Software defined network topology discovery process optimization method and device, equipment and medium

Abstract

The invention relates to a software defined network topology discovery process optimization method and device, equipment and a medium. According to the method, dynamic optimization of the topology discovery time of the software defined network is realized by constructing a ternary coupling model of a topology detection period, topology discovery time and a control message proportion and designing a three-stage optimization framework of feature extraction, model prediction and constraint solution. A topological feature driven double-task neural network prediction mechanism is adopted, and the complex nonlinear relation between the network state and the performance index is effectively captured; self-adaptive adjustment of periodic parameters is realized in combination with an improved hybrid genetic algorithm, and limitation of a static configuration mode is broken through; the topology convergence time delay is obviously reduced on the premise of ensuring the control message proportion constraint, and the control plane efficiency and the data plane load are both considered; for network topologies of different scales and structures, a global optimal decision is realized by dynamically adjusting a detection period, stable performance is kept in a heterogeneous scene, and the method has good environmental adaptability, decision real-time performance and engineering expandability.

Description

Technical Field

[0001] This invention belongs to the field of software-defined networking technology, and relates to a method, device, and medium for optimizing the topology discovery process in software-defined networks. Background Technology

[0002] With the rapid development of internet technology and the continuous emergence of new applications, traditional network architectures face numerous challenges, including insufficient flexibility, complex management, and long innovation cycles. Software-Defined Networking (SDN), as a novel network architecture, separates the control plane from the data plane, achieving network programmability and centralized control, providing a completely new approach to solving the pain points of traditional networks. In emerging application scenarios such as large-scale data centers, cloud computing, and 5G, SDN's flexibility and efficiency make it an excellent solution for network architecture. For example, after deploying SDN in its B4 backbone network, Google achieved link utilization exceeding 95%, significantly improving the efficiency of network resource utilization.

[0003] In SDN architecture, the controller acts as the "brain" of the entire network, responsible not only for network topology discovery and maintenance but also for core functions such as traffic scheduling, path calculation, and security policy enforcement. As network scale expands and business complexity increases, the controller faces ever-growing processing pressure, highlighting its performance bottlenecks. Insufficient controller performance can lead to network congestion, degraded service quality, and even system crashes. Particularly in critical sectors such as finance, healthcare, and industrial control, network latency and instability can result in severe economic losses and security risks. Therefore, research on SDN controller performance optimization is crucial for promoting the large-scale deployment of SDN technology in real-world network environments.

[0004] International standardization organizations such as the Open Networking Foundation (ONF) have pointed out that topology discovery time is a key factor affecting the performance of SDN controllers. During SDN network operation, topology discovery is the foundation and prerequisite for the controller to understand and master the network structure. Topology discovery time directly affects the speed and accuracy of the controller's perception of network status, thus impacting the effectiveness of key functions such as route calculation, traffic scheduling, and fault recovery.

[0005] Currently, most SDN controllers use the OpenFlow Discovery Protocol (OFDP) for topology discovery. However, as network scale increases, the traditional OFDP protocol faces significant challenges in terms of time efficiency and control overhead. Firstly, there is a trade-off between efficiency and timeliness: while short cycles can quickly capture topology changes, high-frequency probes lead to a surge in control messages, crowding out data plane bandwidth; long cycles, while reducing overhead, may cause routing black holes or load imbalances due to topology update delays. Secondly, there is insufficient topology sensitivity; fixed cycles cannot adapt to changes in network size and structure. For example, in tree topologies, long-path links require more frequent probes to avoid end-to-end latency accumulation; while in dense mesh topologies, redundant probes may lead to resource waste.

[0006] Taking data center networks as an example, when network equipment fails or topology changes, the controller needs to promptly detect and update the network topology to recalculate routing paths. If topology discovery takes too long, it can lead to interrupted packet forwarding, application service unavailability, and other problems, negatively impacting user experience and business operations. On the other hand, control messages generated during topology discovery also consume network bandwidth and processing resources. In large SDN networks, control messages can consume more than 10% of network bandwidth, increasing the risk of network congestion. Therefore, while optimizing topology discovery time, it is also necessary to consider the balance of control overhead and find the optimal trade-off between time efficiency and resource consumption.

