Layered active reliable routing method for space-air-ground integrated computing power network

Abstract

The invention provides a space-air-ground integrated computing power network layered active reliable routing method, and the core of the method lies in a hybrid backup path management and control mechanism: a control plane calculates a main path based on an intelligent algorithm, and actively pre-configures a local backup forwarding instruction based on segment routing for a main path node; at the same time, a global backup path instruction set involving non-primary path nodes is pre-computed and stored. And the data plane senses the fault in real time through an enhanced detection mechanism and reports the fault. And when the control plane responds, on-line recalculation is not needed, a pre-stored global backup instruction is directly searched and issued, and the report node completes rapid switching based on segment routing on the data plane. According to the method, the calculation overhead and the recovery process are decoupled through pre-calculation and on-demand triggering, the flow table storage overhead of a data plane is greatly reduced while sub-second fault recovery is guaranteed, the optimal balance of the storage efficiency and the recovery speed is achieved, and the routing quality and toughness in a dynamic network are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of computer network communication technology, specifically to a hierarchical active reliable routing method for an integrated air-space-ground computing power network. Background Technology

[0002] With the development of sixth-generation mobile communication systems and computing networks, integrated air-space-ground networks that combine satellites, aviation platforms, and ground facilities have become a key infrastructure for achieving full coverage and ubiquitous computing. These networks, through the collaboration of heterogeneous nodes, aim to provide continuous, high-quality computing services for latency-sensitive applications such as autonomous driving and holographic communication.

[0003] However, the inherent dynamism of integrated air-space-ground networks poses a core challenge. The high-speed movement of low-Earth orbit satellites, the mobile deployment of airborne platforms, and the mobility of terminals cause network topology to change drastically at the millisecond level. This highly time-varying characteristic poses a serious challenge to traditional routing protocols based on static or periodic convergence assumptions (such as OSPF and BGP): their high signaling overhead, slow convergence speed, and dependence on path stability easily lead to frequent routing oscillations, path failures, and update delays, ultimately resulting in service interruptions and degradation, failing to meet the stringent requirements of computing power tasks for continuity and reliability.

[0004] To enhance routing resilience in dynamic networks, the industry has introduced programmable data plane technology, particularly IPv6-based segment routing (SRv6). SRv6 allows packets to carry explicit forwarding instructions, enabling the data plane to perform rapid path switching and providing new underlying capabilities for fault recovery. However, efficiently utilizing this capability still requires addressing two interrelated fundamental challenges at the upper-layer routing mechanism level:

[0005] First, there's the dilemma of backup path distribution strategies, namely the trade-off between storage overhead and recovery latency. Existing robust routing solutions mainly fall into two categories:

[0006] Proactive strategy: Pre-calculate backup paths for all potential failure scenarios and distribute them to forwarding devices in advance. This strategy can achieve rapid failover, but its fatal flaw is that in a highly dynamic network, it requires maintaining a massive number of backup forwarding table entries, leading to a sharp expansion or even exhaustion of storage resources, resulting in low practical deployment feasibility.

[0007] Passive strategy: Backup path calculation and deployment are only triggered after a failure occurs. While this strategy saves storage space, it adds the latency of calculation, control plane interaction, and deployment to the fault recovery process, resulting in an excessively long overall recovery time, making it difficult to support services requiring millisecond-level recovery.

[0008] Secondly, there is a bottleneck in the computation of backup paths, namely the contradiction between "path optimization quality" and "real-time decision-making." Traditional graph-based centralized algorithms (such as Dijkstra's algorithm) require global recalculation after topology changes, resulting in excessively high computational complexity and failing to meet the real-time decision-making requirements of dynamic environments. In recent years, intelligent routing methods based on deep reinforcement learning (such as DQN and DDQN) have made decisions through online learning, but they still have significant limitations in practical applications: First, the algorithms generally suffer from Q-value overestimation, causing the policy to converge to a suboptimal solution; second, existing models fail to fully model and utilize the structured information of the network topology itself, resulting in insufficient generalization ability and convergence speed when the topology changes frequently; third, most research focuses on path computation in the control plane, failing to deeply coordinate with the fast switching mechanism of the data plane, leading to limited end-to-end recovery performance.

