Intelligent link sensing platform design and implementation method based on fragment retransmission
By using the fragmented status monitoring and dynamic fragmented retransmission strategy of the intelligent link sensing platform, the problem of accurate positioning and coordination of link sensing and fragmented retransmission in dynamic environments is solved, and efficient and reliable data transmission is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-25
- Publication Date
- 2026-04-07
AI Technical Summary
Existing link awareness and fragmented retransmission solutions cannot adapt to dynamic and harsh link environments, resulting in coarse-grained link awareness, static fragmented retransmission strategies, lack of precise positioning and dynamic adjustment, and no coordination between the awareness module and the retransmission module, which cannot form an optimization closed loop, leading to low data transmission reliability and efficiency.
Design an intelligent link awareness platform based on fragmented retransmission. Through fragmented link status awareness, dynamic fragmentation decision-making, intelligent retransmission execution, and policy closed-loop optimization, it realizes fragmented status monitoring, dynamic fragmentation adjustment, and selective retransmission. The policy parameters are optimized by combining Q-Learning reinforcement learning algorithm.
It improves link awareness accuracy, shortens fault diagnosis time, increases fault location accuracy, reduces retransmission overhead, enhances data transmission reliability and efficiency, and adapts to complex and ever-changing network environments.
Smart Images
Figure CN121814682A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of data transmission and communication technology, specifically relating to a design and implementation method of an intelligent link sensing platform based on fragmented retransmission. Background Technology
[0002] With the development of fields such as the Industrial Internet and satellite communications, data transmission scenarios are gradually shifting from "stable wired links" to "unstable wireless / long-distance links." These links generally suffer from high packet loss rates (10%-30%), high latency jitter (50ms-500ms), and large bandwidth fluctuations, placing stringent demands on the reliability and real-time performance of data transmission. Traditional link awareness platforms only monitor overall link metrics (such as packet loss rate and average latency) and fail to delve into the "fragmentation layer" of data transmission. This means they cannot obtain the transmission status of individual data fragments (whether fragments are lost, the number of retransmissions, and transmission latency). In other words, traditional platforms only know that "a certain data packet is lost," but cannot pinpoint "a specific fragment within the data packet is lost, or whether the transmission latency of the lost fragment is abnormal." This results in low accuracy in problem diagnosis and a lack of precise basis for retransmission strategies.
[0003] Secondly, traditional fragmentation retransmission strategies are mostly static strategies, using a "fixed parameter" design, that is, the fragment size is fixed. This leads to increased protocol overhead due to excessive fragmentation on high-bandwidth links, and severe single packet loss due to excessively large fragments on low-bandwidth / high packet loss links. The retransmission range is fixed, that is, the entire data packet is retransmitted each time, resulting in serious bandwidth waste. The retransmission timing is fixed, without taking into account the real-time latency adjustment of the link, resulting in retransmission too early (ineffective) on high-latency links and retransmission too late (affecting real-time performance) on low-latency links.
[0004] In addition, in traditional solutions, the link awareness module and the fragment retransmission module are independent of each other and lack coordination. The awareness module only outputs the overall status of the link and does not participate in the adjustment of the retransmission strategy; the retransmission module only executes according to the preset rules and does not provide feedback on the retransmission effect to the awareness module.
[0005] In summary, existing link awareness and fragmented retransmission solutions cannot adapt to dynamic and harsh link environments. There is an urgent need for an integrated platform that integrates "fragment-level awareness, dynamic decision-making, intelligent execution, and closed-loop optimization" to improve the reliability and efficiency of data transmission.
