A method for predicting performance of an SDN network based on a multivariate graph structure

Abstract

The application provides a kind of SDN network performance prediction method based on multivariate graph structure, it is related to the technical field of network performance prediction, the present SDN performance prediction method lacks time dimension modeling, information is excessively smooth, time sequence feature loss and other problems, six kinds of meta graphs representing the key elements of SDN network are constructed from structure, capacity, congestion, competition, time and other dimensions, the exchange matrix is generated based on the meta graph, and the source node and destination node features are extracted by matrix decomposition;Adopt source attention mechanism and destination attention mechanism to give adaptive weight to the features, input the weighted features and node embedding vector after splicing into multilayer perceptron, realize the accurate prediction of SDN network delay, jitter, packet loss rate and other performance indicators.This method can capture the instantaneous behavior of the network and the competition relationship between flows, retain the time sequence features of data packets, improve the topology generalization ability and prediction accuracy, and meet the real-time performance evaluation and dynamic optimization requirements of SDN network.

Description

Technical Field

[0001] This invention relates to the technical field of network performance prediction, and more particularly to a method for predicting SDN network performance based on a multi-element graph structure. Background Technology

[0002] Current SDN network performance prediction mainly relies on two types of methods: One is white-box methods based on simulators, such as ns-3 and OMNeT++. These methods can provide accurate performance metrics, but they are extremely slow, unsuitable for real-time optimization, and lack generalization ability, especially when topology changes. The second type is black-box methods based on learning, including machine learning regression methods such as RouteNet and RouteNet-Fermi. These methods have better topology generalization ability; the constructed models can be directly applied to unseen topologies, routing policies, and traffic intensity, and have high prediction accuracy. Compared to the first method, the inference speed is faster, and real-time performance evaluation can be performed in conjunction with twin networks. However, this method still has shortcomings: (1) In the message passing process of GNN, although the entire network can be covered after multiple iterations, as the number of iterations increases, the information will be over-smoothed, making it difficult to accurately capture the competitive relationship between streams that are far apart.

[0003] (2) Input features compress traffic into a histogram, which is better than the average, but still loses the precise time sequence of packet arrivals. For highly sensitive micro-burst traffic, this compression may lead to prediction bias.

[0004] Therefore, there is an urgent need for a predictive evaluation method that can comprehensively consider the impacts on SDN network performance. Summary of the Invention

[0005] To address the aforementioned problems in the existing technology, this invention provides a method for predicting the performance of SDN networks based on a multi-element graph structure.

[0006] The objective of this invention can be achieved through the following technical solutions: A method for predicting the performance of SDN networks based on multi-dimensional graph structures includes the following steps: Based on the characteristics of SDN networks, six path structures of SDN network metagraphs are constructed. The SDN network metagraphs are directed graphs, and each SDN network metagraph includes at least one source node with an in-degree of 0 and one destination node with an out-degree of 0. By extracting features from the SDN network metagraph and calculating the exchange matrix of the SDN network metagraph based on the adjacency matrix of any two adjacent nodes in the SDN network metagraph, the decomposed source feature matrix and destination feature matrix are obtained through a matrix decomposition algorithm. By employing source attention mechanism and destination attention mechanism, weights are assigned to the source feature matrix and the destination feature matrix respectively, to obtain weighted source node weighted feature vector and destination node weighted feature vector; The preset source node embedding vector, destination node embedding vector, source node weighted feature vector, and destination node weighted feature vector are concatenated and then input into a pre-trained multilayer perceptron model to obtain the prediction performance index of the SDN network.