[0009] In summary, existing technologies suffer from fragmentation and shortcomings when addressing the routing reliability challenges of integrated air-space-ground networks: they either sacrifice storage efficiency for speed or recovery speed for efficiency. Furthermore, path calculation methods lack dynamic adaptability and optimization capabilities, and are not tightly integrated with the rapid recovery capabilities of the data plane. Therefore, this field urgently needs an innovative routing mechanism that can: at the system level, break the trade-off between storage and latency to achieve rapid recovery with low overhead; at the algorithm level, achieve high-quality real-time path decision-making in highly dynamic environments; and at the implementation level, achieve efficient synergy between control intelligence and data plane forwarding capabilities. Summary of the Invention

[0010] To address the aforementioned issues, this invention proposes a hierarchical proactive reliable routing method for integrated air-space-ground computing networks. Through a hybrid backup path management mechanism based on SRv6 and pre-computation, it deeply integrates intelligent path computation with rapid data plane recovery capabilities. Simultaneously, enhanced fault detection enables real-time topology awareness. While ensuring sub-second fault recovery, it significantly reduces the flow table storage overhead of the data plane, achieving a multi-dimensional balance of reliability, low latency, and high efficiency in dynamic networks. The technical solution provided by this invention is as follows:

[0011] A hierarchical active reliable routing method for an integrated air-space-ground computing network includes the following steps:

[0012] Step 1: The application plane receives user service requests and corresponding quality of service (QoS) requirement types, and forwards the requests and requirement types to the control plane;

[0013] Step 2: Based on the network topology state it maintains, the control plane pre-calculates the corresponding path segments for all node pairs and generates the corresponding SRv6 segment routing header information; the pre-calculation is performed using the GAT-DDQN algorithm that fuses graph attention networks and deep dual-Q networks.

[0014] Step 3: The control plane selects the complete sequence from the source node to the destination node as the main path from the pre-calculation results, and sends the SRv6 segment routing header information corresponding to the main path to the source node; at the same time, it actively pre-installs the backup path SRv6 segment routing header information from itself to the destination node into the respective data plane forwarding table for the nodes on the main path.

[0015] Step 4: The control plane stores the pre-calculated SRv6 segment routing header information of other paths involving non-primary path nodes as a backup path instruction set in the control plane.

[0016] Step 5: Nodes in the data plane monitor the link status of their ports by running an enhanced bidirectional forwarding detection mechanism. The detection mechanism determines the status by using a guaranteed port and a fault counter. When the port status switches between "valid" and "faulty", the node updates its local port status register.

[0017] Step 6: When the port status changes, the node will immediately report the information, including the port status change identifier and the destination node of the service, to the control plane.

[0018] Step 7: If the node reporting the fault is the primary path node and the fault point is itself, then the node directly switches its data plane to the backup path SRv6 segment routing header information pre-installed for it in Step 3, so as to restore traffic forwarding based on the SRv6 mechanism.

[0019] Step 8: If the fault point is located downstream of the node that reported the fault, after receiving the reporting information, the control plane searches for the matching pre-calculated SRv6 segment routing header information from the backup path instruction set described in Step 4 based on the fault node and the service destination node, and sends it to the node that reported the fault; the node installs the received SRv6 segment routing header information in its data plane, and switches to the path after encapsulating the subsequent service traffic with the SRv6 segment routing header information.

[0020] Preferably, the agent architecture of the GAT-DDQN algorithm in step 2 is: a Q-value function approximator composed of GATs, trained using the DDQN framework, including the following units:

[0021] The dynamic topology-aware state construction unit is used to extract a subgraph containing the current node and its first-order neighbor nodes from the global network topology based on the current state, and to construct the node feature matrix in the subgraph.

[0022] The graph attention network feature extraction unit, consisting of multiple graph attention layers, is used to aggregate the neighbor information of nodes in the subgraph through a multi-head attention mechanism.

[0023] The Q-value calculation and decision unit is used to linearly map the node features extracted by the graph attention network and output the Q-value corresponding to each actionable node in the current state. It also adopts a dual deep Q-network framework, which uses an online network to select actions and uses a target network to evaluate Q-values ​​to generate training targets.

[0024] The priority experience replay training unit is used to store experience data generated by agent interaction, and to assign sampling priority to each experience according to the temporal difference error. Sampling is performed according to the priority ratio, and the bias is corrected by importance sampling weight during training to optimize the parameters of the Q value calculation and decision unit.

[0025] Preferably, the main path distribution and local backup pre-installation in step 3 are specifically as follows:

[0026] First, select different topology databases based on the requirements, and then obtain the source nodes from them. to the destination node Optimal route path The control plane then converts it into an SRv6 segment routing header and sends it to the source node in the form of a flow table; for intermediate nodes Retrieve from topology database Optimal path for node pairs And actively install the corresponding SRv6 segment routing header information to the node. In the routing table.