[0006] This invention aims to address the shortcomings of existing link awareness and data transmission technologies. These include: coarse-grained link awareness (monitoring only the overall link status, lacking monitoring of individual data fragments in terms of packet loss, latency, retransmission count, etc.), resulting in inaccurate identification of transmission problems; static fragmentation retransmission strategies (fixed fragment size, fixed retransmission range, fixed retransmission timing), failing to dynamically adjust based on real-time link status (bandwidth, packet loss rate, latency), leading to bandwidth waste or low retransmission efficiency; and lack of collaboration between the link awareness module and the fragmentation retransmission module, with awareness data not guiding retransmission strategies and retransmission effects not being fed back to the awareness module, thus failing to form an optimization loop. This invention proposes a link awareness platform design and implementation method based on fragmentation retransmission, achieving fine-grained link awareness, intelligent fragmentation based on real-time link status, and deep collaboration between the fragmentation strategy and the link awareness module. This significantly improves the reliability and efficiency of data transmission under high packet loss / high latency links, and is suitable for scenarios sensitive to link quality, such as the Industrial Internet and satellite communications. Summary of the Invention
[0007] (a) Technical problems to be solved
[0008] The technical problem to be solved by this invention is how to provide a design and implementation method for an intelligent link awareness platform based on fragmented retransmission, so as to solve the problem that existing link awareness and fragmented retransmission schemes cannot adapt to dynamic and harsh link environments.
[0009] (II) Technical Solution
[0010] To address the aforementioned technical problems, this invention proposes a design and implementation method for an intelligent link awareness platform based on fragmented retransmission, which includes the following steps:
[0011] Step 1, Segment-level Link Status Awareness: The link status is collected and perceived through steps such as segment identifier implantation, segment status feedback, multi-dimensional indicator calculation, and real-time indicator reporting.
[0012] The second step, dynamic fragmentation decision: Based on the link status input, the optimal fragmentation size is calculated using the double constraint algorithm. After deciding whether to adopt a fragmentation redundancy strategy, a transmission strategy is formed and distributed.
[0013] The third step, intelligent retransmission execution: Receive the transmission strategy formed in the previous step, parse it, execute the retransmission strategy through data fragmentation processing and packet loss fragmentation detection, and adopt a redundancy supplementary retransmission mechanism to avoid retransmission failure, and feed back the retransmission parameters.
[0014] Step 4: Policy closed-loop optimization: The Q-Learning reinforcement learning algorithm is used to iteratively optimize the policy parameters, and the policy is distributed through a canary release strategy to further improve the transmission success rate and further reduce bandwidth overhead and latency.
[0015] (III) Beneficial Effects
[0016] This invention proposes a design and implementation method for an intelligent link awareness platform based on fragmented retransmission. Compared with existing technologies, it achieves fragment-level status monitoring, improves link awareness accuracy, shortens fault diagnosis time, and enhances fault location accuracy. It employs a "dynamic fragmentation + selective retransmission + redundancy on-demand activation" mechanism, intelligently adjusting the size and fragmentation method of data packets according to current network conditions and application requirements to ensure optimal transmission performance in different network environments. By monitoring fragmented reception status, it only retransmits lost data fragments, avoiding unnecessary resource waste and significantly reducing retransmission overhead. Through adaptive network conditions, it automatically activates the redundancy mechanism in unstable or high packet loss situations to increase data transmission reliability, while reducing redundancy and improving transmission efficiency when network conditions are good. This enables stable and efficient data transmission services in complex and ever-changing network environments. Attached Figure Description
[0017] Figure 1 This is a flowchart of the processing method of the present invention. Detailed Implementation
[0018] To make the objectives, contents, and advantages of the present invention clearer, the specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples.
[0019] To address the aforementioned technical problems, this invention proposes a design and implementation method for a link-aware platform based on fragmented retransmission, which includes the following steps:
[0020] Step 1, Segment-level Link Status Awareness: The link status is collected and perceived through steps such as segment identifier implantation, segment status feedback, multi-dimensional indicator calculation, and real-time indicator reporting.
[0021] The second step, dynamic fragmentation decision: Based on the link status input, the optimal fragmentation size is calculated using the double constraint algorithm. After deciding whether to adopt a fragmentation redundancy strategy, a transmission strategy is formed and distributed.
[0022] The third step, intelligent retransmission execution: Receive the transmission strategy formed in the previous step, parse it, execute the retransmission strategy through data fragmentation processing and packet loss fragmentation detection, and adopt a redundancy supplementary retransmission mechanism to avoid retransmission failure, and feed back the retransmission parameters.
[0023] Step 4: Policy closed-loop optimization: The Q-Learning reinforcement learning algorithm is used to iteratively optimize the policy parameters, and the policy is distributed through a canary release strategy to further improve the transmission success rate and further reduce bandwidth overhead and latency.
