SDN (Software Defined Network) progressive upgrading method

Abstract

The invention discloses a progressive upgrading method for an SDN (Software Defined Network). An SDN controller periodically acquires a historical traffic matrix of a hybrid network in a hybrid acquisition mode through management connection; performing order polynomial fitting on the traffic time sequence by using a least square method to generate a predicted traffic matrix; taking a prediction traffic matrix and a network topology as input, constructing an optimization model taking maximization of an SDN controllable traffic scale as a core target and taking upgrading budget and a link utilization rate as constraints, and efficiently solving and outputting an upgrading priority list of traditional nodes through a dual rising method of dual weights; and executing upgrading and evaluating the effect according to the list to form closed-loop optimization. According to the method, the characteristic of partial path controllability of the SDN is fully utilized, the contribution of node upgrading to the controllable flow is quantified, the SDN control range is rapidly expanded on the premise of ensuring the network performance, and the core value of progressive migration is improved.

Description

A progressive upgrade method for SDN networks Technical Field

[0001] This invention relates to the field of software-defined networking (SDN) technology, and specifically to a progressive upgrade method for SDN networks. Background Technology

[0002] Traditional IP networks employ distributed control protocols (such as OSPF and BGP), with control logic dispersed across various network nodes. This leads to complex network management, rigid policy configuration, and difficulties in expansion, making it challenging to dynamically and finely balance and optimize traffic load based on global status. In contrast, Software-Defined Networking (SDN), by separating the control plane and data plane, achieves centralized network management and flexible programming, demonstrating significant advantages such as simplified management, increased flexibility, and intelligent traffic control. However, for existing large-scale network infrastructures, replacing all nodes based on traditional architecture with a pure SDN architecture in one go faces practical constraints such as high costs, significant technical risks, and potential disruption to existing services. To overcome this contradiction, hybrid SDN networks, which combine the existing foundation of traditional networks with the advantages of SDN technology, have become an ideal evolutionary direction. This involves gradually introducing SDN nodes into the traditional network and deploying a centralized controller, enabling the controller to indirectly perceive network status and traffic dynamics, thereby executing optimization strategies. Correspondingly, a gradual migration strategy is commonly adopted in practice, that is, upgrading traditional network nodes to SDN nodes in batches and selectively, gradually expanding the scope of SDN management to achieve a smooth network evolution.

[0003] The core challenge of incremental migration lies in optimizing the upgrade sequence. Specifically, given a limited upgrade budget (e.g., the proportion of nodes that can be upgraded in each phase), the challenge is to select the optimal subset of nodes from a large pool of traditional nodes to enable the SDN controller to maximize overall network performance (e.g., maximizing controllable traffic volume, minimizing maximum link utilization, maximizing network throughput, and improving load balancing). The complexity of this problem lies in: Limited observation methods: The controller cannot directly read the flow tables and counters of the switch or actively send probe packets through protocols such as OpenFlow. It can only obtain topology and traffic information by passively listening to network traffic (such as port mirroring), analyzing network protocols (such as LLDP, STP BPDU), or indirectly inferring. Coupling: Upgrade decisions (which nodes to select) are coupled with traffic engineering decisions (how to schedule controllable traffic), and the unknown set of upgrades leads to an uncertain set of available traffic scheduling paths. Dynamic nature: Network traffic exhibits significant time-varying characteristics, with both short-term bursts and long-term patterns coexisting. Upgrade decisions must combine the ability to optimize the current state with the ability to adapt to future trends. Computational complexity: This problem can be modeled as a mixed-integer nonlinear programming problem, which is an NP-hard problem and is difficult to solve directly for the optimal solution in large networks.

