Space-air heterogeneous network cross-domain link selection and message distribution method

Abstract

The application provides a space-air heterogeneous network cross-domain link selection and message distribution method, and belongs to the field of communication ad hoc networks. The method collects cross-domain path capabilities and traffic conditions of each subnet through heterogeneous network situation awareness technology, and the obtained network situation information provides a basis for subsequent link planning. The method evaluates the capabilities of each path through the heterogeneous network cross-domain link selection technology, comprehensively considers link delay, packet loss and network load, selects an optimal cross-domain path, meets the communication requirements of services, and realizes network load balancing. The method realizes cross-domain message distribution through protocol format conversion of cross-domain data packets by a routing node through a segmented routing technology. The application can realize interconnection and intercommunication of space-air heterogeneous networks, reduces network overhead of message forwarding between heterogeneous networks, fully utilizes limited network resources, achieves load balancing, and provides an effective networking technical solution for emergency rescue and other fields.

Description

Technical Field

[0001] This invention belongs to the field of ad hoc communication networks, specifically relating to a method for cross-domain link selection and message distribution in heterogeneous space-air networks. Background Technology

[0002] Following natural disasters, basic communication infrastructure such as base stations and access points is highly vulnerable to damage. The shutdown of this infrastructure can paralyze communication networks in disaster-stricken areas, severely hindering rescue and relief efforts. Since such infrastructure cannot be repaired quickly, rapidly and accurately deploying communication relays to establish temporary relief communication networks is a feasible and efficient method in emergencies. By using handheld devices by rescue personnel and deploying aerial units to cover communication blind spots, mobile ad hoc networks can be rapidly established to ensure communication support for post-disaster reconstruction.

[0003] However, the construction of the aforementioned temporary communication network requires all communication units to be equipped with the same basic communication equipment. In emergency scenarios such as disaster relief, communication units may come from various sectors of society, and their communication equipment varies. It is impractical to uniformly equip and deploy communication infrastructure for them in large-scale emergency scenarios. Therefore, communication units from different equipment vendors and service providers will build their own subnets. Subnets can communicate seamlessly with each other, but due to differences in communication systems, routing protocols, and other basic configurations, efficient information exchange between subnets will be difficult. Traditional methods achieve cross-subnet communication through route redistribution, but this method incurs significant network overhead. Furthermore, due to limited communication resources in emergency scenarios, after solving the fundamental problem of heterogeneous network interconnection, how to efficiently and rationally utilize limited communication resources and achieve load balancing is also an important issue that needs to be considered.

[0004] In summary, developing an application technology that is easy to deploy and scalable to enable message flow across subnets, achieve the integration of heterogeneous networks, and adaptively adjust forwarding paths according to network conditions to achieve load balancing has high practical value and is an important research task at present. Summary of the Invention

[0005] In the process of heterogeneous network integration, traditional methods for interconnecting heterogeneous subnets under different communication systems suffer from high network overhead and the inability of cross-domain packet forwarding paths to adapt to changes in network status and transmission requirements. To address these issues, this invention proposes a cross-domain link selection and message distribution method for aerospace heterogeneous networks. By constructing a virtual backbone network to perform status detection and cross-domain link path planning for each subnet, and by using segmented routing technology to achieve cross-domain message forwarding, heterogeneous network integration is realized.

[0006] This invention provides a method for cross-domain link selection and message distribution in heterogeneous aerospace networks, applicable to network scenarios composed of multiple heterogeneous subnets. The method first selects gateway nodes in the heterogeneous network to construct a virtual backbone network. Each gateway node is equipped with the protocol stack of its respective subnet. Then, functional modules implementing this method are deployed between the application layer and transport layer of the TCP / IP protocol stack at the gateway node. The method includes the following steps:

[0007] Step 1: Perform subnet situational awareness, including traffic probing, latency probing, and packet loss rate probing. Traffic probing obtains the traffic of data packets sent within each subnet, and latency probing and packet loss rate probing obtain the link-level latency and packet loss rate between adjacent subnets. Perform subnet traffic prediction, including dividing each subnet into blocks and using a neural network to predict traffic for each subnet. The input of the neural network is the historical traffic matrix of the subnet, and the output is a predicted future traffic matrix. The matrix records the traffic between each block of the subnet.

