A method and system for in-stream detection for a DTN network

Abstract

The application discloses a kind of methods and systems of detecting along with stream for DTN network.The system is deployed in deep space time-varying network environment, including four core modules of detecting along with stream activation, multistage time stamp acquisition, adaptive compression addition and trajectory reconstruction and diagnosis.The method is run on the system, including: encapsulation node is based on stream strategy matching intercept target Bundle message, in its main block after dynamic cascade hang IOAM extension block of custom type;Target message carrying extension block flows through transfer node, in bundle protocol layer, multistage time stamp cross-layer asynchronous acquisition mechanism is executed, obtains the multiple event time information of message in node internal processing process;Variable-length compression mechanism is used to encode node identifier and relative offset, generate node data slot and dynamically add;Destination node or network management center extracts node identifier to reconstruct the topology running trajectory of message actual flow and obtains the protocol stack processing overhead and storage queuing overhead of each transfer node.

Description

Technical Field

[0001] This invention belongs to the field of delay-tolerant network communication technology, specifically relating to a flow detection method and system for DTN networks. Background Technology

[0002] As deep space exploration missions become increasingly complex, space time-varying network environments face challenges from extreme physical conditions such as extremely long transmission delays, highly asymmetrical link heights, and frequent disconnections. To ensure reliable data transmission in such harsh environments, space time-varying network environments commonly employ a delay-tolerant network (DTN) architecture based on store-and-forward mechanisms and a bundle protocol (BP). Against this backdrop, to monitor the network health status of onboard nodes in real time, it is necessary to combine in-band flow measurement technology with an asynchronous management protocol (AMP) to achieve fine-grained monitoring of network performance.

[0003] However, existing flow-following detection schemes are mainly designed for low-latency, fully interconnected ground-based networks. If directly applied to deep space delay-tolerant networks, they suffer from the following serious drawbacks:

[0004] First, it is extremely dependent on high-precision clock synchronization across the entire network. Deep space nodes are affected by relativistic effects and extremely long signal propagation delays, making it very difficult to achieve high-precision absolute clock synchronization. Directly adopting the ground-based delay measurement method based on the subtraction of absolute timestamps will distort the measurement results due to physical clock differences between nodes, and will not accurately reflect network performance.

[0005] Second, the overhead of streaming is too high, consuming valuable bandwidth. Resources for intermittent deep-space links are severely constrained. Earth-Moon communication link rates are typically within tens of megabits per second, while the streaming uplink rate for long-distance Earth-Mars deep-space communication is often limited to several thousand to several hundred kilobits per second. Under such low-speed physical channels, traditional solutions add an absolute 32-bit or 64-bit timestamp to each node. As the number of forwarding hops increases, the message metadata expands linearly, which is unacceptable for the extremely bandwidth-constrained deep-space links.

[0006] Third, it cannot distinguish between network congestion and connection failure / queueing. Traditional network management systems can only measure the total end-to-end latency. However, in latency-tolerant network architectures, packet latency includes processor protocol stack processing overhead as well as memory queuing overhead caused by connection failures. Existing solutions cannot finely isolate and decouple these two at the data plane, making it difficult for the network management center to accurately locate network performance bottlenecks.

[0007] Furthermore, existing standard-defined flow detection mechanisms are primarily geared towards continuously connected internet environments, and their time correlation analysis is typically based on a unified time reference, making them unsuitable for latency-tolerant network scenarios operating on a store-and-forward basis. In summary, achieving high-precision topology trajectory reconstruction and latency performance isolation and decoupling in latency-tolerant networks without relying on network-wide clock synchronization and with minimal flow bandwidth overhead is a pressing technical bottleneck that needs to be addressed. Summary of the Invention

[0008] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a flow detection method and system for DTN networks. This invention does not rely on high-precision absolute clock synchronization between nodes for performance measurement. It utilizes multi-level relative offsets and variable-length compression mechanisms for local asynchronous cross-layer acquisition. While effectively compressing the network-wide hop count overhead, it can precisely decouple the protocol stack processing overhead and storage queuing overhead of spatial nodes, and completely reverse-reconstruct the actual running topology trajectory of packets.

