Edge computing network service function chain embedding method based on K shortest circuit

Abstract

The invention discloses an edge computing network service function chain embedding method based on a K shortest circuit, and designs a service function embedding algorithm based on a K shortest circuit algorithm aiming at a time delay constrained distributed service function chain embedding problem under the condition of meeting end-to-end time delay constraints and aiming at minimizing deployment overhead. The method comprises the following steps: firstly, providing a service function embedding problem modeling method based on a weighted hierarchical graph, and converting a service function chain embedding problem into a delay constrained minimum overhead path problem in the weighted hierarchical graph; secondly, designing a time delay constrained service function chain embedding mechanism based on the weighted hierarchical graph and a K-linear metric aggregation algorithm; compared with the prior art, the method has better high performance and operability, and has practical significance for complex network application in the edge computing network.

Description

Technical Field

[0001] This invention relates to a method for embedding edge computing network service function chains based on K-shortest paths, belonging to the field of communication technology. Background Technology

[0002] Network slicing technology, based on mobile edge computing and network function virtualization, models compute-intensive applications (such as connected vehicles, intelligent warfare, and security protection chains) as service function chains (SFCs) composed of ordered virtual network functions (VNFs). By rapidly instantiating VNFs on edge service nodes close to the terminal and orderly scheduling data flows between instances, it can effectively resolve the contradiction between limited terminal resources and high service quality requirements faced by complex network applications when deployed in mobile networks, thus realizing Infrastructure as a Service.

[0003] Rapidly and efficiently deploying Service Function Chains (SFCs) requests in the physical network is a key research area for application-driven MEC development. Efficient deployment of service function chains in MEC faces many challenges. First, for each SFC request, not only data flow routing but also the deployment location of each VNF in the service chain must be decided. Second, due to the randomness and dynamism of service request arrivals, rapid response to service requests is crucial for improving system throughput. Third, service orchestration decisions involve constraints on various resources (e.g., communication, computing, and storage resources) and quality of service (e.g., communication rate, latency, and reliability). Given the QoS requirements of a request and physical network resource constraints, rapidly and efficiently deploying the request is an NP-hard and complex decision-making process. Existing research addresses this problem. Some studies only consider resource constraints while relaxing QoS constraints, while others only consider QoS constraints and simplify resource constraints, designing heuristic algorithms for approximate optimal solutions by modeling the problem as a mixed-integer linear programming model. Other studies have explored empty sets... Considering the complexity of the decision-making factors and the difficulty of model construction, we adopted the widely researched deep reinforcement learning method to solve the problem. Most of these methods only address parts of the three core issues of SFC orchestration and do not provide a fast and efficient SFC embedding mechanism. Summary of the Invention

[0004] Purpose of the invention: To address the problems and shortcomings of existing technologies, this invention provides an edge computing network service function chain embedding method based on K-shortest path. The main differences from existing research are: firstly, it adopts a more realistic scenario assumption, namely, that the requested VNF can be deployed in a distributed manner on different service nodes; secondly, it comprehensively considers service latency requirements, resource overhead, and network resource state constraints; and finally, it aims to accelerate request decision-making speed and improve service throughput and service quality as the overall objective.

[0005] Technical solution: A method for embedding edge computing network service function chains based on K-shortest paths, including the following: 1) To model the segmented routing features embedded in SFC, a weighted network hierarchical graph modeling method is proposed. By mapping and reverse mapping from physical network topology to hierarchical network topology, the originally complex segmented routing problem can be simplified into a more general shortest path problem.

[0006] 2) Based on the weighted network layered graph, with the goal of minimizing resource overhead under request delay constraints, the SFC embedding problem is modeled as a delay-constrained shortest path problem using the linear aggregation metric method. In order to solve the delay-constrained shortest path problem, an SFC embedding mechanism based on Lagrange relaxation and K-shortest path algorithm is designed.

[0007] Given an SDN network G = ( V , E ),in V It is a switching node that supports software-defined functions. E Represents the physical links between nodes. Set of physical nodes. V There exists a set of service function nodes that can support VNF ​​deployment. V VNF VNF at nodes v ∈ V VNF ⊆ V Deploying in the middle will bring resource overhead and processing latency. C v This represents the total computing resources of a node. Data is on the link. e ∈ E Transmitting data online will incur bandwidth overhead and transmission latency. B e This represents the total bandwidth of the link. Let each node... v ∈ V VNF There are sufficient resources to deploy VNFs. The SDN controller is able to sense the network resource status and deploy SFC requests. Any SFC request in the network. r It can be described as a 5-tuple:r = ( s , t , F , B , D ).in B Indicates the requested bandwidth. F = [ f 1, f 2, …, f N ] represents a sequence of service functions, where N = | F | indicates the length of the service chain. D It is an SFC request. r Maximum end-to-end latency requirement. Service request. r Data from the source node s The process needs to be followed in sequence. F The VNF in the middle reaches the destination node t .

[0008] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the edge computing network service function chain embedding method based on K-shortest path as described above.

[0009] A computer-readable storage medium having a computer program / instructions thereon that, when executed by a processor, implements the steps of the K-shortest path-based edge computing network service function chain embedding method as described above.