[0008] Step 2: The cross-domain forwarding of data packets is divided into the selection of the gateway node and the forwarding of nodes within the subnet. The gateway node of the subnet where the source node of the data packet is located plans the cross-domain link for the data packet according to the virtual backbone network topology it maintains, and determines the next transit gateway node. When the data packet is transited at each gateway node, it is formatted according to the routing protocol of the subnet to which it is to be transferred before being forwarded.

[0009] Step 3: When there are more than one cross-domain forwarding path for a data packet, score each cross-domain forwarding path by probing the link-level latency and packet loss rate between subnets and the predicted subnet traffic matrix, and select the path with the highest score as the cross-domain forwarding path for the data packet.

[0010] Step 1, which involves subnet situational awareness, includes: (1) Traffic probing: Nodes within each subnet periodically acquire the data packet traffic and target node addresses sent by their own application layer, and report them to the subnet's gateway node; (2) Latency probing: Probing the link-level latency between logically adjacent gateway nodes; Each gateway node maintains a probing table, storing the latest time of sending data packets to logically adjacent target gateway nodes; If the sending time of a target gateway node in the table exceeds a set duration, the gateway node actively initiates a latency maintenance data packet to measure the link-level latency; When a gateway node generates a data packet that arrives at a target gateway node in the table, it adds a header to the data packet. The frame is used to record the sending time of the packet. When the data packet is forwarded to the target gateway node, the target gateway node calculates the link-level delay from the previous gateway node to this node. Logically adjacent gateway nodes refer to two nodes that cross only one subnet between them. (3) Packet loss rate detection: Each gateway node counts the data packets that arrive at each logically adjacent gateway node and counts the data packets received from each logically adjacent gateway node. Each gateway node periodically sends an ACK message to each logically adjacent gateway node. The message records the total number of data packets received by this node from the other party in the previous period and restarts the data packet counting process. After each gateway node receives the ACK message, it calculates the packet loss rate.

[0011] The gateway node of the present invention plans and selects cross-domain forwarding paths for different target subnets in advance in each period through periodic updates, and generates and stores a routing table. Whenever a data packet is sent, the cross-domain forwarding path is obtained by directly querying the routing table. When the period is updated, the cross-domain forwarding paths for different target subnets are re-planned and selected, and the routing table is updated.

[0012] Compared with existing technologies, the advantages and positive effects of the method of this invention are as follows:

[0013] (1) The method of the present invention summarizes the cross-domain path capabilities and traffic of each subnet through a virtual backbone network composed of gateway nodes, detects the network operation status, and provides a data foundation for subsequent link planning; through segmented routing technology, the cross-domain forwarding of data packets is divided into the selection of the gateway node and forwarding within several subnet segments, achieving interconnection between heterogeneous subnets running under different protocol stacks with low network overhead; when the network status changes, a path adaptive adjustment mechanism is introduced to select the optimal path under the current status, ensuring the cross-domain flow of business messages.

[0014] (2) The method of the present invention realizes the protocol format conversion of cross-domain data packets through the gateway node, which significantly reduces the network overhead of cross-domain message distribution; through cross-domain link selection technology, cross-domain forwarding links are flexibly selected according to business needs and actual network conditions, and cross-domain path configuration of data packets is optimized in a targeted manner; through network situation awareness and cross-domain link selection to jointly optimize path configuration, when network congestion or abnormality occurs, the path configuration can be changed in a timely manner according to the network situation, so as to achieve load balancing of limited network resources.