[0009] Technical solution:

[0010] To achieve the above objectives, the present invention provides a flow detection method for DTN networks, comprising the following steps:

[0011] S1. The encapsulation node intercepts the target Bundle message based on the flow policy and dynamically cascades a custom type of In-situ Operations, Administration and Maintenance (IOAM) extension block after the main block of the target Bundle message; it then forwards the target Bundle message carrying the IOAM extension block to the next node.

[0012] S2. When a target Bundle message carrying an IOAM extension block flows through a relay node in a spatial time-varying network environment, the relay node extracts the IOAM extension block and executes a multi-level timestamp cross-layer asynchronous acquisition mechanism within the bundle protocol layer to obtain event time information characterizing the processing of the target Bundle message within the relay node.

[0013] S3. The relay node uses a variable-length compression mechanism to encode the relative offset corresponding to the node identifier and event time information, generating a node data slot in a unified format. The node data slot is then dynamically appended to the block-type-specific data field of the IOAM extension block according to the physical concatenation order of the message flow, forming a concatenated IOAM extension block.

[0014] S4. The destination node or network management center receives and extracts the IOAM extension blocks that are concatenated along the flow in the target Bundle message, extracts the node identifier in reverse order according to the physical concatenation order of the node data slots, reconstructs the actual running trajectory of the message, and calculates the protocol stack processing overhead and storage queuing overhead of each relay node based on the difference decoupling calculation according to the hop-by-hop relative offset.

[0015] The multi-level timestamp cross-layer asynchronous acquisition mechanism described in S2 of this invention specifically includes: the relay node, through cross-layer asynchronous embedding, collects the time when the target Bundle message is received in the BP protocol layer's receiving queue, the time when it is handed over due to link disconnection and / or congestion and written to non-volatile memory, the time when it is reread from non-volatile memory due to the opening of the contact window, and the time when the route re-encapsulation is completed and delivered to the lower layer, respectively denoted as stack entry time T1, storage entry time T2, storage exit time T3, and stack exit time T4;

[0016] The variable-length compression mechanism described in S3 specifically includes: using short ID mapping and binary stream variable-length encoding to adaptively compress and encode the node identifier and local relative time difference, generating the relative offsets including the root reference offset ΔT1, the slot relative processing offset ΔT2, the slot storage queuing relative offset ΔT3, and the slot relative transmission offset ΔIOAMT4; and performing boundary overflow detection in real time during appending, and triggering the adaptive variable-length expansion mechanism to allocate a higher bit width encoding space when the slot storage queuing relative offset ΔT3 is detected to exceed the current encoding space limit;

[0017] The actual running trajectory of the reconstructed message described in S4 specifically includes: without relying on control plane routing logs, forwarding table records and / or additional topology databases, the node identifiers of each relay node are extracted in reverse order based solely on the physical concatenation order of the data slots of each node carried along with the flow, and the node sequence of the actual flow of the message in the spatial time-varying network environment is directly restored.

[0018] In step S1, when mounting a custom type of in-band measurement and management IOAM extension block, the user payload block of the target Bundle message is not modified, the user payload block maintains the integrity verification state of the original data, and the block processing control flag in the header of the IOAM extension block is explicitly set to the transparent forwarding state when the block is unknown.

[0019] In step S2, the multi-level timestamp cross-layer asynchronous acquisition mechanism obtains event timing information representing the processing of the target Bundle message within the node, specifically including:

[0020] Push time This is the moment when the BP layer core interface receives the complete target Bundle message;

[0021] Warehouse entry time This refers to the moment when the target Bundle message is transferred to local non-volatile memory for hibernation queuing due to a disconnection and / or congestion of the forward link;

[0022] Outbound time The moment when the communication contact window opens, causing the target Bundle message to be reread from non-volatile memory into the memory transmission queue;

[0023] Off-stack time This is the moment when the BP layer completes route re-encapsulation and delivers the transport protocol to the lower layer.