[0010] Beneficial Effects: Compared to existing technologies, this invention addresses the SFC embedding problem with time-delay constraints, aiming to minimize resource overhead. It models routing using a weighted hierarchical network graph, and then models SFC embedding as a shortest path problem using a linear aggregation metric. Finally, it solves for the optimal deployment scheme using Lagrange relaxation and the K-shortest path algorithm. Compared to existing SFC embedding schemes that only consider resource constraints, this invention is more consistent with real-world scenarios and provides an efficient method for solving deployment schemes. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the SFC segmented routing model based on a weighted hierarchical graph; Figure 2 This is a comparison chart of the time cost of calculating the approximate optimal path; Figure 3 This is a comparison chart of the time cost of calculating the approximate optimal path; Figure 4 This is a comparison chart of online algorithm SFC request admission under different cost functions. Figure 5 This describes the impact of different β values ​​on the number of requests received by the algorithm ONLINE. Figure 6 This describes the impact of different σ values ​​on the number of requests received by the ONLINE algorithm. Detailed Implementation

[0012] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. After reading the present invention, any modifications of the present invention in various equivalent forms by those skilled in the art will fall within the scope defined by the appended claims.

[0013] A method for embedding edge computing network service function chains based on K-shortest paths includes the following: 1) The segmented routing features embedded in SFC are modeled based on the weighted network hierarchical graph modeling method. By mapping and reverse mapping from physical network topology to hierarchical network topology, the originally complex segmented routing problem can be simplified into a more common shortest path problem.

[0014] 2) Based on the weighted network layered graph, with the goal of minimizing resource overhead under request delay constraints, the SFC embedding problem is modeled as a delay-constrained shortest path problem using the linear aggregation metric method. The SFC embedding mechanism based on Lagrange relaxation and K-shortest path algorithm is used to find the delay-constrained shortest path.

[0015] Given an SDN network G = ( V , E ),in V It is a switching node that supports software-defined functions. E Represents the physical links between nodes. Set of physical nodes. V There exists a set of service function nodes that can support VNF ​​deployment. V VNF VNF at nodes v ∈ V VNF V Deploying in the middle will bring resource overhead and processing latency. C v This represents the total computing resources of a node. Data is on the link. e ∈ E Transmitting data online will incur bandwidth overhead and transmission latency. B e This represents the total bandwidth of the link. Let each node... v ∈ V VNFThere are sufficient resources to deploy VNFs. The SDN controller is able to sense the network resource status and deploy SFC requests. Any SFC request in the network. r It can be described as a 5-tuple: r = ( s , t , F , B , D ).in B Indicates the requested bandwidth. F = [ f 1, f 2, …, f N ] represents a sequence of service functions, where N = | F | indicates the length of the service chain. D It is an SFC request. r Maximum end-to-end latency requirement. Service request. r Data from the source node s The process needs to be followed in sequence. F The VNF in the middle reaches the destination node t .

[0016] The routing method where the data stream of an SFC request passes sequentially through the VNF deployment nodes is called segmented routing. r The segmented routing is modeled by introducing a hierarchical graph model. G L Given a request r = ( s , t , F , B , D and physical network topology G ( V , E ), Express support f i The hierarchical graph of the deployed node set is constructed as follows: Step 1: Create G of N One copy G 0, G 1,…, G N , N = | F For any copy G i = ( V i , E i ),nodev ∈ V exist G i The mapping is defined as v i ,link e ∈ E exist G i The mapping is defined as e i .

[0017] Step 2: For any node , i >0, it is in G i The mapping in is v i ,exist G i-1 The mapping in is v i-1 Then in G L Add directed edges v i-1 , v i >. Definition = {< v i-1 , v i >, v ∈ , i ≥1}.

[0018] By performing the above two steps, we can... G Generate a hierarchical graph G L = ( V L , E L ), in V L = , E L = .

[0019] Each VNF requested by an SFC request requires computational and storage resources to deploy on the physical nodes of the network, and traffic transmission requires link bandwidth resources. To minimize the impact of arbitrary SFC requests... r = ( s , t , F, B , D The deployment overhead of VNFs needs to be considered when designing an SFC deployment scheme, taking into account the computational resource overhead of deploying VNFs on the service node and the SFC grouping overhead from the source node. s to the destination node d The bandwidth resources consumed during routing are constrained. If the request... r It also has end-to-end latency requirements, and the data processing latency at each VNF deployment node and the transmission latency on each link need to be considered. For service function sequences... F The first in i Virtual network function f i It should be deployed to a set of nodes that support this type of VNF deployment. a certain node v In this context, resource overhead is defined as... c v This indicates that the program is running on this node. f i The required resource overhead. For any node. Its operation f i Operating overhead c v As shown in equation (1), where In v ( f i ) indicates at node v Create and run f i The overhead required for virtual machines.

[0020] (1) (2) This indicates that a virtual network function is running on node v. The required resource costs This indicates that the virtual network function f is running on node v. i Required data processing latency, node data processing latency d v The definition of is similar to that in equation (1), and its definition is shown in equation (2). Wherein, Indicates at node v Create on f iThe time required by the virtual machine. Queuing latency occurs when the service node processes SFC requests. Since it is assumed that each incoming request must be responded to immediately with a rejection or acceptance, and that each service node has sufficient computing and storage resources to meet the deployment requirements of the requests, the queuing latency of the nodes is negligible compared to the data processing latency.

[0021] For each link e ∈ G The cost of running an SFC request c e With request r The bandwidth consumed on the link follows a linear relationship, as shown in equation (3). ρ e It is a constant representing the overhead per unit bandwidth of transmission, and the delay per unit data transmission of the link. d e It can be represented by a constant.