[0015] (3) The method of this invention significantly reduces the network overhead of cross-domain message distribution and achieves load balancing of limited network resources, which has important theoretical significance and wide application value. This invention is applicable to cross-domain message forwarding tasks in heterogeneous mobile ad hoc networks in space and air, and provides an effective networking technology solution for fields such as emergency rescue. Attached Figure Description

[0016] Figure 1 This is an implementation architecture diagram of the cross-domain link selection and message distribution method for heterogeneous aerospace networks of the present invention;

[0017] Figure 2 This is a schematic diagram illustrating the deployment of the cross-domain link selection and message distribution method for heterogeneous aerospace networks of the present invention in a specific scenario.

[0018] Figure 3 This is a diagram illustrating a specific scenario of cross-domain link latency measurement according to the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0020] like Figure 1 As shown, the implementation of the cross-domain link selection and message distribution method for heterogeneous aerospace networks of the present invention mainly includes two parts: heterogeneous network situational awareness technology and heterogeneous network cross-domain link selection and message distribution technology. The heterogeneous network situational awareness technology constructs a virtual backbone network to undertake situational awareness tasks, providing support for cross-domain link selection and message distribution. The heterogeneous network cross-domain link selection and message distribution technology is also deployed in the backbone network nodes, selecting forwarding links for cross-domain data packets based on situational awareness information and constructing cross-domain paths through segmented routing technology. The method of the present invention realizes the interconnection and interoperability of heterogeneous aerospace networks, reduces the network overhead of message forwarding between heterogeneous networks, and simultaneously achieves full utilization of limited network resources, achieving load balancing.

[0021] like Figure 2 As shown, one deployment scenario of this invention consists of five heterogeneous subnets: A, B, C, D, and E. The gateway node selected among the cross-subnet nodes is as follows: Figure 2 The red nodes in the middle form a virtual backbone network. The method of this invention is implemented as an application software or device embedded between the application layer and the transport layer of the TCP / IP protocol stack of the virtual backbone network node, which undertakes the situational awareness of each subnet and the cross-domain link planning task, and realizes the management of the entire network.

[0022] The heterogeneous network situational awareness technology of this invention comprises three main parts: cross-domain virtual backbone network construction, subnet situational awareness, and subnet traffic prediction. The virtual backbone network, as the main body, undertakes the situational awareness task of the entire network and is responsible for subsequent link planning; subnet situational awareness statistically analyzes the cross-domain link performance of each subnet; and subnet traffic prediction statistically analyzes the traffic level within the subnet.

[0023] Step 1.1: Constructing a Virtual Backbone Network. To achieve situational awareness in heterogeneous networks, several nodes need to be selected within the network to undertake awareness tasks. Each subnet selects a gateway node, and they are interconnected through multi-hop links within the subnets to jointly form a virtual backbone network to assume responsibilities. The selection and maintenance rules for gateway nodes are as follows: the node with the highest connectivity in the intersection of nodes in adjacent subnets is selected as the channel between two adjacent subnets, i.e., the gateway node. A gateway node is located in multiple subnets simultaneously, equipped with the protocol stacks of each subnet, and can communicate with each subnet it belongs to, undertaking the responsibility of cross-subnet communication. The gateway node periodically notifies the other nodes within each of its subnets of its gateway identity.

[0024] Step 1.2: Subnet Situation Awareness. Heterogeneous subnets operate in complex environments, and their overall situation is constantly changing. To achieve network load balancing and improve overall network throughput during periods of high communication traffic, it is first necessary to collect information on the network's operational status. Information on heterogeneous subnets is mainly collected in two parts: node attributes and edge attributes. Node attributes include the traffic generation of nodes in the recent period; edge attributes include edge latency and packet loss rate. Specifically, node traffic data is used as the raw traffic data for subsequent traffic prediction; multi-hop link latency between gateway nodes is used as a speed indicator for subsequent evaluation of cross-domain subnets; and multi-hop link packet loss rate between gateway nodes is used as a reliability indicator for subsequent evaluation of cross-domain subnets. In this embodiment of the invention, subnet situation awareness mainly includes the following three aspects.