[0007] The challenge in addressing these issues and shortcomings lies in achieving a dynamic balance between topology discovery time and control overhead in a dynamically changing network environment. On one hand, real-time changes in network load, topology, and service demands require a highly adaptive topology discovery mechanism capable of dynamically adjusting the probe cycle based on different network states to ensure rapid and efficient discovery of network topology changes under any circumstances. On the other hand, a complex nonlinear coupling exists between topology discovery time and the proportion of control messages, a relationship difficult to accurately describe and optimize using simple linear models. For example, when the probe cycle is too short, a surge in control message volume can lead to switch queue congestion and nonlinear increases in transmission delay, thus prolonging the actual topology discovery time. Conversely, when the probe cycle is too long, while control overhead decreases, the delay in topology updates can affect network real-time performance and quality of service. Furthermore, existing research largely focuses on protocol improvements or message optimization, failing to fundamentally resolve the contradiction between static mechanisms and dynamic environments. For instance, fixed cycle parameters cannot detect network expansion or topology changes; redundant probing in tree topologies leads to resource waste, while insufficient probing in dense networks may prevent timely detection of topology changes. Summary of the Invention

[0008] To address the problems existing in the traditional methods mentioned above, this invention proposes a software-defined network topology discovery process optimization method, a software-defined network topology discovery process optimization device, a computer device, and a computer-readable storage medium, which can reduce the topology discovery time as much as possible in the software-defined network environment while ensuring that user bandwidth traffic is not crowded.

[0009] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions: On the one hand, a method for optimizing the topology discovery process in software-defined networks is provided, characterized in that the method includes: Step S101: Extract the structural features of the network adjacency matrix and perform standardization processing to obtain a standardized topological feature vector; the standardized topological feature vector includes network size features, connectivity features, path features, and stability features; Step S102: After concatenating the standardized topology feature vector with the current topology discovery cycle, feature extraction is performed using a shared coding layer to obtain high-order coding features, and the topology discovery time and control message ratio are predicted based on the high-order coding features. Step S103: Based on the topology discovery time and the proportion of control messages, an improved hybrid genetic algorithm is used to solve for the topology discovery cycle that satisfies the constraints, while continuously generating new topology discovery cycles. Then, proceed to step S102 for multiple iterations to finally obtain the optimal topology discovery cycle for the current scenario. The controller plane adjustment cycle value is Afterwards, it can optimize the topology discovery time while ensuring that the proportion of control messages does not exceed the threshold.

[0010] On the other hand, a software-defined network topology discovery process optimization apparatus is also provided, the apparatus comprising: The standardized topology feature vector extraction module is used to extract the structural features of the network adjacency matrix and perform standardization processing to obtain a standardized topology feature vector; the standardized topology feature vector includes network size features, connectivity features, path features, and stability features. The topology discovery time and control message ratio prediction module is used to concatenate the standardized topology feature vector with the current topology discovery cycle and then use a shared coding layer to extract features to obtain high-order coding features, and predict the topology discovery time and control message ratio based on the high-order coding features respectively. The topology discovery cycle optimization module is used to solve for the topology discovery cycle that satisfies the constraints based on the topology discovery time and the proportion of control messages, while continuously generating new topology discovery cycles. Then, the process jumps to the topology discovery time and control message ratio prediction module for multiple iterations, ultimately obtaining the optimal topology discovery cycle for the current scenario. The controller plane adjustment cycle value is Afterwards, it can optimize the topology discovery time while ensuring that the proportion of control messages does not exceed the threshold.

[0011] In another aspect, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the software-defined network topology discovery process optimization method described above.

[0012] Furthermore, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the software-defined network topology discovery process optimization method described above.