[0024] Step S3 specifically includes the following steps:

[0025] S3.1 All relay nodes in the network are uniformly mapped to unsigned integer short IDs, and a variable-length compression mechanism is used for adaptive encoding. If the short ID is no greater than 23, it only occupies 1 byte of encoding space.

[0026] S3.2. Each relay node calculates the difference between the timestamps of multiple key events captured, generating a relative offset. The relative offset includes: root baseline offset. Relative processing offset within the slot relative offset of storage queue within slot and the relative transmission offset within the slot ;in The message creation time;

[0027] S3.3. Concatenate the short ID and each relative offset sequentially into a node data slot, and determine the queued relative offset stored in the slot in real time during appending. If the current byte encoding space limit is exceeded, an adaptive variable-length expansion mechanism is triggered to allocate a higher bit-width encoding space.

[0028] In step S4, the protocol stack processing overhead The specific calculation formula is as follows: ;

[0029] The storage queuing overhead The specific calculation formula is as follows: .

[0030] This invention also provides a flow-following detection system for DTN networks. This system is deployed in various nodes within a spatially time-varying network environment and includes a flow-following detection activation module, a multi-level timestamp acquisition module, an adaptive compression and appending module, and a trajectory reconstruction and diagnosis module, wherein:

[0031] The following detection activation module is deployed in the BP protocol layer of the encapsulation node. It is used to intercept target Bundle messages based on flow policy matching, and dynamically cascade custom type in-band measurement and management (IOAM) extension blocks after the main block of the target Bundle message, and forward the target Bundle message carrying the IOAM extension blocks to the next node.

[0032] The multi-level timestamp acquisition module is deployed in the BP protocol layer of the relay node. It is used to extract the IOAM extension block when the target Bundle packet carrying the IOAM extension block flows through, and to collect the stacking time of the packet in the node through a cross-layer asynchronous instrumentation mechanism. , Warehouse entry time Outbound time Stack time ;

[0033] The adaptive compression appending module is deployed in the BP protocol layer of the relay node. It is used to calculate the relative offset between multiple key event timestamps, compress and encode node data slots through short ID mapping and variable length compression mechanism, and dynamically cascade the node data slots to the IOAM extension block to form a cascaded IOAM extension block.

[0034] The trajectory reconstruction and diagnosis module is deployed at the destination node or network management center. It is used to receive and extract the IOAM extension blocks that are concatenated with the flow, extract the node identifiers of each relay node in reverse order according to the concatenation order, reconstruct the actual running topology trajectory of the message, and calculate the protocol stack processing overhead and storage queuing overhead of each node based on the difference decoupling formula.

[0035] Beneficial effects:

[0036] (1) Reduce reliance on high-precision clock synchronization across the entire network. This invention constructs a relative offset measurement system based on the node's local time axis, collecting the relative difference of message flow events locally at the node. The network management center only needs to use local differences for decoupling calculations, which eliminates the rigid dependence of the flow measurement process on the microsecond-level high-precision absolute clock synchronization protocol between all network nodes from a mathematical perspective, and avoids the cumulative distortion error caused by the inherent clock difference of the physical clocks of heterogeneous nodes in the deep space disconnection.

[0037] (2) Achieving lightweight variable-length compression of streaming metadata overhead. This invention utilizes short ID mapping and a variable-length compression mechanism to compress timestamps and topology information under normal conditions to a byte-level encoding space. Simultaneously, for deep-space long-latency link breakage scenarios, an adaptive variable-length expansion mechanism is proposed. When the current encoding space limit is exceeded, the encoding bit width is automatically expanded, which ensures extremely low streaming overhead for regular bandwidth and prevents time value overflow caused by long link breaks.