[0024] Example 1:
[0025] Figure 1 This is the main flowchart of the technical solution of this invention. For example... Figure 1 As shown, the link-aware platform design and implementation method based on fragmented retransmission proposed in this invention includes the following steps:
[0026] Step 1: Segment-level link status awareness
[0027] This step addresses the issue of "coarse sensing granularity," enabling precise data acquisition and index calculation at the slice level. It is executed in four steps:
[0028] S11. Segmentation Identification Implantation (Source Collection Preparation)
[0029] For data packets transmitted using the TCP protocol, the sending end uses the Linux kernel TCP Option extension field to insert "unique fragment identifier (4-byte fragment ID), sending timestamp (8 bytes, accurate to milliseconds), link identifier (4 bytes, distinguishing different links), and checksum (4 bytes)" into each TCP fragment, totaling 20 bytes;
[0030] For application layer protocols, a 12-byte extended field is added to the header of the application layer data packet. The format is "total number of fragments (2 bytes) + current fragment ID (2 bytes) + sending timestamp (6 bytes) + packet loss flag (2 bytes, initially 0)" to ensure that the receiving end can identify individual fragments.
[0031] S12, Fragmentation Status Feedback (Receiver Response)
[0032] After parsing the fragment identifier, the receiving end reports the status through the "fragment ACK message": In the TCP scenario, the TCP ACK packet is reused, and the Option field carries "fragment ID, reception timestamp (8 bytes), and packet loss identifier (1 byte, 0 = normal reception, 1 = packet loss)";
[0033] If the fragmentation verification fails, meaning the data is corrupted, the receiving end directly marks "packet loss flag = 1" to avoid invalid retransmission.
[0034] S13. Multi-dimensional indicator calculation (data processing)
[0035] The perception layer triggers indicator calculations every 100ms, generating two types of core indicators: segment-level and link-level.
[0036] 1) Segment-level metrics:
[0037] The granularity is a single fragment, and the transmission delay is equal to the receive timestamp minus the send timestamp. The number of retransmissions is equal to the number of times the "packet loss flag = 1" is returned for the same fragment ID. The single-link fragment loss rate is equal to the number of lost fragments on the link / the total number of fragments on the link.
[0038] 2) Link-level metrics
[0039] The granularity is a single link, aggregating fragmented data. The average link latency jitter = the standard deviation of fragmented latency (reflecting latency fluctuations); the link bandwidth utilization = the actual total size of transmitted fragments / the theoretical bandwidth of the link (within 100ms); the retransmission bandwidth overhead = the total size of retransmitted fragments / the total size of transmitted fragments on the link.
[0040] S14. Real-time reporting of indicators (data output)
[0041] The status reporting module encapsulates data in the format of "link identifier + timestamp + indicator set" and reports it to the dynamic sharding decision module of the decision layer in real time via MQTT or gRPC protocol. If the reporting fails, local caching is enabled, and data is resent in batches after the link is restored to avoid data loss.
[0042] Step Two: Dynamic Segmentation Decision Making (Core Process for Decision-Making)
[0043] This step addresses the issue of "static strategy," dynamically adjusting shard size and redundancy based on perception layer metrics, and is executed in 5 steps:
[0044] S21, Link Status Input (Data Reception)
[0045] The dynamic sharding decision module listens to the metrics reported by the perception layer, stores them according to "link identifier," and updates the link status snapshot every 100ms. Key input parameters include:
[0046] Current link bandwidth B (unit: kbps);
[0047] Current fragmentation packet loss rate L (unit: %) of the link;
[0048] Link average delay jitter D (unit: ms).
[0049] S22. Calculate the optimal partition size using the double-constraint algorithm.
[0050] The "dual-constraint adaptive algorithm" is adopted to balance "real-time performance" and "packet loss resistance" to determine the fragment size S (unit: bytes), as shown in the following formula:
[0051]
[0052] Constraint 1 is Ensure that the fragment transmission time is ≤ 1 / 5 of the latency jitter (taken from industry experience value) to avoid timeout retransmission due to latency fluctuations (avoid misjudging fragment packet loss due to excessive transmission time when the link latency fluctuates).