[0004] Existing technical solutions are mostly based on static network snapshots (such as topology and traffic at a certain moment) or simplified traffic models (such as assuming the traffic matrix is ​​constant) for heuristic search. They lack fine modeling of the dynamic characteristics of traffic and fail to effectively decouple the complex relationship between upgrade decision and path decision, resulting in poor upgrade effect or poor algorithm convergence. Summary of the Invention

[0005] The present invention addresses the problem that existing incremental upgrade methods for SDN networks lack detailed modeling of dynamic traffic characteristics, and provides an incremental upgrade method for SDN networks.

[0006] To solve the above problems, the present invention is achieved through the following technical solution:

[0007] A progressive upgrade method for SDN networks includes the following steps:

[0008] Step 1: The controller obtains the complete topology of the hybrid network through the link layer discovery protocol, and periodically acquires the network traffic information using a differentiated collection strategy; then, it constructs a network traffic matrix based on the network topology and network traffic information.

[0009] Step 2: The controller uses the least squares method to perform multi-order polynomial fitting on the time series of the collected historical network traffic matrix, and predicts future traffic trends to obtain the network predicted traffic matrix.

[0010] Step 3: The controller takes the whole network predicted traffic matrix and the whole network topology as input, constructs an optimization problem with the main objective of maximizing the controllable traffic scale of SDN, and uses the dual ascent method to solve the optimization problem to obtain the upgrade priority score of each traditional node, and then obtains the upgrade recommendation list.

[0011] Step 4: Based on the upgrade recommendation list output by the controller, network operations and maintenance personnel perform a physical upgrade operation to replace traditional nodes with SDN nodes.

[0012] The specific steps in step 1 above, which involve periodically acquiring network traffic information using a differentiated collection strategy, are as follows:

[0013] For SDN nodes in a hybrid network, the controller periodically collects flow table statistics of SDN nodes through the SDN southbound interface, and combines Packet-In message analysis to directly obtain the link transmission traffic in which the SDN node participates.

[0014] For traditional nodes in a hybrid network, the controller periodically collects the link inbound and outbound traffic of traditional nodes through port mirroring, and obtains port statistics based on the Simple Network Management Protocol. Combining the principle of traffic conservation, the controller indirectly infers the traffic distribution among traditional nodes.

[0015] The specific process of step 2 above is as follows:

[0016] Step 2.1: For each historical time point, first count the traffic of each node pair in the traffic matrix to obtain the total network traffic, and then calculate the proportion of traffic of each node pair relative to the total network traffic to obtain the traffic proportion of each node pair.

[0017] Step 2.2: Based on the total network traffic at each historical time point obtained in Step 2.1, select the total network traffic as the core fitting target, with time as the independent variable and the total network traffic as the dependent variable. Use a multi-order polynomial to perform least squares fitting to obtain the optimal multi-order polynomial coefficients, and then fit the multi-order polynomial.

[0018] Step 2.3: Calculate the goodness of fit of the fitted multi-order polynomial: If the goodness of fit is greater than or equal to the set reliability threshold, the fitted multi-order polynomial is the final multi-order polynomial, and proceed to step 2.4; otherwise, supplement the historical flow matrix and return to step 2.1 or adjust the order of the multi-order polynomial and return to step 2.2.

[0019] Step 2.4: First, use the final multi-order polynomial to predict the total network traffic at the next time point. Then, use the moving average of the traffic share of each node pair at recent historical time points as the predicted traffic share of each node pair at the next time point. Finally, use the predicted total network traffic at the next time point and the predicted traffic share of each node pair at the next time point to obtain the predicted network traffic matrix at the next time point.

[0020] In step 3 above, the optimization problem is:

[0021]

[0022] In the formula, Represents node pairs Predicted transmission traffic, A set of SDN nodes; For traditional node sets, Represents a node Decision variables , Represents a node Upgraded Represents a node No upgrade Indicates the upgrade budget; Represents node pairs The minimum link capacity on the path, Indicates the maximum allowed link utilization; Represents nodes Connected links, Represents a node The set of outgoing edges, Represents a node The set of incoming edges, Indicates link Controllable flow Indicates link Uncontrollable traffic, A set of nodes; Represents node pairs Controllable flow Represents node pairs Uncontrollable traffic.