[0025] (1) Traffic Probing: Due to limited network resources, in order to allocate as many communication resources as possible to service traffic, the principle of minimizing the amount of traffic measured should be followed when probing network information. Since the module designed in this invention is located in the middle layer of the communication protocol stack of the gateway node, when the application layer generates a data packet, the module can obtain the target node of the data packet. In this embodiment of the invention, the period is set to 1 second. When the application layer sends a data packet, the functional module of this invention records all data packets sent by the application layer within 1 second from the current time and their target node addresses. Nodes in the subnet report the currently recorded traffic and target addresses to the gateway node in the subnet periodically, so as to provide raw data for the subsequent generation of the traffic matrix.

[0026] (2) Delay Probing: To minimize measurement traffic, service traffic should be fully utilized. When a node in a subnet generates a service data packet destined for another gateway node, a frame recording the packet's transmission time is added to the packet header. When the data packet is forwarded to the target gateway node, the target gateway node can obtain this timestamp and calculate the link-level delay from the previous gateway node to itself. However, due to the randomness of service packet generation and the diversity of transmission paths, there may be no service packet exchange between two gateway nodes for a long time. To address link delay probing between gateway nodes with no service exchange for extended periods, each gateway node in the subnet should maintain a data packet transmission list, storing data packet transmission information with other gateway nodes in its respective subnet. The entries include the addresses of the other gateway nodes and the latest time the data packet was sent to the corresponding gateway node. Figure 3 Gateway node 1 in the intersection of network A and network B stores the transmission information of its two logically adjacent gateway nodes. Logically adjacent nodes mean that the nodes are separated by only one subnet. Figure 3 For example, gateway node 1 is logically adjacent to gateway nodes 2 and 3, but gateway nodes 2 and 3 are not logically adjacent. If gateway node 1 does not communicate with its adjacent gateway nodes for a set time (PacketTIME), gateway node 1 will proactively initiate a latency maintenance data packet to measure the link latency.

[0027] (3) Packet Loss Rate Detection: A timed feedback mechanism is introduced. When a gateway node generates a data packet destined for another gateway node (i.e., a logically adjacent gateway node) within its subnet, that node starts counting, incrementing the count by 1 for each data packet generated. Simultaneously, as a gateway node, it should also count data packets received from other gateway nodes within its subnet. Whenever the set duration (PacketTIME) is reached, all gateway nodes send an ACK (acknowledgment) message to their logically adjacent gateway nodes, recording the total number of data packets received from the other party within the previous period, and simultaneously restarting the data packet counting process. Upon receiving the ACK message from its adjacent gateway node, the gateway node can determine the reception status of its recorded sent data packets and calculate the packet loss rate. This achieves the statistical analysis of transmission success rates across subnet links, i.e., links between gateway nodes.

[0028] Step 1.3: Subnet Traffic Prediction. The subnet probing method has successfully captured the overall network situation. When a heterogeneous network operates under high load, constructing static cross-domain traffic paths can lead to all cross-domain traffic being concentrated on pre-defined paths, causing congestion or even packet loss. To plan the paths for cross-domain data packets, the detected traffic information should be used to predict future traffic patterns within each subnet, thereby planning cross-domain forwarding paths, achieving load balancing, and improving network throughput. Therefore, this invention designs a subnet traffic prediction algorithm: first, the subnets are divided into blocks using a graph partitioning algorithm; second, a traffic matrix is ​​generated using the detected subnet traffic information; and finally, a neural network is deployed within each subnet, using the historical traffic matrix as input to predict the future traffic matrix. This process mainly includes the following steps:

[0029] Step 1.3.1. Perform graph partitioning on the entire network topology: Use the LDG (Linear Deterministic Greedy) algorithm to partition adjacent nodes in each subnet, dividing a single subnet into several blocks. The LDG algorithm considers placing neighboring nodes together as much as possible during partitioning to reduce edge cutting in the network topology. It uses a greedy algorithm to place a node in the subgraph containing the most of its neighbors, while ensuring balanced node load in each subgraph.