[0022] c e = ρ e • B , (3) Given an SFC request r Hierarchical graph model G L = ( V L , E L For any link e ∈ E L If the relevant edge weights are defined w ( e ), then it is called for r The weighted hierarchical graph determines the weights based on network link latency and bandwidth. w ( e ).

[0023] Figure 1 The diagram shows a service function chain for a request. F = { f 1, f 2 Example of a weighted hierarchical graph. Its starting point is... A The destination node is J Layer 0 represents the original topology, while layers 1 and 2 represent its copies. f 1 can be deployed in { C , O ,H , I On the node, f 2 can be deployed in { C , M , G Therefore, there exists a directed edge on the node: C 0 → C 1, O 0 → O 1 , H 0 → H 1, I 0 → I 1, C 1→ C 2, M 1 → M 2 , G 1 → G 2 The weights on the edges represent data such as latency and resource overhead. In this hierarchical graph, an edge satisfying { f 1, f 2 The routing scheme for sequence requirements can be described as follows: A 0 → O 0 → O 1 → E 1 → M 1 → M 2 → I 2 → J 2 (Orange lines in the image).

[0024] Based on the graph G L The operation overhead and latency definitions for nodes and edges defined above can be applied to requests respectively. r Weighted hierarchical graph The edge weights are set as follows: for any edge e ∈ G , its in G i The mapping link on is e i Its cost weight is defined as Its delay weight is defined as For any node v ∈ V , its in G i The i Mapping nodes at level -1 v i-1 and the i Layer mapping v i For directed edges < v i-1 , v i > Its cost weight is defined as Its delay weight is defined as Based on the weight definition method and the hierarchical graph construction method, for any SFC request... r The algorithm for constructing its weighted hierarchical graph is shown in Table 1: Table 1 SFC Request r Weighted hierarchical graph construction algorithm

[0025] Regarding the request r and its weighted hierarchical diagram =( V L , E L , c e , d e ), any edge e ∈ E L .set up P ( s 0, t N )express All from the source point s 0 to destination node t N The set of paths express P ( s 0, t N Satisfying the delay in ) D The constrained path, from the source point s 0 to destination node t N Minimum cost path with delay constraints DCLC The definition is shown in equation (4): (4) This is an NP-complete problem, and there is no polynomial-time solution algorithm. p represents One of the paths in the middle.

[0026] Based on linear aggregation metric ( LAM The method is to solve DCLC An efficient heuristic method for the problem. This algorithm transforms the original problem into a single-weighted shortest path problem by aggregating multiple edge weights into a single weight. For any edge... e ∈ E L Its aggregate weight metric is defined as shown in equation (5): w e ( λ ) = c e + λ • d e , λ ≥ 0 (5) For ease of description, for any path from the source node to the destination node... p ,definition D ( p The total path delay is represented by (6), as shown in equation (6): (6) C ( p The total cost of the path is represented by the cost weight. (7) W λ ( p ) indicates the aggregate weight w e ( λ ) next path p Total weight: W λ ( p ) = (8) For a given hyperparameter λ (Used to adjust the weight of latency in the overall weighting), based on the defined aggregate weight. w e ( λ Find the shortest path If the shortest path Total latency Then by increasing λ To improve latency w e ( λ The proportion of ) in. Pλ Indicates in the given λ The following w e ( λ The edge weights are calculated to obtain the first... k The set of shortest paths P λ ={ p 1, p 2,…, p k}, W λ ( p 1) ≤ W λ ( p 2) ≤ … ≤ W λ ( p k ) ≤ W ( p ), p P λ ,in W λ ( p i ) = .set up It is a set of paths that satisfy the time delay constraint. = { p ∈ P λ : D ( p ) ≤ D} W ( p ) represents the weight of any path other than the first K shortest paths.

[0027] For any path from the source node to the destination node p ,definition To find the shortest path that satisfies the time delay constraint p The objective function.

[0028] Use the objective function to construct a new boundary function Boundary ( λ As shown in equation (9): (9) in, (10) (11) For a given λ ,Depend onBoundary ( λ The selected path is defined as follows: p λ Based on time delay constraints D and D ( p ), p ∈ P λ Relationship, set P λ They can be divided into three types: Type I is arbitrary. p ∈ P λ The delay constraint is not met. ,at this time P λ A set is called p c -type; Type II is arbitrary. p ∈ P λ All satisfy the time delay constraint, at this time ,at this time P λ A set is called p d -type; In the case of Class III, P λ The intermediate path includes two types: Type I and Type II. P λ A set is called p mixed -type.

[0029] For any λ , Boundary ( λ Given a lower bound for the DCLC problem, solve... Boundary ( λ The maximum value of ) can solve the optimal path for the DCLC problem. Define the optimal path that makes equation (12) hold. Value ,Right now: (12) based on The obtained path is the approximate optimal path for the DCLC problem. Based on this idea, the algorithm shown in Table 2 was designed. For randomly arriving SFC requests... r In its corresponding weighted hierarchical diagram The minimum cost path with time constraints is found in the middle.

[0030] Table 2 shows the shortest path with time constraints in the weighted hierarchical graph.

