] With the booming development of network function virtualization (NFV) and software defined network (SDN) technologies, service function chaining (SFC) has become a key technology for building flexible and efficient network services. SFC implements complex network services by connecting a series of virtual network functions (VNFs) in sequence. However, with the expansion of network scale and the increasing complexity of business needs, the deployment of SFC has gradually expanded from single-domain networks to multi-domain network environments. Cross-domain SFC deployment faces many new challenges, among which reliability, latency, cost and privacy protection are four crucial aspects.

[0003] Existing technologies mainly focus on optimizing deployment costs, latency, and resource utilization in cross-domain SFC deployment. For example, some studies are dedicated to finding the most cost-effective cross-domain path to reduce operators' CAPEX and OPEX; other studies focus on how to optimize resource allocation and improve the utilization efficiency of network resources; and some studies focus on reducing end-to-end latency and improving user experience. These studies have achieved certain results in their respective focuses, but they generally have the following defects:

[0004] Insufficient attention to reliability: Existing technologies often ignore the reliability of cross-domain SFC deployment, especially in complex multi-domain environments, where the risk of link and node failures increases and reliability issues become more prominent. The reason is that traditional deployment methods do not fully consider reliability factors in cross-domain environments and lack effective reliability assessment and assurance mechanisms. Simply optimizing cost or latency may result in the selected path passing through multiple domains or links with lower reliability, thereby reducing the overall reliability of the SFC.

[0005] Delay constraints are difficult to guarantee: Existing technologies have limitations in dealing with delay constraints. Although some methods take delay into consideration, they often simply optimize end-to-end delay without fine-tuning the specific delay requirements between VNFs. In a cross-domain environment, the delay of inter-domain links may be high and unstable. If the delay requirements between VNFs are strict, traditional deployment methods may find it difficult to guarantee service quality. The reason is that there is a lack of a mechanism that can adaptively adjust the deployment strategy according to the delay requirements between VNFs, which cannot effectively address the delay challenges in a cross-domain environment.

[0006] Insufficient privacy protection: Traditional centralized cross-domain SFC deployment methods also face challenges in privacy protection. The centralized method requires the multi-domain controller (MDC) to collect and process sensitive data such as network topology and resource information of all domains, which not only increases the burden on the MDC, but also brings the risk of privacy leakage within the domain. In a multi-domain network environment, each domain is often managed by different operators or organizations, which has higher requirements for privacy protection. The reason is that the centralized architecture naturally has the risk of privacy leakage, and lacks a mechanism that can protect privacy within the domain while ensuring cross-domain collaboration.

[0007] Low deployment efficiency: Traditional resource allocation and path selection methods may be inefficient and difficult to adapt to dynamically changing network environments and large-scale SFC deployment requirements. The reason is that there is a lack of a mechanism that can intelligently and efficiently allocate resources and select paths, which cannot effectively cope with the challenges brought by large-scale SFC deployment. Summary of the invention

[0008] The present invention provides a cross-domain service function chain deployment method and system, which solves the technical problems of the background technology.

[0009] The present invention provides a cross-domain service function chain deployment method, comprising the following steps:

[0010] S1: Divide the service function chain into at least one service function chain segment according to the delay requirement of the virtual network function in the service function chain;

[0011] S2: Evaluate the reliability of physical nodes in the domain and calculate the overall reliability of the domain based on the reliability of physical nodes;

[0012] S3: Based on the segmentation results and reliability evaluation results, select at least one domain for deploying the service function chain;

[0013] S4: In the selected domain, use deep learning to allocate resources and deploy virtual network functions in the service function chain.

[0014] In the above scheme, through domain reliability assessment and reliability-based domain selection, the probability of faulty nodes in the SFC deployment path is reduced, the overall reliability of SFC is improved, and the risk of link and node failure in complex multi-domain environments is effectively dealt with; through SFC segmentation and combining the average delay of inter-domain links with the delay requirements between VNFs for segmentation, it is possible to handle delay constraints more finely, deploy delay-sensitive VNFs in the same domain, reduce the delay caused by cross-domain links, and effectively ensure service quality; using deep learning methods for intra-domain resource allocation and deployment, it is possible to intelligently and efficiently allocate resources and select paths, adapt to dynamically changing network environments and large-scale SFC deployment requirements, and improve deployment efficiency.