[0023] In step 3 above, the solution process of the dual ascent method is as follows:

[0024] Step 3.1, Initialization:

[0025] For each link Assigning initial dual link weights with performance constraints :

[0026] ,

[0027] For each node pair Assign initial dual flow weights to controllable flow targets :

[0028] ,

[0029] Step 3.2: Iterative Evaluation:

[0030] Calculate the node value score:

[0031] ,

[0032] Weight update:

[0033] Dual link weight update:

[0034] ,

[0035] Dual link weight update:

[0036] ,

[0037] Convergence and Output:

[0038] When all dual link weights change Less than the set threshold And the changes in weights of all dual links Less than the set threshold When the maximum number of iterations is reached, the node value score calculated in the final iteration round is taken as its overall upgrade priority.

[0039] In the formula, It is a node pair The source node or the destination node, It is with nodes Directly connected links, For link The initial dual link weights, This is the link ratio coefficient. Links estimated using the traditional shortest path allocation method Initial link traffic, For link The capacity; For node pairs The initial dual traffic weights, This is the flow rate ratio coefficient. For node pairs Predicted transmission traffic; For nodes The Round value score, For node pairs The Wheel-to-wheel traffic weighting For node pairs The Controllable flow rate For node pairs The Controllable flow rate For link The Wheel-to-dual link weights, Links estimated using the traditional shortest path allocation method The Round-robin traffic, To maximize the allowed link utilization, This indicates the non-negation operation; For link The Wheel-to-dual link weights, This refers to the link step size parameter; For node pairs The Wheel-to-wheel traffic weighting This is the flow step size parameter.

[0040] The core idea of ​​this invention is to transform the complex mixed integer programming problem into an optimization problem that captures the dynamic patterns of historical traffic data by fitting a function to the data, and constructs an optimization problem based on the fitting results with the main objective of maximizing the controllable traffic scale of SDN. By using the dual ascent method to efficiently solve the continuous relaxation problem, a near-optimal node upgrade priority ranking is obtained.

[0041] Compared with the prior art, the present invention has the following characteristics:

[0042] 1. Aligned with SDN engineering practice: Fully utilizes the core feature of controllable partial paths, abandons the excessive constraints of full SDN paths, is more in line with the actual scenario of gradual migration, and greatly improves the efficiency of expanding controllable traffic scale.

[0043] 2. Precisely quantify the value of upgrades: Through the dual ascending method and double dual weight design, the contribution of node upgrades to new controllable traffic is directly quantified, avoiding ineffective upgrades (such as upgrading nodes through which no critical traffic passes) and improving the accuracy of decision-making.

[0044] 3. Balancing control and performance: While maximizing controllable traffic, the system ensures network stability through link utilization constraints and cost penalty mechanisms, thus balancing the expansion of control scope with operational stability and meeting the needs of project implementation.

[0045] 4. Dynamically adapt to traffic changes: Based on the least squares method to fit the dynamic traffic pattern, the upgrade decision is aligned with future traffic trends, ensuring the continuity and stability of newly added controllable traffic and improving the foresight of the decision. Attached Figure Description

[0046] Figure 1 is an overall flowchart of a progressive upgrade method for SDN networks.

[0047] Figure 2 is a flowchart of the dual ascent method.

[0048] Figure 3 shows the module structure of the SDN network progressive upgrade system. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific examples and the accompanying drawings.

[0050] Referring to Figure 1, a progressive upgrade method for SDN networks includes the following steps:

[0051] S1: Topology Discovery and Traffic Matrix Acquisition. The controller obtains the complete topology of the hybrid network through the link layer discovery protocol, and periodically acquires network-wide traffic information using a differentiated acquisition strategy; then, it constructs a network-wide traffic matrix based on the network topology and traffic information.