[0031]

[0032] The main purpose of Algorithm 2 is to handle SFC requests. r The corresponding weighted layered graph Searching for from s mapping nodes s 0 to t mapping nodes t N The algorithm finds the path that satisfies the delay constraint and has the shortest path cost. Lines 1-7 of the algorithm first set the linear aggregation metric weights according to equation (5). w e ( λ = 0) = c e + 0∙ d e That is, setting λ =0, only the cost weight of the edge is considered (row 1). Considering only the cost weight of the edge, the shortest path algorithm Dijkstra is first used to calculate the weight of the edge from... s 0 to t N path (Line 2) If the latency of this path meets the requirements, then this path is the optimal path, and the function is returned. (Line 3). Otherwise, continue under this weight, using the calculation prior to... k Algorithm for the shortest path k SP calculation path set P 0, and utilize the functions defined by equations (9), (10), and (11). Boundary ( λ Function selection path p 0, if the set of paths is p mixed -type, otherwise return the path. p 0 (lines 4 to 7). Lines 9 to 15 of the algorithm set the linear aggregation metric weights according to equation (5). w e ( λ = )= d e That is, setting λ = Ignoring link costs, we only consider edge delay weights (line 9). Under these weights, we use the Dijkstra's shortest path algorithm to calculate the delay from... s 0 to t N path (Line 10) If the delay of the path meets the requirements, then there are no more paths in the system that meet the conditions, and an empty set is returned. (Line 11). Otherwise, continue under this weight, using the calculation prior to... k Algorithm for the shortest path k SP calculation path set And using the functions defined by equations (9), (10), and (11) Boundary ( λ Function selection path If the path set yes p mixed -type will return the path (Lines 12 to 15). Lines 17 to 24 calculate using an iterative method. λ (Line 18), and set the linear aggregation metric weights based on that value. w e And under this weight, utilize Dijkstra's and k SP algorithm calculates the shortest path p λ and before k A set of paths P λ (Lines 19-20), once P λ yes p mixed If the path is a set of paths of type -type, then the loop will be broken and the current optimal path will be returned. p λ (Line 21). Otherwise if P λ yes p c -type, then update p 0← p λ (Line 22); if P λ yes p d -type, then update ← p λ (Line 23), repeat the above process. The time complexity of this algorithm mainly comes from utilizing... k The SP algorithm iteratively calculates and solves for the optimal solution. λ The worst-case time complexity of Algorithm 2 designed in this invention is... O ( m log 3 m ∙( m + n logn + k log k )),in m = | E L |, n = | V L |

[0033] Based on Algorithms 1 and 2, an SFC admission and VNF embedding algorithm was designed, as shown in Table 3. Table 3 shows the SFC request admission and embedding algorithm.

[0034]

[0035] algorithm SFC_Embedding_Algorithm ( G , r Based on Algorithm 1 and Algorithm 2, implement the handling of randomly arriving service requests. r Make rapid orchestration decisions. For arriving requests... r The algorithm first constructs its weighted hierarchical graph quickly based on Algorithm 1. At the same time, return the request r Source node s and destination node t Mapping nodes in the hierarchical graph: , s 0, t N ← Origin _ Graph _ to _ Weighted _ Layered _ Graph ( G , F , s , t , c e , d e , c v , d v (Line 1). In In the middle, the approximate optimal path with time delay constraints is calculated based on Algorithm 2: p ( s 0, t N ) ← Finding _ Delay _ Constrained _ Shortest _ path ( , s 0, t N , D (line 2). If the returned path p ( s 0, t N Not for , path p ( s 0, t N ) converted to The corresponding node sequence S = [ s 0, v 1, v 2,…, v K , t N At the same time, initialize a new node sequence. S’ ← S Create a sequence of physical nodes for SFC functionality. V VNF ← [ , ,…, (Lines 3-5). The main purpose of lines 6-17 is to backmap the paths calculated in the weighted hierarchical graph back to the original physical network. G In the middle. For links in the original physical network. e Verify separately S’ The first jump in the middle s 0, v 1>(line 8), the last jump< v K , t N >(line 11), and any jump< v i , v i+1 >,1< i < K Is it e exist If the mapping in the text (lines 13-15) is correct, replace it with a link. e At the same time, increment the counter for that edge by 1. The main purpose of lines 18-25 is to complete the VNF sequence. F Mapping to physical nodes. Based on the generation characteristics of weighted hierarchical graphs, different subgraphs... G i-1 and G i Only exist in space Gi-1 arrive G i The directed edges can be traversed in order. S ,like S Nodes in v i and v i+1 yes G Middle node v In subgraph G i-1 and G i If the corresponding mapping in the mapping is found, then the node must be a deployment node of a certain VNF. This process is repeated sequentially. V VNF The nodes in the sequence are replaced to obtain the final result. F The actual physical location of each VNF in the graph. Lines 28-32 of the algorithm are mainly responsible for verifying the graph. G any edge in e Does the remaining bandwidth meet the deployment requirements? e . count * B > B e This indicates that the current link can no longer meet the conditions, therefore the request is rejected. r Access is allowed, otherwise it is permitted. r The algorithm's operation mainly involves finding the shortest path with latency constraints in the weighted hierarchical graph; therefore, the time complexity of Algorithm 3 is also... O ( m log 3 m ∙( m + n log n + k log k )),in m = | E L |, n = | V L |

[0036] In a dynamic orchestration environment, SFC requests always arrive one after another in sequence, i.e., slots. T Within, SFC requests arrive in order. r 0, r 1, r 2, …, r k , r k+1Therefore, a dynamic online SFC orchestration mechanism needs to be designed to immediately decide whether to allow or deny each newly arriving SFC request based on the current network resource situation. The ultimate goal of online SFC orchestration is to orchestrate as many SFC requests as possible within a given time slot. To meet this dynamic requirement and objective, the dynamic nature of network nodes and links must be considered. Based on the designed SFC embedding mechanism based on a weighted hierarchical graph, an online embedding mechanism with a controllable contention ratio can be achieved by modifying the weight parameters of edges and nodes.