[0027] Preferably, the enhanced bidirectional forwarding detection mechanism in step 5 is implemented by maintaining a local port status register list and associated guaranteed port registers and fault counters. Its port state switching logic is formally defined by the following rules:

[0028] a. Data Structure Definition: Suppose a switch has N physical ports, and its logical state vector is represented as follows: ,in , This indicates that the link where port p is located is valid; otherwise, it indicates that the link where the port is located is invalid. A guarantee port register is maintained for each port. Record its most recently confirmed valid peer port number, and a suspected fault counter. The initial value is 0, which is used to count the suspected faults of the port.

[0029] b. State transition rules: Determining whether rule 1 is valid or invalid:

[0030] If at time t, the guaranteed port of port p A valid probe packet was received, and the following conditions were met:

[0031]

[0032] in This is the timestamp of the last time port p successfully received a probe packet. If the preset delay threshold is used, then:

[0033]

[0034] If after the update Then a state transition is triggered:

[0035]

[0036] At the same time, a fault reporting event is generated. And send it to the control plane;

[0037] Determining whether Rule 2 is invalid or valid:

[0038] If port p is currently in a certain state If a valid probe data packet is received directly at time t, then execute immediately:

[0039]

[0040] At the same time, a recovery reporting event is generated. And send it to the data plane;

[0041] c. Probe message sending rules: For all current states For ports with a value of 0, a probe message is broadcast to them periodically T. For the state A port with a value of 1 relies on periodic bidirectional probing with the peer port to maintain its valid state.

[0042] Preferably, step 6 is implemented as follows:

[0043] If and only if the port's logical state According to the state transition rules, the event transitions from valid to invalid or from invalid to valid. or When triggered, the data plane hardware immediately generates a fault reporting message. ;

[0044] The message The data structure is constructed as a tuple:

[0045]

[0046] in, This is the unique identifier for this node; The port identifier p where the state transition occurred; The destination node identifier for the currently affected service flow; Timestamp is the timestamp t of the state transition time. This is an event type identifier, a constant representing "failure" or "recovery".

[0047] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0048] 1. Breaks the performance trade-off dilemma. By adopting a hybrid backup strategy of "proactive pre-installation + instruction set pre-store", the computational overhead and recovery process are decoupled, which can significantly reduce the storage scale of data plane flow tables while ensuring sub-second fast fault recovery, perfectly solving the fundamental contradiction that storage overhead and recovery latency cannot be achieved simultaneously.

[0049] 2. Improved intelligent decision-making quality. The adopted GAT-DDQN algorithm fully utilizes the graph structure information of the network topology, enabling it to output better and more stable primary and backup paths, significantly improving routing quality and long-term performance in dynamic environments.

[0050] 3. Achieved efficient cross-layer collaboration. The enhanced detection mechanism ensures the real-time and accuracy of fault perception, and the SRv6 data plane enables rapid programming and switching of traffic. The control plane and data plane collaborate deeply through a lightweight event interface, forming a highly efficient autonomous system with "global optimization and local execution," possessing excellent scalability and engineering practicality. Attached Figure Description

[0051] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0052] Figure 1 It is a constructed time-varying space-ground integrated network topology;

[0053] Figure 2 This is a data plane architecture diagram of the present invention;

[0054] Figure 3 This is a diagram of the GAT-DDQN intelligent agent network architecture of the present invention;

[0055] Figure 4 This is a packet capture screenshot of the backup path information that was proactively installed after a critical link failure.

[0056] Figure 5 This is a packet capture screenshot of the backup path information from the hybrid passive installation after the extended link failed. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0058] To make the above-mentioned objectives, features and effects of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0059] To verify and demonstrate the effectiveness of the method of this invention, it can be deployed in a Mininet-based simulation platform or a real network environment. The system includes a control plane (such as an SDN controller based on P4Runtime) and a data plane (a programmable switch supporting SRv6 and enhanced BFD, such as a BMv2 software switch). The network topology consists of heterogeneous nodes including ground nodes, airborne base stations, and satellite nodes, and its link states can change dynamically. The topology diagram is shown below. Figure 1 As shown in Table 1, the node information is as follows.

[0060] Table 1 Node Information Table

[0061]

[0062] A hierarchical active reliable routing method for an integrated air-space-ground computing network includes the following steps:

[0063] Step 1, Service Request Reception and Forwarding. The application plane receives user service requests and corresponding quality of service (QoS) requirement types, and forwards the requests and requirement types to the control plane.