[0052] S1.1: Topology Discovery. The controller obtains the complete topology of the hybrid network through the Link Layer Discovery Protocol (LLDP), including all nodes (switches) and links, identifying all nodes in the network and the interconnections between them. The total number of nodes recorded is... And assign a unique identifier to each node. For a set of nodes, For a set of links.

[0053] S1.2: Multi-source traffic data acquisition. The controller employs a differentiated acquisition strategy to obtain network-wide traffic information:

[0054] (1) For upgraded SDN nodes, the controller periodically collects the flow table statistics of SDN nodes through the SDN southbound interface (such as OpenFlow), and directly obtains the link transmission traffic in which the SDN node participates by combining Packet-In message analysis.

[0055] (2) For traditional nodes, the inbound and outbound traffic of the links of traditional nodes is collected periodically through port mirroring technology. At the same time, the port statistics of traditional nodes, such as port bandwidth and packet forwarding volume, are obtained based on the Simple Network Management Protocol (SNMP). Combined with the principle of traffic conservation, the traffic distribution between traditional nodes is indirectly inferred.

[0056] S1.3: Construction of the Network-wide Traffic Matrix. Based on the network topology obtained in S1.1, the direct traffic data and indirect statistical data collected in S1.2 are fused and inferred to construct the network-wide traffic matrix among all network nodes. .

[0057] Full network traffic matrix A single element in is denoted as , indicating node pair Transmission traffic (unit: bits / second or data packets / second, which can be adjusted according to actual network monitoring needs).

[0058]

[0059] S2: Traffic pattern fitting based on least squares method. The controller uses the least squares method to perform multi-order polynomial fitting on the time series of the collected historical network traffic matrix, and predicts future traffic trends to obtain the network-wide predicted traffic matrix.

[0060] S2.1: Feature Extraction. Statistical analysis of the total network traffic matrix at each historical time point. All node pairs The transmission traffic is used to obtain the total network traffic. :

[0061]

[0062] S2.2: Fitting. Based on the total network traffic at each historical time point obtained in 2.1. Select the total network traffic As the core fitting objective, time Independent variable: Total network traffic As the dependent variable, use polynomial model Perform least squares fitting and solve... Obtain the optimal polynomial coefficients Thus, the fitted result is obtained. A polynomial of order n. Wherein The value ranges from 2 to 4. The optimal order is determined through cross-validation to avoid overfitting or underfitting.

[0063] S2.3: Evaluation. Calculate the fitted value. Goodness of fit of order polynomial ,when Greater than or equal to the reliability threshold (e.g.) When the condition is met, the fitted multi-order polynomial is the final result. polynomial of order If the flow rate is below the reliable threshold, supplement the historical flow matrix and return S 2.1, or adjust the order of the multi-order polynomial. And return S 2.2.

[0064] Goodness of fit The calculation formula is:

[0065]

[0066] In the formula, the sum of squared residuals Total Sum of Squares , and These are the actual observed values ​​and the model predicted values, respectively. This is the observed mean. The closer to 1, the higher the model fit.

[0067] S2.4: Prediction.

[0068] First, using the final polynomial of order Predict the next time point Predicting total network traffic .

[0069]

[0070] Then, the traffic percentage of each node at recent historical time points (e.g., taking the most recent M time points) is calculated. The moving average is used as the predicted flow percentage for each node pair at the next time point. .

[0071] Each node at each historical point in time Transmission traffic Total network traffic The proportion of nodes is obtained. Traffic share for:

[0072]

[0073] Finally, the predicted total network traffic at the next time point will be... The predicted flow percentage of each node pair at the next time point Multiply by this to obtain the predicted flow of each node pair at the next time point. This allows for the reconstruction of the predicted network traffic matrix for the next time point. .

[0074]

[0075] S3: Solving for node upgrade priority based on the duality ascending method. The controller uses the predicted flow matrix obtained in S2. Using the network topology obtained from S1 as input, we construct an optimization problem with the main objective of maximizing the controllable traffic scale of SDN, and use the dual ascent method to efficiently solve the optimization problem, output the upgrade priority score of each traditional node, and then obtain the upgrade recommendation list.