[0015] Preferably, in step S1, dividing the service function chain into at least one service function chain segment comprises:

[0016] Calculate the average delay of all inter-domain links;

[0017] comparing the latency requirements between adjacent virtual network functions in the service function chain with the average latency of the inter-domain links;

[0018] If the delay requirement is greater than the average delay of the inter-domain link, segmenting between the adjacent virtual network functions;

[0019] The location information of all segmentation points is recorded to generate a segmentation topology map.

[0020] In the above scheme, by comparing the delay requirement with the average delay of the inter-domain link, the service function chain segmentation based on the delay requirement is realized, making the segmentation strategy more accurate and able to more effectively meet the delay requirements of different SFCs; providing clear segmentation judgment criteria, making the segmentation process more operational and easy to implement; generating a segmentation topology map, providing structured segmentation information for subsequent domain selection and deployment, and facilitating the execution of subsequent steps.

[0021] Preferably, the calculation of the average delay of all inter-domain links is specifically as follows:

[0022]

[0023] Among them, D m and D n Represented as the mth and nth network domains in a multi-domain network; D m,n Indicated as D m and D n The physical link delay between them; E represents the set of physical links.

[0024] In the above scheme, a clear formula for calculating the average delay of inter-domain links is provided, making the segmentation method more specific and executable, and facilitating engineering implementation; the average delay of inter-domain links is calculated by the arithmetic mean, which provides an accurate benchmark value for subsequent delay comparison and improves the accuracy of segmentation judgment.

[0025] Preferably, in step S2, the reliability of the physical nodes in the evaluation domain includes:

[0026] The objective function is used to maximize the reliability of physical nodes, specifically:

[0027]

[0028] Among them, λ is an integer used to balance the weight between cost and reliability. represents the comprehensive reliability of the physical nodes selected for deploying virtual network functions; δ is a weight parameter used to balance the importance of the time dimension and the space dimension. for Reliability in the time dimension, for Reliability of spatial dimensions, Indicates whether the i-th virtual network function in r is mapped to a physical node in domain Dm Indicates whether the kth virtual link in the rth request is mapped to an intra-domain link Indicates whether the kth virtual link in the rth request is mapped to the inter-domain link e m,n ;

[0029] The objective function includes the following constraints:

[0030] Each virtual network function must and can only be mapped to one physical node;

[0031] The computing resources of each node will not exceed its capacity limit:

[0032] Each virtual link must and can only be mapped to one physical link;

[0033] There are bandwidth requirement constraints on both intra-domain links and inter-domain links;

[0034] For any virtual link, the sum of the actual delays of the mapped physical links cannot exceed the maximum allowed delay between the virtual network functions at both ends of the virtual link;

[0035] It has load balancing constraints and cost budget constraints.

[0036] In the above scheme, the reliability of physical nodes is quantitatively evaluated through objective functions and weight parameters, which provides a reliable basis for subsequent domain reliability calculation and domain selection. The objective function distinguishes the reliability of time and space dimensions, making the reliability evaluation more comprehensive and accurate. Through multiple constraints, the rationality and feasibility of VNF deployment are guaranteed, avoiding resource conflicts, capacity overload, delay overruns and other problems. By selecting the most reliable physical nodes for VNF deployment, the overall reliability of SFC is ultimately improved and the risk of service interruption is reduced.

[0037] Preferably, in step S2, obtaining the overall reliability of the domain based on the reliability of the physical nodes includes:

[0038] The overall reliability of a domain includes the inherent reliability of physical nodes and the remaining resources of the nodes;

[0039] The remaining resources of the node include domain Dm The remaining available resource ratio of the i-th physical node in , specifically including the remaining computing resource ratio and the remaining bandwidth resource ratio;

[0040] According to the remaining available resources of the node and the reliability value of the node, the weighted average method is used to define the domain D m reliability.

[0041] In the above scheme, the inherent reliability of physical nodes and the remaining resources of nodes are comprehensively considered, making the domain reliability assessment more comprehensive and closer to the actual situation; introducing the remaining resources of nodes as a factor in reliability assessment can more accurately reflect the reliability level of the domain and avoid the one-sidedness brought by only considering the inherent reliability of nodes; through a more comprehensive and accurate domain reliability assessment, a more reliable basis is provided for subsequent domain selection, further improving the overall reliability of SFC.

[0042] Preferably, in step S3, selecting at least one domain for deploying the service function chain based on the segmentation result and the reliability evaluation result includes:

[0043] Construct a directed weighted graph, where each domain is a node, the overall reliability of the domain is the node weight, and the cost of the inter-domain link is the edge weight;

[0044] In the directed weighted graph, a path with the lowest cost and the highest reliability is found, and the nodes in the path are the domains used to deploy the service function chain.