[0037] All link resources are dynamically changing. To control the online algorithm contention ratio, the weights of each node and link must be assigned considering both their current utilization and the dynamic nature of resource changes. That is, for the k-th SFC request, the resource status faced by the current k-th SFC request needs to be determined after considering the resource consumption of the previous k-1 requests. For the SFC request sequence [ r 0, r 1, r 2, …, r k , r k+1 The first of ,…] k One request r k When constructing its weighted hierarchical graph, for any edge e ∈ G The cost weight is modified as shown in equation (13).

[0038] (13) in, Indicates hyperparameters, β >0, B e ( k ) = B e ( k - 1) - B k , indicates the first k One request r k Upon arrival, the link e The remaining bandwidth capacity, B k Indicates a request r k The communication rate. B e (0) = B e This represents the total bandwidth capacity of the link. Since it is assumed that the nodes have sufficient resources to meet deployment requirements, node overhead remains constant.

[0039] To ensure a high competition ratio between online and offline algorithms, the following strategy was adopted for requests: For any request... r If the total cost of the routing scheme generated by Algorithm 3 is less than a given threshold σ, then the request is allowed to access the network; otherwise, it is rejected. This admission mechanism ensures that the contention ratio of the online algorithm is within a certain range. β = 2| V |At worst-case scenario O (log| V |).

[0040] Performance verification The performance of the method of this invention was evaluated using numerical simulation, and the impact of key parameters on the algorithm performance was investigated. The host computer was configured with a 4GHz Intel i7 Quad-core CPU and 64GB RAM. All algorithms were implemented in Python, and the open-source graph algorithm library Networkx was used. The topology generation tool GT-ITM was used to construct the experimental topology.

[0041] (1) Comparison of static single SFC orchestration performance First, the orchestration performance of different methods for the same SFC request was compared under the same network parameters. The verification process compared the proposed method with a method based on Lagrange Relaxation based Aggregated Cost (LARAC) for solving the DCLC problem. In the LARAC-based orchestration scheme, when solving the DCLC dual problem, the boundary function... Boundary ( λ The solution to this problem only considers the shortest path, resulting in a limited set of path candidates. For simplicity, the mechanism designed in this invention is referred to as... K_UNICAST_DELAY The method for solving DCLC problems based on the Lagrange relaxation method is described as follows: UNICAST_DEALY To compare the performance of the methods, two comparison metrics were defined: (a) Path cost ratio PCR The main purpose of the designed method is to calculate the minimum cost path under time delay constraints. Therefore, by comparing the path costs calculated by different methods under the same conditions, the differences in the methods in solving the optimal solution can be reflected. Thus, the PCR index is defined, as shown in Equation (14). The smaller the index, the better the performance of the mechanism involved in this method.

[0042] (14) cost(pK_UNICAST_DELAY () represents the total cost of the path solved by the algorithm designed in this invention; cost ( p UNICAST_DELAY ) represents the total cost of the path solved by the comparison algorithm.

[0043] (b) Average runtime delay ART This indicator reflects the computation time spent by the method in finding the optimal solution. The shorter the time, the better the method is.

[0044] In this experiment, the network parameters were set as follows: The topology generation tool GT-ITM was used to generate five Mesh and Waxman network topologies with network sizes of [50, 100, 150, 200, 250] nodes. For each network topology, 30% of the nodes were randomly selected as service deployment nodes, and each node could deploy one of eight random VNF types. Node and link costs were normalized and set to a uniform distribution conforming to [5, 10], link latency was set to a uniform distribution of [2ms, 10ms], and link bandwidth was uniformly set to 500Mbps.

[0045] For SFC requests r The settings are as follows: Request sequence F The length is set to 8, including all types of VNFs, and the starting node is randomly selected in the topology. s and target node t bandwidth requirements B Set to 100Mbps, latency constraint D The settings are defined in the following way: Generate request r After calculating the weighted hierarchical graph corresponding to the physical topology, the path with the minimum cost is calculated for each. p LC and shortest path with delay p LD , r The time delay constraint is shown in formula (15).

[0046] (15) d ( p LC ) represents the path with the least cost. p LC The time delay, d ( p LD () represents the shortest path with the shortest delay. p LD The time delay; for K_ UNICAST_DELAY The method is set to select the top 50 shortest paths as candidate paths, i.e.k = 50. Table 4 and Figure 2 The mechanism designed in this invention was implemented under different network topologies. K_UNICAST_DELAY Comparison mechanism UNICAST_DEALY In comparison indicators PCR and ART The comparison was performed as follows: The same algorithm was repeated 100 times on each topology, and the average value of the evaluation metrics was used as the comparison value.

