A TSN scheduling optimization method for merging after enhancing flow characteristics

Abstract

This application relates to the field of TSN network scheduling optimization technologies, and particularly to a TSN scheduling optimization method with merged enhanced flow characteristics, which includes: a scheduling optimization unit receives flow characteristic description information sent by an application end, and dynamically quantifies time sensitivity, priority, and traffic fluctuation tolerance through an enhanced flow characteristic analysis module; generates an optimized scheduling scheme based on the enhanced flow characteristics matching a predefined gating list template; when detecting traffic fluctuations, dynamically adjusts the gating configuration to ensure transmission determinism. This application significantly reduces the problem complexity under large-scale traffic and complex topologies by introducing a flow characteristic enhancement algorithm and a dynamic gating adjustment mechanism, quickly generates an optimized scheme, and effectively guarantees the real-time performance and transmission reliability of the TSN network.

Description

Technical Field

[0001] The present invention belongs to the technical field of time-sensitive network (TSN) scheduling optimization, and specifically relates to a TSN scheduling optimization method with enhanced and merged flow characteristics. Background Art

[0002] Time-sensitive network (TSN), as a network technology supporting high real-time and deterministic transmission, is widely used in fields such as industrial automation, vehicle communication, and multimedia transmission. In TSN, the design of the scheduling algorithm is crucial for ensuring the low-latency and high-reliability transmission of time-triggered (TT) flows. As the core element of time-aware shaper (TAS) scheduling, the gate control list (GCL) realizes precise control of network transmission by specifying which data flows are allowed to pass through the switch port in different time periods. However, in practical applications, TSN scheduling faces many challenges, especially in scenarios with large-scale traffic and complex network topologies, and existing technologies are difficult to meet the requirements of real-time and determinism.

[0003] Currently, the satisfiability modulo theories (SMT) solver is widely used to solve constraint problems and has achieved certain results in TT flow scheduling. For example, the backtracking algorithm based on SMT defines the frame offset and queue allocation as integer variables, explores valid solutions by incrementally adding constraints, and thus schedules each flow one by one. Li et al. further optimized the SMT algorithm by replacing the original link hyperperiod with the greatest common divisor of all TT flows in the link, reducing the number of time slots in the GCL, simplifying the configuration, and accelerating the calculation speed of the scheduling result. However, these methods still have significant deficiencies when facing complex network topologies and large-scale traffic. Traditional deterministic solving algorithms (such as the branch and bound method, integer programming algorithm, etc.) usually have exponential or factorial computational complexity, and as the problem scale increases, the solving time increases sharply, making it difficult to meet the strict real-time requirements of the TSN network.

[0004] In addition, the composite scheduling method attempts to achieve a better scheduling goal by combining multiple algorithms. For example, the transmission path planning problem is transformed into the maximum independent set problem in a conflict graph, and the composite method of integer linear programming (ILP) and the maximum independent set traverses the configuration vertices (cvertices) of each flow, gradually expanding the conflict graph and searching for the maximum independent set containing all flows. Although this method can theoretically provide a more comprehensive solution, its computational complexity also rises rapidly with the increase of the network scale, resulting in low solving efficiency.

[0005] What is even more difficult is that in actual applications, traffic transmission fluctuations can cause uncertainty problems. Specifically, the traffic sent by the application end sometimes cannot be transmitted according to the GCL gating in the current cycle, but instead enters another gating. This phenomenon not only destroys the integrity of the scheduling plan, but also triggers a series of chain reactions, seriously affecting the transmission certainty of the entire network topology. Therefore, how to deal with the uncertainty caused by traffic fluctuations while ensuring real-time performance has become a key issue that needs to be solved in the field of TSN scheduling.

[0006] In summary, the existing TSN scheduling technology has problems such as high computational complexity, long solution time, and insufficient adaptability to traffic fluctuations when facing large-scale traffic and complex network topology. To address these problems, a scheduling optimization method is urgently needed that can effectively reduce computational complexity, improve solution efficiency, and enhance adaptability to traffic fluctuations to meet the strict requirements of TSN networks for real-time and determinism. Summary of the invention

[0007] An embodiment of the present invention provides a TSN scheduling optimization method for merging enhanced flow features, comprising: a scheduling optimization unit deployed in a time-sensitive network TSN receives flow feature description information sent by an application end; the scheduling optimization unit dynamically quantifies the time sensitivity and priority of each data flow based on a flow feature enhancement algorithm; the scheduling optimization unit matches the enhanced flow feature information with a predefined gating list GCL template to generate an optimized scheduling scheme; when the TSN network detects actual traffic transmission fluctuations, the gating configuration in the current cycle is dynamically adjusted according to the enhanced flow features.