[0013] One of the above technical solutions has the following advantages and beneficial effects: The aforementioned software-defined network (SDN) topology discovery process optimization method, device, and medium are described above. This method constructs a ternary coupled model of topology detection cycle, topology discovery time, and control message ratio, and designs a three-stage optimization framework of feature extraction, model prediction, and constraint solving to achieve dynamic optimization of SDN topology discovery time. Centralized deployment at the controller end results in low maintenance costs. A topology feature-driven dual-task neural network prediction mechanism effectively captures the complex nonlinear relationship between network state and performance indicators. An improved hybrid genetic algorithm is used to adaptively adjust the cycle parameter, overcoming the limitations of static configuration. It significantly reduces topology convergence latency while ensuring control message ratio constraints, balancing control plane efficiency and data plane load. For network topologies of different sizes and structures, globally optimal decisions are achieved by dynamically adjusting the detection cycle. Stable performance is maintained in heterogeneous scenarios such as tree topologies and large-scale sparse networks, demonstrating good environmental adaptability, real-time decision-making, and engineering scalability. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application or the conventional technology, the drawings used in the description of the embodiments or the conventional technology will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of the overall architecture and process of dynamic optimization of software-defined network topology discovery time in one embodiment; Figure 2A flowchart of a software-defined network topology discovery time dynamic optimization method in one embodiment; Figure 3 Here is a flowchart of the feature extraction stage in one embodiment; Figure 4 Here is a flowchart of the model prediction stage in one embodiment; Figure 5 This is a flowchart of the constraint solving stage in one embodiment. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0017] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0018] It should be noted that, in this document, the reference to "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The presentation of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will understand that the embodiments described herein can be combined with other embodiments. The term "and / or" as used herein refers to any combination of one or more of the associated listed items, and all possible combinations, including such combinations.

[0019] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0020] like Figure 1 , Figure 2 As shown, the software-defined network topology discovery time dynamic optimization method in this embodiment includes the following steps: Step S101: Extract the structural features of the network adjacency matrix and perform standardization processing to obtain the standardized topological feature vector; the standardized topological feature vector includes network size features, connectivity features, path features and stability features.

[0021] Specifically, step S101 is the feature extraction step: extracting the network adjacency matrix. Transformed into a 7-dimensional standardized feature vector that includes network size, connectivity, and stability. Z-score standardization is used to eliminate dimensional differences, ensuring the numerical stability of subsequent model training. Step S102: After concatenating the standardized topology feature vector with the current topology discovery cycle, feature extraction is performed using a shared coding layer to obtain high-order coding features. Based on the high-order coding features, the topology discovery time and the proportion of control messages are predicted respectively.

[0022] Specifically, step S102 is the model prediction step: standardizing topological feature vectors and concatenating the topology discovery cycle. Then, the resulting 8-dimensional vector As model input, it undergoes three levels of dimensionality reduction (256). 128 High-order features are extracted from a shared coding layer (64 nodes), and then topology discovery time is predicted separately through separate fully connected networks. and control message ratio .

[0023] Step S103: Based on the topology discovery time and the proportion of control messages, an improved hybrid genetic algorithm is used to solve for the topology discovery cycle that satisfies the constraints, while continuously generating new topology discovery cycles. Then, proceed to step S102 for multiple iterations to finally obtain the optimal topology discovery cycle for the current scenario. The controller plane adjustment cycle value is Afterwards, it can optimize the topology discovery time while ensuring that the proportion of control messages does not exceed the threshold.

[0024] Specifically, step S103 is the constraint solution step: based on the model prediction obtained in the previous step... An improved hybrid genetic algorithm is used to discover the cycle of topologies that satisfy the constraints. The solution is performed, and new cycles are continuously generated. Then, proceed to step S102 for multiple iterations to finally obtain an optimal topology discovery cycle for the current scenario. The controller plane adjustment cycle value is Afterwards, the topology discovery time can be optimized while ensuring that the proportion of control messages does not exceed the threshold.

[0025] In the aforementioned dynamic optimization of topology discovery time in software-defined networks, the method includes: extracting the resulting features of the network adjacency matrix and performing standardization processing to obtain a standardized topology feature vector; the standardized topology feature vector includes network size features, connectivity features, path features, and stability features; concatenating the standardized topology feature vector with the current topology discovery cycle and then using a shared coding layer for feature extraction to obtain higher-order coding features, and predicting the topology discovery time and control message ratio based on the higher-order coding features; and using an improved hybrid genetic algorithm to solve for the topology discovery cycle that satisfies the constraints based on the topology discovery time and control message ratio, while continuously generating new topology discovery cycles. It then jumps to the model prediction step and performs multiple iterations to finally obtain the optimal topology discovery cycle for the current scenario. The controller plane adjustment cycle value is Subsequently, it can optimize topology discovery time while ensuring that the proportion of control messages does not exceed a threshold. This method breaks the deadlock caused by the inability of fixed periodic parameters to detect network expansion or topology changes, resulting in resource waste due to redundant probing in tree topologies, while in dense networks, insufficient probing may fail to detect topology changes in a timely manner. It achieves joint optimization of topology discovery time and control overhead, while taking into account the network's real-time performance, stability, and resource utilization efficiency.