[0038] (3) A dual-dimensional decoupling measurement mechanism for node dwell time in DTN scenarios is proposed. This invention accurately decomposes the total dwell time of nodes into protocol stack processing overhead in the data plane. ) and storage queuing overhead ( This provides quantitative data support for the network management center to make precise judgments about network bottlenecks, determining whether the bottlenecks are due to insufficient computing power of relay nodes or delays caused by the disconnection of physical links in space.

[0039] (4) Achieve lightweight reverse reconstruction of message trajectory. This invention does not rely on control plane routing logs, forwarding table records or additional topology databases. It can directly restore the actual node sequence of message flow in the spatial time-varying network environment based solely on the physical concatenation order of the data slots of each node carried along with the flow, which significantly reduces the complexity of topology reconstruction and storage overhead of the network management center. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the macroscopic topology of a spatial time-varying network environment provided in an embodiment of the present invention;

[0041] Figure 2 This is a block diagram of a virtual module of the flow detection system provided in an embodiment of the present invention;

[0042] Figure 3 This is a schematic diagram of multi-level timestamp embedding and storage flow within a single-node protocol stack provided in an embodiment of the present invention;

[0043] Figure 4 This is a flowchart illustrating the flow performance and topology measurement method provided in this embodiment of the invention.

[0044] Figure 5 This is a structural diagram of a lightweight IOAM extension block with a three-level cascaded nested structure provided in an embodiment of the present invention. Detailed Implementation

[0045] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0046] This invention provides a flow detection method and system for DTN networks. The spatially time-varying network nodes use a bundle protocol as the basis for mesh transmission.

[0047] See Figure 1This illustration shows a macroscopic topology diagram of a space time-varying network provided by an embodiment of the present invention. In deep space communication applications such as Mars exploration, due to obstruction caused by celestial motion or periodic changes in distance, the physical links between encapsulation nodes (such as Mars rovers), relay nodes A, B, and C (such as onboard relay satellites), and the destination node / network management center (such as Earth ground stations) are time-varying discontinuous links. When messages are transmitted in the network, an end-to-end reachable path cannot be guaranteed, and hop-by-hop routing must be performed relying on the "store-and-forward" mechanism of relay nodes.

[0048] See Figure 2 To deploy a measurement mechanism in this spatially time-varying network environment, this invention provides a virtual module diagram architecture. The system mainly consists of a flow detection activation module deployed on the encapsulation node, a multi-level timestamp acquisition module and an adaptive compression appending module deployed on each relay node, and a trajectory reconstruction and diagnosis module deployed on the destination node or network management center. Each module is embedded in the BP layer of each node, and measurement is completed through the collaboration of the control plane and the data plane.

[0049] The following is combined with Figure 4 The flowchart shown illustrates the complete lifecycle of the lightweight flow performance and topology metric method of this invention.

[0050] Step S1: The encapsulation node intercepts the target Bundle message based on the flow policy and dynamically cascades a custom type of in-band measurement and management (IOAM) extension block after the main block of the target Bundle message; it then forwards the target Bundle message carrying the IOAM extension block to the next node, as follows:

[0051] First, when the BP protocol layer of the encapsulation node generates or intercepts a service bundle message that is about to be sent, its internal flow detection activation module will read the source endpoint identifier (EID) or destination EID in the primary block of the message.

[0052] The system matches the extracted EID against the list of flow detection policies preset by the network management agent. If no policy is matched, normal transparent forwarding is performed; if a policy is matched, the flow detection probe is triggered, and a custom-type in-band measurement and management (IOAM) extension block is dynamically cascaded after the primary block of the target bundle packet. During this process, the module keeps the user payload block data intact to maintain the integrity verification status of the original service data.