[0008] Furthermore, the scheduling optimization unit divides the network into multiple logical domains, each domain independently generates a scheduling plan, and synchronizes the global state through a coordinator. For ultra-large-scale networks, the method divides the network into multiple logical domains, each domain independently runs a scheduling algorithm, and the domains synchronize the global state through a coordinator and generate cross-domain scheduling rules to avoid resource conflicts.

[0009] In an exemplary embodiment, the scheduling optimization unit includes a time-aware shaper TAS, an enhanced stream feature analysis module ESAM, and a gated adaptive controller GAC.

[0010] Furthermore, the scheduling optimization unit also includes a traffic prediction module, which predicts the traffic fluctuation trend in the future period through a long short-term memory network (LSTM). If the traffic is predicted to exceed the limit, the gating configuration is adjusted in advance to reserve resources.

[0011] In an exemplary embodiment, the enhanced flow feature information includes a time sensitivity weight, a priority queue allocation weight, and a traffic fluctuation tolerance threshold.

[0012] In an exemplary embodiment, before the scheduling optimization unit receives the flow feature description information sent by the application side, it further includes: the application side generates flow feature description information including time window constraints according to historical traffic statistics information and real-time traffic requirements; the application side transmits the flow feature description information to the scheduling optimization unit through control plane signaling.

[0013] In an exemplary embodiment, the process of the application side generating the flow feature description information includes: the application side collects the historical transmission delay distribution, bandwidth occupancy rate, and jitter range of each data flow; the application side calculates the initial weight of the flow feature based on the collected data and reports it to the scheduling optimization unit.

[0014] In an exemplary embodiment, before the scheduling optimization unit generates an optimized scheduling scheme, it further includes: the enhanced flow feature analysis module ESAM performs enhanced calculation on the feature weight W of each data flow based on the formula

[0015] W f =α·T s +β·P q +γ·F t

[0016] where T f represents the time sensitivity score, P s represents the priority queue allocation score, F q represents the traffic fluctuation tolerance score, and α, β, γ are the corresponding weight coefficients respectively, and satisfy α + β + γ = 1. t Further, the weight coefficients α, β, γ are dynamically optimized through a reinforcement learning model. The reinforcement learning model takes the real-time state of the network (including delay, jitter, bandwidth occupancy rate) as the input, and the scheduling performance metrics (delay reduction rate, throughput) as the reward function, and periodically updates the weight combination through the Q-Learning algorithm to achieve adaptive adjustment of the flow feature weights.

[0017] In an exemplary embodiment, before the scheduling optimization unit dynamically adjusts the gating configuration within the current period, it further includes: the time-aware shaper TAS detects the actual transmission status of each data flow within the current period and generates a gating adjustment request according to the detection result; the gating adaptive controller GAC receives the adjustment request and recalculates the gating configuration in combination with the enhanced flow feature information.

[0018] Further, the scheduling optimization unit introduces an energy consumption weight factor η, and the optimization objective function is Minimize η·E+(1 - η)·D, where E is the energy consumption and D is the delay, and η is dynamically adjusted according to the network load.

[0019] ​

[0020] In an exemplary embodiment, the process by which the gating adaptive controller (GAC) recalculates the gating configuration includes: The GAC calculates the gating closing time C based on the formula

[0021] C g = max(T w - D f , 0)

[0022] where T g represents the current time window length and D w represents the actual transmission delay of the data stream; if C f > 0, the current gating open state is continued; otherwise, the gating is closed and switched to the next cycle. g

[0023] Furthermore, when the gating closing time C g = 0, the GAC enables a backup gating list (Backup GCL) to reserve dedicated time slots for high-priority flows, ensuring that critical traffic is not affected by interruptions.

[0024] In an exemplary embodiment, it further includes: When the TSN network detects that the traffic has returned to normal, the scheduling optimization unit restores the current gating configuration to the initial state or maintains the optimized gating configuration.

