A region-constrained parallel self-evolving network topology generation method

Abstract

The present invention discloses a method for generating a regionally constrained parallel self-evolving network topology map, comprising the following steps: Step 1: Performing topology detection on the target area to obtain an IP-level topology; Step 2: Performing topology recovery to obtain router-level topology data; Step 3: Classifying the router-level topology data into three hierarchical categories: access layer routers, convergence layer routers, and backbone layer routers; Step 4: Using a neural network classification method to obtain key network nodes and set them as anchor points; Step 5: Dividing a complex network into multiple sub-regions based on administrative regions or interactive optimization requirements; Using a divide-and-conquer approach to perform a parallel layout for each sub-region; and Further dividing the layout evolution results into a core layer, convergence layer, and access layer. The present invention divides a complex network into multiple sub-regions and adopts a divide-and-conquer strategy to perform a parallel self-evolving layout for each sub-region, thereby achieving balanced node distribution and adaptability to geographical constraints.

Description

Technical Field

[0001] The present invention belongs to the field of computer technology, and in particular relates to a method for generating a region-constrained parallel self-evolving network topology graph. Background Art

[0002] With the rapid development of computer networks, Internet topology diagrams have become a crucial tool for understanding and managing complex network structures. Network topology diagrams graphically display the connections between various devices (such as routers and switches) on the Internet. Generating accurate and intuitive network topology diagrams is crucial for network planning, monitoring, and troubleshooting, especially in networks spanning multiple geographic regions or with multiple logical partitions.

[0003] Early research on topological graph generation focused on small datasets. Graph layout algorithms were primarily based on force-directed layout models and their derivatives, such as the Fruchterman-Reingold (FR) model and the Kamada-Kawai (KK) model. These models, driven by natural rule models (spring or charge forces), offer visually appealing visualizations of the overall graph structure and demonstrate excellent versatility. However, with the explosive growth of network data, traditional approaches have encountered bottlenecks in processing large datasets. The ability to depict and understand topological graphs far outstrips the speed of data generation, and they struggle to maintain a balance between readability and aesthetics. To address this challenge, multi-scale layout strategies, such as On Demand Layout (ODL) and the FM3 algorithm, have been designed to accelerate layout computations for large-scale networks. Meanwhile, domestic researchers are actively exploring hierarchical layout schemes, attempting to optimize layout performance through graph matching techniques and improved force-directed algorithms. Furthermore, some research has abstracted the layout problem into a function optimization problem, employing genetic algorithms and subsequent simulated annealing algorithms to explore optimal solutions, thereby overcoming local optima and improving computational efficiency. To solve the problem of representation confusion caused by large data scale, scholars such as Holten pioneered the "edge bundling" technology and proposed a layout algorithm that emphasizes both aesthetics and functionality. Especially with the assistance of graph data visualization tools such as Gephi, the ability to display graph layout results has reached a new high.

[0004] Although these network topology graph generation methods have been widely used in network topology visualization, they only focus on the logical layout of the network topology. Specifically, the existing methods have the following shortcomings:

[0005] (1) The network layout is only considered from the perspective of topological connection relationship and aesthetics, and fails to fully consider the node's own attributes, such as IP geographical location, subnet information, etc., which may cause the generated network topology map to deviate from the actual network distribution.

[0006] (2) Existing accelerated layout algorithms (such as multi-scale or hierarchical layout) sacrifice detailed information of the network at different levels of abstraction, resulting in the loss of key local structures and connection characteristics in high-level views, and the iterative layout is unstable.

[0007] (3) To handle the dynamic layout of large-scale networks and achieve dual optimization of layout in terms of aesthetics and functionality, the edge bundling technology and other methods currently introduced inevitably bring higher computational complexity, resulting in a slow overall layout generation process, which is not conducive to real-time interactive presentation and dynamic adjustment of large-scale networks.

[0008] (4) Existing layout presentation methods usually lack effective interactive means, and users cannot adjust the layout presentation of the network topology diagram according to specific needs, which affects the user's in-depth analysis and understanding of the network structure.

[0009] Traditional network topology layout methods often fail to accurately reflect the geographic location of nodes, resulting in discrepancies in the relative positions and lengths of nodes and edges within the layout. This discrepancy can mislead network managers and researchers in their understanding of network topology. Furthermore, as network scale increases, network layout becomes increasingly complex and challenging. While some accelerated layout methods, such as multi-scale or hierarchical layout techniques, have emerged, these sacrifice network details at different levels of abstraction and suffer from unstable iterative layouts. Furthermore, methods that balance aesthetics and functionality often suffer from high complexity, resulting in low layout efficiency and slow generation. Furthermore, existing layouts lack interactivity, resulting in a poor user experience and hindering in-depth analysis and understanding of the network structure. Therefore, a new network topology generation method is needed. By constraining the layout based on realistic node locations, a parallel, efficient topology generation method is designed that is both adaptable to large-scale networks and fully preserves key details. This method provides a stable and efficient layout while enhancing interactivity to improve the user experience and support in-depth network structure analysis. Summary of the Invention

[0010] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide a method for generating a parallel self-evolving network topology graph with regional constraints. By combining heuristic graph growth and dynamic force balance strategy, the complex network is divided into multiple sub-regions. A divide-and-conquer strategy is adopted to perform parallel self-evolving layout on each sub-region, so as to achieve the balance of node distribution and adaptability to geographical constraints.

[0011] The object of the present invention is achieved through the following technical solution: a method for generating a parallel self-evolving network topology graph with regional constraints, comprising the following steps:

[0012] Step 1: Perform topology detection on the target area to obtain the IP-level topology: Use the detection toolkit Scamper to obtain initial network measurement data and obtain the initial IP-level topology of the target area. The IP-level topology of the target area includes the node IP addresses and the path relationships between nodes.