[0053] To ensure forward compatibility across the entire network, the following detection activation module explicitly sets a specific bit in the "Block Processing Control Flags" of the custom IOAM extension block header to the "transparent forwarding when the block is unknown" state. See also Figure 5 In the first-level cascaded nested structure of this Bundle layer, if some heterogeneous relay nodes through which the flow passes do not have a flow detection program deployed, these non-IOAM nodes will not parse or report errors on the extension block after reading the control flag bit, but will directly pass it through and forward it, achieving excellent forward compatibility.

[0054] Step S2: When the target Bundle message carrying the IOAM extension block flows through a relay node in a spatially time-varying network environment, the relay node extracts the IOAM extension block and executes a multi-level timestamp cross-layer asynchronous acquisition mechanism within the bundle protocol layer to obtain event time information characterizing the processing of the target Bundle message within the relay node; specifically as follows:

[0055] When a target Bundle message carrying an IOAM extension block is forwarded hop-by-hop in a space time-varying network environment, each time it arrives at a satellite relay node that supports flow detection (such as relay node A), the multi-level timestamp collection module of that node will collect the local timestamps of the four core events of the message flowing within the node stack through a cross-layer asynchronous instrumentation mechanism.

[0056] See Figure 3 This demonstrates the flow between multi-level timestamp tracking points and physical storage entities within a single-node protocol stack. The collection locations and definitions of these four key event timestamps are as follows:

[0057] 1. Push time When the BP layer core rendezvous layer interface receives the complete underlying target Bundle message and sends it to the memory receive queue, the data plane interception thread immediately captures the local system clock and records it as . .

[0058] 2. Warehousing time In deep space scenarios with long-delay link interruptions, if the BP layer routing resolver detects a broken forward physical link (e.g., Earth or the next-hop satellite is not visible) or severe congestion in the queue, preventing immediate packet forwarding, the routing scheduler transfers the bundle packet to local non-volatile memory, such as onboard flash memory or a radio (SDR) database, for hibernation and queuing. The local clock is captured and recorded as the ingestion time at the moment of successful writing to the storage entity. .

[0059] 3. Outbound time When the communication contact window reopens and the physical link becomes available again due to celestial motion, the routing scheduling mechanism reads the dormant target bundle packet from non-volatile memory and re-reads it into the memory transmission queue. At the instant the reading is completed, the local clock is captured and recorded as the outgoing time. .

[0060] 4. Off-stack time The moment the BP layer completes route re-encapsulation, next-hop EID rewriting, and finally delivers the target Bundle packet to the underlying rendezvous layer protocol for slice transmission, the local clock is captured and recorded as the off-stack time. .

[0061] Step S3: The relay node uses a variable-length compression mechanism to encode the relative offsets corresponding to the node identifier and event time information, generating node data slots in a unified format. These node data slots are then dynamically appended to the block-specific data field of the IOAM extension block according to the physical concatenation order of the message flow, forming a concatenated IOAM extension block, as detailed below:

[0062] After obtaining the local After the timestamp, the adaptive compression append module of the relay node abandons recording the high-overhead absolute timestamp and instead converts it into a relative offset on the local timeline.

[0063] This module calls the local interpolation algorithm to calculate four relative offsets:

[0064] Root reference offset: ;

[0065] Relative processing offset within the slot: Reflects the overhead of node protocol stack parsing;

[0066] In-slot storage queue relative offset: Reflects the incurred costs due to supply chain disruptions;

[0067] Relative transmission offset within the slot: This reflects the overhead of repackaging from the library to the stack.

[0068] To address the extremely valuable bandwidth of space-constrained links, the system performs short mapping on the current node's ID, uniformly mapping it to an unsigned integer short ID, and applies Concise Binary Object Representation (CBOR, based on RFC 8949 standard) adaptive stream coding. When the node's short... At that time, after encoding, it only occupies 1 byte in the binary stream.