[0064] Step 2, Global Topology Pre-calculation. Based on the network topology it maintains, the control plane pre-calculates the corresponding path segments for all node pairs and generates the corresponding SRv6 segment routing header (SRH) information. This pre-calculation is performed using the GAT-DDQN algorithm, which combines a Graph Attention Network (GAT) and a Deep Double-Q Network. Its architecture diagram is shown below. Figure 3 As shown, the core is a Q-value function approximator composed of GAT, and it is trained using the DDQN framework.

[0065] Dynamic topology-aware state building units are used to construct states based on the current state. (Current node) and destination node From the global network topology Extracting contents The subgraph of the subgraph and its first-order neighbor nodes; and the node feature matrix of the subgraph are constructed. , where nodes eigenvectors Defined as:

[0066]

[0067] in, Represents a node One-hot encoded vector, Indicates from arrive The normalized shortest path distance, This is a binary indicator function, where 1 indicates the existence of an adjacency relationship. express The set of neighboring nodes, Represents the zero vector; the edge relationships of the subgraph are directly inherited from the global topology. Adjacency relationship.

[0068] The graph attention network feature extraction unit, composed of multiple graph attention layers, is used to aggregate the neighbor information of nodes in the subgraph through a multi-head attention mechanism. For node i, the output feature of its single-head attention is calculated as follows:

[0069]

[0070] Where W is the learnable weight matrix, σ is the non-linear activation function, and the attention coefficient is... Calculated in the following way:

[0071]

[0072] in It is a learnable attention vector, and || denotes the concatenation operation. Multi-head attention concatenates or averages multiple single-head outputs to enhance the model's expressive power.

[0073] The Q-value calculation and decision unit is used to linearly map the node features extracted by the graph attention network and output the Q-value corresponding to each actionable node in the current state. It employs a dual-depth Q-network framework, utilizing an online network to select actions and a target network to evaluate the Q-value to generate training objectives. The formula for calculating the Q-value is as follows:

[0074]

[0075] in, These are online network parameters. These are the target network parameters. The online network is responsible for selecting the next state. The optimal action is determined by the network, while the target network is responsible for evaluating the Q-value of that action. This effectively alleviates the problem of overestimation of Q-value.

[0076] The priority experience replay training unit is used to store experience data generated by agent interaction, and to assign sampling priority to each experience according to the temporal difference error. Sampling is performed according to the priority ratio, and the bias is corrected by importance sampling weight during training to optimize the parameters of the Q value calculation and decision unit.

[0077] Step 3: Primary Path Distribution and Local Backup Pre-installation. The control plane selects the complete sequence from the source node to the destination node as the primary path from the pre-calculation results, and distributes the SRv6 SRH information corresponding to the primary path to the source node; at the same time, it actively pre-installs the backup path SRv6 SRH information from itself to the destination node into the respective data plane forwarding table for the nodes on the primary path.

[0078] Based on the requirements, first select different topology databases to obtain the optimal routing path from the source node to the destination node. The optimal path here is H1-S4-S2-S3-S7-H2, which is then transformed into SRv6 SRH (S4-S2-S3-S7) by the control plane and distributed to the source node in the form of a flow table. Here it is S4; for intermediate nodes Retrieve from topology database Optimal path for node pairs And actively install its corresponding SRv6 SRH information to the node. In the routing table.

[0079] Step 4, Pre-store backup path instruction set. The control plane stores the pre-calculated SRv6 SRH information of other paths involving non-primary path nodes as a backup path instruction set. Here, nodes outside the primary path may also become selected backup path nodes for this route, and their backup paths pointing to the final destination H2 are stored in the control plane to await installation using the hybrid passive backup method.

[0080] Step 5, Enhanced Fault Detection and State Management. Nodes in the data plane monitor the link status of their ports by running an enhanced bidirectional forwarding detection mechanism. This detection mechanism determines the status using a guaranteed port and a fault counter. When the port status switches between "valid" and "faulty," the node updates its local port status register. The data plane architecture diagram is shown below. Figure 2 As shown.

[0081] The enhanced bidirectional forwarding detection mechanism is implemented by maintaining a list of local port status registers and associated guaranteed port registers and fault counters. Its port state switching logic is formally defined by the following rules:

[0082] a. Data Structure Definition

[0083] Suppose a switch has N physical ports, and its logical state vector is represented as follows: ,in , This indicates that the link containing port p is valid; otherwise, it indicates that the link containing port p is invalid. A guarantee port register is maintained for each port. Record its most recently confirmed valid peer port number, and a suspected fault counter. The initial value is 0, and it is used to count the suspected faults of the port.