[0025] Furthermore, the scheduling optimization unit further includes a traffic prediction module that predicts the traffic fluctuation trend in future cycles through a long short-term memory network (LSTM). If traffic overrun is predicted, the gating configuration is adjusted in advance to reserve resources.

[0026] In the above embodiments of the present invention, by introducing a flow feature enhancement algorithm and a dynamic gating adjustment mechanism, in the face of large-scale traffic and complex network topologies, the problem complexity can be significantly reduced, and an optimized scheduling scheme can be quickly generated, thus effectively ensuring the real-time performance and transmission determinacy of the TSN network. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a schematic flowchart of a TSN scheduling optimization method with enhanced and merged flow features provided by an embodiment of the present invention.

[0028] Figure 2 It is a schematic flowchart of the enhanced flow feature analysis module (ESAM) in an embodiment of the present invention for enhancing the calculation of the data stream feature weights.

[0029] Figure 3 It is a schematic structural diagram of the time-aware shaper (TAS) and the gating adaptive controller (GAC) working together in an embodiment of the present invention. ​

[0030] Figure 4 It is a timing schematic diagram of the dynamic gating adjustment mechanism in the embodiment of the present invention. Detailed implementation manners

[0031] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only for the purpose of illustrating and explaining the present invention, and are not used to limit the present invention.

[0032] The present invention provides a TSN scheduling optimization method for post-merging enhanced flow features. The core lies in solving the problem of ensuring the real-time performance of the TSN network in a large-scale traffic and complex network topology environment by introducing a flow feature enhancement algorithm and a dynamic gating adjustment mechanism. The following will be described in detail with reference to the attached Figure 1 to the attached Figure 4 the specific implementation manners of the present invention.

[0033] In an embodiment of the present invention, the entire scheduling optimization process starts from the application side generating flow feature description information and ends with the scheduling optimization unit generating an optimized scheduling scheme and dynamically adjusting the gating configuration. As shown in the attached Figure 1 figure, this process includes multiple key steps: the application side generates flow feature description information, the scheduling optimization unit receives and processes the flow feature description information, the enhanced flow feature analysis module (ESAM) performs flow feature weight enhancement calculation, the time-aware shaper (TAS) detects the transmission status, and the gating adaptive controller (GAC) recalculates the gating configuration. These steps together constitute a complete scheduling optimization system.

[0034] First, the application side generates flow feature description information including time window constraints based on historical traffic statistics information and real-time traffic requirements. This process involves data collection and initial weight calculation. The application side calculates the initial weights of flow features by collecting the historical transmission delay distribution, bandwidth occupancy rate, and jitter range of each data flow and analyzing these data. For example, for a specific data flow, its historical transmission delay distribution may show that its average delay is 10 milliseconds, the jitter range is 2 milliseconds, and the bandwidth occupancy rate is 50%. After comprehensively analyzing these parameters, the application side calculates the initial weights and transfers the flow feature description information to the scheduling optimization unit through control plane signaling. The key to this link is to ensure that the flow feature description information can accurately reflect the traffic characteristics in the current network environment, so as to provide a reliable basis for subsequent scheduling optimization.

[0035] When the scheduling optimization unit receives the flow feature description information sent by the application side, the enhanced flow feature analysis module (ESAM) dynamically quantifies the time sensitivity and priority of each data flow. As shown in the attached Figure 2 figure, ESAM is based on the formula W f = α·T s + β·Pq +γ·F t enhances the feature weights W of each data stream f by calculation, where T s represents the time sensitivity score, P q represents the priority queue allocation score, F t represents the traffic fluctuation tolerance score, and α, β, γ are the corresponding weight coefficients respectively, and satisfy α + β + γ = 1.

[0036] For example, in a certain application scenario, if the time sensitivity score T s = 0.8, the priority queue allocation score P q = 0.6, the traffic fluctuation tolerance score F t = 0.4, and the weight coefficients α = 0.5, β = 0.3, γ = 0.2, then the enhanced feature weight W f = 0.5·0.8 + 0.3·0.6 + 0.2·0.4 = 0.7. In this way, ESAM can accurately quantify the features of each data stream, laying a foundation for subsequent scheduling optimization.