[0076] S3.1: Problem Modeling. Let... For a set of nodes, , For SDN node set, This is a traditional set of nodes. Decision variables ,in Represents a node Upgraded Represents a node No upgrade; SDN node set after upgrade .

[0077] For each node pair Predicted transmission traffic If node or node Any node in the network is an SDN node, that is This will predict the transmission traffic. Define it as controllable traffic; otherwise, predict the transmission traffic. Defined as uncontrollable traffic. This definition mathematically constructs a more operational optimization problem: by setting the controllability criterion as requiring the data path to traverse at least one SDN node, rather than requiring end-to-end full-path SDN, the constraint strength of the decision space is significantly reduced. This allows the dual ascent method to efficiently identify bottleneck nodes that act as convergence points for multiple traffic flows, prioritizing the upgrading of such nodes. This instantly brings multiple originally scattered and uncontrollable data flows passing through these nodes into a centralized management scope, thereby leveraging the maximization of the global controllable traffic scale (i.e., the objective function) with minimal upgrade actions (i.e., the lowest decision variable cost), perfectly aligning with the core principle of pursuing optimal marginal benefits in incremental migration.

[0078] The core optimization goal (maximizing the controllable traffic scale of SDN) is:

[0079]

[0080] In the formula, For node pairs Predicted transmission traffic, This is a set of SDN nodes.

[0081] The constraints are:

[0082] Upgrade Budget Constraint: The number of traditional nodes in this upgrade round shall not exceed the upgrade budget. ,Right now:

[0083]

[0084] In the formula, For traditional node sets, Represents a node Decision variables Represents a node Upgraded Represents a node No upgrade Indicates the number of elements in the set. This indicates the budget for the upgrade.

[0085] Network performance constraints: Link utilization shall not exceed the preset maximum allowable link utilization. (default To avoid link congestion, i.e.:

[0086]

[0087] In the formula, For node pairs Predicted transmission traffic, For node pairs Minimum link capacity on the path, This represents the maximum allowed link utilization.

[0088] Flow conservation constraint: Both types of flow satisfy the condition that the flow into a node equals the flow out of a node, and the total flow of a node pair equals the sum of the controllable flow component and the uncontrollable flow component.

[0089] Controllable flow conservation constraint:

[0090]

[0091] Uncontrollable flow conservation constraint:

[0092]

[0093] Overall constraints:

[0094]

[0095] In the formula, Represents nodes Connected links, Represents a node The set of outgoing edges, Represents a node The set of incoming edges, Indicates link Controllable flow Indicates link Uncontrollable traffic, A set of nodes; Represents node pairs Controllable flow Represents node pairs Uncontrollable traffic, Represents node pairs Predicted transmission traffic.

[0096] The total traffic is divided into controllable traffic and traditional uncontrollable traffic. These two formulas ensure that these two types of traffic follow the law of traffic conservation in network transmission, that is, traffic is generated at the source, absorbed at the destination, and the inflow equals the outflow at intermediate nodes, thus providing a physically feasible basis for network optimization.

[0097] S3.2: Solving using the dual ascent method. The optimization problem is transformed into its dual problem, and then solved iteratively using the dual ascent method.

[0098] In the solution framework of the dual ascent method, its mathematical modeling relies on a set of key symbols. To achieve efficient solution, this invention introduces two sets of core dual variables: one is the link dual weights of the link performance constraints. It dynamically reflects each link The degree of congestion; and secondly, the flow dual weight of the flow target constraint. It represents the pair of unlocked nodes. The urgency of controllable flow. Based on the introduction of two sets of core dual variables, the solution process of the dual ascent method of this invention is shown in Figure 2, which specifically includes the following steps:

[0099] S3.2.1: Initialization:

[0100] For each link Assigning performance-constrained dual link weights ,in:

[0101]

[0102] For each node pair Assign dual flow weights to controllable flow targets ,in:

[0103]

[0104] In the formula, This is the link scaling factor (recommended value range: 0.05-0.2). The traffic ratio coefficient (recommended value range 0.1-0.3) ensures that the controllability of high-traffic node pairs is given priority. Links estimated using the traditional shortest path allocation method Initial link traffic, For link capacity; For node pairs Predicted transmission traffic.