[0013] Step 2: Use the alias resolution method to recover the IP-level topology of the target area and obtain the router-level topology data;

[0014] Step 3: Collect the attributes and connection attributes of each node in the router-level topology data. Then, use the k-shell decomposition method in complex networks to divide the router-level topology data into three hierarchical categories: access layer routers, aggregation layer routers, and backbone layer routers. The core routers of the three categories are defined as follows:

[0015] Access layer router: refers to the router that is directly connected to the user terminal in the network topology. It is located at the edge of the network and accesses the network through a single link.

[0016] Aggregation layer router: refers to a router in the middle layer of the network topology that is connected to the access layer router. This type of router acts as an intermediary between the access layer and the backbone layer routers, and is responsible for processing and forwarding network traffic from the access layer routers.

[0017] Backbone router: refers to the router at the core position in the network topology;

[0018] After the division is completed, the routers are assigned weights based on their level and importance in the topology: backbone routers are assigned a weight of 3, aggregation routers are assigned a weight of 2, and access routers are assigned a weight of 1. The weight of an edge is defined as the sum of the weights of the routers connected to it at both ends.

[0019] Step 4: Use the neural network classification method to determine the key mechanism nodes and set them as anchor points;

[0020] Step 5: Construct a parallel self-evolving network topology layout model: Divide the complex network into multiple sub-regions based on administrative regions or interaction optimization requirements. Adopt a divide-and-conquer approach to perform parallel layout for each sub-region. The layout evolution results are further divided into the core layer, aggregation layer, and access layer.

[0021] Step 6. Develop an interactive visualization tool for the network topology.

[0022] The topologically irregular convex polygon region adaptive layout algorithm verifies and adjusts node positions through multiple iterations to ensure that all nodes are located in the corresponding sub-regions and that the node distribution fits the polygon region. The algorithm includes three steps: parallel iterative layout, symmetric mapping correction, and layout result optimization.

[0023] Parallel iterative layout: The number of points in the region determines whether the region needs parallel processing. If parallel processing is required, a heuristic graph growth strategy is used to divide the region into multiple subgraphs. Each subgraph is then laid out in parallel. After processing each subgraph, it is merged into a large graph to form the entire region. If parallel processing is not required, the entire region is directly processed serially.

[0024] Symmetrical mapping correction: After each iterative layout, it is determined whether any nodes are located outside the overall area. If such a node exists, the nearest polygon boundary is found. Then, using this boundary as the axis of symmetry, the node is symmetrically mapped to the other side of the boundary. This new position is assigned to the node, replacing the previous one, and the next iteration is carried out.

[0025] Layout optimization: Perform a second layout to optimize the distribution of non-anchor points and avoid overlap. This process is based on the following steps:

[0026] Step 1: After the initial parallel iterative layout is completed, the non-anchor points with overlapping positions will be recorded while outputting the results;

[0027] Step 2: For all non-anchor points that overlap, a second layout is performed using a conventional force-directed algorithm based on the position. The center of the layout is the overlapping position, and the layout objects are all non-anchor points at that position.

[0028] The third step is to use the new position after the second layout to overwrite the original position of the overlapping points in the result file.

[0029] The beneficial effects of the present invention are:

[0030] 1. Based on the restored routing-level topology structure, the present invention adopts node IP positioning technology to add geographic attribute information to nodes and extract key nodes to form anchor points in the layout, so that the subsequent network layout can reflect the mapping relationship with the real location under regional constraints, thereby improving the accuracy and credibility of the network topology map.

[0031] 2. The present invention proposes a method for generating a parallel self-evolving network topology graph with regional constraints. Combining heuristic graph growth and dynamic force balance strategies, the complex network is divided into multiple sub-regions. A divide-and-conquer strategy is adopted to perform a parallel self-evolving layout on each sub-region, so as to achieve balanced node distribution and adaptability to geographical constraints.

[0032] 3. This invention has developed an interactive network topology visualization tool. By subdividing the layout into core, aggregation, and access layers, the tool displays the hierarchical logical structure and functional characteristics of the topology. Furthermore, the tool supports flexible region-based topology display and provides various interactive operations, such as zooming, panning, node highlighting, and path tracing, to enhance users' understanding of the network topology. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A schematic diagram of the structure of the Internet topology map generation framework of the present invention;

[0034] Figure 2 This is a flow chart of a method for generating a region-constrained parallel self-evolving network topology graph according to the present invention;

[0035] Figure 3 Query flow chart for IP geographic information database;

[0036] Figure 4 Flowchart of the graph partitioning algorithm;

[0037] Figure 5 Generate a flow chart for the parallel self-evolving network topology graph with region constraints;

[0038] Figure 6 This is a visualization result of the parallel self-evolving network topology graph with region constraints. DETAILED DESCRIPTION

[0039] The present invention aims to propose a method for generating a parallel self-evolving network topology map with regional constraints. This method first detects the network in the target area and uses topology recovery techniques to obtain the router-level topology infrastructure. Then, combined with technologies such as IP location, network node attributes (such as IP location and subnet information) are added. Based on the node attributes and connection (topology) properties, neural network classification and other methods are used to identify key organizational nodes, define them as anchor points, and construct an anchor point information database. These anchor points serve to fix the positions of important nodes during the self-evolution of the layout algorithm, ensuring the accuracy of the locations of important nodes. Next, a parallel self-evolving network topology map layout model is constructed. Using the anchor points as a benchmark, the complex network is divided into multiple sub-regions based on administrative regions or interactive optimization requirements. Using a divide-and-conquer approach, each sub-region is laid out in parallel. The self-evolving network topology map layout algorithm combines heuristic graph growth and dynamic force balancing strategies to self-iterate and optimize multiple target nodes, ensuring a balanced node distribution that conforms to the geographically constrained polygonal area. The layout evolution results were then further divided into the core layer, aggregation layer, and access layer, corresponding to different functions and node degree characteristics, to enhance the logical hierarchy of the topology diagram. Finally, an interactive visualization tool was developed for the network topology, allowing users to freely switch between global and local perspectives, customize display content, and deeply explore network elements, greatly enriching the analysis dimension and convenience.