[0026] This method significantly improves the performance and adaptability of Software-Defined Networking (SDN) in dynamic and complex environments. By optimizing the balance between topology discovery time and control overhead, it not only accelerates network state awareness and improves the efficiency of route calculation and traffic scheduling, but also effectively reduces the bandwidth consumption of control messages, lowering the risk of network congestion and thus enhancing overall network stability and quality of service. Furthermore, this optimization method provides more reliable network support for SDN in critical application scenarios such as large-scale data centers, cloud computing, and 5G, promoting the widespread deployment and application of SDN technology.

[0027] In one embodiment, step S101 includes: extracting the number of switch nodes and the number of edges to obtain network scale features; extracting the average node degree and network density to obtain connectivity features; extracting the maximum connected subgraph diameter and the average diameter length to obtain path features; extracting global clustering coefficients to obtain stability features; and performing Z-score standardization on the network scale features, connectivity features, path features, and stability features to obtain a 7-dimensional standardized feature vector.

[0028] like Figure 3 As shown, in step S101 of this embodiment, extracting the 7-dimensional feature vector includes: S1011, Extract scale features from the network adjacency matrix, including the number of switch nodes. Number of sides ; S1012, extract connectivity features from the network adjacency matrix, including average node degree. and network density ; S1013, Extract path features from the network adjacency matrix, including the diameter of the maximum connected subgraph. and average path length ; S1014 extracts stability features, i.e., global clustering coefficients, from the network adjacency matrix. ; S1015: Z-score normalization is performed on the extracted raw features from the network adjacency matrix to eliminate dimensional differences, ensure the numerical stability of subsequent model training, and form the normalized feature vector that will be input to subsequent modules. The Z-score standardization calculation method is as follows:

[0029] in, To represent the standardized z-score value, The original data, , These are the mean and standard deviation of the training set, respectively.

[0030] In one embodiment, the shared coding layer includes: a three-level fully connected unit; each fully connected unit includes: a linear transformation layer, layer normalization, and a ReLU activation function, with hidden layer dimensions of 256, 128, and 64 respectively, and Dropout is applied to prevent overfitting; step S102 includes: concatenating the standardized topology feature vector with the current topology discovery cycle to obtain an input feature vector; performing high-dimensional feature extraction and dimensionality reduction on the input feature vector using a three-level fully connected unit to obtain high-order coding features; using a first prediction branch to predict the topology discovery time for the high-order coding features; the first prediction branch includes a linear transformation layer and an activation function; using a second prediction branch to predict the control packet proportion for the high-order coding features; the second prediction branch is symmetrical to the first prediction branch, but the parameters are independent.

[0031] like Figure 4 As shown, in step S102, inputting the standardized topological feature vector into the model and obtaining the prediction result includes: S1021, Receive the standardized topological feature vector from the topological feature extraction module. and current topology discovery cycle parameters C Together they form the input feature vector ; S1022, for the input feature vector High-dimensional feature extraction and dimensionality reduction are performed. This step consists of three fully connected units, each containing linear transformation, layer normalization, and ReLU activation function. The hidden layer dimensions are 256. 128 64, and Dropout(0.2) is applied to prevent overfitting, finally outputting high-order encoded features. ; S1023, based on the higher-order features output in the previous step An independent prediction branch (containing a 32-dimensional hidden layer and an output layer) is constructed, and the topology discovery time is predicted through linear transformation and activation function. ; S1024, symmetrical to the temporal prediction branch but with independent parameters, is based on higher-order features output from the same coding layer. Constructing the proportion of predictive control messages in independent branches .