[0053] Constraint 2 is The parameter 0.1 (representing a packet loss rate of 10%) is introduced to reduce the fragment size under high packet loss rates, thereby reducing the number of bytes lost in a single packet loss and lowering retransmission bandwidth overhead; under low packet loss rates, the fragment size can be appropriately increased to reduce protocol header overhead.
[0054] In addition, the upper limit of the fragment size is set to 1024×1024 bytes, or 1MB, to avoid excessively large fragments that would increase protocol overhead; at the same time, a lower limit of 1KB is set to avoid excessively small fragments that would cause the header overhead to exceed the proportion.
[0055] S23, Fragmentation Redundancy Decision (Enhanced Packet Loss Resistance) A redundancy strategy is dynamically adjusted based on the fragmentation packet loss rate L to balance "packet loss resistance" and "bandwidth overhead": When the packet loss rate L > 20% (high packet loss link), RS coding (Reed-Solomon coding) is enabled to generate redundant fragments. The fragments are generated according to the redundancy ratio (number of redundant fragments / number of original fragments, represented by R. The higher the R value, the stronger the data's ability to resist loss or damage) 1:1 (1 original fragment + 1 redundant fragment). This ensures that when the packet loss rate is ≤ 30%, the receiving end can recover the data from any fragment (original fragment or redundant fragment) without retransmission. When 10% < packet loss rate L ≤ 20% (medium packet loss link), the redundancy ratio is reduced to 2:1 (2 original fragments + 1 redundant fragment). When the packet loss rate L≤10% (low packet loss link): disable redundant fragmentation and transmit only the original fragments to save bandwidth.
[0056] S24. Strategy Issuance (Execution Command Output) The dynamic sharding decision module encapsulates "link identifier + optimal shard size S + redundancy R + effective timestamp" into a policy message, which is then sent to the sharding processing module in the execution layer via a message queue. At the same time, the policy parameters are stored in the policy log library in the storage and visualization layer for subsequent auditing and optimization.
[0057] Step 3: Intelligent retransmission execution (core process of execution layer) This step addresses the problem of "low retransmission efficiency" by implementing "selective retransmission + redundancy supplementation," and is executed in 6 steps: S31. Policy Reception and Parsing (Command Reception) The sharding processing module of the execution layer subscribes to the message queue, receives the policies issued by the decision layer, and updates the local policy cache according to the "link identifier". If there is no new policy for a certain link, the most recent policy is used by default (the maximum validity period is 5 minutes, and a policy request is triggered after the timeout).
[0058] S32. Data fragmentation processing (executed at the source) The sending end processes the data to be transmitted according to the following rules: 1) Splitting The original data packet is used to calculate the "optimal fragment size S" using a double-constraint algorithm, and then split into N original fragments. Each fragment is assigned a consecutive fragment ID (incrementing from 1). 2) Generate redundant fragments The original fragments are RS encoded to generate redundant fragments. For example, when the redundancy ratio is 1:1, N original fragments correspond to N redundant fragments, and the redundant fragment ID is "original fragment ID + N". 3) Segmented packaging Insert "link identifier, fragment ID, and a marker for original or redundant fragment (0=original, 1=redundant)" into the header of each fragment to facilitate identification by the receiving end.
[0059] S33. Packet Loss and Fragmentation Detection (State Monitoring) A "sliding window mechanism" is used to monitor the fragment reception status. The window size is set to 10 (configurable), and the window contains 10 consecutive fragment IDs (e.g., the current window is [1-10]). Fragment IDs with "packet loss flag = 1" fed back by the receiver are added to the "packet loss fragment queue" in real time. When all fragments in the window are fed back (regardless of whether there is packet loss), the window slides to the next 10 fragments (e.g., [11-20]) to avoid window accumulation.
[0060] S34, Execute retransmission strategy The selective retransmission module performs retransmission decisions according to the following rules: Retransmission scope: Only fragments in the "Packet Loss Fragment Queue" are added to the retransmission queue, and the entire data packet is not retransmitted (e.g., if only fragments 3 and 5 are lost in window [1-10], then only fragments 3 and 5 will be retransmitted). Retransmission timing: Based on the "average link latency T" reported by the perception layer, the retransmission timeout is set to 2.5×T (2.5 times is taken from industry experience value to avoid early retransmission of high latency links and late retransmission of low latency links); if no feedback of "packet loss flag = 0" (no packet loss) is received after the timeout, retransmission is triggered.