[0040] This paper proposes for the first time a method for generating a region-constrained parallel self-evolving network topology graph, which mainly solves the following technical problems:

[0041] (1) Mapping constraint problem between network topology and actual geographic location: Traditional network topology layout methods cannot reflect the geographic spatial relationship of nodes. The present invention is based on the restored routing-level topology structure and adopts node IP positioning technology to add geographic attribute information to nodes and extract key nodes to form anchor points in the layout, so that the subsequent network layout can reflect the mapping relationship with the actual location under regional constraints, thereby improving the accuracy and credibility of the network topology map.

[0042] (2) Problem of efficient layout of network topology under irregular convex polygonal region constraints: In order to achieve efficient topological layout of large-scale networks that conforms to the actual geographical distribution, the present invention proposes a region-constrained parallel self-evolving network topology graph generation method. Combining heuristic graph growth and dynamic force balance strategy, the complex network is divided into multiple sub-regions. A divide-and-conquer strategy is adopted to perform parallel self-evolving layout on each sub-region, so as to achieve the balance of node distribution and adaptability to geographical constraints.

[0043] (3) Problems with the lack of readability and interactivity in network topology views: This paper develops a set of interactive network topology view tools. On the one hand, by subdividing the layout into the core layer, aggregation layer, and access layer for display, the hierarchical logical structure and functional characteristics of the topology map are reflected. On the other hand, it supports flexible region-based topology display and provides a variety of interactive operations such as zooming, panning, node highlighting, and path tracing to enhance users' understanding of the network topology structure.

[0044] In the Internet topology map generation framework of the present invention, a region-constrained parallel self-evolutionary network topology map generation method is implemented, and multiple databases are deployed to efficiently store key data in the process and provide necessary information support for each task.

[0045] The framework aims to generate high-quality network topology visualization effects. The framework structure is as follows Figure 1 As shown. Specifically, the framework includes the following modules: a topology detection and recovery module, which is used to perform IP traceroute detection and alias resolution on the target area; a network node information enhancement module, which is used to obtain the geographical location, subnet and other information of the network equipment, and associate it with the network topology hierarchy. By dividing the network topology according to regions or subnets, this module simplifies the topology structure, making it easier for network managers to obtain clear information when observing the whole and the part; the regional constrained parallel self-evolution network topology map generation and visualization module completes the visualization effect generation of the network topology map through parallel iterative layout, symmetric mapping, result optimization and inter-regional connection steps; the final framework output will display the visualization results of the network topology in the form of a map. The processed and analyzed network topology map will be displayed in the form of a map, which will help users better manage the network.

[0046] The technical solution of the present invention is further described below with reference to the accompanying drawings.

[0047] like Figure 2 As shown, the method for generating a region-constrained parallel self-evolving network topology graph of the present invention comprises the following steps:

[0048] Step 1: Perform topology detection on the target area to obtain the IP-level topology: Use the traceroute and ping-based detection toolkit Scamper to obtain initial network measurement data and obtain the initial IP-level topology of the target area. The IP-level topology of the target area includes the node IP and the path relationship between the nodes.

[0049] The detection part uses CAIDA's Scamper to perform large-scale Internet measurements. Scamper is a parallelized packet detector that can use a variety of different measurement techniques to perform large-scale Internet measurements.

[0050] Specifically, Scamper probes the input IP list using common detection methods such as traceroute and ping. It works by sending a probe packet to the target IP address and recording the packet's path. Using this path information, it determines the path from the probe source node to the target node.

[0051] Traceroute is a widely used technology for Internet exploration that can effectively reveal the path from a source node to a destination node. This tool primarily uses the Internet Control Message Protocol (ICMP) to collect information about each router node passed through, including their IP addresses and round-trip time (RTT).

[0052] Traceroute implements path tracing by specifically setting the Time to Live (TTL) value of the probe packets it sends. The TTL value of a packet determines the number of routers it can pass through in the network. Every time a packet passes through a router, its TTL value is reduced by 1. Once the TTL value of a packet drops to 0, it is discarded by the last router that processed it, and the router generates an ICMP Time Exceeded message and sends it back to the source address of the packet. In this way, by increasing the TTL value and sending a series of probe packets, traceroute can gradually reveal the complete path information taken to reach the destination network. This method not only provides network administrators with an effective means to diagnose network connection problems, but also has important significance for network security analysis and network performance evaluation.

[0053] Step 2: Preprocess the data: Use the alias resolution method to recover the IP-level topology of the target area and obtain the router-level topology data;

[0054] After acquiring a large amount of IP information for the target area, despite having the data foundation, it's still not possible to directly perform a topology analysis on the IP level. This is because this IP information may contain IP addresses belonging to different interfaces of the same router. Directly applying a topology algorithm, due to its random nature, can result in excessive distances between IP addresses on different interfaces of the same router, which is inconsistent with actual conditions. Therefore, topology recovery of this IP information is necessary.