[0069] See Figure 5 The third-layer node data field structure, wherein the short ID is relative to each calculated variable-length offset. Data slots are concatenated and assembled according to the field definitions, and then encapsulated into a unified format Node DataSlot. Because network outages in spatially time-varying networks can last for days or even weeks, this leads to… The value can be extremely large, and the adaptive compression append module will initiate boundary overflow detection in real time during appending. Once it is found that the storage queuing offset exceeds the upper limit of the currently allocated encoding space, the adaptive variable-length expansion mechanism will be automatically triggered to adaptively allocate a compliant higher-bit-width encoding space, fundamentally preventing timestamp wraparound and counter overflow errors caused by long-chain breaks in deep space. Finally, the module cascades the node data slot into the block-specific data field of the IOAM extension block in an append manner.

[0070] Step S4: The destination node or network management center receives and extracts the IOAM extension blocks concatenated along the flow in the target Bundle packet. It then extracts the node identifiers in reverse order of the physical concatenation of node data slots, reconstructs the actual trajectory of the packet, and calculates the protocol stack processing overhead and storage queuing overhead of each relay node based on the hop-by-hop relative offset and difference decoupling, as detailed below:

[0071] When the target Bundle message traverses a multi-hop time-varying link in space and is finally delivered to the destination node or Earth network management center, its internal trajectory reconstruction and diagnosis module extracts the IOAM extension block that is cascaded along with the flow.

[0072] First, the network management center reads the concatenated CBOR ordered array from the specific data field within the extended block. Since each relay node dynamically appends its metadata during forwarding, the physical arrangement of node data slots in the data stream forms a one-to-one mapping with the actual topological concatenation order of the packet in the network. The trajectory reconstruction and diagnostic module, based on the physical concatenation position of the data slots, reverse-engineers the short node ID for each hop, thereby directly reconstructing the actual trajectory of the Bundle packet in the spatially time-varying network environment.

[0073] Next, for each hop node's data slot, this module extracts... Using the aforementioned difference decoupling formula, the operating performance indicators of this node are calculated independently:

[0074] Node protocol stack processing overhead (CPU computing power bottleneck indicator): ;

[0075] Node storage queuing overhead (Link physical disconnection indicator): .

[0076] Since this formula is entirely based on the relative offset of the difference collected locally at the node (e.g. - All of these are local clocks of relay node A. In terms of mathematical principles, this eliminates the absolute physical clock asynchrony error between encapsulation nodes, relay nodes, and destination nodes caused by physical distance or relativistic effects, and does not require any high-precision clock synchronization protocol interaction across the entire network.

[0077] Typical Application Case Data Flow Description

[0078] To further verify the feasibility and lightweight effect of the embodiments of the present invention, a complete numerical application case based on a time-varying link for Mars exploration is given below:

[0079] Assuming the encapsulated node (such as a Mars rover) operates in its local time... A Bundle message was created. This message flows through onboard relay node A. It is known that the physical hardware clock of relay node A has a time difference compared to the standard clock due to crystal aging and other reasons. The inherent absolute clock error.

[0080] 1. When a message arrives at relay node A and enters its bundle protocol stack BP layer, relay node A collects its local time stack stamp as... At this point, the compression append module calculates and records the root reference offset. .

[0081] 2. Subsequently, relay node A detects that the forward physical link towards Earth has been broken due to Mars' rotation. The BP scheduler immediately transfers the message to the onboard Flash memory for suspension. At this time, the local storage timestamp of acquisition node A is... The relative processing offset within the slot was calculated. .

[0082] 3. The message remained dormant in Flash memory, waiting. After a considerable period, the Mars relay satellite re-established visibility contact with the Earth ground station, the window opened, and the message was reread into the memory transmission buffer queue. At this time, the local output timestamp of acquisition node A was... The relative offset of the storage queue within the slot is calculated. Because the overhead of the broken link is relatively large, the CBOR-based variable-length compression mechanism detects it and automatically expands the storage space from the usual single-byte to a double-byte encoding space for safe storage.

[0083] 4. Finally, the BP layer completes the routing encapsulation of the packet and delivers it to the transport layer. At this point, the local off-stack timestamp of data collection node A is... Calculate the relative transmission offset. Relay node A will send the short ID and... The compressed data is encapsulated and appended to the node data slot before being sent out.