[0030] Step 1.3.2. Flow Matrix Generation: Using the graph partitioning algorithm, assuming the subnet topology is divided into k blocks, the flow matrix is ​​a k-dimensional square matrix. Let the flow matrix be TM, and its structure is as follows:

[0031]

[0032] Among them throughput ijThe total traffic from block i to block j is represented by the following formula:

[0033]

[0034] Where N i N j These are the number of nodes in blocks i and j, respectively. Indicates the nth element in block i i The nth node of the arrival block j is sent by the node. j Traffic of each node.

[0035] Step 1.3.3. Network Traffic Prediction: A GRU (Gated Recurrent Unit) neural network is used for time series prediction. The prediction model is trained using historical measurement data, and the traffic matrix of the previous second is used as input to predict the traffic matrix of the next second.

[0036] After the above steps, the trained subnet traffic prediction model can predict future network traffic. At the same time, it can be continuously corrected during the prediction process, which helps with cross-domain link planning and achieve network-wide load balancing.

[0037] In heterogeneous network cross-domain link selection and message distribution technologies, the main components include cross-domain segmentation routing technology and cross-domain path selection technology. The cross-domain segmentation routing algorithm of this invention operates between the application layer and the network layer, avoiding manipulation of network layer protocols and adapting to various heterogeneous networks. The cross-domain path selection technology, addressing the differences in cross-domain business communication needs, adapts to the differentiated requirements of business data based on the aforementioned subnet situational awareness and traffic prediction results. Simultaneously, it introduces an adaptive dynamic adjustment mechanism to avoid network congestion and achieve load balancing.

[0038] In heterogeneous subnets, each subnet operates under different routing protocols, such as wired network routing protocols for terrestrial backbone networks and ad hoc network routing protocols for space-based data link networks. Cross-network communication between heterogeneous subnets under different communication systems is difficult. This invention designs a cross-domain segmented routing technology, dividing the cross-domain forwarding of data packets into the selection of gateway nodes and forwarding within several subnet segments, achieving interconnection between heterogeneous subnets with lower network overhead. The cross-domain segmented routing process for data packets in this invention includes the following steps 2.1 to 2.5.

[0039] Step 2.1. Send data packets to the gateway: Nodes within the source subnet send data packets to the gateway node within their subnet.

[0040] Step 2.2. Gateway Node Processing Data Packets: After receiving a data packet, the gateway node unpacks the data packet to the upper layer, finds that its real target node is located in another subnet, plans a cross-domain link for it based on the backbone network topology it maintains, determines the next transit gateway node, converts the data packet according to the new subnet routing protocol, and sends it to the next transit gateway node through the new subnet.

[0041] Step 2.3. Multi-hop forwarding of data packets within the subnet: After being sent by the gateway node, the data packet is relayed within the subnet and sent to the planned next gateway node.

[0042] Step 2.4. Repeat steps 2.2 and 2.3 to forward the message to the gateway node of the subnet where the target node is located.

[0043] Step 2.5. Subnet processing of data packets within the target subnet: The gateway node of the target subnet forwards the data packets to the target node, completing the data packet segmentation and routing.

[0044] By introducing segmented routing operations above the network layer, the process of routing data packets across subnets is transformed into the selection of gateway nodes and routing within each subnet along the way. This avoids the flooding of a large number of non-business data packets caused by route redistribution, and only requires the necessary cross-domain flow information to be synchronized between the gateway nodes of each subnet.

[0045] When multiple alternative cross-domain paths exist for cross-domain data packets, the method of this invention evaluates the capabilities of each path based on the performance and traffic of each subnet's cross-domain path obtained through subnet situational awareness surveys. It comprehensively considers link latency, packet loss, and network load to select the optimal cross-domain path to meet the communication needs of the service and achieve network load balancing. When the network situation changes, an adaptive path adjustment mechanism is introduced to select the optimal path under the current situation, ensuring the cross-domain flow of service messages.