[0045] In the above scheme, the complex domain selection problem is transformed into a mature graph theory optimal path search problem, which can be solved using existing graph algorithms, simplifying the domain selection process; by taking domain reliability as node weight and link cost as edge weight, it is possible to consider both cost and reliability in the optimal path search process of the graph, thus achieving comprehensive optimization of the two; using graph algorithms for domain selection can improve the efficiency and intelligence level of domain selection, and adapt to the SFC deployment requirements in large-scale multi-domain network environments.

[0046] Preferably, the node weight is normalized using the inverse of the domain reliability value; and the edge weight is a linear combination of the link deployment cost and the transmission delay.

[0047] In the above scheme, specific node weight and edge weight calculation methods are provided, making the domain selection method more specific and executable, and facilitating engineering implementation; the node weights are processed by inverse normalization, cleverly converting "highest reliability" into "lowest weight", so that the reliability weight can be effectively combined with the graph algorithm to search for the optimal path; the edge weight adopts a linear combination of cost and delay, and the weight relationship between cost, reliability and delay in domain selection can be flexibly adjusted by adjusting the weight coefficient to meet the deployment requirements of different SFCs.

[0048] Preferably, in step S4, the using deep learning to allocate resources for virtual network functions in the service function chain and deploying them includes:

[0049] Construct a dual deep Q network DDQN model, where the state space includes information about physical nodes in the domain and information about SFC requests, and the action space includes deployment locations and paths;

[0050] The DDQN model is guided to learn the optimal deployment strategy through a reward function that takes into account resource utilization, deployment cost, and link reliability factors.

[0051] According to the trained DDQN model, the optimal deployment location and path are selected for the virtual network functions within the service function chain segment.

[0052] In the above scheme, the deep reinforcement learning method is used to make the resource allocation and deployment process more intelligent, and the deployment strategy can be adaptively adjusted according to the dynamic changes of the network environment; the deep learning model can make decisions quickly, improve the efficiency of resource allocation and deployment, and adapt to the needs of large-scale SFC deployment; through the design of the reward function, the DDQN model can learn the optimal deployment strategy that takes into account resource utilization, deployment cost and link reliability, and achieve multi-objective optimization; the deep reinforcement learning model has strong generalization ability, can adapt to the dynamically changing network environment, and maintain the effectiveness of the deployment strategy.

[0053] Preferably, the state space of the DDQN model is specifically:

[0054] S={V,Φ,B,D,N m ,CR m ,R m ,CC m ,B m ,B.C. m ,a}

[0055] Where V represents the set of virtual network functions, Φ and B represent the set of computing and bandwidth resource requirements of virtual network functions, D represents the set of latency requirements, and N m Represents the set of available physical nodes, CR mRepresents the reliability value of the node, R m and B m represents the available computing and bandwidth resources, CC m and B.C. m represents the unit computing and bandwidth resource cost; a = [a 1 ,a 2 ,...,a L ] is an allocation vector, indicating the domain number to which the virtual network function in the service function chain request is allocated, and L is the preset maximum number of virtual network functions;

[0056] The action space of the DDQN model is specifically:

[0057]

[0058] in, Indicates whether the i-th virtual network function in r is mapped to Indicates whether the kth virtual link in the rth request is mapped to an intra-domain link

[0059] The reward function is specifically:

[0060]

[0061] Among them, w R 、w C 、w D are the weight factors of reliability, cost and delay, respectively, and C cost is the deployment cost of nodes and links.

[0062] In the above scheme, the state space and action space of the DDQN model are clearly defined, the input and output of the model are clarified, and the design and implementation of the model are convenient; through the design of the reward function, reliability, cost and delay are quantified as reward values, and the weight relationship between the three in the deployment decision can be flexibly adjusted by adjusting the weight factor to achieve multi-objective optimization; the specific state space, action space and reward function are defined, which provides a clear goal and direction for the training of the DDQN model and improves the trainability and effectiveness of the model.

[0063] Preferably, a cross-domain service function chain deployment system applies a cross-domain service function chain deployment method, including a hierarchical control architecture, the hierarchical control architecture including a multi-domain controller and multiple intra-domain controllers;

[0064] The steps S1 and S3 are executed by a multi-domain controller, and the steps S2 and S4 are executed by a controller within a domain.