[0061] S35, Redundancy supplementary retransmission (anti-retransmission failure) If the packet retransmission attempt fails after ≥2 attempts, meaning the receiver continuously reports "Packet Loss Flag = 1": If the fragment has redundant fragments (R>0), the redundancy supplementation module directly sends the corresponding redundant fragments (e.g., the redundant fragments of the original fragment 3 are N+3), and the receiving end can recover the data by RS decoding (inputting either the original fragment or the redundant fragment); If there is no redundant fragment, mark the fragment as "unrecoverable", trigger an alarm to the dashboard, and skip the fragment (to avoid infinite retransmission and bandwidth consumption).
[0062] S36. Retransmission effect feedback (closed-loop input) The selective retransmission module encapsulates "link identifier, retransmission fragment ID, number of retransmissions, retransmission success rate (number of successfully retransmitted fragments / total number of retransmitted fragments), and number of redundancy replenishment" into feedback messages, which are then reported to the policy optimization module of the decision layer through the message queue, providing data support for subsequent policy iterations.
[0063] Step 4: Optimize the closed-loop strategy (collaborative process of decision-making layer + storage layer) This process addresses the issue of insufficient coordination between sensing and retransmission. It iteratively optimizes strategy parameters based on historical data and is executed in 5 steps: S41. Historical data storage (data accumulation) Use a time-series database to store core data from the past 30 days in the following structure: Table 1 shows slice_metrics (slice-level data): tag (link identifier, slice ID), field (transmission delay, number of retransmissions, packet loss identifier), and timestamp (time stamp). Table 2 shows the link_strategy (strategy data): tag (link identifier), field (fragment size S, redundancy R, retransmission timeout coefficient), and timestamp (time stamp). Table 3 shows the transmission_effect data: tag (link identifier), field (transmission success rate, retransmission bandwidth overhead, end-to-end delay), and timestamp. A data retention strategy is adopted to automatically clean up expired data after 30 days to avoid storage overflow.
[0064] S42. Strategy Effectiveness Evaluation (Effect Analysis) The strategy optimization module triggers an effect evaluation every day at midnight, and statistics are compiled by "link scenario" (e.g., "packet loss rate L=20%-30%, link bandwidth B=1Mbps-5Mbps, latency jitter D=300ms-500ms"). Calculate the average transmission success rate P, average retransmission bandwidth overhead C, and average end-to-end delay T for each scenario. If P < 95% or C > 25% or T > 500ms, mark the strategy corresponding to this scenario as "strategy to be optimized" and extract the historical strategy parameters (S, R, retransmission timeout coefficient) and effect data for this scenario.
[0065] S43. Optimize parameter iteration (algorithm optimization) The Q-Learning reinforcement learning algorithm is used to iteratively optimize the policy parameters. The core design is as follows: State space: Link scenario parameters (L, B, D); Motion space: Parameters to be optimized (adjustment step size of S ±10%, adjustment options of R (0, 1:2, 1:1), adjustment step size of retransmission timeout coefficient ±0.1). Design a reward function: Reward = 0.6 × P − 0.3 × C − 0.1 × (T / 1000) (maximize transmission success rate, minimize bandwidth overhead and latency). Algorithm iteration: Through 1000 simulation training, the optimal parameter combination for each "scenario to be optimized" is obtained (for example, the original S=5KB is adjusted to 4.5KB, and the retransmission timeout coefficient is adjusted from 2.5 to 2.3).
[0066] S44. Optimize the strategy for canary release (risk control) To avoid the risks associated with directly deploying the new strategy to the entire system, a "canary release" mechanism will be adopted: Phase 1: Select 10% of the link nodes in the same scenario to apply the new strategy and monitor them continuously for 24 hours; Phase 2: If the average transmission success rate P≥98%, the average retransmission bandwidth overhead C≤20%, and the average end-to-end delay T≤450ms in Phase 1, expand to 50% of nodes and continue monitoring for 12 hours. Phase 3: If the results of Phase 2 are satisfactory, the new strategy will be rolled out to all nodes in the same scenario; if the results of any phase are unsatisfactory, the strategy will be rolled back to the original strategy and iterated and optimized again.