[0055] Topology recovery involves aggregating IP addresses belonging to the same router, converting IP-level topology data into router-level topology data. Before topology recovery, each node in the data represents a single IP address; after topology recovery, each node in the data represents a router or terminal device, which may contain multiple IP addresses.

[0056] The function of topology recovery is to restore the IP-level data obtained by the probe to router-level data, so that it is more consistent with the actual network structure.

[0057] Step 3: Enhance Node Information: When generating a topology map, a core challenge is how to constrain the positioning of key nodes based on a force-directed layout, ensuring that their locations meet both geographic accuracy and visual aesthetics. To address this challenge, the first step is to collect the individual and connection properties of each node in router-level topology data. In city-level network topologies, the IP addresses, IP subnets, and corresponding geographic locations of government agencies, medical institutions, higher education institutions, and major commercial entities are generally readily available. These nodes play a crucial role in generating the city network topology.

[0058] At the same time, to address the problem of limiting the location of important nodes in force-directed layout, the present invention supports users with the ability to customize anchor points, allowing them to be manually added or adjusted. The anchor points remain fixed in position during the iterative process of the layout algorithm, ensuring the accuracy of the geographic locations of important nodes in the network topology. Generally speaking, in a city's router-level network topology layout, the number of anchor points accounts for approximately one-tenth of the total number of nodes. Therefore, during the layout process, approximately one-tenth of the anchor points are fixed, while the remaining nodes are iteratively adjusted using a parallel layout algorithm.

[0059] The properties and connection properties of these Internet nodes are all stored in the IP geographic information database. Its main function is to provide the geographic information of the required IP for each query. These geographic information include not only the latitude and longitude of the IP, but also the country, province, city and ISP (Internet service provider) information to which it belongs. During the processing, after obtaining a new node, for the IP information to be laid out, the point IP list will first be matched in the IP geographic information database. If the point information exists in the database, information is added to the point and the sub-area to which the point belongs is determined. If the information of the point does not exist, or the information is incomplete, the system will access the Internet IP information database through the interface query to query the point information. If the query is successful (the information of the IP is obtained), information is added to the point and the point information is stored in the database, so that the information in the database becomes richer and richer. If the query is unsuccessful, the point is discarded. In this way, the information in the IP geographic information database will be continuously updated and enriched each time a task is executed. The specific process is as follows Figure 3 shown.

[0060] On this basis, in order to make the nodes and edges in large-scale graph data have a clearer hierarchical division, the present invention is based on the three-tier architecture system of the network and uses the k-shell decomposition method in complex networks to divide the router-level topology data into three hierarchical categories: access layer routers, aggregation layer routers, and backbone layer routers; the three categories of core routers are defined as follows:

[0061] Access layer router: refers to the router that is directly connected to the user terminal in the network topology. It is at the edge of the network and accesses the network through a single link. Its main function is to enable user terminals to access the network, and to realize the aggregation and distribution of multi-terminal network flows and basic routing functions. It mainly consists of gateway routers for small local area networks.

[0062] Aggregation layer router: refers to a router in the middle layer of the network topology that is connected to the access layer router. This type of router acts as an intermediary between the access layer and backbone layer routers, responsible for processing and forwarding network flows from the access layer routers. It is often embedded with high-performance routing algorithms to achieve efficient packet forwarding and is mainly composed of high-performance enterprise-level routers.

[0063] Backbone routers: These routers are located at the core of the network topology, handling large, core services. These routers handle network services that the aggregation layer can't handle and connect to core servers (such as DNS root servers). Routers often feature reliability, efficiency, redundancy, fault tolerance, and network management capabilities. They primarily consist of large backbone routers.

[0064] After the division is completed, weights are assigned to routers according to their level and importance in the topology: backbone layer routers are assigned a weight of 3, aggregation layer routers are assigned a weight of 2, and access layer routers are assigned a weight of 1. Based on the weight values ​​of the routers at both ends of the connection, weight values ​​are assigned to the connection relationships (i.e., edges) in the topology data. The weight value of an edge is defined as the sum of the weight values ​​of the routers connected to it at both ends, ranging from 6 to 1.

[0065] Through this method, the original router-level topology data is transformed from an unweighted graph to a weighted graph, and the weights of nodes and edges in the graph are distinguished by color. The higher the weight of the nodes and edges in the graph, the greater their importance in the entire topology.

[0066] Therefore, by identifying the degree of nodes in the regional topology, we can categorize them into different layers. Node IP addresses are presented as IP segments. By comparing these segments with router-level topology data, we can identify key IP addresses in the network topology and assign them geographic location information.

[0067] Step 4: Based on the node's inherent and connectivity properties, such as geographic location, IP subnet information, and neighboring nodes, a neural network classification method is used to identify key organizational nodes and set them as anchor points. The anchor points remain fixed during the iterations of the layout algorithm, ensuring the accuracy of the geographic locations of important nodes in the network topology. Generally speaking, in a city's router-level network topology, the number of anchor points accounts for approximately one-tenth of the total number of nodes. Therefore, during the layout process, approximately one-tenth of the anchor points are fixed, while the remaining nodes are iteratively adjusted using a parallel layout algorithm.

[0068] Step 5: Construct a regionally constrained, parallel, self-evolving network topology layout model: Divide the complex network into multiple sub-regions based on administrative regions or interaction optimization requirements. Adopt a divide-and-conquer approach to perform parallel layouts for each sub-region. The layout evolution results are further divided into the core layer, aggregation layer, and access layer.

[0069] The region-constrained parallel self-evolving network topology map generation and visualization module receives two sets of input data: one is the router-level topology data obtained in the topology detection and recovery module, and the other is the data from the hierarchical interactive region partitioning and layout module, as well as the anchor point information and sub-region layout data of each sub-region after partitioning.