[0065] In the above scheme, the method is applied to an actual hierarchical control architecture, verifying the feasibility and effectiveness of the method in an actual system; a hierarchical control architecture is adopted to separate global control and intra-domain control, which reduces the control complexity, improves the scalability and efficiency of the system, and can adapt to large-scale multi-domain network environments; the intra-domain controller is responsible for intra-domain operations, and the multi-domain controller does not need to obtain detailed information within the domain, which reduces the risk of privacy leakage within the domain and meets the needs of privacy protection in a multi-domain network environment; the multi-domain controller and the intra-domain controller work together to realize multi-domain collaborative control and intra-domain autonomous management, and improve the flexibility and robustness of the system.

[0066] Compared with the prior art, the present invention has the following beneficial effects:

[0067] The present invention discloses a cross-domain service function chain deployment method and system. Through a delay segmentation strategy, SFC is decomposed into smaller segments to better adapt to the cross-domain delay characteristics; through domain reliability evaluation, a domain with higher reliability is selected for deployment to improve the overall reliability of SFC; and combined with a deep reinforcement learning algorithm, efficient resource allocation and deployment within the domain are achieved; at the same time, the system adopts a hierarchical control architecture to ensure privacy and security within the domain. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 It is a schematic diagram of a cross-domain service function chain deployment method provided by an embodiment of the present invention;

[0069] Figure 2 is a flow chart of a highly reliable cross-domain deployment algorithm provided by an embodiment of the present invention;

[0070] Figure 3 It is a schematic diagram of a cross-domain service function chain deployment system architecture provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0071] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0072] like Figure 1 As shown, the present application provides a cross-domain service function chain deployment method, comprising the following steps:

[0073] S1: Divide the service function chain into at least one service function chain segment according to the delay requirement of the virtual network function in the service function chain;

[0074] S2: Evaluate the reliability of physical nodes in the domain and calculate the overall reliability of the domain based on the reliability of physical nodes;

[0075] S3: Based on the segmentation results and reliability evaluation results, select at least one domain for deploying the service function chain;

[0076] S4: In the selected domain, use deep learning to allocate resources and deploy virtual network functions in the service function chain.

[0077] In the above scheme, in response to the problem of "delay constraints are difficult to guarantee" mentioned in the background technology, step S1 segments the SFC by analyzing the delay requirements between VNFs in the SFC; if the delay requirements between adjacent VNFs are relatively strict, they may be deployed in the same domain to avoid the high delay and instability caused by cross-domain links. This segmentation strategy can handle delay constraints more finely and improve service quality; in response to the problem of "lack of attention to reliability" mentioned in the background technology, step S2 introduces a domain reliability evaluation mechanism. By evaluating the reliability of physical nodes in the domain and aggregating them into the overall reliability of the domain, a reliability basis is provided for subsequent domain selection. This shows that this method not only focuses on cost and delay, but also attaches more importance to the reliability of cross-domain deployment, which can reduce the risk of SFC service interruption caused by domain or link failures; step S3 is the key link to connect the segmentation results and reliability evaluation. Based on the segmentation results of step S1, the deployment requirements of the SFC segment are determined. At the same time, combined with the domain reliability evaluated in step S2, a domain with higher reliability is selected for deployment. This selection strategy can give priority to domains with high reliability while meeting the latency requirements, further improving the overall reliability of SFC. After the domain is selected, step S4 uses deep learning methods to allocate and deploy resources within the domain. The application of deep learning aims to improve the efficiency of resource allocation and path selection and solve the problem of "low deployment efficiency" mentioned in the background technology. The deep learning model can learn the dynamic changes of the network environment and intelligently allocate resources to meet the needs of large-scale SFC deployment.

[0078] Preferably, in step S1, dividing the service function chain into at least one service function chain segment comprises:

[0079] Calculate the average delay of all inter-domain links;

[0080] comparing the latency requirements between adjacent virtual network functions in the service function chain with the average latency of the inter-domain links;

[0081] If the delay requirement is greater than the average delay of the inter-domain link, segmenting between the adjacent virtual network functions;

[0082] The location information of all segmentation points is recorded to generate a segmentation topology map.

[0083] In the above scheme, in order to make full use of the resources in each domain, the SFC needs to be segmented first. For each SFC request, it can be segmented according to the average delay of the inter-domain link to ensure that the delay requirements between VNFs are met when deployed across domains.

[0084] Preferably, the calculation of the average delay of all inter-domain links is specifically as follows:

[0085]

[0086] Among them, D m and D n Represented as the mth and nth network domains in a multi-domain network; D m,n Indicated as D m and D n The physical link delay between them; E represents the set of physical links.