[0084] b. State transition rules

[0085] Rule 1: Determining whether a rule is valid or invalid

[0086] If at time t, the guaranteed port of port p A valid probe packet was received, and the following conditions were met:

[0087]

[0088] in This is the timestamp of the last time port p successfully received a probe packet. If the preset delay threshold is used, then:

[0089]

[0090] If after the update ( If the preset counting threshold is reached, a state transition is triggered:

[0091]

[0092] At the same time, a fault reporting event is generated. And send it to the control plane.

[0093] Rule 2: Determining whether an invalid rule is valid

[0094] If port p is currently in a certain state If a valid probe data packet is received directly at time t, then execute immediately:

[0095]

[0096] At the same time, a recovery reporting event is generated. And send it to the data plane.

[0097] c. Probe message sending rules

[0098] Detection message The transmission follows the following strategy: for all current states For ports with a value of 0, broadcast messages to them at period T. Regarding the state A port with a value of 1 relies on periodic bidirectional probing with the peer port to maintain its valid state.

[0099] Step 6: Real-time reporting of topology change information. When a port status changes, the node will immediately report information including the port identifier of the status change and the destination node of the service to the control plane.

[0100] If and only if the port's logical state According to the state transition rules, an event transitions from valid to invalid (fault) or from invalid to valid (recovery). or When triggered, the data plane hardware immediately generates a fault reporting message. .

[0101] The message The data structure is constructed as a tuple:

[0102]

[0103] in, This is the unique identifier for this node; The port identifier p where the state transition occurred; The destination node identifier for the currently affected service flow; Timestamp is the timestamp t of the state transition time. This is an event type identifier, a constant representing "failure" or "recovery".

[0104] Step 7, Primary Path Node Self-Recovery. If the node reporting the fault is the primary path node and the fault point is itself, then the node directly switches its data plane to the backup path SRv6 SRH information pre-installed in Step 3 to restore traffic forwarding based on the SRv6 mechanism. For example, when a link S2-S3 on the critical path fails, it detects the fault through the bidirectional forwarding detection mechanism and records it in its local register. Since it is a link on the critical path, it directly recovers through the pre-installed backup path. Figure 4 As shown, it directly installs the active backup path S6-S7-H2, which points from the current location S2 to the final destination H2;

[0105] Step 8, downstream fault recovery on demand. If the fault point is located downstream of the node that reported the fault, the control plane, upon receiving the reporting information, searches for matching pre-calculated SRv6 SRH information from the backup path instruction set described in Step 4, based on the fault node and the service destination node, and sends it to the node that reported the fault. This node installs the received SRv6 SRH information on its data plane and encapsulates subsequent service traffic with the SRv6 SRH information before switching to this path. For example, if the link S6-S7 on the backup path is disconnected due to the relative movement of satellite group nodes, after receiving the fault feedback, the control plane sends out a pre-calculated and pre-stored backup path S5-S3-S7 from the current node S6 to the destination H2. Figure 5 As shown, after reaching the destination node and adjacent to switch S7, the SRv6 related messages are unloaded to achieve transparent transmission of service data packets.

[0106] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A hierarchical active reliable routing method for an integrated space-air-ground computing network, characterized in that, Includes the following steps: Step 1: The application plane receives user service requests and corresponding quality of service (QoS) requirement types, and forwards the requests and requirement types to the control plane; Step 2: Based on the network topology state it maintains, the control plane pre-calculates the corresponding path segments for all node pairs and generates the corresponding SRv6 segment routing header information; the pre-calculation is performed using the GAT-DDQN algorithm that fuses graph attention networks and deep dual-Q networks. Step 3: The control plane selects the complete sequence from the source node to the destination node as the main path from the pre-calculation results, and sends the SRv6 segment routing header information corresponding to the main path to the source node; at the same time, it actively pre-installs the backup path SRv6 segment routing header information from itself to the destination node into the respective data plane forwarding table for the nodes on the main path. Step 4: The control plane stores the pre-calculated SRv6 segment routing header information of other paths involving non-primary path nodes as a backup path instruction set in the control plane. Step 5: Nodes in the data plane monitor the link status of their ports by running an enhanced bidirectional forwarding detection mechanism. The detection mechanism determines the status by using a guaranteed port and a fault counter. When the port status switches between "valid" and "faulty", the node updates its local port status register. Step 6: When the port status changes, the node will immediately report the information, including the port status change identifier and the destination node of the service, to the control plane. Step 7: If the node reporting the fault is the primary path node and the fault point is itself, then the node directly switches its data plane to the backup path SRv6 segment routing header information pre-installed for it in Step 3, so as to restore traffic forwarding based on the SRv6 mechanism. Step 8: If the fault point is located downstream of the node that reported the fault, after receiving the reporting information, the control plane searches for the matching pre-calculated SRv6 segment routing header information from the backup path instruction set described in Step 4 based on the fault node and the service destination node, and sends it to the node that reported the fault; the node installs the received SRv6 segment routing header information in its data plane, and switches to the path after encapsulating the subsequent service traffic with the SRv6 segment routing header information.