[0070] The attribute fields of the node are shown in the following table:

[0071] Field Name illustrate ip_num The number of IP addresses contained in this point ip_list The IP list contained in this point node_type Node type, indicating the level to which the node belongs isp_percentage List of ISPs to which IP addresses in the site belong is_anchor Is this point an anchor point? is_cross_region Is this point a cross-city point? is_cross_isp Is this point a cross-ISP point? is_backbone Is it a backbone node?

[0072] The attribute fields of the edge are shown in the following table:

[0073] Field Name illustrate edge.type The number of anchor points contained in the two endpoints of the edge corresponds to the different colors of the edge in the display edgeIndex Whether the point belongs to the key point in the entire area is_backbone Is it a backbone edge?

[0074] The overall idea of ​​the present invention uses the idea of ​​divide and conquer, which requires dividing the entire layout area into several sub-areas for layout, forming router topology subgraphs of multiple areas. The divide and conquer strategy will not only divide the entire layout area into several sub-areas for layout, but also record cross-regional nodes and their connection information. The basis for this division is usually administrative regions. For example, when laying out a country, it is divided into provinces; when laying out a province, it is divided into cities, from which backbone edges and nodes are extracted as upper-level data, and cross-regional nodes are recorded as the "portal nodes" of the region. The module is laid out in each sub-area after division, and the initial layout is completed by quickly processing sub-area data of different sizes through parallel division of regions. During the integration process, the information that can be obtained by each node is added to the node attributes and edge attributes as much as possible to enrich the information and facilitate network managers to mine detailed data.

[0075] The adaptive layout algorithm for topologically irregular convex polygonal areas verifies and adjusts node positions through multiple iterations to ensure that all nodes are located in the corresponding sub-areas and that the node distribution fits the polygonal area. The algorithm includes three steps: parallel iterative layout, symmetric mapping correction, and layout result optimization.

[0076] Parallel iterative layout: The number of points in the area determines whether the area needs parallel processing. If parallel processing is required, a heuristic graph growth strategy is used to divide the area into multiple sub-graphs, and then each sub-graph is laid out in parallel. After the processing of each sub-graph is completed, it is merged into a large graph to form the overall area. If parallel processing is not required, the entire area is directly processed serially.

[0077] Network topologies requiring parallel processing are typically large. The number of nodes within a region determines whether it warrants parallel processing. A heuristic graph growth strategy is employed to divide the subregion into multiple regions. The layout of each of these smaller regions is then performed in parallel. Nodes in the network iteratively adjust their positions in space by simulating physical forces (such as gravity and repulsion). As the algorithm progresses, the forces between nodes gradually reach equilibrium, resulting in a decrease in the distance moved by the nodes after each iteration.

[0078] Parallel processing essentially uses a graph partitioning algorithm to divide the layout area into multiple subgraphs, process them in parallel, and then merge them into a larger graph after processing. To achieve optimal layout efficiency, the workload on different processes should be as balanced as possible, while communication between them should be minimized. Therefore, when partitioning the network topology, it is necessary to ensure that the sum of node weights in different regions is as balanced as possible and that edge cuts between regions are minimized.

[0079] The graph partitioning algorithm has two stages: initial partitioning and information restoration and refinement. The specific process is as follows: Figure 4 The algorithm considers the balance of the number of nodes between different regions and the edge cuts between different regions. Under the premise of ensuring the balance of the number of nodes in each region, it minimizes the edge cuts between regions to maximize the parallel processing efficiency. The flow chart of the network topology layout of the graph partitioning algorithm is shown in the figure below. Figure 3 shown.

[0080] Since the basis for the layout area division of the present invention is the administrative area boundary, and in order to combine geographic information, the area to which the node belongs is limited by the information in the IP geographic information library. Under this limitation, the efficiency improvement of synchronous processing of sub-areas by parallel processing technology is not great. Because in a region, the number of nodes in central cities such as provincial capitals and capitals may be much larger than that of other ordinary cities, the uneven number of nodes in each sub-area greatly reduces the efficiency improved by parallel processing. Based on this limitation, the present invention applies parallel processing technology to non-anchor points in a single sub-area. When each sub-area is laid out, its efficiency is improved by parallel processing technology, thereby improving the efficiency of the overall layout.

[0081] First, the number of points in the region is used to determine whether the region requires parallel processing. If so, the region division k is calculated. The network topology is then simplified. Network topologies that require parallel processing are usually large in scale, and performing region division directly on the original image takes a long time. Therefore, the network topology needs to be simplified before performing region division. Whether region division is necessary, the present invention tested the processing speed of server hardware for a parallel self-evolving network topology map generation and visualization module that deploys region constraints. Without considering inter-region communication, it was found that the efficiency of a single process processing points is highest when the number is around 5000, but the specific efficiency fluctuates depending on the number of edges. Therefore, when laying out each sub-region, the number of points in the sub-region can be divided by 5000 to obtain the optimal processing efficiency division number k for that sub-region. For sub-regions with more than 5000 points, an algorithm is used to perform k region division; for sub-regions with less than 5000 points, serial processing is performed directly. In this way, the processing speed of the sub-region can be maximized.

[0082] In the coarsening stage, the network topology is simplified by node matching, and the two matched nodes are merged to form the graph G. i Simplify to G i+1 From G i Select an edge set M from i , for the edge set M i For any two edges e(u, v) and e(m, n), nodes u, v, m and n are different nodes. i There is no one who can meet the conditions to join M i When the edge set M i It is called the maximum edge set. Due to different strategies for selecting edges, there may be different M i . Get the maximum edge set M i Then, for the edge set M i For any edge e(u, v), merge nodes u and v, and the weight value of the merged node u′ is w′ uIt is the sum of the weight values ​​of the two nodes u and node v before the merger. The edge weight between the merged node u′ and the adjacent node v′ is the sum of the edge weights of the corresponding nodes before the merger.