[0087] In the above scheme, for each SFC request r, check the delay requirement D between VNFs r,i Is it greater than the average delay of the inter-domain link? Set an empty list segments to record the location of the segment. If D r,i > average delay, then record position i in segments.

[0088] Preferably, in step S2, the reliability of the physical nodes in the evaluation domain includes:

[0089] The objective function is used to maximize the reliability of physical nodes, specifically:

[0090]

[0091] Among them, λ is an integer used to balance the weight between cost and reliability. represents the comprehensive reliability of the physical nodes selected for deploying virtual network functions; δ is a weight parameter used to balance the importance of the time dimension and the space dimension. for Reliability in the time dimension, for Reliability of spatial dimensions, Indicates whether the i-th virtual network function in r is mapped to a physical node in domain Dm Indicates whether the kth virtual link in the rth request is mapped to an intra-domain link Indicates whether the kth virtual link in the rth request is mapped to the inter-domain link e m,n ;

[0092] The objective function includes the following constraints:

[0093] Each virtual network function must be mapped to and can only be mapped to one physical node:

[0094]

[0095] The computing resources of each node will not exceed its capacity limit:

[0096]

[0097] Each virtual link must be mapped to only one physical link:

[0098]

[0099] There are bandwidth requirement constraints on both intra-domain links and inter-domain links;

[0100] For intra-domain links:

[0101] For inter-domain links:

[0102] For any virtual link, the sum of the actual delays of the mapped physical links cannot exceed the maximum allowed delay between the virtual network functions at both ends of the virtual link:

[0103]

[0104] With load balancing constraints and cost budget constraints:

[0105]

[0106] Preferably, in step S2, obtaining the overall reliability of the domain based on the reliability of the physical nodes includes:

[0107] The overall reliability of a domain includes the inherent reliability of physical nodes and the remaining resources of the nodes;

[0108] The remaining resources of the node include domain D m The remaining available resource ratio of the i-th physical node in , specifically including the remaining computing resource ratio and the remaining bandwidth resource ratio;

[0109] According to the remaining available resources of the node and the reliability value of the node, the weighted average method is used to define the domain D m reliability.

[0110] In one embodiment provided in the present application, during the SFC deployment process, selecting a domain with high reliability is crucial to ensure service continuity and performance. In order to accurately evaluate and select the most appropriate domain, when defining the reliability of the domain, it is necessary not only to consider the inherent reliability of the physical node, but also to comprehensively consider the remaining resources of the node. The more remaining resources, the more capacity the node has to handle burst traffic or perform additional tasks, thereby improving the flexibility and robustness of the overall system;

[0111] Domain D m The remaining available resource ratio of the i-th physical node in:

[0112] Remaining computing resource ratio:

[0113]

[0114] Remaining bandwidth resource ratio:

[0115]

[0116] Then the node The ratio of the average remaining available resources to:

[0117]

[0118] According to the remaining available resources of the node and the reliability value of the node, the weighted average method is used to define the domain D m Reliability:

[0119]

[0120] Preferably, in step S3, selecting at least one domain for deploying the service function chain based on the segmentation result and the reliability evaluation result includes:

[0121] Construct a directed weighted graph, where each domain is a node, the overall reliability of the domain is the node weight, and the cost of the inter-domain link is the edge weight;

[0122] In the directed weighted graph, a path with the lowest cost and the highest reliability is found, and the nodes in the path are the domains used to deploy the service function chain.

[0123] In one embodiment provided in the present application, each domain D m As a node, its weight is the reliability of the domain, and each edge e m,n The weight is the cost of the inter-domain link BC m,nFirst, select the domain with the highest reliability as the deployment domain of the first segment, filter out the next domain that meets the conditions according to the delay and bandwidth requirements between segments, and build a directed weighted graph WG = (D', L'), where D' is the set of feasible domains and L' is the set of inter-domain links that meet the conditions. Starting from the source node, use the depth-first algorithm to recursively explore all possible paths, and track the total cost and total reliability of each path during the exploration process to find a path with the lowest cost and the highest reliability. The nodes in this path are the domains assigned to each SFC segment.

[0124] Preferably, the node weight is normalized using the inverse of the domain reliability value; and the edge weight is a linear combination of the link deployment cost and the transmission delay.

[0125] Preferably, in step S4, the using deep learning to allocate resources for virtual network functions in the service function chain and deploying them includes:

[0126] Construct a dual deep Q network DDQN model, where the state space includes information about physical nodes in the domain and information about SFC requests, and the action space includes deployment locations and paths;

[0127] The DDQN model is guided to learn the optimal deployment strategy through a reward function that takes into account resource utilization, deployment cost, and link reliability factors.