[0105] Initial link traffic For predictive flow matrix Initial link traffic estimated using the traditional shortest path allocation method. Initial link traffic. Essentially, it simulates the predicted traffic matrix before SDN upgrades and optimizations. When all traffic in a network is forwarded according to traditional distributed routing protocols (such as OSPF shortest path), each link... The load that will be borne by the network. The calculation process can be summarized as follows: First, based on the network topology, assign each node a certain load. Determine its flow Find the shortest path that should be taken; then, traverse all node pairs and calculate the predicted flow for each node pair. The traffic is accumulated across all links included in its shortest path; ultimately, the total traffic accumulated on each link is the initial traffic estimate for that link. Its mathematical model is derived from the formula Precise description, in which As an indicator function, when the link Belongs to node pairs shortest path The value is 1 if the condition is met, and 0 otherwise. This estimate serves as a baseline for algorithm initialization, used to identify potentially congested links under traditional routing.

[0106] In this embodiment, link weight ( Based on the traditional shortest path allocation estimation using T', the initial utilization rate of the core link is approximately 75%; traffic weight. High-traffic node pairs in the core-convergence layer It is significantly higher than the access layer.

[0107] S3.2.2: Iterative Evaluation: Perform Round iteration ( To preset the maximum number of iterations, a value of 100-200 is recommended.

[0108] Node scoring under fixed weights:

[0109] The simulated SDN controller can optimally schedule traffic and compute the data for each node. The Wheel Virtual Upgrade Value Score The mathematical expression is:

[0110]

[0111] In the formula, It is a node pair The source node or the destination node, It is with nodes Directly connected physical links, For node pairs The Wheel-to-wheel traffic weighting For node pairs The Controllable flow rate For node pairs The Controllable flow rate For link The Wheel-to-dual link weights, Links estimated using the traditional shortest path allocation method The Round-robin traffic, For link capacity, To maximize the allowed link utilization, This indicates the non-negation operation.

[0112] No. Wheel Virtual Upgrade Value Score The first item To add value to controllable traffic, the second item This comes at the cost of performance constraints. Core logic: Nodes The upgrade value is positively correlated with "the total amount of newly controllable traffic × traffic weight" and negatively correlated with "the cost of new congestion", which ensures both the expansion of the control scope and the maintenance of network stability.

[0113] Weight update:

[0114] Dual link weight update:

[0115]

[0116] In the formula, This is the link step size parameter (recommended value range: 0.01-0.1). If the link... The virtual load exceeds The weight is increased to penalize the path, guiding subsequent iterations to optimize traffic allocation.

[0117] Dual link weight update:

[0118]

[0119] In the formula, This is the flow step size parameter (recommended value range: 0.05-0.2). This is to fully realize controllable high-flow node pairs. With increased weight, the guiding algorithm prioritizes unlocking its controllability.

[0120] In this embodiment, in each iteration, the module calculates the values ​​of 100 traditional nodes. Aggregation layer node Connecting 10 access layer nodes and 5 core SDN nodes, after the upgrade, access layer-core layer traffic (accounting for 40% of the traffic associated with that node) that previously needed to be forwarded through a traditional aggregation node can now be... Accessing the SDN network transforms traffic into controllable traffic, adding 38Gbps of controllable traffic, with the performance cost of its associated links being only 1.8 (not exceeding...). ),therefore It consistently ranked first throughout the iterations.

[0121] Convergence and Output: When all dual weights change and The algorithm terminates when the maximum number of iterations is reached. The node value score calculated in the final iteration round is then taken. As a priority for overall upgrades, the higher the score, the greater the contribution of node upgrades to "expanding the controllable traffic scale".