[0046] Based on the above process, the link-level latency and packet loss rate of each cross-domain multi-hop link within each subnet have been measured, and the traffic matrix of the subnet for the next time period has been predicted using intelligent algorithms. Let the average cross-domain link latency and packet loss rate of the j-th subnet be obtained from subnet situational awareness and traffic prediction, and used as indicators of subnet transmission rate and reliability, respectively. Simultaneously, the elements of the predicted traffic matrix are summed and normalized using the subnet's nominal capacity W to obtain the subnet load evaluation index W. j The calculation formula is as follows:

[0047]

[0048] Among them W j Here, k is the load evaluation metric for the j-th subnet, k is the total number of blocks in the j-th subnet, and throughput is... xyIt is the element located in the x-th row and y-th column of the traffic matrix of the j-th subnet.

[0049] When multiple cross-domain paths exist from the source subnet to the target subnet of a cross-domain data packet, each path should be scored, and the path with the highest score should be selected as the cross-domain transfer path for the message. For the i-th path, the latency and packet loss rate of each segment of the path are obtained. Each segment of the path corresponds to two gateway nodes: a sending gateway and a receiving gateway. The latency of the path segment is obtained by detecting the link-level latency from the sending gateway to the receiving gateway, and the packet loss rate of the path segment is obtained by detecting the packet loss rate from the sending gateway to the receiving gateway. The load evaluation index of the subnet where each segment of the path is located is calculated. In this embodiment of the invention, the average link-level latency and average packet loss rate of the most recent several periods can also be obtained as the latency and packet loss rate of the corresponding path segment.

[0050] The formula for calculating the score of the i-th path is as follows:

[0051]

[0052] Among them, delay i,j This represents the delay of the j-th segment within the i-th path. P represents the total latency of cross-domain path i; i,j This represents the packet loss rate of the j-th segment in cross-domain path i. W represents the total packet loss rate of cross-domain path i; i,j This represents the traffic of the subnet traversed by the j-th segment of the cross-domain path i; ω1 and ω2 are set weights, with values ​​ranging from 0 to 1. In this embodiment of the invention, the path "Gateway Node 1 - Gateway Node 2 - Gateway Node 3" is used as an example to illustrate path segmentation. The first segment is "Gateway Node 1 - Gateway Node 2", and the second segment is "Gateway Node 2 - Gateway Node 3". In this embodiment of the invention, each path may have multiple possible transit paths, that is, the source gateway node and the destination gateway node are the same, but the specific transit gateway nodes are not completely the same, but there are some identical transit gateway nodes. Such two paths are considered to belong to one path. Therefore, for each path, These are the minimum latency and minimum packet loss rate among all possible relay paths for path i, respectively, which are used as the latency and packet loss rate for path i. It is the largest subnet load evaluation index value in the i-th path.

[0053] This invention employs a periodic update approach to select cross-domain paths. During each period, the gateway node plans the cross-domain paths and generates and stores routing table entries. Whenever a data packet needs to be sent to a specified target subnet, it is directly queried from the routing table. When the update cycle arrives, the scores of each path are recalculated, and the routing table is updated, thereby achieving adaptive adjustment of cross-domain paths. In addition to periodic corrections, when network traffic prediction detects impending congestion in a subnet, path correction can also be triggered to avoid network congestion.

[0054] In summary, this invention successfully proposes and implements a heterogeneous network cross-domain link selection and message distribution technology to address the networking problems of heterogeneous networks under various conditions. This technology effectively enables information flow between heterogeneous subnets and, with the assistance of subnet situational awareness, performs path planning for cross-domain message forwarding. The method of this invention does not require routing table synchronization between subnets, and the network maintenance message size is smaller, reducing cross-domain network communication overhead. Furthermore, it can adaptively adjust paths according to the actual needs of the task and the current network conditions to achieve load balancing of network resources.