[0067] S45. Optimization effect archiving (closed loop complete) After full release, the strategy optimization module will store the "parameter comparison before and after optimization, effect comparison (P, C, T change rate)" in the strategy log library; if the optimization effect of a certain scenario meets the standard for 3 consecutive times, the strategy of that scenario will be marked as a "stable strategy" and the next evaluation cycle will be extended (from 1 day to 7 days) to reduce the consumption of computing resources.
[0068] Beneficial effects: This invention proposes a link-aware platform design and implementation method based on fragmented retransmission. Compared with existing technologies, it achieves fragment-level status monitoring, improves link-awareness accuracy, shortens fault diagnosis time, and enhances fault location accuracy. It employs a "dynamic fragmentation + selective retransmission + redundancy on-demand activation" mechanism, intelligently adjusting the size and fragmentation method of data packets according to current network conditions and application requirements to ensure optimal transmission performance in different network environments. By monitoring fragmented reception status, it only retransmits lost data fragments, avoiding unnecessary resource waste and significantly reducing retransmission overhead. Through adaptive network conditions, it automatically activates the redundancy mechanism in unstable or high packet loss situations to increase data transmission reliability, while reducing redundancy and improving transmission efficiency when network conditions are good. This enables stable and efficient data transmission services in complex and ever-changing network environments.
[0069] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.
Claims
1. A design and implementation method for an intelligent link awareness platform based on fragmented retransmission, characterized in that, The method includes the following steps: Step 1, Segment-level Link Status Awareness: The link status is collected and perceived through steps such as segment identifier implantation, segment status feedback, multi-dimensional indicator calculation, and real-time indicator reporting. The second step, dynamic fragmentation decision: Based on the link status input, the optimal fragmentation size is calculated using a double-constraint algorithm. After deciding whether to adopt a fragmentation redundancy strategy, a transmission strategy is formed and distributed. The third step, intelligent retransmission execution: Receive the transmission strategy formed in the previous step, parse it, execute the retransmission strategy through data fragmentation processing and packet loss fragmentation detection, and adopt a redundancy supplementary retransmission mechanism to avoid retransmission failure, and feed back the retransmission parameters. Step 4: Policy closed-loop optimization: The Q-Learning reinforcement learning algorithm is used to iteratively optimize the policy parameters, and the policy is distributed through a canary release strategy to further improve the transmission success rate and further reduce bandwidth overhead and latency.
2. The design and implementation method of the intelligent link awareness platform based on fragmented retransmission as described in claim 1, characterized in that, The first step includes: S11, Segmentation Identifier Implantation For data packets transmitted using the TCP protocol, the sending end uses the TCP Option extension field to insert "fragment unique identifier, sending timestamp, link identifier, and checksum" into each TCP fragment; For application layer protocols, add an extended field to the application layer data packet header, in the format of "total number of fragments + current fragment ID + sending timestamp + packet loss flag", to ensure that the receiving end can identify individual fragments; S12, Fragmentation Status Feedback After parsing the fragment identifier, the receiving end reports the status through a "fragment ACK message": In a TCP scenario, the TCP ACK packet is reused, and the Option field carries "fragment ID, reception timestamp, and packet loss identifier"; a packet loss identifier of 0 indicates normal reception, and 1 indicates packet loss; S13, Multi-dimensional indicator calculation The perception layer periodically triggers the calculation of indicators, generating two types of core indicators: segment-level and link-level. 1) Segment-level metrics: The granularity is a single fragment; transmission delay = receive timestamp - send timestamp; retransmission count = the number of times "packet loss flag = 1" is returned for the same fragment ID; single-link fragment packet loss rate = number of lost fragments on the link / total number of fragments on the link; 2) Link-level metrics The granularity is a single link, aggregating fragmented data. The average link latency jitter = the standard deviation of fragmented latency; the link bandwidth utilization = the total actual size of transmitted fragments / the theoretical bandwidth of the link; the retransmission bandwidth overhead = the total size of retransmitted fragments / the total size of transmitted fragments on the link. S14, Real-time reporting of indicators The status reporting module encapsulates data in the format of "link identifier + timestamp + indicator set" and reports it to the dynamic sharding decision module of the decision layer in real time via MQTT or gRPC protocol. If the reporting fails, local caching is enabled, and data is resent in batches after the link is restored to avoid data loss.