[0122] In this embodiment, after 138 iterations, the change in all dual weights was less than 0.01, indicating algorithm convergence. Ultimately, 19 aggregation layer nodes and 1 high-traffic access layer node entered the top 20 priority nodes. (Highest).

[0123] S3.3: Generate a priority list. After the algorithm converges, the value scores of each traditional node calculated in the final iteration will be used. As its upgrade priority. According to Sort the nodes from highest to lowest and select the top B (budget amount) nodes as the recommended upgrade node list for this round.

[0124] S4: Upgrade Execution. Network operations personnel execute the physical upgrade operation of replacing legacy nodes with SDN nodes based on the upgrade recommendation list output by the controller.

[0125] The aforementioned progressive SDN network upgrade method is deployed within the SDN controller. The SDN controller establishes a management connection with the hybrid network (containing both legacy and SDN network nodes) to determine the next batch of legacy network nodes to be upgraded. During the initial deployment phase, the SDN controller must first connect to the management plane of the hybrid network. Specifically, the controller connects via network cable to an access point on a pre-configured out-of-band management network port or management VLAN for the network. All legacy network nodes in the network are reachable through this management plane (i.e., the controller's IP address is routable to the management interface IP addresses of each node). The controller is pre-configured with the management IP addresses and access credentials (such as SNMP Community Strings) for these legacy nodes.

[0126] As shown in Figure 3, the SDN controller includes a topology discovery module, a traffic information module, and a deployment and upgrade module. The topology discovery module discovers the topology of the hybrid network using the LLDP protocol. It periodically collects, constructs, and stores the entire network's historical traffic matrix sequence through a hybrid approach combining SDN node flow statistics, traditional node port mirroring, SNMP statistics, and traffic inference. The traffic information module has a built-in least squares fitting unit to extract macroscopic features of the historical traffic matrix. It performs least squares fitting using a d-order polynomial (2≤d≤4), verifies the model's reliability based on the goodness-of-fit R² (R²≥0.8), and finally generates a predicted traffic matrix. The deployment and upgrade module has a built-in dual-ascending solver. Taking the predicted traffic matrix and network topology as input, it initializes link dual weights, iteratively calculates node upgrade value scores, updates link weights, and determines convergence conditions. It outputs the upgrade priority of each traditional node and a recommended list of upgrade nodes.

[0127] It should be noted that although the embodiments described above are illustrative, they are not intended to limit the invention. Therefore, the invention is not limited to the specific embodiments described above. Any other embodiments obtained by those skilled in the art under the guidance of this invention without departing from its principles are considered to be within the protection scope of this invention.

Claims

1. A progressive upgrade method for SDN networks, characterized in that, The process includes the following steps: Step 1: The controller obtains the complete topology of the hybrid network through the link layer discovery protocol, and periodically acquires network-wide traffic information using a differentiated collection strategy; then, it constructs a network-wide traffic matrix based on the network topology and network-wide traffic information; Step 2: The controller uses the least squares method to perform multi-order polynomial fitting on the time series of the collected historical network-wide traffic matrix, and predicts future traffic trends to obtain a predicted network-wide traffic matrix; Step 3: The controller uses the predicted network-wide traffic matrix and the network topology as input to construct an optimization problem with the main objective of maximizing the controllable traffic scale of SDN, and uses the dual ascent method to solve the optimization problem to obtain the upgrade priority score of each traditional node, thereby obtaining an upgrade recommendation list; Step 4: Network maintenance personnel perform physical upgrade operations to replace traditional nodes with SDN nodes according to the upgrade recommendation list output by the controller.