[0007] Compared to existing technologies, the SDN network performance prediction method based on a multi-element graph structure of this invention avoids the problem of excessive information smoothing caused by multiple iterations of GNNs by using matrix factorization and attention mechanisms based on the metagraph. It can accurately represent the competitive relationship between long-distance flows and improve the fine granularity of performance prediction. By assigning adaptive weights to different metagraph features through a source / destination attention mechanism, it strengthens the contribution of key features and weakens redundant information, significantly improving the model convergence speed and prediction accuracy. The model used can be directly applied to unseen network topologies, routing strategies, and traffic intensity, with fast inference speed, which can meet the engineering requirements of real-time performance evaluation and dynamic optimization of SDN networks. It can simultaneously output multiple key performance indicators such as latency, jitter, and packet loss rate, covering the core needs of SDN network operation and maintenance and traffic scheduling.

[0008] In some embodiments, the SDN network metagraph is represented as ,in, Represents the nodes in the SDN network metagraph. Represents the edges of the SDN network metagraph. This represents the relationships between nodes in the SDN network metagraph. , , Let V be the set of nodes and E be the set of edges. It is a set of relationships between nodes.

[0009] In some embodiments, the SDN network metagraph with six path structures includes six path structures: M1, M2, M3, M4, M5, and M6. The connection structure of M1 is s→R→R→...→d; the connection structure of M2 is s→R→L. w →R→d;The connection structure of M3 is s→R→L d →R→d; The connection structure of M4 is s→R→Q h →R→d; The connection structure of M5 is s→R←F→R→d; The connection structure of M6 is s→R→T→R→d; Where s represents the source node; d represents the destination node; R represents the router / switch; and L represents the link, where L... w and L dThese represent high-bandwidth links and low-latency links, respectively; F represents a flow; Q represents a queue, where Q... h This represents a high-congestion queue, where T represents a time slot; Where s→R represents the starting connection, R→R represents the topology connection, R→d represents the ending connection, and R→L and R→L represent the topology connection. w and R→L d Indicates switch-link, L→R, L w →R and L d →R represents a link-switch, R→Q and R→Q h Represents queue mappings, Q→R and Q h →R indicates that the output of the queue re-enters the router forwarding process, F→R indicates that the flow passes through the router, R→T indicates the state of the router at time t, and T→R indicates the subsequent forwarding behavior affected by the state at that moment.

[0010] In some embodiments, the switching matrix of the SDN network metagraph includes: ; in, The exchange matrix represents the metagraph of an SDN network. For any two adjacent nodes in the SDN network metagraph under the same path structure and The adjacency matrix, This represents the total number of nodes in the SDN network metagraph.

[0011] In some embodiments, the exchange matrix of each path in the SDN network metagraph is decomposed using the matrix factorization (MF) algorithm to obtain the source feature matrix S and the destination feature matrix D.

[0012] In some embodiments, the weighting of the source attention mechanism and the destination attention mechanism includes: ; ; in, This represents the weighted feature vector of the source node. The weight matrix represents the source feature matrix S in the i-th SDN network metagraph; This represents the weighted feature vector of the target node. The weight matrix represents the target feature matrix D in the i-th SDN network metagraph; This represents the dot product between matrices.

[0013] In some embodiments, after obtaining the weighted feature vectors of the source node and the target node, they are concatenated with the embedding vectors of the source node and the target node, as follows: ; in, Embed vectors for source nodes; Embed a vector for the target node; This indicates vector concatenation.

[0014] In some embodiments, the loss function used to train the multilayer sensing model as follows: ; in, This is a performance prediction value. This represents the actual performance value, where t is time, N is the total number of time windows, and P is the set of node pairs, where the performance prediction value is... The predicted performance metrics include latency, jitter, and packet loss rate. Attached Figure Description

[0015] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0016] Figure 1 This is a flowchart illustrating an SDN network performance prediction method based on a multi-element graph structure according to the present invention. Figure 2 This is a schematic diagram of the six SDN network metagraph structures of this invention; Figure 3 This is a schematic diagram illustrating the construction of the multilayer sensing model of the present invention; Figure 4 This is a schematic diagram of the source and target attention mechanism training network of the present invention. Detailed Implementation

[0017] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided.