[0032] In one embodiment, step S103 includes: setting the elite pool size. Initialize the population population Includes N randomly generated individuals; sets the initial value of the evolutionary iteration counter to 1; performs a dynamic range pruning operation on each individual in the current population, dynamically adjusting the period according to the formula. The search range is determined; the fitness of each individual is calculated based on the topology discovery time and the proportion of control messages; and individuals are selected from the current population based on their fitness using non-dominated sorting. Elite individuals are added to the elite pool; a tournament selection strategy is used to select parent populations from the current population, and adaptive crossover is performed on the parent populations to generate offspring populations; guided Gaussian mutation is performed on the offspring populations to generate a mutant population; simulated annealing is performed on each individual in the elite pool to optimize the solution according to the annealing formula; the elite pool, offspring population, parent population, and mutant population are merged to obtain a new population; the fitness of all individuals in the new population is calculated; based on the fitness of individuals in the new population, a population of size of [missing information] is selected from the new population. N For the next generation of the population, the evolution iteration counter is updated to the value of the evolution iteration counter incremented by 1. If the value of the evolution iteration counter is not greater than the preset iteration count threshold, the next generation of the population is used as the current population to continue the evolution iteration. This process continues until the value of the evolution iteration counter is greater than the preset iteration count threshold, thus obtaining the optimal topology discovery cycle for the current scenario. .

[0033] In one embodiment, the fitness of each individual is calculated based on the topology discovery time and the proportion of control messages, including: if the proportion of control messages... The range satisfies: Then the fitness function is used. Calculate the fitness value of an individual; where, Set a preset threshold for the proportion of control messages; otherwise, set the fitness of the individual to infinity.

[0034] In one embodiment, the optimal topology discovery cycle in the current scenario for:

[0035] in, This represents the optimal topology discovery cycle for the current scenario. To preset the iteration count threshold, C This refers to the topological discovery cycle of individuals within a population. This is a function for calculating individual fitness.

[0036] In one embodiment, the formula for dynamically adjusting the crossover weights in the adaptive crossover operation is:

[0037] in For the first Standard deviation of the gene generation; The formula for dynamically adjusting the mutation intensity during Gaussian mutation is:

[0038] in, For gene standard deviation, t This represents the number of evolutionary iterations.

[0039] like Figure 5 As shown, in step S103, a hybrid genetic algorithm is used to solve for the optimal topology discovery cycle. The process specifically includes: S1031: Set the size of the elite pool ,in N Population size; S1032: Initialize the population ,Include N Each randomly generated individual consists of two elements: a topology discovery period randomly generated between 100ms and 3000ms. and a weight factor randomly initialized in the range of 0 to 1 ; S1033: Set the generation counter ; S1034: If If so, execute S1035-S10314; otherwise, jump to S10315. S1035: For the current population Each individual element performs a dynamic range clipping operation, dynamically adjusting the cycle according to the formula. C The search range; the specific formula is:

[0040] in, This indicates the search range for period C. The mean of the population cycle in generation t is... ms is the dynamic radius. and These represent the minimum and maximum thresholds of the period, which can be set to 100ms and 3000ms respectively in practice. Through a constraint space shrinkage mechanism, the system gradually focuses on the region of high-quality solutions, balancing search efficiency and accuracy.

[0041] S1036: Calculate the fitness of each individual: Extract the adjacency matrix The structural characteristics are used to predict the output cycle of the model. and control message ratio ;like Then according to the fitness function Calculate the fitness value; otherwise, set it to infinity. S1037: Based on non-dominated sorting from Selected from Elite individuals deposited ; S1038: Adopting a tournament selection strategy from Screening of parental populations ; S1039: Yes An adaptive crossover operation is performed, with the crossover probability and individual weight factors adjusted dynamically to generate the offspring population. The crossover probability and individual weighting factor are calculated using the following dynamic formula:

[0042] in, This represents the individual weighting factor of the offspring. This indicates the topological discovery cycle of offspring. , This refers to the two parent individuals used to produce offspring. , These are the weighting factors for the two parent individuals used to generate offspring. Represents the crossover probability, i.e. Specifically:

[0043] in For the first Standard deviation of the gene generation; S10310: Yes Perform guided Gaussian mutation operation, mutation strength Adjust according to the dynamic formula to generate a mutant population. The formula for dynamically adjusting the intensity of variation is:

[0044] in, For gene standard deviation, t This represents the number of evolutionary iterations.

[0045] S10311: Elite Pool Each individual in the process performs a simulated annealing optimization operation, updating the solution according to the annealing formula; S10312: Merging populations ; S10313: Yes The fitness calculation for all individuals in the process is re-executed using S1036; S10314: From the environment selection operation The screening scale is The next generation of population ,make And return to S1034; S10315: Output the globally optimal cycle time .