[0083] Figure G i The set of all edges E i The sum of all edge weights is W(E i ), Figure G i+1 The set of all edges E i+1 The sum of all edge weights is W(E i+1 ), maximum matching M i The sum of the weights of all edges in is W(M i ), since graph G i+1 It is from Figure G i Directly merge matching M i All edges in are obtained, so W(E i+1 )=W(E i )-W(M i ), in order to obtain smaller edge cuts during the initial partitioning, graph G i+1 The sum of the edge weights W(E i+1 ) is as small as possible, and it is necessary to select the edges with larger weights and meet the conditions to join M i .

[0084] Maximum matching M i The strategy for selecting the middle edge is: from graph G i Randomly select a node v that does not match M i On any edge of v, select an edge e(v,u) with the largest weight from the adjacent edges of v, and node u is not in the matching M i On any edge of M, add edge e(v,u) to M i , repeat from graph G i Select nodes until M i For graph G i The maximum matching of .

[0085] After obtaining the coarsened network topology, the entire network topology needs to be initially partitioned. Since vertex and edge weights are merged into new nodes and edges during the coarsening process, performing regional partitioning on the coarsened graph does not lose the vertex and edge weight properties of the original graph. This paper uses a heuristic graph growth algorithm to perform k-region partitioning on the subregions. It then performs layouts on each of the smaller regions in parallel, and finally restores the layout to that of the subregion.

[0086] The idea of ​​the heuristic graph growth algorithm is that when there are nodes in the graph G that have been divided into regions and nodes that have not been divided into regions, the nodes in the divided regions belong to k regions. The algorithm continuously selects points that can reduce the most edge cuts from the nodes in the undivided regions and adds them to one of the k regions. At the same time, in order to ensure the balance of point weights between different regions, it is necessary to set an additional balance parameter c so that the sum of the points in each region is less than Finally, the graph G is divided into k regions. Assume that a k-region partition already exists for the graph G, the region where point v is located is P(v), and e(u, v) is an edge from v to any other point u connected to v. Then, the edge cuts that can be reduced by moving point v to the region numbered i are shown in the following formula:

[0087]

[0088] The pseudo code of the heuristic graph growth algorithm is as follows:

[0089]

[0090] The heuristic graph growth algorithm needs to allocate each node in the network topology and update the gain value of other nodes after a single node is allocated. Therefore, its time complexity is O(n 2 k), where n represents the number of nodes and k represents the number of regions to be divided. Based on the heuristic graph growth algorithm, the non-anchor point region division in the subregion can be directly obtained, and this division can ensure that the sum of the point weights between each region is as balanced as possible.

[0091] In the process of restoring the coarsened network topology to the original network topology, the boundary nodes need to be re-divided into regions. Gain(v,i) is still used as the main parameter to consider when moving node v to other regions. The pseudo code of the refinement process algorithm is shown below.

[0092]

[0093]

[0094] Figure G i+1 Restore to graph G i , for graph G i For nodes on the boundary, recalculate the possible reduction of edge cut values ​​when the nodes are divided into other areas, and consider the point weight balance between different areas during the movement process. When there is no node that can reduce the edge cut, the graph G is obtained. i The partition result P i, repeat the refinement process and finally get the region division result of the original topology. For the heuristic boundary algorithm, when performing node exchange, the candidate nodes are nodes on the region boundary, with a maximum number of n. After selecting a node for region transfer, the node will not be transferred to the region again. At the same time, the gain value of the adjacent node of the node needs to be updated, so its time complexity is O(n 2 k), where n represents the number of nodes and k represents the number of regions to be divided.

[0095] According to the above steps, the k-region partitioning of non-anchor points in the sub-region can be obtained. At the same time, the partitioning can ensure that the sum of the point weights between each region is as balanced as possible, and the node division and initial layout of each region are completed.

[0096] Symmetric Mapping Correction: As networks continue to expand and become increasingly complex, traditional static graphs are no longer sufficient to fully capture the network's full picture. Therefore, this paper builds an interactive system based on the principles of layering and divide-and-conquer, aiming to provide a two-layered perspective and help viewers gain a deeper understanding of network topology. This system not only extracts backbone nodes and edges into a higher-level graph but also provides multiple viewing options, allowing viewers to freely select network sections of interest for in-depth analysis.

[0097] In addition to displaying the core structure in layers, this invention also supports interactive options to provide more detailed information and statistical information. First, the system monitors mouse click events, allowing network administrators to easily obtain detailed information about the desired nodes, including IP addresses, longitude and latitude, and node degrees. Second, during the layout process, nodes and edges are assigned different types and distinguished by visual features such as size, height, and color. This allows network administrators to intuitively identify nodes and edges of the same type, effectively avoiding homogeneity when dealing with large amounts of data.

[0098] Finally, the system optimizes the statistical display of the ISP attributes to which the nodes belong. As a key node attribute, ISPs influence network connection quality, data transmission costs, and network access control and regulation. During the layout operation, the system not only counts and displays the top 10 ISPs with the most nodes in the target area and their affiliated nodes, but also provides options for network managers to select. By clicking on an ISP, users can display only the nodes and edges of that ISP, providing a deeper understanding of its service scope, number of users, and other information. This interactive approach provides network managers with a richer perspective, enabling a more comprehensive understanding of network topology, and providing strong support for network management and optimization.