3. The design and implementation method of the intelligent link awareness platform based on fragmented retransmission as described in claim 2, characterized in that, In step S12, if the fragmentation verification fails, i.e. the data is corrupted, the receiving end directly marks "packet loss flag = 1" to avoid invalid retransmission.
4. The design and implementation method of the intelligent link awareness platform based on fragmented retransmission as described in claim 2, characterized in that, The second step includes: S21, Link Status Input The dynamic fragmentation decision module listens to the metrics reported by the perception layer, stores them according to "link identifier", and periodically updates the link status snapshot. The core input parameters include: current link bandwidth B, current link fragmentation packet loss rate L, and average link latency jitter D. S22. Calculate the optimal partition size using the double-constraint algorithm. A "dual-constraint adaptive algorithm" is adopted to determine the fragment size S, taking into account both "real-time performance" and "packet loss resistance". S23, Fragmentation Redundancy Decision A redundancy strategy is dynamically adjusted based on the fragmentation packet loss rate L to balance "packet loss resistance" and "bandwidth overhead": When the packet loss rate L > 20%, RS coding is enabled to generate redundant fragments. The redundancy ratio R is 1:1, that is, 1 original fragment + 1 redundant fragment. This ensures that when the packet loss rate is ≤ 30%, the receiving end can recover the data from any one fragment without retransmission. When 10% < packet loss rate L ≤ 20%, the redundancy ratio is reduced to 2:
1. When the packet loss rate L≤10%: disable redundant fragmentation and transmit only the original fragments; S24, Strategy Issuance The dynamic sharding decision module encapsulates "link identifier + optimal shard size S + redundancy R + effective timestamp" into a policy message, which is then sent to the sharding processing module in the execution layer via a message queue. At the same time, the policy parameters are stored in the policy log library in the storage and visualization layer for subsequent auditing and optimization.
5. The design and implementation method of the intelligent link awareness platform based on fragmented retransmission as described in claim 4, characterized in that, The formula for the fragment size S is as follows: Constraint 1 is Ensure that the fragment transmission time is ≤ 1 / 5 of the delay jitter; Constraint 2 is By introducing a correction parameter of 0.1, the fragment size can be reduced under high packet loss rate, thereby reducing the number of bytes lost in a single packet loss. The upper limit for the fragment size is set to 1024×1024 bytes, or 1MB, to avoid excessively large fragments leading to increased protocol overhead; at the same time, a lower limit of 1KB is set to avoid excessively small fragments causing header overhead to exceed the proportion.
6. The design and implementation method of the intelligent link awareness platform based on fragmented retransmission as described in claim 4, characterized in that, The third step includes: S31. Policy Reception and Parsing The sharding processing module of the execution layer subscribes to the message queue, receives the policies issued by the decision layer, and updates the local policy cache according to the "link identifier"; if there is no new policy for a certain link, the most recent policy is used by default. S32, Data Fragmentation Processing The sending end processes the data to be transmitted according to the following rules: 1) Splitting The original data packet is used to calculate the "optimal fragment size S" using a double-constraint algorithm, and then split into N original fragments. Each fragment is assigned a consecutive fragment ID. 2) Generate redundant fragments RS encoding is performed on the original fragments to generate redundant fragments; 3) Segmented packaging Insert "link identifier, fragment ID, and markers for original or redundant fragments" into the header of each fragment to facilitate identification by the receiving end; S33, Packet Loss and Fragmentation Detection A "sliding window mechanism" is used to monitor the fragment reception status. The window size is set to 10, and the window contains 10 consecutive fragment IDs. Fragment IDs with "packet loss flag = 1" fed back by the receiver are added to the "packet loss fragment queue" in real time. When all fragments in the window have been fed back, the window slides to the next 10 fragments. S34, Execute retransmission strategy The selective retransmission module performs retransmission decisions according to the following rules: Retransmission scope: Only fragments in the "Packet Loss Fragment Queue" are added to the retransmission queue; the entire data packet is not retransmitted. Retransmission timing: Based on the "average link delay T" reported by the perception layer, the retransmission timeout is set to 2.5×T; if no feedback of "packet loss flag = 0" is received after the timeout, retransmission is triggered. S35, Redundancy Supplement Retransmission If the packet retransmission attempt fails after ≥2 attempts, the receiver will continuously report "Packet Loss Flag = 1": If the fragment has a redundant fragment R>0, the redundancy supplementation module directly sends the corresponding redundant fragment, and the receiving end can recover the data by RS decoding. If there is no redundant fragment, mark the fragment as "unrecoverable", trigger an alarm to the dashboard, and skip the fragment. S36, Retransmission Effect Feedback The selective retransmission module encapsulates "link identifier, retransmission fragment ID, number of retransmissions, retransmission success rate, and number of redundancy replenishments" into feedback messages, which are then reported to the policy optimization module at the decision-making level through a message queue, providing data support for subsequent policy iterations.