[0018] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0019] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0020] Please see Figure 1 The diagram below illustrates a flowchart of an SDN network performance prediction method based on a multi-element graph structure according to the present invention. The method includes the following steps: S1. Construct SDN network metagraphs with six path structures based on the characteristics of SDN networks. The SDN network metagraphs are directed graphs, and each SDN network metagraph includes at least one source node with an in-degree of 0 and one destination node with an out-degree of 0. The SDN network metagraph is a global network view maintained by the SDN controller, constructed between selected source and destination nodes based on the characteristics of SDN networks. It includes information such as nodes, links, and flow tables. By adjusting different paths, such as "low-latency paths" or "high-bandwidth channels," the network controller generates forwarding paths that meet business requirements, realizing application-driven networks. Through centralized control and programmability, the SDN network architecture achieves intelligent, dynamic, and service-aware paths, which is a key foundation for building agile and efficient future networks.

[0021] This invention designs six SDN network metagraphs that embody the characteristics of SDN networks, considering various dimensions such as structure, capacity, congestion, contention, and time. Each of these SDN network metagraphs includes at least one source node with an in-degree of 0 and one destination node with an out-degree of 0. The connection structures of the six network metagraphs respectively include: router / switch, high-bandwidth link, low-latency link, flow, high-congestion queue, and time slot.

[0022] In this embodiment, please refer to Figure 2 The SDN network metagraph with six path structures includes six path structures: M1, M2, M3, M4, M5, and M6. The connection structure of M1 is s→R→R→...→d; the connection structure of M2 is s→R→L. w →R→d;The connection structure of M3 is s→R→L d →R→d; The connection structure of M4 is s→R→Q h →R→d; The connection structure of M5 is s→R←F→R→d; The connection structure of M6 is s→R→T→R→d; Where s represents the source node; d represents the destination node; R represents the router / switch; and L represents the link, where L... w and Ld These represent high-bandwidth links and low-latency links, respectively; F represents a flow; Q represents a queue, where Q... h This represents a high-congestion queue, where T represents a time slot; Where s→R represents the starting connection, R→R represents the topology connection, R→d represents the ending connection, and R→L and R→L represent the topology connection. w and R→L d Indicates switch-link, L→R, L w →R and L d →R represents a link-switch, R→Q and R→Q h Represents queue mappings, Q→R and Q h →R indicates that the output of the queue re-enters the router forwarding process, F→R indicates that the flow passes through the router, R→T indicates the state of the router at time t, and T→R indicates the subsequent forwarding behavior affected by the state at that moment.

[0023] In other embodiments, those skilled in the art can make appropriate adjustments to the SDN network metagraph based on their understanding of the inventive concept of this application.

[0024] S2. By extracting the features of the SDN network metagraph and calculating the exchange matrix of the SDN network metagraph based on the adjacency matrix of any two adjacent nodes in the SDN network metagraph, the source node feature matrix S and the destination node feature matrix D are obtained by matrix decomposition algorithm. Please see Figure 2 In the SDN network metagraph structure, any two adjacent nodes can be s and R, R and d, or R and L. w The exchange matrix of the SDN network metagraph is calculated to decompose and obtain the latent characteristics of the source and destination nodes.

[0025] In this embodiment, by defining Node types in the SDN network metagraph and Given the adjacency matrix, the exchange matrix of each path in the SDN network metagraph is as follows: ; in Represents a commutation matrix. This represents the total number of nodes in the SDN network metagraph.

[0026] For example, for the SDN network metagraph M1, there exists The exchange matrix for each path The feature matrix is ​​obtained by using the matrix factorization (MF) algorithm. and .

[0027] In other embodiments, those skilled in the art, based on their understanding of the inventive concept of this application, can decompose the exchange matrix. It is decomposed into the source feature matrix S and the target feature matrix D.

[0028] S3. Using source attention mechanism and destination attention mechanism, weights are assigned to the source feature matrix and the destination feature matrix respectively to obtain the weighted source node weighted feature vector and the destination node weighted feature vector; The source node embedding vector, the destination node embedding vector, the source node weighted feature vector, and the destination node weighted feature vector are concatenated and then input into a pre-trained multilayer perceptron model to obtain the prediction performance index of the SDN network.