[0128] According to the trained DDQN model, the optimal deployment location and path are selected for the virtual network functions within the service function chain segment.

[0129] In the above scheme, in the SFC deployment problem, since its state space is discrete, it is appropriate to use the DDQN algorithm. Compared with the traditional DQN, DDQN effectively alleviates the problem of Q value overestimation by separating the process of action selection and value evaluation, thereby avoiding possible deviations in the learning process and ensuring the convergence and quality of the strategy. For tasks such as SFC deployment that have strict requirements on resource allocation and path selection, the stability of DDQN is particularly critical. It can help the system make node and link selection decisions more accurately, thereby improving the success rate of SFC requests and optimizing the utilization efficiency of network resources. In addition, DDQN not only shows excellent learning performance in the short term, but also shows superior stability and adaptability in long-term operation.

[0130] Preferably, the state space of the DDQN model is specifically:

[0131] S={V,Φ,B,D,N m ,CR m ,R m ,CC m ,B m,B.C. m ,a}

[0132] Where V represents the set of virtual network functions, Φ and B represent the set of computing and bandwidth resource requirements of virtual network functions, D represents the set of latency requirements, and N m Represents the set of available physical nodes, CR m Represents the reliability value of the node, R m and B m represents the available computing and bandwidth resources, CC m and B.C. m represents the unit computing and bandwidth resource cost; a = [a 1 ,a 2 ,...,a L ] is an allocation vector, indicating the domain number to which the virtual network function in the service function chain request is allocated, and L is the preset maximum number of virtual network functions;

[0133] The action space of the DDQN model is specifically:

[0134]

[0135] in, Indicates whether the i-th virtual network function in r is mapped to Indicates whether the kth virtual link in the rth request is mapped to an intra-domain link

[0136] The reward function is specifically:

[0137]

[0138] Among them, w R 、w C 、w D are the weight factors of reliability, cost and delay, respectively, and C cost is the deployment cost of nodes and links.

[0139] In the above scheme, since the units of node reliability, resource cost and delay are not in the same order of magnitude, directly weighting and summing them may cause some factors to be dominated by other factors. Therefore, these parameters need to be properly normalized to ensure that they are in the same order of magnitude so that the impact of each factor can be fairly reflected in the reward function. Use the minimum-maximum normalization method:

[0140] Preferably, a cross-domain service function chain deployment system applies a cross-domain service function chain deployment method, including a hierarchical control architecture, the hierarchical control architecture including a multi-domain controller and multiple intra-domain controllers;

[0141] The steps S1 and S3 are executed by a multi-domain controller, and the steps S2 and S4 are executed by a controller within a domain.

[0142] like Figure 2 As shown, in one embodiment provided by the present application, the multi-domain controller calculates the average delay between domains based on the abstract graph. If the delay requirement between VNFs is greater than the average delay between domains, segmentation can be performed to ensure that the delay requirement between VNFs is met during cross-domain deployment. After the SFC request is segmented, a directed weighted graph is constructed, and the nodes in the path with the lowest inter-domain link cost are selected as the domains assigned to each segment, and the domains assigned to each VNF are recorded.

[0143] Domain controller: In each domain, the DDQN algorithm is used to select physical nodes and links for the VNFs in the assigned segments.

[0144] like Figure 3 As shown, in one embodiment provided by the present application, the system architecture consists of a multi-domain controller and multiple intra-domain controllers; the architectural design realizes the comprehensive management of delay constraints, reliability evaluation and cost-effectiveness in the SFC deployment process through a hierarchical control mechanism. Among them, the multi-domain controller is responsible for segmenting the SFC based on the delay constraints between VNFs, and selecting domains with high reliability to carry SFC segments according to the historical performance and current status of each domain. After considering the cost of each candidate domain, the MDC finally determines the specific allocation plan of the SFC segment, striving to minimize costs while ensuring service quality. Each domain is configured with an independent domain controller, which uses historical data and real-time monitoring information to evaluate the reliability indicators of all nodes under its jurisdiction, and uses the DDQN method to intelligently select the most suitable physical nodes and links for the VNF in each SFC segment to ensure efficient and reliable resource allocation.