7. The design and implementation method of the intelligent link awareness platform based on fragmented retransmission as described in claim 5, characterized in that, When the redundancy ratio is 1:1, N original fragments correspond to N redundant fragments, and the redundant fragment ID is "original fragment ID + N".
8. The design and implementation method of the intelligent link awareness platform based on fragmented retransmission as described in claim 7, characterized in that, The fourth step includes: S41, Historical Data Storage Use a time-series database to store core data from the past 30 days, categorized by shard level, strategy data, and transmission effect data. S42. Strategy Effectiveness Evaluation The strategy optimization module triggers daily performance evaluations at midnight, with statistics categorized by "link scenarios": Calculate the average transmission success rate P, average retransmission bandwidth overhead C, and average end-to-end delay T for each scenario. If P < 95% or C > 25% or T > 500ms, mark the strategy corresponding to this scenario as "strategy to be optimized" and extract the historical strategy parameters and effect data for this scenario; S43, Optimize parameter iteration The Q-Learning reinforcement learning algorithm is used to iteratively optimize the policy parameters. The core design is as follows: State space: Link scenario parameters L, B, D; Motion space: Parameters to be optimized: S, R, and retransmission timeout coefficient; Design the reward function as Reward = 0.6 × P − 0.3 × C − 0.1 × (T / 1000); Algorithm iteration: The optimal parameter combination for each "scenario to be optimized" is obtained through 1000 simulation training runs; S44, Optimize strategy for canary release To avoid the risks associated with directly deploying the new strategy to the entire system, a "canary release" mechanism will be adopted: Phase 1: Select 10% of the link nodes in the same scenario to apply the new strategy and monitor them continuously for 24 hours; Phase 2: If the average transmission success rate P≥98%, the average retransmission bandwidth overhead C≤20%, and the average end-to-end delay T≤450ms in Phase 1, expand to 50% of nodes and continue monitoring for 12 hours. Phase 3: If the results of Phase 2 are satisfactory, the new strategy will be rolled out to all nodes in the same scenario; if the results of any phase are unsatisfactory, the original strategy will be rolled back and iterated and optimized again. S45. Optimization Effect Archive After full release, the strategy optimization module stores the "parameter comparison and effect comparison before and after optimization" in the strategy log library. If the optimization effect of a certain scenario meets the standard for 3 consecutive times, the strategy for that scenario is marked as a "stable strategy", and the next evaluation cycle is extended, thereby reducing the consumption of computing resources.
9. The design and implementation method of the intelligent link awareness platform based on fragmented retransmission as described in claim 8, characterized in that, In S41, The slice-level data is slice_metrics, which includes: link identifier, slice ID, transmission delay, retransmission count, packet loss identifier, and timestamp; The strategy data is called link_strategy, which includes: link identifier, fragment size S, redundancy R, retransmission timeout coefficient, and timestamp; The transmission effect data is called transmission_effect, which includes: link identifier, transmission success rate, retransmission bandwidth overhead, end-to-end delay, and timestamp.
10. The design and implementation method of the intelligent link awareness platform based on fragmented retransmission as described in claim 8, characterized in that, In S43, the adjustment step size of S is ±10%, the adjustment options of R are 0, 1:2, and 1:1, and the adjustment step size of the retransmission timeout coefficient is ±0.1.