[0055] This invention aims to solve the problems of communication difficulties between heterogeneous data links using different routing protocols and network congestion caused by traditional pre-configured fixed cross-domain links. It achieves heterogeneous network integration and load balancing through subnet situational awareness technology, cross-domain segmented routing technology, cross-domain link planning, and adaptive adjustment technology. This invention implements a method for constructing a network backbone, planning cross-domain links through backbone nodes, and relaying data packets across links to achieve load-balanced heterogeneous network platform integration and communication. It effectively solves the problem of communication difficulties between different subnets in the process of heterogeneous aerospace networking. Simultaneously, by adaptively adjusting cross-domain forwarding paths, it can achieve load balancing of network resources, which has significant practical application value. This method can reduce routing maintenance overhead and improve network resource utilization efficiency while ensuring heterogeneous network interconnection. Furthermore, the modular design and scalability of this method allow it to be customized and expanded according to different application scenarios and needs, demonstrating broad application potential.

[0056] Except for the technical features described in the specification, all other technologies are known to those skilled in the art. Descriptions of well-known components and technologies are omitted in this invention to avoid redundancy and unnecessary limitation. The embodiments described above do not represent all embodiments consistent with this application. Various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this invention are still within the protection scope of this invention.

Claims

1. A method for cross-domain link selection and message distribution in aerospace heterogeneous networks, applied in networks composed of more than one heterogeneous subnet, characterized in that, This method first selects gateway nodes in a heterogeneous network to construct a virtual backbone network, equips each gateway node with the protocol stack of its respective subnet, and then deploys the functional modules that implement the following steps between the application layer and transport layer of the TCP / IP protocol stack in the gateway node: Step 1: Perform subnet situational awareness, including traffic probing, latency probing, and packet loss rate probing. Traffic probing obtains the traffic of data packets sent within each subnet, and latency probing and packet loss rate probing obtain the link-level latency and packet loss rate between adjacent subnets. Perform subnet traffic prediction, including dividing each subnet into blocks and using a neural network to predict traffic for each subnet. The input of the neural network is the historical traffic matrix of the subnet, and the output is a predicted future traffic matrix. The matrix records the traffic between each block of the subnet. Step 2: The cross-domain forwarding of data packets is divided into the selection of the gateway node and the forwarding of nodes within the subnet. The gateway node of the subnet where the source node of the data packet is located plans the cross-domain link for the data packet according to the virtual backbone network topology it maintains, and determines the next transit gateway node. When the data packet is transited at each gateway node, it is formatted according to the routing protocol of the subnet to which it is to be transferred before being forwarded. Step 3: When there are more than one cross-domain forwarding path for a data packet, score each cross-domain forwarding path by probing the link-level latency and packet loss rate between subnets and the predicted subnet traffic matrix, and select the path with the highest score as the cross-domain forwarding path for the data packet.

2. The method according to claim 1, characterized in that, In the method described above, the selection and maintenance rules for gateway nodes are as follows: the node with the highest connectivity is selected from the intersection of communication nodes in adjacent subnets as the gateway node. The gateway node is located in multiple subnets at the same time, equipped with the protocol stack of each subnet, and can communicate with each subnet it is in. The gateway node periodically notifies the other nodes in each subnet of its own gateway identity.