[0099] The task status library plays a crucial role through user interaction. Its core function is to receive status updates from various modules within the system and ensure that this information is reflected in real time on the terminal interface. This design allows users to understand the current progress of tasks anytime, anywhere, and make necessary operations and adjustments based on this real-time information. Throughout the task execution process, users can customize the display area to suit their needs, adapting to different information display requirements. Modules such as topology detection and recovery upload their results and status to the status information library in real time. The status information library then aggregates this status information and outputs it to the terminal interface, ensuring that users can view and monitor task execution at any time, thereby achieving transparent management and instant feedback on the entire task process.

[0100] How can we simplify the topology without losing data, allowing network managers to observe the overall network structure from both a global and regional perspective while also obtaining detailed information when observing local areas? To solve this problem, this paper adopts the divide-and-conquer plus layering approach and improves the force-directed algorithm to adapt it to the irregular sub-region layout after layering.

[0101] After the layout of each sub-area is completed, the core nodes and edges are extracted and displayed in the overall area using a layered approach. Interactive options are provided, allowing viewers to choose which perspective to observe. If the overall perspective is selected, only the core information extracted from the overall layout area is visible, making it easier for network managers to grasp the overall network structure. Selecting a local perspective allows further observation of the entire layout information of each sub-area, allowing network managers to explore detailed information.

[0102] When adopting the hierarchical and divide-and-conquer approach to simplify the network topology of geographic information, the traditional FR force-directed algorithm is no longer applicable. For the traditional FR force-directed algorithm, the displacement vector of each node is calculated based on the net force (the net force of gravity and repulsion, calculated by a given formula) received by the node in each iteration, and the node position is updated. Figure 5 Therefore, an initial temperature is set at the beginning of each iteration. With each iteration, the temperature is lowered according to a specific rule, and the node displacement distance is limited by the temperature. As the iteration progresses, the temperature decreases, and with it, the average node displacement decreases until it reaches the set value, at which point the iteration is exited. Traditional force-directed algorithms produce rectangular or circular layouts, but in a map context, aesthetically pleasing areas defined by administrative divisions must be filled as fully as possible. Therefore, improvements to force-directed algorithms are needed.

[0103] Specifically, after each iteration of layout, the present invention determines whether any node is located outside the overall area. If such a node exists, the method finds the polygonal boundary closest to the node. Using this boundary as the axis of symmetry, the node is symmetrically mapped to the other side of the boundary, and the new position is assigned to the node, replacing the previous one, before the next iteration. The process is as follows:

[0104]

[0105]

[0106] This approach ensures that all nodes are within the designated area after layout. Due to the characteristics of FR layout, the overall node movement distance after each iteration is smaller than that of the previous iteration. Therefore, combined with the correction of symmetric mapping, the node distribution within the sub-area after layout is more closely aligned with polygonal shapes, rather than simple rectangles or circles.

[0107] At this point, the improved force-directed layout algorithm combined with the layered and divide-and-conquer ideas can display the overall network structure while retaining the overall topology information, compared with the traditional topology simplification method, and has both overall and detailed characteristics.

[0108] Optimizing layout results: Combining parallel iterations of the improved force-directed layout algorithm with symmetric mapping corrections, the algorithm generates a network topology with uniform node distribution that conforms to the regional shape. This means that the final layout will not be a simple one where all nodes are evenly distributed within a rectangular or circular area, but rather one where node positions and distributions are more realistic, potentially irregular, geographical or administrative divisions. Overlapping non-anchor points are expanded, and these overlapping points are also found when the initial layout is traversed and the results are generated.

[0109] At this point, the improved force-directed layout for this sub-region has been completed. However, in actual operation, considering that a large number of nodes may overlap when laying out a large number of nodes in an area of ​​limited size, if the results are output directly, the data may differ from the amount of graph data intuitively perceived in the visualization results. To solve this problem and make all non-anchor points distributed in different positions as much as possible, we introduced a second layout. The purpose of the second layout is to optimize the distribution of non-anchor points and avoid position overlap; this process is based on the following steps:

[0110] Step 1: After the initial parallel iterative layout is completed, the non-anchor points with overlapping positions will be recorded while outputting the results;

[0111] In the second step, for all non-anchor points that overlap, a second layout is performed using the conventional force-directed algorithm based on the position. The center of the layout is the overlapping position, and the layout objects are all non-anchor points at that position. Repeat the above process for all overlapping positions.

[0112] The third step is to use the new position after the second layout to overwrite the original position of the overlapping points in the result file.

[0113] Through the above steps, the second layout ensures the reasonable distribution of non-anchor points, making the visualization results more intuitive and the data display more accurate, and solving the visualization deviation caused by point overlap. The visualization results of the parallel self-evolution network topology with regional constraints are as follows: Figure 6 shown.

[0114] The force-directed algorithm provides a solution model for network graph layout in a general sense through adaptive iteration, and makes the drawing of network topology have relatively good aesthetic characteristics. Subsequent practical applications and improvements of the force-directed algorithm combine the structural characteristics of the network topology itself, such as community, so that the special topological structure of the network data itself can be better displayed on the basis of satisfying the aesthetic effect.

[0115] Combining the parallel iterations of the improved force-directed layout algorithm with symmetric mapping corrections, the algorithm generates a network topology with uniform node distribution that conforms to the regional shape. This means that the final layout will not be a simple one where all nodes are evenly distributed within irregular regions, but rather one where the node locations and distribution are more consistent with realistic, potentially irregular, geographical or administrative divisions. If a sub-region uses parallel layout, the layout is restored to the actual layout of the sub-region. After obtaining the actual sub-region layout structure, the layout algorithm specifically handles the connection information of cross-region nodes. The cross-region nodes recorded during the partition process are marked as "portal nodes" for the region, and the relationship edges between regions are completed based on the recorded connection information.