[0046] In summary, this embodiment constructs a ternary coupled model of topology detection period, topology discovery time, and control message ratio, and designs a three-stage optimization framework of feature extraction, model prediction, and constraint solving to achieve dynamic optimization of topology discovery time in software-defined networks. This invention achieves centralized deployment at the controller end, resulting in low maintenance costs; it employs a topology feature-driven dual-task neural network prediction mechanism to effectively capture the complex nonlinear relationship between network state and performance indicators; it combines an improved hybrid genetic algorithm to achieve adaptive adjustment of periodic parameters, overcoming the limitations of static configuration modes; it can significantly reduce topology convergence latency while ensuring control message ratio constraints, balancing control plane efficiency and data plane load; and it achieves globally optimal decision-making by dynamically adjusting the detection period for network topologies of different sizes and structures, maintaining stable performance in heterogeneous scenarios such as tree topologies and large-scale sparse networks, demonstrating good environmental adaptability, real-time decision-making, and engineering scalability.

[0047] In one embodiment, a software-defined network topology discovery process optimization apparatus is provided, the apparatus comprising: The standardized topology feature vector extraction module is used to extract the structural features of the network adjacency matrix and perform standardization processing to obtain standardized topology feature vectors. The standardized topology feature vectors include network size features, connectivity features, path features, and stability features. The topology discovery time and control message ratio prediction module is used to concatenate the standardized topology feature vector with the current topology discovery cycle and then use a shared coding layer to extract features to obtain high-order coding features. Based on the high-order coding features, the topology discovery time and control message ratio are predicted respectively. The topology discovery cycle optimization module is used to solve for the topology discovery cycle that satisfies the constraints based on the topology discovery time and the proportion of control messages, while continuously generating new topology discovery cycles. Then, the process jumps to the topology discovery time and control message ratio prediction module for multiple iterations, ultimately obtaining the optimal topology discovery cycle for the current scenario. The controller plane adjustment cycle value is Afterwards, it can optimize the topology discovery time while ensuring that the proportion of control messages does not exceed the threshold.

[0048] In one embodiment, the standardized topology feature vector extraction module is also used to extract the number of switch nodes and edges to obtain network size features; extract the average node degree and network density to obtain connectivity features; extract the maximum connected subgraph diameter and average diameter length to obtain path features; extract the global clustering coefficient to obtain stability features; and perform Z-score standardization on the network size features, connectivity features, path features, and stability features to obtain a 7-dimensional standardized feature vector.

[0049] In one embodiment, the shared coding layer includes: a three-level fully connected unit; each fully connected unit includes: a linear transformation layer, layer normalization, and a ReLU activation function, with hidden layer dimensions of 256, 128, and 64 respectively, and Dropout is applied to prevent overfitting; the topology discovery time and control message proportion prediction module is also used to concatenate the standardized topology feature vector and the current topology discovery cycle to obtain an input feature vector; the input feature vector is subjected to high-dimensional feature extraction and dimensionality reduction processing by the three-level fully connected unit to obtain high-order coding features; the high-order coding features are used to predict the topology discovery time using a first prediction branch; the first prediction branch includes a linear transformation layer and an activation function; the high-order coding features are used to predict the control message proportion using a second prediction branch; the second prediction branch is symmetrical to the first prediction branch, but the parameters are independent.

[0050] In one embodiment, the topology discovery cycle optimization module is also used to set the elite pool size. Initialize the population population Includes N randomly generated individuals; sets the initial value of the evolutionary iteration counter to 1; performs a dynamic range pruning operation on each individual in the current population, dynamically adjusting the period according to the formula. The search range is determined; the fitness of each individual is calculated based on the topology discovery time and the proportion of control messages; and individuals are selected from the current population based on their fitness using non-dominated sorting. Elite individuals are added to the elite pool; a tournament selection strategy is used to select parent populations from the current population, and adaptive crossover is performed on the parent populations to generate offspring populations; guided Gaussian mutation is performed on the offspring populations to generate a mutant population; simulated annealing is performed on each individual in the elite pool to optimize the solution according to the annealing formula; the elite pool, offspring population, parent population, and mutant population are merged to obtain a new population; the fitness of all individuals in the new population is calculated; based on the fitness of individuals in the new population, a population of size of [missing information] is selected from the new population. N For the next generation of the population, the evolution iteration counter is updated to the value of the evolution iteration counter incremented by 1. If the value of the evolution iteration counter is not greater than the preset iteration count threshold, the next generation of the population is used as the current population to continue the evolution iteration. This process continues until the value of the evolution iteration counter is greater than the preset iteration count threshold, thus obtaining the optimal topology discovery cycle for the current scenario. .