[0047] Table 4 Comparison of Approximate Optimal Path Costs

[0048] Table 4 shows the utilization of different numbers of physical nodes. K_UNICAST_DELAY Mechanism and UNICAST_DEALY The comparison of shortest path costs with time constraints calculated by the mechanism is shown in Table 4. As can be seen from Table 4, the PCR ratio is generally less than 100% in different network topologies, indicating that... K_UNICAST_DELAY The resulting DCLC path is better because UNICAST_ DEALY The mechanism is equivalent to k = 1 K_UNICAST_DELAY The algorithm has a smaller set of possible paths, therefore the calculated path cost must be less than or equal to... K_UNICAST_DELAY The path. Meanwhile, as the number of network nodes increases, the PCR shows a decreasing trend, because as the number of nodes increases, the node sizes of the generated hierarchical graphs are [50*8, 100*8, 150*8, 200*8, 250*8]. In this case... K_UNICAST_DELAY The method includes the previous K The set of shortest paths is more likely to contain the optimal path, while UNICAST_DEALY The algorithm used obtains the optimal value of the Lagrange duality, while the algorithm obtained under stricter time delay constraints results in a greater difference in the optimal path cost between the two methods.

[0049] Figure 2 This invention demonstrates the SFC orchestration method designed under different network topology types and different numbers of network nodes. K_ UNICAST_DELAY Comparison Method UNICAST_DEALY In solving DCLC The computation time spent on the path. From Figure 2It can be seen that as the network size increases, both methods exhibit a non-linear increase in the time cost of calculating the optimal path. This is because the increase in node size is accompanied by an increase in the number of links. Furthermore, the construction of the network layering graph also increases the number of links between layered networks. According to the time complexity analysis of Algorithm 3, the algorithm's computation time has an exponential relationship with the number of network nodes and links; therefore, the increasing latency trend is consistent with reality. Comparing Mesh and Waxman networks, the computation time of both methods is longer in Mesh networks at the same network size. This is because Mesh is a fully connected network; with the same number of network nodes, the number of links between nodes is greater, thus increasing the computation time for solving the problem. K The shortest path algorithm and Dijkstra's algorithm have larger path search spaces and require more iterations, resulting in higher computational costs and longer processing times.

[0050] In the case of a medium-sized network, it can be seen K_UNICAST_DELAY Comparison UNICAST_DEALY The difference in computational latency is small, therefore it is suitable for practical applications. K_UNICAST_DELAY The increased latency overhead is entirely acceptable compared to the benefits gained in calculating the optimal path.

[0051] K_UNICAST_ DELAY Compared to LARAC-based UNICAST _DEALY Its greatest advantage lies in calculating linear aggregation metrics. LAM The front of the bottom K The set of shortest paths increases the selection range for shortest paths under the time delay constraint, thus increasing... K The increased number of [paths] increases the likelihood of choosing a path closer to the optimal solution, but also increases [the chances of failure]. K The method requires more time to calculate the path set. Figure 3 This demonstrates the performance of a mesh network with 200 nodes. K_UNICAST_ DELAY Different settings K The comparison shows the changes in the cost of SFC requesting the optimal path and the time cost of solving the optimal path when the value is set.

[0052] from Figure 3 As can be seen, for a given network topology and SFC request, by increasing K This approach can achieve a better minimum-cost path, but the computational cost of the algorithm also increases accordingly. However, for a specific number of networks... K After the value reaches a certain value, it can be increased. K The value is no longer meaningful for solving the optimal path, therefore for K The selection needs to be configured according to the actual network.

[0053] (2) Performance comparison of online orchestration mechanisms This section primarily tests the contention ratio of the online orchestration mechanism designed in this invention. The experimental topology was constructed using the topology generation tool GT-ITM to generate a Mesh network with network node sizes of [50, 100, 150, 200, 250]. The bandwidth of each link follows a uniform distribution of [100Mbps, 1000Mbps]. The number of service nodes supporting VNF deployment was set to 30% of the network node size, and the number of computing units that each service node could use was set to 50% of the total resources of the service node. The link latency of the network topology was set to a uniform distribution of [2ms, 10ms]. Fifteen VNF types were available for selection in the experiment, and the operational overhead required for each type of VNF was set to a uniform distribution of [5, 10] units of resource.

[0054] For each SFC request r = ( s , t , F , B , D Configure the network as follows: G ( V , E Two nodes are randomly selected from the set of nodes as source nodes. s and destination node t . F The length follows a random distribution of [4, 10], and the type is randomly selected from 15 types of VNFs. Bandwidth requirements. B Set to conform to a uniform distribution within the range of [20Mbps, 100Mbps], latency D It is set to conform to a uniform distribution within the range [40ms, 400ms]. Online orchestration mechanisms using nonlinear link cost functions are marked as ONLINE, while offline orchestration methods using linear cost functions are marked as LINEAR.

[0055] Figure 4 The study compared the number of SFC (Service Controller) access requests between ONLINE and LINEAR under different network scales. K All values ​​are 30. For ONLINE, its parameters are... β Set to 2| V |

[0056] from Figure 4In reality, regardless of the link cost assignment function, the number of SFC requests that can be admitted increases with the increase in network size. This is because as the network size increases, the number of optional paths also increases, and the capacity of SFC requests increases accordingly. However, under the same network size, ONLINE can accommodate more SFC requests. This is because ONLINE considers not only the current network state but also possible future changes when assigning link weights.