2. The progressive upgrade method for SDN networks according to claim 1, characterized in that, Step 1 employs a differentiated collection strategy to periodically acquire network-wide traffic information. Specifically, for SDN nodes in a hybrid network, the controller periodically collects flow table statistics of SDN nodes through the SDN southbound interface and, in conjunction with Packet-In message analysis, directly obtains the link transmission traffic involved by the SDN nodes. For traditional nodes in a hybrid network, the controller periodically collects the link inbound and outbound traffic of traditional nodes through port mirroring, and simultaneously obtains port statistics based on the Simple Network Management Protocol, indirectly inferring the traffic distribution among traditional nodes by combining the principle of traffic conservation.

3. The progressive upgrade method for SDN networks according to claim 1, characterized in that, The specific process of step 2 is as follows: Step 2.1: For each historical time point, first count the traffic of each node pair in the traffic matrix to obtain the total network traffic, and then calculate the proportion of traffic of each node pair relative to the total network traffic to obtain the traffic proportion of each node pair; Step 2.2: Based on the total network traffic of each historical time point obtained in step 2.1, select the total network traffic as the core fitting target, with time as the independent variable and the total network traffic as the dependent variable, and use a multi-order polynomial to perform least squares fitting to obtain the optimal multi-order polynomial coefficients, and then fit the multi-order polynomial; Step 2.3: Calculate the goodness of fit of the fitted multi-order polynomial: If the goodness of fit is greater than or equal to the set reliability threshold, the fitted multi-order polynomial is the final multi-order polynomial, and proceed to step 2.4; otherwise, supplement the historical traffic matrix and return to step 2.1 or adjust the order of the multi-order polynomial and return to step 2.2; Step 2.4: First, use the final multi-order polynomial to predict the predicted total network traffic at the next time point, then use the moving average of the traffic share of each node pair at the recent historical time points as the predicted traffic share of each node pair at the next time point, and finally use the predicted total network traffic at the next time point and the predicted traffic share of each node pair at the next time point to obtain the predicted network traffic matrix at the next time point.

4. The progressive upgrade method for SDN networks according to claim 1, characterized in that, In step 3, the optimization problem is: In the formula, Represents node pairs Predicted transmission traffic, A set of SDN nodes; For traditional node sets, Represents a node Decision variables ， Represents a node Upgraded Represents a node No upgrade Indicates the upgrade budget; Represents node pairs The minimum link capacity on the path, Indicates the maximum allowed link utilization; Represents nodes Connected links, Represents a node The set of outgoing edges, Represents a node The set of incoming edges, Indicates link Controllable flow Indicates link Uncontrollable traffic, A set of nodes; Represents node pairs Controllable flow Represents node pairs Uncontrollable traffic.

5. The progressive upgrade method for SDN networks according to claim 1, characterized in that, In step 3, the solution process of the dual ascent method is as follows: Step 3.1, Initialization: For each link Assigning initial dual link weights with performance constraints ： For each node pair Assign initial dual flow weights to controllable flow targets ： Step 3.2: Iterative evaluation: Calculate the node value score: ， Weight update: Dual link weight update: Dual link weight update: ， Convergence and Output: When all dual link weights change Less than the set threshold And the changes in weights of all dual links Less than the set threshold When the maximum number of iterations is reached, the node value score calculated in the final iteration round is taken as its comprehensive upgrade priority, where, It is a node pair The source node or the destination node, It is with nodes Directly connected links, For link The initial dual link weights, This is the link ratio coefficient. Links estimated using the traditional shortest path allocation method Initial link traffic, For link The capacity; For node pairs The initial dual traffic weights, This is the flow rate ratio coefficient. For node pairs Predicted transmission traffic; For nodes The Round value score, For node pairs The Wheel-to-wheel traffic weighting For node pairs The Controllable flow rate For node pairs The Controllable flow rate For link The Wheel-to-dual link weights, Links estimated using the traditional shortest path allocation method The Round-robin traffic, To maximize the allowed link utilization, This indicates the non-negation operation; For link The Wheel-to-dual link weights, This refers to the link step size parameter; For node pairs The Wheel-to-wheel traffic weighting This is the flow step size parameter.

Images