[0051] In one embodiment, the topology discovery cycle optimization module is further configured to control the packet proportion. The range satisfies: Then the fitness function is used. Calculate the fitness value of an individual; where, Set a preset threshold for the proportion of control messages; otherwise, set the fitness of the individual to infinity.

[0052] In one embodiment, the optimal topology discovery cycle in the current scenario is defined in the topology discovery cycle optimization module. for:

[0053] in, This represents the optimal topology discovery cycle for the current scenario. To preset the iteration count threshold, C This refers to the topological discovery cycle of individuals within a population. This is a function for calculating individual fitness.

[0054] In one embodiment, the formula for dynamically adjusting the crossover weight in the adaptive crossover operation of the topology discovery cycle optimization module is:

[0055] in For the first Standard deviation of genes This represents the standard deviation of the genes in generation 0.

[0056] During the Gaussian mutation operation, the formula for dynamically adjusting the mutation intensity in the topology discovery cycle optimization module is as follows:

[0057] in, For gene standard deviation, t This represents the number of evolutionary iterations.

[0058] It is understood that for a detailed explanation of the software-defined network topology discovery process optimization device, please refer to the corresponding explanations of the various embodiments of the software-defined network topology discovery process optimization method above, and will not be repeated here. Each module in the above-described software-defined network topology discovery process optimization device can be implemented entirely or partially through software, hardware, or a combination thereof. Each module can be embedded in hardware or independently of a device with data processing capabilities, or it can be stored in software in the memory of the aforementioned device, so that the processor can call and execute the operations corresponding to each module. The aforementioned device can be, but is not limited to, various types of data processing computer devices already existing in the art.

[0059] In one embodiment, a computer device is also provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the processing steps in the above-described embodiment of the software-defined network topology discovery process optimization method.

[0060] It is understood that, in addition to the memory and processor mentioned above, the computer equipment described above also includes other hardware and software components not listed in this specification. The specific components can be determined according to the model of the image processing computer in different application scenarios, and will not be listed and described in detail in this specification.

[0061] In one embodiment, when the processor executes the computer program, it can also implement the steps or sub-steps added in the various embodiments of the software-defined network topology discovery process optimization method described above.

[0062] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), memory bus DRAM (RDRAM), and interface DRAM (DRDRAM), etc.

[0063] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0064] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of protection of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and all such modifications and improvements fall within the scope of protection of this application.

Claims

1. A method for optimizing the topology discovery process in software-defined networks, characterized in that, The method includes: Step S101: Extract the structural features of the network adjacency matrix and perform standardization processing to obtain a standardized topological feature vector; the standardized topological feature vector includes network size features, connectivity features, path features, and stability features; Step S102: After concatenating the standardized topology feature vector with the current topology discovery cycle, feature extraction is performed using a shared coding layer to obtain high-order coding features, and the topology discovery time and control message ratio are predicted based on the high-order coding features. Step S103: Based on the topology discovery time and the proportion of control messages, an improved hybrid genetic algorithm is used to solve for the topology discovery cycle that satisfies the constraints, while continuously generating new topology discovery cycles. Then, proceed to step S102 for multiple iterations to finally obtain the optimal topology discovery cycle for the current scenario. The controller plane adjustment cycle value is Afterwards, it can optimize the topology discovery time while ensuring that the proportion of control messages does not exceed the threshold.

2. The software-defined network topology discovery process optimization method according to claim 1, characterized in that, Step S101 includes: The number of switch nodes and edges are extracted from the network adjacency matrix to obtain network size characteristics; The average node degree and network density are extracted from the network adjacency matrix to obtain connectivity features; The path features are obtained by extracting the maximum connected subgraph diameter and average diameter length from the network adjacency matrix; Global clustering coefficients are extracted from the network adjacency matrix to obtain stability features; The network size feature, connectivity feature, path feature, and stability feature are subjected to Z-score normalization to obtain a 7-dimensional normalized feature vector.