[0057] The settings of the ONLINE parameters can affect algorithm performance. The parameters were tested. β And the impact of the threshold σ on the algorithm. By... β The value is from 2| V |Changes to 8| V |To evaluate parameters β The impact on the proposed ONLINE performance. From Figure 5 As can be seen, with β With the increase in [data / data], ONLINE will accept fewer requests. For example, when [data / data]... β = 8 n At that time, the number of accepted requests is no greater than the number of requests accepted at that time. β =2 n 70% of the requests were accepted at that time because β The larger the value, the higher the cost of using overloaded resources, leading to more conservative resource usage.

[0058] Finally, the impact of the admission threshold σ on the performance of the proposed ONLINE with and without the threshold was evaluated to examine the importance of access control. Figure 6 As can be seen, without admission control, the number of SFC requests that ONLINE can accept decreases. Furthermore, the larger the network size, the higher the admission control threshold. σ The greater the impact on the number of accepted requests, the more bandwidth resources are needed to accept them. This is because in large networks, the distance between the source and destination of a request can be very large. Given a threshold, the proposed online algorithm can reject such requests, thereby enabling it to accept future requests and achieve higher network throughput.

[0059] This invention studies the distributed service function chain embedding problem with latency constraints. Under the condition of satisfying end-to-end latency constraints, and with the objective of minimizing deployment overhead, a service function embedding algorithm based on the K-shortest path algorithm is designed. First, this invention proposes a service function embedding problem modeling method based on a weighted hierarchical graph, transforming the service function chain embedding problem into a minimum-cost path problem with latency constraints in a weighted hierarchical graph. Second, based on the weighted hierarchical graph and... KA service function chain embedding mechanism with time-delay constraints was designed using a linear metric aggregation algorithm. Finally, a simulation environment was built to compare the performance of the designed mechanism with that of the comparative algorithm, demonstrating that the designed mechanism has better performance. The mechanism designed in this invention has good operability and practical significance for complex network applications in edge computing networks.

[0060] Obviously, those skilled in the art should understand that the steps of the edge computing network service function chain embedding method based on K-shortest path in the above embodiments of the present invention can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using device-executable program code, thereby storing them in a storage device for execution by the computing device. Furthermore, in some cases, the steps shown or described can be performed in a different order than presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the embodiments of the present invention are not limited to any particular hardware and software combination.

Claims

1. A method for embedding edge computing network service function chains based on K-shortest paths, characterized in that, Includes the following: 1) Modeling the segmented routing features embedded in SFC based on the weighted network hierarchical graph modeling method; First, generate the physical network. G Hierarchical graph model G L = ( V L , E L ); V L Represents the set of nodes in a hierarchical graph. E L Represents the edge set of the hierarchical graph; secondly, for any link... e ∈ E L Define the relevant edge weights w e The hierarchical graph with edge weights is used as an SFC request. r Weighted network hierarchical diagram ; 2) Based on the weighted network layered graph, with the goal of minimizing resource overhead under request delay constraints, the SFC embedding problem is modeled as a delay-constrained shortest path problem using the linear aggregation metric method. The SFC embedding mechanism based on Lagrange relaxation and K-shortest path algorithm is used to find the delay-constrained shortest path.

2. The edge computing network service function chain embedding method based on K-shortest path according to claim 1, characterized in that, In the hierarchical graph model that generates the physical network G, as described in 1), SFC request r The routing method where data flows sequentially through VNF deployment nodes is called segmented routing; in order to handle SFC requests r The segmented routing is modeled by introducing a hierarchical graph model. G L Physical network G = ( V , E ),in V It is a switching node that supports software-defined functions. E Represents the physical link between nodes; a set of physical nodes. V There exists a set of service function nodes that can support VNF ​​deployment. V VNF The SDN controller senses the network resource status and deploys SFC requests; any SFC request in the network r Described as a 5-tuple: r = ( s , t , F , B , D ),in B Indicates the requested bandwidth. F = [ f 1, f 2, …, f N [] indicates a sequence of service functions. N = | F | indicates the length of the service chain. D It is an SFC request. r Maximum end-to-end latency requirement; SFC request r Data from the source node s The process needs to be followed in sequence. F The VNF in the middle reaches the destination node t ; Express support f i Deployed node set, physical network G The method for constructing the hierarchy diagram is as follows: Step 1: Create G of N One copy G 0, G 1,…, G N , N = | F |;For any copy G i = ( V i , E i ),node v ∈ V exist G i The mapping is defined as v i ,link e ∈ E exist G i The mapping is defined as e i ; Step 2: For any node , i > 0, it is in G i The mapping in is v i ,exist G i-1 The mapping in is v i-1 Then in G L Add directed edges v i-1 , v i >; Definition = {< v i-1 , v i >, v ∈ , i ≥1}; Through the operations described in Steps 1-2 above, it is possible to... G Generate a hierarchical graph G L = ( V L , E L ), in V L = , E L = .

3. The edge computing network service function chain embedding method based on K-shortest path according to claim 1, characterized in that, In step 2, in order to minimize arbitrary SFC requests r = ( s , t , F , B , D The deployment overhead of VNFs needs to be considered when designing an SFC deployment scheme, taking into account the computational resource overhead of deploying VNFs on the service node and the SFC grouping overhead from the source node. s to the destination node d The bandwidth resource constraints consumed during routing; if the request r It also has end-to-end latency requirements, and the data processing latency at each VNF deployment node and the transmission latency on each link need to be considered; for service function sequences F The first in i Virtual network function f i It should be deployed to a set of nodes that support VNF ​​deployment. a certain node v In this context, resource overhead is defined as... c v , indicating that at this node v running f i The required resource costs; for any node Its operation f i resource consumption c v As shown in equation (1); (1) (2) This indicates that a virtual network function is running on node v. The required resource costs This indicates that the virtual network function f is running on node v. i Required data processing latency, node data processing latency d v The definition is shown in equation (2); where, Indicates at node v Create on f i The time required for the virtual machine.