2. The hierarchical active reliable routing method for an integrated air-space-ground computing network according to claim 1, characterized in that, The agent architecture of the GAT-DDQN algorithm described in step 2 is as follows: a Q-value function approximator composed of GATs, trained using the DDQN framework, including the following units: The dynamic topology-aware state construction unit is used to extract a subgraph containing the current node and its first-order neighbor nodes from the global network topology based on the current state, and to construct the node feature matrix in the subgraph. The graph attention network feature extraction unit, consisting of multiple graph attention layers, is used to aggregate the neighbor information of nodes in the subgraph through a multi-head attention mechanism. The Q-value calculation and decision unit is used to linearly map the node features extracted by the graph attention network and output the Q-value corresponding to each actionable node in the current state. It employs a dual-deep Q-network framework, utilizing an online network to select actions and a target network to evaluate Q-values ​​to generate training targets; The priority experience replay training unit is used to store experience data generated by agent interaction, and to assign sampling priority to each experience according to the temporal difference error. Sampling is performed according to the priority ratio, and the bias is corrected by importance sampling weight during training to optimize the parameters of the Q value calculation and decision unit.

3. The hierarchical active reliable routing method for an integrated air-space-ground computing network according to claim 1, characterized in that, Step 3, the main path distribution and local backup pre-installation, specifically involves: First, select different topology databases based on the requirements, and then obtain the source nodes from them. to the destination node Optimal route path The control plane then converts it into an SRv6 segment routing header and sends it to the source node in the form of a flow table; for intermediate nodes Retrieve from topology database Optimal path for node pairs And actively install the corresponding SRv6 segment routing header information to the node. In the routing table.

4. The hierarchical active reliable routing method for an integrated air-space-ground computing network according to claim 3, characterized in that, In step 5, the enhanced bidirectional forwarding detection mechanism is implemented by maintaining a list of local port status registers and associated guaranteed port registers and fault counters. Its port state switching logic is formally defined by the following rules: a. Data Structure Definition: Suppose a switch has N physical ports, and its logical state vector is represented as follows: ,in , This indicates that the link where port p is located is valid; otherwise, it indicates that the link where the port is located is invalid. A guarantee port register is maintained for each port. Record its most recently confirmed valid peer port number, and a suspected fault counter. The initial value is 0, which is used to count the suspected faults of the port. b. State transition rules: Determining whether rule 1 is valid or invalid: If at time t, the guaranteed port of port p A valid probe packet was received, and the following conditions were met: ； in This is the timestamp of the last time port p successfully received a probe packet. If the preset delay threshold is used, then: ； If after the update Then a state transition is triggered: ； At the same time, a fault reporting event is generated. And send it to the control plane; Determining whether Rule 2 is invalid or valid: If port p is currently in a certain state If a valid probe data packet is received directly at time t, then execute immediately: ； At the same time, a recovery reporting event is generated. And send it to the data plane; c. Probe message sending rules: For all current states For ports with a value of 0, a probe message is broadcast to them periodically T. For the state A port with a value of 1 relies on periodic bidirectional probing with the peer port to maintain its valid state.

5. A hierarchical active reliable routing method for an integrated air-space-ground computing network according to claim 4, characterized in that, The specific implementation of step 6 is as follows: If and only if the port's logical state According to the state transition rules, the event transitions from valid to invalid or from invalid to valid. or When triggered, the data plane hardware immediately generates a fault reporting message. ; The message The data structure is constructed as a tuple: ； in, This is the unique identifier for this node; The port identifier p where the state transition occurred; The destination node identifier for the currently affected service flow; Timestamp is the timestamp t of the state transition time. This is an event type identifier, a constant representing "failure" or "recovery".

Images