[0029] The source node embedding vector and the destination node embedding vector are used to represent the low-dimensional vector representation of two nodes in the graph (such as the start and end points of an edge). Each node is mapped to a vector in a low-dimensional continuous vector space to preserve its structural or semantic information.

[0030] This invention uses an attention mechanism to assign different weights to various features.

[0031] In this embodiment, by constructing as follows Figure 3 The feature extraction steps for the multilayer perceptron model shown are as follows: For a network with m sources and n destinations, by constructing paths M1-M6, and generating the exchange matrix for each path... Then, k-dimensional features of each source and destination are extracted through matrix decomposition, that is, for any commutation matrix... All of them have: ; By matrix decomposition, the commutative matrix is ​​obtained. It is decomposed into the source feature matrix S and the target feature matrix D.

[0032] We employ source attention and destination attention mechanisms to assign weights to the source node feature matrices and destination node feature matrices obtained from the SDN network metagraph decomposition and then perform a weighted summation, as follows: In this embodiment, source and destination attention mechanisms are used to determine the influence weights of different features, and the network is trained accordingly. Figure 4 As shown. In this process, the source (destination) vector and the feature vectors of each meta-graph are first concatenated. The concatenated vector is then passed through two fully connected layers, two ReLU layers, and one Softmax layer, finally outputting the weight matrix of the feature matrix S and the weight matrix D of the i-th meta-graph. and ,have: ; ; in, This represents the weighted feature vector of the source node. The weight matrix represents the source feature matrix S in the i-th SDN network metagraph; This represents the weighted feature vector of the target node. The weight matrix represents the target feature matrix D in the i-th SDN network metagraph; This is the dot product between matrices, also known as the Hadamard product.

[0033] Finally, embedding vectors of source nodes Target node embedding vector Weighted eigenvectors and The data is stitched together and used as input to a multilayer perceptron (MLP). The MLP is then used to predict the performance metrics at that moment. ; in, Embed vectors for source nodes; Embed a vector for the target node; This indicates vector concatenation.

[0034] In this embodiment, during the training of the multilayer sensing model, the present invention takes into account minimizing the difference between the performance prediction value and the true value of each node, and the loss function used in training the multilayer sensing model is... as follows: ; in, This is a performance prediction value. This represents the actual performance value, where t represents time, N represents the total number of time windows, and P represents the set of node pairs.

[0035] In this embodiment, the performance prediction value The predicted performance metrics include latency, jitter, and packet loss rate.

[0036] In this embodiment, six metagraphs are designed to represent dimensions such as structure, capacity, congestion, contention, and time, thereby unifying the modeling of heterogeneous elements such as routers, links, flows, queues, and time slots. This addresses the shortcomings of traditional methods that only focus on a single dimension of topology or traffic, and provides a complete characterization of the SDN network's operational status.

[0037] In other embodiments, those skilled in the art can make appropriate adjustments to the attention mechanism or assign weights to the source node feature matrix and the destination node feature matrix in other ways based on their understanding of the inventive concept of this application.

[0038] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0039] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the algorithm. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0040] In the embodiments provided by this invention, it should be understood that the disclosed apparatus / terminal devices and methods can be implemented in other ways. For example, the apparatus / terminal device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0041] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0042] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0043] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms.