4. The edge computing network service function chain embedding method based on K-shortest path according to claim 1, characterized in that, For each link e ∈ G Run SFC request r Cost c e With SFC request r The bandwidth consumed on the link follows a linear relationship, as shown in equation (3); where ρ e It is a constant representing the overhead per unit bandwidth of transmission, defining the unit data transmission delay of the link. d e It is a constant; c e = ρ e ∙ B , (3)。 5. The edge computing network service function chain embedding method based on K-shortest path according to claim 1, characterized in that, For SFC requests r and its weighted hierarchical diagram = ( V L , E L , c e , d e ), any edge e ∈ E L ; set up P ( s 0, t N )express All from the source point s 0 to destination node t N The set of paths express P ( s 0, t N Satisfying the delay in ) D The constrained path, from the source point s 0 to destination node t N Minimize resource overhead path with latency constraints DCLC The definition is shown in equation (4): (4) p represents One of the paths in the middle.

6. The edge computing network service function chain embedding method based on K-shortest path according to claim 5, characterized in that, The method of minimizing resource cost is based on linear aggregation metric to solve the path, which aggregates multiple edge weight metrics into a single weight; for any edge e ∈ E L Its aggregate weight metric is defined as shown in equation (5): w e ( λ ) = c e + λ ∙ d e , λ ≥ 0 (5) For any path from the source node to the destination node p ,definition D ( p The total path delay is represented by (6), as shown in equation (6): (6) C ( p The total cost of the path is represented by the cost weight. (7) W λ ( p ) indicates the aggregate weight w e ( λ ) next path p Total weight: W λ ( p ) = (8) For a given hyperparameter λ Based on the defined aggregation weight w e ( λ Find the shortest path If the shortest path Total latency Then by increasing λ To improve latency w e ( λ The proportion of ) in; P λ Indicates in the given λ The following w e ( λ The edge weights are calculated to obtain the first... k The set of shortest paths P λ = { p 1, p 2,…, p k }, W λ ( p 1) ≤ W λ ( p 2) ≤… ≤ W λ ( p k ) ≤ W ( p ), p ∉ P λ ,in W λ ( p i ) = ;set up It is a set of paths that satisfy the time delay constraint. = { p ∈ P λ : D ( p ) ≤ D }; W ( p Except for p1 to p k The weights of all other paths besides; For any path from the source node to the destination node p ,definition To find the shortest path that satisfies the time delay constraint p The objective function; Use the objective function to construct a new boundary function Boundary ( λ As shown in equation (9): (9) in, (10) (11) For a given λ ,Depend on Boundary ( λ The selected path is defined as follows: p λ Based on time delay constraints D and D ( p ), p ∈ P λ Relationship, set P λ They are divided into three types: Type I is arbitrary. p ∈ P λ The delay constraint is not met. ,at this time P λ A set is called p c -type; Type II is arbitrary. p ∈ P λ All satisfy the time delay constraint, at this time ,at this time P λ A set is called p d -type; In the case of Class III, P λ The intermediate path includes two types: Type I and Type II. P λ A set is called p mixed -type; For any λ , Boundary ( λ Given a lower bound for the DCLC problem, solve... Boundary ( λ The maximum value of ) can solve the optimal path for the DCLC problem. Define the optimal path that makes equation (12) hold. Value ,Right now: (12) based on The obtained path is the approximate optimal path for the DCLC problem.

7. The edge computing network service function chain embedding method based on K-shortest path according to claim 5, characterized in that, Based on the weighted hierarchical graph and the path generated by the K-shortest path algorithm, reverse mapping of links and service functions is performed in the original network topology to determine the links and nodes that the service passes through; and based on the bandwidth consumption of the mapped links and the computing resource consumption of the nodes, it is determined whether the service can access the network. For any request r If the total cost of the path generated by the weighted hierarchical graph and the K-shortest path algorithm is less than the given threshold σ, then the request is allowed to access the network; otherwise, it is rejected.

8. The edge computing network service function chain embedding method based on K-shortest path according to claim 5, characterized in that, To control the online algorithm contention ratio, the weights assigned to each node and link must consider both their current utilization rate and the dynamic changes in resources. For the SFC request sequence... r 0, r 1, r 2, …, r k , r k+1 The first of ,…] k One request r k When constructing its weighted hierarchical graph, for any edge e ∈ G The cost weight is modified as shown in equation (13); (13) in, Indicates hyperparameters, β > 0, B e ( k ) = B e ( k - 1) - B k , indicates the first k One request r k Upon arrival, the link e The remaining bandwidth capacity, B k Indicates a request r k The communication rate; B e (0) = B e This represents the total bandwidth capacity of the link; node overhead remains constant.

9. A computer device, characterized in that: The computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the edge computing network service function chain embedding method based on any one of claims 1-8.

10. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that: When the computer program / instructions are executed by the processor, they implement the steps of the edge computing network service function chain embedding method based on any one of claims 1-8.

Images