[0145] The multi-domain network G = (N, E) is modeled as an undirected graph, where N is the set of all physical nodes and E is the set of all physical links. There are m network domains, and the set of domains can be expressed as D = {D 1 ,D 2 ,...,D m}. Assume that network domain D m =(N m ,E m ), N m ∈N represents the domain D m The set of all physical nodes in E m Domain D m The set of physical links in the domain; physical node Total computing resource capacity The available computing resources are The unit computing resource cost is Physical Link The maximum bandwidth is The available bandwidth is The unit bandwidth cost is The inherent delay is Assume that there is only one physical link between the two domains, network domain D m and D n The physical link between m,n The maximum bandwidth of ∈E is MB m,n , the available bandwidth is B m,n , the unit bandwidth cost is BC m,n , the intrinsic delay is D m,n .

[0146] Each domain D m As an aggregation node, construct an abstract view G'=(D,E'), E'={(D i ,D j )|D i ,D j ∈Dand D i ≠D j} is a collection of inter-domain links.

[0147] Abstract Model:

[0148] ①Domain as a node: Each domain D m ∈D is regarded as a node, D m It contains a set of physical nodes and links, but the multi-domain coordinator does not need to know these details;

[0149] ②Inter-domain links as edges: Each edge e m,n ∈E' represents two domains D m and D n Direct connection between.

[0150] The abstract graph is represented by an adjacency matrix, where the matrix element A' m,n and B' m,n Respectively represent domain D m and D n The link properties between them.

[0151]

[0152] Among them, A' m,n Indicates domain D m and domain D n The link delay between m,n Indicates domain D m and domain D n The available bandwidth of the link between m,n Indicates domain D m and domain Dn The unit bandwidth cost of the link between them.

[0153] S represents the set of all SFC requests, V r is the set of VNFs in the rth SFC request; each SFC request r is represented by a tuple, r = (V r ,Φ r ,B r ,D r ,C budget ), V r represents the VNF set, Φ r,i Represents VNF i The computing resource requirements of B r,i For VNF i To VNF i+1 The bandwidth requirement, D r,i For VNF i To VNF i+1 Cbudget represents the maximum allowed delay, and Cbudget represents the fee paid by the user for deploying SFC.

[0154] In another embodiment provided in the present application, the performance of the proposed algorithm is evaluated, and the experimental results are analyzed and discussed.

[0155] Network topology settings: This embodiment selects four network topologies, namely Aarnet, Abilene, Chinanet and Interoute, assuming that there is only one directly connected link between the two network domains. The computing resource capacity of the nodes in each domain is randomly distributed between 600 and 800 units, and the link bandwidth capacity is randomly distributed between 800 and 1200 units. The unit computing resource cost follows the uniform distribution U(0.1,0.8), and the unit bandwidth cost also follows the uniform distribution, ranging from U(1,3). The delay of the intra-domain link is fixed between 1 and 6 milliseconds, and the delay of the inter-domain link is higher, ranging from 5 to 20 milliseconds.

[0156] SFC settings: Each SFC consists of 5 to 10 VNFs, and its arrival process follows a Poisson process with parameter λ=5. The computing resource requirements of each VNF are randomly selected from the uniform distribution U(5,10), and the bandwidth resource requirements are also from the uniform distribution U(10,20). In order to more accurately reflect real-world applications, the delay requirements between VNFs are set to the uniform distribution U(1,10), taking into account both services with high real-time requirements and applications that are not very sensitive to delay. The survival time of each SFC request (i.e., the time from creation to termination) follows an exponential distribution, simulating the unpredictable service duration in the actual environment.

[0157] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A cross-domain service function chain deployment method, characterized in that: The following steps are involved: S1: Divide the service function chain into at least one service function chain segment according to the delay requirement of the virtual network function in the service function chain; S2: Evaluate the reliability of physical nodes in the domain and calculate the overall reliability of the domain based on the reliability of physical nodes; S3: Based on the segmentation results and reliability evaluation results, select at least one domain for deploying the service function chain; S4: In the selected domain, use deep learning to allocate resources and deploy virtual network functions in the service function chain.

2. A cross-domain service function chain deployment method according to claim 1, characterized in that: In step S1, dividing the service function chain into at least one service function chain segment includes: Calculate the average delay of all inter-domain links; comparing the latency requirements between adjacent virtual network functions in the service function chain with the average latency of the inter-domain links; If the delay requirement is greater than the average delay of the inter-domain link, segmenting between the adjacent virtual network functions; The location information of all segmentation points is recorded to generate a segmentation topology map.