3. The method according to claim 1, characterized in that, Step 1, which involves subnet situational awareness, includes: Traffic detection: Nodes within each subnet periodically obtain the data packet traffic and the destination node address of the data packets sent by their own application layer, and report them to the subnet's gateway node; Latency Probing: Probing the link-level latency between logically adjacent gateway nodes; each gateway node maintains a probe table, storing the latest time of sending data packets to logically adjacent target gateway nodes; if the sending time of a target gateway node in the table exceeds a set duration, the gateway node actively initiates a latency maintenance data packet to that target gateway node to measure the link-level latency; when a gateway node generates a data packet that arrives at a target gateway node in the table, it adds a frame to the packet header to record the packet's sending time; when the data packet is forwarded to the target gateway node, the target gateway node calculates the link-level latency from the previous hop gateway node to itself; logically adjacent gateway nodes are those that are separated by only one subnet. Packet loss rate detection: Each gateway node counts the packets generated and arrive at each logically adjacent gateway node, and counts the packets received from each logically adjacent gateway node. Each gateway node periodically sends an ACK message to each logically adjacent gateway node. This message records the total number of packets received by the node from the other party in the previous period, and restarts the packet counting process. After receiving the ACK message, each gateway node calculates the packet loss rate.

4. The method according to claim 1 or 3, characterized in that, Step 1, subnet traffic prediction, includes: First, dividing each subnet into blocks using the LDG algorithm; second, representing the traffic between subnet blocks as a traffic matrix. If the subnet topology is divided into k blocks, then the traffic matrix TM of that subnet is a k-dimensional square matrix, as shown below: Among them, throughput ij The total traffic from block i to block j is represented by the following formula: N i N j These are the number of nodes in the i-th block and the j-th block, respectively. It is the nth block in the i-th block i The nth node sends a message to the j-th block. j Traffic of each node; Then, a GRU neural network is used as the network traffic prediction model. The historical traffic matrix of the subnet is input into the GRU neural network in time sequence, and the traffic matrix of the subnet at the next moment is predicted and output continuously. GRU stands for Gated Recurrent Unit.

5. The method according to claim 1, characterized in that, In step 2, cross-domain forwarding of data packets includes: Step 2.1: The source node of the data packet sends the data packet to the gateway node of its subnet; Step 2.2: The gateway node unpacks the received data packet, obtains the target node of the data packet, plans cross-domain links for forwarding the data packet based on the subnet where the target node is located and the virtual backbone network topology it maintains, determines the next transit gateway node, converts the data packet according to the routing protocol of the next transit subnet, and then sends it to the next transit gateway node. Step 2.3: The data packet is relayed and forwarded within the subnet and sent to the next relay gateway node determined in Step 2.2; Step 2.4: Repeat steps 2.2 and 2.3 to forward the data packet to the gateway node of the subnet where the target node is located; Step 2.5: The gateway node of the subnet where the target node is located forwards the data packet to the target node, completing the data packet segmentation routing.

6. The method according to claim 1, characterized in that, In step 3, each cross-domain forwarding path is scored, including: For the i-th path, obtain the latency and packet loss rate of each segment of the path. Each segment of the path corresponds to two gateway nodes: the sending gateway and the receiving gateway. The latency of the path segment is obtained by detecting the link-level latency from the sending gateway to the receiving gateway, and the packet loss rate of the path segment is obtained by detecting the packet loss rate from the sending gateway to the receiving gateway. Calculate the load evaluation index of the subnet where each segment of the path is located: obtain the index by summing the elements in the predicted traffic matrix and then normalizing it using the nominal capacity of the subnet itself. Then calculate the score for the i-th path. Among them, delay i,j This represents the delay of the j-th segment within the i-th path. P represents the total delay of the i-th path; i,j This represents the packet loss rate of the j-th segment within the i-th path. W represents the total packet loss rate of the i-th path; i,j ω1 represents the load evaluation index value of the subnet traversed by the j-th segment of the i-th path; ω1 and ω2 are the set weights, with values ​​ranging from 0 to 1.

7. The method according to claim 1 or 2, characterized in that, In the method described above, the gateway node plans and selects cross-domain forwarding paths for different target subnets in advance in each period through periodic updates, and generates and stores a routing table. Whenever a data packet is sent, the gateway node directly queries the routing table to obtain the cross-domain forwarding path. When the period is updated, the gateway node plans and selects cross-domain forwarding paths for different target subnets again and updates the routing table.

Images