3. The software-defined network topology discovery process optimization method according to claim 1, characterized in that, The shared coding layer includes three levels of fully connected units; each fully connected unit includes a linear transformation layer, layer normalization, and ReLU activation function, with hidden layer dimensions of 256, 128, and 64 respectively, and Dropout is applied to prevent overfitting; step S102 includes: The standardized topological feature vector is concatenated with the current topological discovery cycle to obtain the input feature vector; The input feature vector is subjected to high-dimensional feature extraction and dimensionality reduction processing using a three-level fully connected unit to obtain high-order encoded features; The topology discovery time is predicted using a first prediction branch on the higher-order encoded features; the first prediction branch includes a linear transformation layer and an activation function. The proportion of control messages is predicted using a second prediction branch for the higher-order coding features; the second prediction branch is symmetrical to the first prediction branch, but the parameters are independent.

4. The software-defined network topology discovery process optimization method according to claim 1, characterized in that, Step S103 includes: Set the elite pool size ; Initialize population population Includes N randomly generated individuals; Set the initial value of the evolution iteration counter to 1; Perform dynamic range pruning on each individual in the current population, and dynamically adjust the period according to the formula. The search scope; The fitness of each individual is calculated based on the topology discovery time and the proportion of control messages. Based on individual fitness, individuals are selected from the current population using non-dominated ranking. Each elite individual is stored in the elite pool; A tournament selection strategy is used to select parent populations from the current population. An adaptive crossover operation is performed on the parent population to generate the offspring population; Perform guided Gaussian mutation on the offspring population to generate a mutant population; Perform simulated annealing optimization on each individual in the elite pool and update the solution according to the annealing formula; The elite pool, the offspring population, the parent population, and the mutant population are merged to obtain a new population. Fitness is calculated for all individuals in the new population; The size of the new population is selected based on the fitness of individuals within the new population. N The next generation of the population will update the value of the evolution iteration counter to the value of the evolution iteration counter plus 1; If the value of the evolution iteration counter is not greater than the preset iteration count threshold, the next generation population is used as the current population to continue the evolution iteration; this continues until the value of the evolution iteration counter is greater than the preset iteration count threshold, thus obtaining the optimal topology discovery cycle for the current scenario. .

5. The software-defined network topology discovery process optimization method according to claim 4, characterized in that, Based on the topology discovery time and the proportion of control messages, the fitness of each individual is calculated, including: If the control message accounts for The range satisfies: Then the fitness function is used. Calculate the fitness value of an individual; where, Set a preset threshold for the proportion of control messages; otherwise, set the fitness of the individual to infinity.

6. The software-defined network topology discovery process optimization method according to claim 4, characterized in that, Optimal topology discovery cycle in the current scenario for: in, This represents the optimal topology discovery cycle for the current scenario. To preset the iteration count threshold, C This refers to the topological discovery cycle of individuals within a population. This is a function for calculating individual fitness.

7. The software-defined network topology discovery process optimization method according to claim 4, characterized in that, The formula for dynamically adjusting the crossover weights in adaptive crossover operations is: in For the first Standard deviation of the gene generation; The formula for dynamically adjusting the mutation intensity during Gaussian mutation is: in, For gene standard deviation, t This represents the number of evolutionary iterations.

8. A software-defined network topology discovery process optimization device, characterized in that, The device includes: The standardized topology feature vector extraction module is used to extract the structural features of the network adjacency matrix and perform standardization processing to obtain a standardized topology feature vector; the standardized topology feature vector includes network size features, connectivity features, path features, and stability features. The topology discovery time and control message ratio prediction module is used to concatenate the standardized topology feature vector with the current topology discovery cycle and then use a shared coding layer to extract features to obtain high-order coding features, and predict the topology discovery time and control message ratio based on the high-order coding features respectively. The topology discovery cycle optimization module is used to solve for the topology discovery cycle that satisfies the constraints based on the topology discovery time and the proportion of control messages, while continuously generating new topology discovery cycles. It then jumps to the topology discovery time and control message ratio prediction module for multiple iterations, ultimately obtaining the optimal topology discovery cycle for the current scenario. The controller plane adjustment cycle value is Afterwards, it can optimize the topology discovery time while ensuring that the proportion of control messages does not exceed the threshold.

9. A computer device, comprising a memory and a processor, characterized in that, The memory stores a computer program, and when the processor executes the computer program, it implements the steps of the software-defined network topology discovery process optimization method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the software-defined network topology discovery process optimization method as described in any one of claims 1 to 7.

Images