[0084] When the message is finally delivered to the Earth Network Management Center, the center extracts the data slot of relay node A in reverse process. The protocol stack processing overhead of node A is then directly calculated using the difference decoupling formula. Storage queuing overhead The entire calculation process is based entirely on the relative offset difference. The network management center does not need to perceive or correct the +500ms absolute clock error that exists locally in relay node A to obtain the same node performance measurement results as in the case of no clock deviation. This shows that the present invention can achieve accurate fine-grained performance measurement without establishing high-precision absolute clock synchronization between nodes.

[0085] It should be noted that the statement that high-precision absolute clock synchronization between nodes is unnecessary refers to the process of performance measurement during flow. When calculating the overhead of protocol stack processing and storage queuing at relay nodes in the computational space, the acquisition of each timestamp and the calculation of relative offsets are all completed within the node's local timeline. Therefore, there is no need to construct a high-precision unified time reference for performance measurement. Furthermore, this invention does not exclude system time maintenance mechanisms used by delay-tolerant networks to meet functions such as message lifecycle management, contact planning scheduling, or network operation and maintenance; rather, it reduces the dependence of the performance measurement process on high-precision time synchronization between nodes.

Claims

1. A flow detection method for DTN networks, characterized in that, Includes the following steps: S1. The encapsulation node intercepts the target Bundle message based on the flow policy matching, and dynamically cascades a custom type of in-band measurement and management (IOAM) extension block after the main block of the target Bundle message; it then forwards the target Bundle message carrying the IOAM extension block to the next node. S2. When the target Bundle message carrying the IOAM extension block flows through the relay node in the spatial time-varying network environment, the relay node extracts the IOAM extension block and executes a multi-level timestamp cross-layer asynchronous acquisition mechanism in the bundle protocol layer to obtain event time information characterizing the processing process of the target Bundle message inside the relay node. S3. The relay node uses a variable-length compression mechanism to encode the relative offset corresponding to the node identifier and event time information, generating a node data slot in a unified format, and dynamically appends the node data slot to the block-specific data field of the IOAM extension block according to the physical concatenation order of the message flow to form a concatenated IOAM extension block. S4. The destination node and / or network management center receive and extract the IOAM extension blocks concatenated along the flow in the target Bundle message, extract the node identifier in reverse order according to the physical concatenation order of the node data slots, reconstruct the actual running trajectory of the message, and obtain the protocol stack processing overhead and storage queuing overhead of each relay node based on the difference decoupling calculation according to the relative offset of each hop. The multi-level timestamp cross-layer asynchronous acquisition mechanism mentioned in S2 refers to the following: the relay node collects the target Bundle message in the BP protocol layer's receive queue time, the time of handover due to link disconnection and / or congestion and writing to non-volatile memory, the time of rereading from non-volatile memory due to the opening of the contact window, and the time of route re-encapsulation and delivery to the lower layer through cross-layer asynchronous embedding, respectively denoted as stack entry time T1, storage entry time T2, storage exit time T3, and stack exit time T4; The variable-length compression mechanism described in S3 refers to: using short ID mapping and binary stream variable-length encoding to adaptively compress and encode the node identifier and local relative time difference, generating the relative offsets including the root reference offset ΔT1, the slot relative processing offset ΔT2, the slot storage queuing relative offset ΔT3, and the slot relative transmission offset ΔT4; and performing boundary overflow detection in real time during appending, and triggering the adaptive variable-length expansion mechanism in the variable-length compression mechanism to allocate a higher bit width encoding space when the slot storage queuing relative offset ΔT3 is detected to exceed the current encoding space limit; The actual running trajectory of the reconstructed message described in S4 refers to: without relying on control plane routing logs, forwarding table records and / or additional topology databases, the node identifiers of each relay node are extracted in reverse order based solely on the physical concatenation order of the node data slots carried along with the flow, and the node sequence of the message actually flowing in the time-varying topology is directly restored.