3. A cross-domain service function chain deployment method according to claim 2, characterized in that: The calculation of the average delay of all inter-domain links is specifically as follows: Among them, D m and D n Represented as the mth and nth network domains in a multi-domain network; D m,n Indicated as D m and D n The physical link delay between them; E represents the set of physical links.

4. A cross-domain service function chain deployment method according to claim 2, characterized in that: In step S2, the reliability of the physical nodes in the evaluation domain includes: The objective function is used to maximize the reliability of physical nodes, specifically: Among them, λ is an integer used to balance the weight between cost and reliability. represents the comprehensive reliability of the physical nodes selected for deploying virtual network functions; δ is a weight parameter used to balance the importance of the time dimension and the space dimension. for Reliability in the time dimension, for Reliability of spatial dimensions, Indicates whether the i-th virtual network function in r is mapped to a physical node in domain Dm Indicates whether the kth virtual link in the rth request is mapped to an intra-domain link Indicates whether the kth virtual link in the rth request is mapped to the inter-domain link e m,n ; The objective function includes the following constraints: Each virtual network function must and can only be mapped to one physical node; The computing resources of each node will not exceed its capacity limit: Each virtual link must and can only be mapped to one physical link; There are bandwidth requirement constraints on both intra-domain links and inter-domain links; For any virtual link, the sum of the actual delays of the mapped physical links cannot exceed the maximum allowed delay between the virtual network functions at both ends of the virtual link; It has load balancing constraints and cost budget constraints.

5. A cross-domain service function chain deployment method according to claim 4, characterized in that: In step S2, obtaining the overall reliability of the domain based on the reliability of the physical nodes includes: The overall reliability of a domain includes the inherent reliability of physical nodes and the remaining resources of the nodes; The remaining resources of the node include domain D m The remaining available resource ratio of the i-th physical node in , specifically including the remaining computing resource ratio and the remaining bandwidth resource ratio; According to the remaining available resources of the node and the reliability value of the node, the weighted average method is used to define the domain D m reliability.

6. A cross-domain service function chain deployment method according to claim 5, characterized in that: In step S3, the selecting at least one domain for deploying the service function chain based on the segmentation result and the reliability evaluation result includes: Construct a directed weighted graph, where each domain is a node, the overall reliability of the domain is the node weight, and the cost of the inter-domain link is the edge weight; In the directed weighted graph, a path with the lowest cost and the highest reliability is found, and the nodes in the path are the domains used to deploy the service function chain.

7. A cross-domain service function chain deployment method according to claim 6, characterized in that: The node weight is normalized by the inverse of the domain reliability value; the edge weight is a linear combination of the link deployment cost and the transmission delay.

8. A cross-domain service function chain deployment method according to claim 7, characterized in that: In step S4, the use of deep learning to allocate resources for virtual network functions in the service function chain and deploy them includes: Construct a dual deep Q network DDQN model, where the state space includes information about physical nodes in the domain and information about SFC requests, and the action space includes deployment locations and paths; The DDQN model is guided to learn the optimal deployment strategy through a reward function that takes into account resource utilization, deployment cost, and link reliability factors. According to the trained DDQN model, the optimal deployment location and path are selected for the virtual network functions within the service function chain segment.

9. A cross-domain service function chain deployment method according to claim 8, characterized in that: The state space of the DDQN model is specifically: S={V,Φ,B,D,N m ,CR m ,R m ,CC m ,B m ,BC m ,a} Where V represents the set of virtual network functions, Φ and B represent the set of computing and bandwidth resource requirements of virtual network functions, D represents the set of latency requirements, and N m Represents the set of available physical nodes, CR m Represents the reliability value of the node, R m and B m represents the available computing and bandwidth resources, CC m and B.C. m represents the unit computing and bandwidth resource cost; a=[a1,a2,...,a L ] is an allocation vector, indicating the domain number to which the virtual network function in the service function chain request is allocated, and L is the preset maximum number of virtual network functions; The action space of the DDQN model is specifically: in, Indicates whether the i-th virtual network function in r is mapped to Indicates whether the kth virtual link in the rth request is mapped to an intra-domain link The reward function is specifically: Among them, w R 、w C 、w D are the weight factors of reliability, cost and delay, respectively, and C cost is the deployment cost of nodes and links.

10. A cross-domain service function chain deployment system, applying the cross-domain service function chain deployment method according to claim 1, characterized in that: including a hierarchical control architecture, the hierarchical control architecture including a multi-domain controller and a plurality of intra-domain controllers; The steps S1 and S3 are executed by a multi-domain controller, and the steps S2 and S4 are executed by a controller within a domain.

Images