[0116] Step 6. Develop an interactive viewing tool for network topology, allowing users to freely switch between global and local perspectives, customize display content, and deeply explore network elements, greatly enriching the analysis dimension and convenience. By self-evolutionarily generating complex regional constraint network structures and integrating user interaction mechanisms, the practicality and flexibility of visualization are enhanced. The tool supports detailed three-level (backbone layer, aggregation layer, and access layer) network structure display and various interactive operations such as zooming, panning, node highlighting, and path tracing. Through global, local, and customized view overviews, the tool hierarchically displays the core layer nodes of the network and their key connection paths, the configuration details of nodes and edges in the region, and the density of the region. In addition to standard presets, users can also customize settings based on custom anchor points and custom regions.

[0117] This paper presents a method for generating a parallel self-evolving network topology map based on regional constraints. To enhance the usability and flexibility of network topology maps, a set of interactive visualization tools is designed, allowing users to select different viewing angles and levels of detail based on analysis needs. This method enables the classification of connection relationships between nodes in the network, with the following advantages:

[0118] Select the Global View: This view shows the core architecture of the entire network, including backbone nodes and the main connections between them. This view is suitable for quickly understanding the overall structure of the network and the main traffic paths.

[0119] Select a local view: Drill down into a specific sub-area to observe the detailed layout of all nodes and edges within that area.

[0120] Dynamically switch perspectives: Users can dynamically switch between global and local views as needed to understand both the macrostructure and microbehavior of the network.

[0121] Customized views: In addition to the preset global and local views, users can also customize views based on specific analysis objectives, selectively displaying or hiding certain network elements.

[0122] Interactive exploration: Interactive options also include exploration of network elements. For example, users can click on a backbone node to see its connections and impact range in the network.

[0123] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and it should be understood that the scope of protection of the present invention is not limited to such specific descriptions and embodiments. Those skilled in the art can make various other specific variations and combinations based on the technical teachings disclosed in the present invention without departing from the essence of the present invention, and such variations and combinations are still within the scope of protection of the present invention.

Claims

1. A method for generating a region-constrained parallel self-evolving network topology graph, characterized in that: The steps are: Step 1: Perform topology detection on the target area to obtain the IP-level topology: Use the detection toolkit Scamper to obtain initial network measurement data and obtain the initial IP-level topology of the target area. The IP-level topology of the target area includes the node IP addresses and the path relationships between nodes. Step 2: Use the alias resolution method to recover the IP-level topology of the target area and obtain the router-level topology data; Step 3: Collect the attributes and connection attributes of each node in the router-level topology data. Then, use the k-shell decomposition method in complex networks to divide the router-level topology data into three hierarchical categories: access layer routers, aggregation layer routers, and backbone layer routers. The core routers of the three categories are defined as follows: Access layer router: refers to the router that is directly connected to the user terminal in the network topology. It is located at the edge of the network and accesses the network through a single link. Aggregation layer router: refers to the router in the middle layer of the network topology that is connected to the access layer router; This type of router acts as an intermediary between access layer and backbone layer routers, responsible for processing and forwarding network flows from access layer routers; Backbone router: refers to the router at the core position in the network topology; After the division is completed, the routers are assigned weights based on their level and importance in the topology: backbone routers are assigned a weight of 3, aggregation routers are assigned a weight of 2, and access routers are assigned a weight of 1. The weight of an edge is defined as the sum of the weights of the routers connected to it at both ends. Step 4: Use the neural network classification method to determine the key mechanism nodes and set them as anchor points; Step 5: Construct a parallel self-evolving network topology layout model: Divide the complex network into multiple sub-areas based on administrative regions or interaction optimization requirements; adopt a divide-and-conquer approach to perform parallel layout for each sub-area; and further divide the layout evolution results into the core layer, aggregation layer, and access layer.

2. The method for generating a region-constrained parallel self-evolving network topology graph according to claim 1, characterized in that: The parallel self-evolving network topology layout model adopts an adaptive layout algorithm for topologically irregular convex polygonal regions. Through multiple iterations, the node positions are verified and adjusted to ensure that all nodes are located in the corresponding sub-regions and that the node distribution fits the polygonal region. The algorithm includes three steps: parallel iterative layout, symmetric mapping correction, and layout result optimization. Parallel iterative layout: The number of points in the region determines whether the region needs parallel processing. If parallel processing is required, a heuristic graph growth strategy is used to divide the region into multiple subgraphs. Each subgraph is then laid out in parallel. After processing each subgraph, it is merged into a large graph to form the entire region. If parallel processing is not required, the entire region is directly processed serially. Symmetrical mapping correction: After each iterative layout, it is determined whether any nodes are located outside the overall area. If such a node exists, the nearest polygon boundary is found. Then, using this boundary as the axis of symmetry, the node is symmetrically mapped to the other side of the boundary. This new position is assigned to the node, replacing the previous one, and the next iteration is carried out. Layout optimization: Perform a second layout to optimize the distribution of non-anchor points and avoid overlap. This process is based on the following steps: Step 1: After the initial parallel iterative layout is completed, the non-anchor points with overlapping positions will be recorded while outputting the results; Step 2: For all non-anchor points that overlap, a second layout is performed using a conventional force-directed algorithm based on the position. The center of the layout is the overlapping position, and the layout objects are all non-anchor points at that position. The third step is to use the new position after the second layout to overwrite the original position of the overlapping points in the result file.

Images