2. The flow detection method for DTN networks according to claim 1, characterized in that, In step S1, when mounting a custom type of in-band measurement and management (IOAM) extension block, the user payload block of the target Bundle message is not modified. The user payload block maintains the integrity verification state of the original data, and the block processing control flag in the header of the IOAM extension block is explicitly set to the transparent forwarding state when the block is unknown.

3. A flow detection method for DTN networks according to claim 1 or 2, characterized in that, In step S2, the multi-level timestamp cross-layer asynchronous acquisition mechanism obtains event timing information characterizing the processing of the target Bundle message within the node, specifically including: Push time This is the moment when the BP layer core interface receives the complete target Bundle message; Warehouse entry time , is the moment when the target Bundle message is transferred to local non-volatile memory for hibernation queuing due to forward link disconnection and / or congestion; Outbound time The moment when the communication contact window opens, causing the target Bundle message to be reread from non-volatile memory into the memory transmission queue; Off-stack time This is the moment when the BP layer completes route re-encapsulation and delivers the transport protocol to the lower layer.

4. The flow detection method for DTN networks according to claim 3, characterized in that, Step S3 specifically includes the following steps: S3.1 All relay nodes in the network are uniformly mapped to unsigned integer short IDs, and the variable-length compression mechanism is used for adaptive encoding. If the short ID is not greater than 23, it occupies only 1 byte of encoding space. S3.2 Each relay node calculates the difference between the timestamps of multiple key events captured, generating a relative offset, wherein the relative offset includes: root reference offset. Relative processing offset within the slot relative offset of in-slot storage queue and the relative transmission offset within the slot ;in The message creation time; S3.

3. Concatenate the short ID and each relative offset sequentially into a node data slot, and determine the queued relative offset stored in the slot in real time during appending. If the current byte encoding space limit is exceeded, the adaptive variable-length expansion mechanism in the variable-length compression mechanism is triggered to allocate a higher bit-width encoding space.

5. The flow detection method for DTN networks according to claim 4, characterized in that, In step S4, the protocol stack processing overhead The specific calculation formula is as follows: ； The storage queuing overhead The specific calculation formula is as follows: 。 6. A flow detection system for DTN networks, the system being used to perform the method according to any one of claims 1 to 5, characterized in that: The system is deployed in various nodes within a spatial time-varying network environment, and includes a flow-following detection activation module, a multi-level timestamp acquisition module, an adaptive compression and appending module, and a trajectory reconstruction and diagnosis module, wherein: The following detection activation module is deployed in the BP protocol layer of the encapsulation node. It is used to intercept target Bundle packets based on flow policy matching, and dynamically cascade a custom type of in-band measurement and management (IOAM) extension block after the main block of the target Bundle packet, and forward the target Bundle packet carrying the IOAM extension block to the next node. The multi-level timestamp acquisition module is deployed in the BP protocol layer of the relay node. It is used to extract the IOAM extension block when the target bundle packet carrying the IOAM extension block flows through it, and to collect the stacking time of the packet flowing within the node through a cross-layer asynchronous instrumentation mechanism. , Warehouse entry time Outbound time Stack time ; The adaptive compression append module is deployed in the BP protocol layer of the relay node. It is used to calculate the relative offset between multiple key event timestamps, compress and encode node data slots through short ID mapping and the variable length compression mechanism, and dynamically cascade the node data slots to the IOAM extension block to form a cascaded IOAM extension block. The trajectory reconstruction and diagnosis module is deployed at the destination node and / or network management center. It is used to receive and extract the IOAM extension blocks that are cascaded along the flow, extract the node identifiers of each relay node in reverse order according to the cascading order, reconstruct the actual running topology trajectory of the packet, and calculate the protocol stack processing overhead and storage queuing overhead of each node based on the difference decoupling formula.

Images