[0044] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for predicting the performance of SDN networks based on a multi-element graph structure, characterized in that, Includes the following steps: Based on the characteristics of SDN networks, six path structures of SDN network metagraphs are constructed. The SDN network metagraphs are directed graphs, and each SDN network metagraph includes at least one source node with an in-degree of 0 and one destination node with an out-degree of 0. By extracting features from the SDN network metagraph and calculating the exchange matrix of the SDN network metagraph based on the adjacency matrix of any two adjacent nodes in the SDN network metagraph, the decomposed source feature matrix and destination feature matrix are obtained through a matrix decomposition algorithm. By employing source attention mechanism and destination attention mechanism, weights are assigned to the source feature matrix and the destination feature matrix respectively, to obtain weighted source node weighted feature vector and destination node weighted feature vector; The preset source node embedding vector, destination node embedding vector, source node weighted feature vector, and destination node weighted feature vector are concatenated and then input into a pre-trained multilayer perceptron model to obtain the prediction performance index of the SDN network.

2. The SDN network performance prediction method based on a multi-element graph structure according to claim 1, characterized in that: The SDN network metagraph is represented as follows: ,in, Represents the nodes in the SDN network metagraph. Represents the edges of the SDN network metagraph. Represents the relationships between nodes in the SDN network metagraph. , , Let V be the set of nodes and E be the set of edges. It is a set of relationships between nodes.

3. The SDN network performance prediction method based on a multi-element graph structure according to claim 1, characterized in that: The SDN network metagraph with the six path structures includes six path structures: M1, M2, M3, M4, M5, and M6. The connection structure of M1 is s→R→R→...→d; the connection structure of M2 is s→R→L. w →R→d;The connection structure of M3 is s→R→L d →R→d; The connection structure of M4 is s→R→Q h →R→d; The connection structure of M5 is s→R←F→R→d; The connection structure of M6 is s→R→T→R→d; Where s represents the source node; d represents the destination node; R represents the router / switch; and L represents the link, where L... w and L d These represent high-bandwidth links and low-latency links, respectively; F represents a flow; Q represents a queue, where Q... h This represents a high-congestion queue, where T represents a time slot; Where s→R represents the starting connection, R→R represents the topology connection, R→d represents the ending connection, and R→L and R→L represent the topology connection. w and R→L d Indicates switch-link, L→R, L w →R and L d →R represents a link-switch, R→Q and R→Q h Represents queue mappings, Q→R and Q h →R indicates that the output of the queue re-enters the router forwarding process, F→R indicates that the flow passes through the router, R→T indicates the state of the router at time t, and T→R indicates the subsequent forwarding behavior affected by the state at that moment.

4. The SDN network performance prediction method based on a multi-element graph structure according to claim 1, characterized in that, The exchange matrix of the SDN network metagraph includes: ； in, The exchange matrix represents the metagraph of an SDN network. For any two adjacent nodes in the SDN network metagraph under the same path structure and The adjacency matrix, This represents the total number of nodes in the SDN network metagraph M.

5. The SDN network performance prediction method based on a multi-element graph structure according to claim 1, characterized in that: The exchange matrix of each path in the SDN network metagraph M is decomposed using the matrix factorization (MF) algorithm to obtain the source feature matrix S and the destination feature matrix D.

6. The SDN network performance prediction method based on a multi-element graph structure according to claim 1, characterized in that: The weights assigned to the source attention mechanism and the destination attention mechanism include: ； ； in, This represents the weighted feature vector of the source node. The weight matrix represents the source feature matrix S in the i-th SDN network metagraph; This represents the weighted feature vector of the target node. The weight matrix represents the target feature matrix D in the i-th SDN network metagraph; This represents the dot product between matrices.

7. The SDN network performance prediction method based on a multi-element graph structure according to claim 6, characterized in that: After obtaining the weighted feature vectors of the source node and the target node, they are concatenated with the embedding vectors of the source node and the target node, as follows: ； in, Embed vectors for source nodes; Embed a vector for the target node; This indicates vector concatenation.

8. The SDN network performance prediction method based on a multi-element graph structure according to claim 7, characterized in that: The loss function used to train the multilayer perceptron model as follows: ； in, This is a performance prediction value. This represents the actual performance value, where t is time, N is the total number of time windows, and P is the set of node pairs, where the performance prediction value... The predicted performance metrics include latency, jitter, and packet loss rate.

Images