[0037] The scheduling optimization unit divides the network into multiple logical domains, independently generates scheduling schemes within each domain, and synchronizes the global state through a coordinator. For a super-large-scale network, the method divides the network into multiple logical domains, runs the scheduling algorithm independently within each domain, and synchronizes the global state and generates cross-domain scheduling rules between domains to avoid resource conflicts. For example, in a factory network with 200 switches, 10 logical domains are divided, and the scheduling time within each domain is reduced from 15 ms to 8 ms. The inter-domain coordinator ensures the end-to-end determinism of cross-domain traffic through timestamp alignment.

[0038] The weight coefficients α, β, γ are dynamically optimized through a reinforcement learning model. The reinforcement learning model takes the real-time network state (including delay, jitter, bandwidth occupancy rate) as input and the scheduling performance metrics (delay reduction rate, throughput) as the reward function, and periodically updates the weight combination through the Q-Learning algorithm to achieve adaptive adjustment of the flow feature weights. In this embodiment, the scheduling optimization unit incorporates a lightweight reinforcement learning inference engine, and updates the weight coefficients according to the real-time network state every period.

[0039] Next, the scheduling optimization unit matches the enhanced flow feature information with the predefined gating list (GCL) template to generate an optimized scheduling plan. The gating list template is a set of predefined rules that guide the transmission behavior of each data flow in the TSN network. By matching the enhanced flow feature information with the gating list template, the scheduling optimization unit can quickly generate a scheduling plan that meets the current network conditions. For example, for data flows with high time sensitivity, the scheduling optimization unit will give priority to assigning them high priority queues and set a shorter gating opening time; for data flows with large traffic fluctuations, the gating opening time will be appropriately extended to improve their transmission stability. The core of this process is to significantly reduce the complexity of the problem through the flow feature enhancement algorithm, so as to quickly generate an optimized scheduling plan.

[0040] The scheduling optimization unit also includes a traffic prediction module, which uses a long short-term memory network (LSTM) to predict traffic fluctuation trends in future cycles. If traffic is predicted to exceed the limit, the gating configuration is adjusted in advance to reserve resources. For example, the traffic prediction module predicts that the bandwidth demand for video streams in the next cycle will increase by 30% based on historical data. In this case, GAC allocates an additional 5ms time slot for the video stream in advance to avoid queue congestion caused by burst traffic.

[0041] The scheduling optimization unit introduces an energy consumption weight factor η, and the optimization objective function is Minimizeη·E+(1-η)·D, where E is energy consumption, D is delay, and η is dynamically adjusted according to the network load. During low-load periods (such as at night), η=0.8 is set, and redundant ports are closed first to reduce energy consumption; during high-load periods, η=0.2 is set to fully ensure real-time performance.

[0042] In actual operation, the TSN network may cause traffic transmission fluctuations due to external factors. At this time, the Time Aware Shaper (TAS) detects the actual transmission status of each data flow in the current cycle and generates a gating adjustment request based on the detection results. Figure 3 As shown in the figure, TAS determines whether the current gating configuration needs to be adjusted by monitoring the transmission delay, bandwidth occupancy and other parameters of each data stream. For example, if the actual transmission delay of a data stream exceeds the preset threshold, TAS will generate a gating adjustment request and pass it to the Gating Adaptive Controller (GAC). After receiving the adjustment request, GAC recalculates the gating configuration based on the enhanced stream feature information. Figure 4 As shown, GAC is based on formula C g =max(T w -D f ,0), calculate the gate closing time C g , where T w Indicates the length of the current time window, D f Indicates the actual transmission delay of the data stream;

[0043] If C g > 0, the current gating state is maintained; otherwise, the gating is closed and switched to the next cycle.

[0044] For example, assume that the current time window length T w = 20 milliseconds, and the actual transmission delay D of a certain data stream f = 25 milliseconds. Then the gating closing time C g = max(20 - 25, 0) = 0. At this time, the gating will be closed and switched to the next cycle. In this way, GAC can dynamically adjust the gating configuration according to the actual transmission status, so as to ensure that the traffic is transmitted according to the predetermined gating rules.

[0045] In addition, when the TSN network detects that the traffic has returned to normal, the scheduling optimization unit will restore the current gating configuration to the initial state or maintain the optimized gating configuration. This process aims to ensure that the network can quickly return to the normal operating state after traffic fluctuations. For example, when the actual transmission delay of a certain data stream drops below the preset threshold, TAS will stop generating gating adjustment requests, and GAC will decide whether to restore to the initial state according to the current gating configuration. If the current gating configuration has been optimized, the optimized configuration can be maintained to improve network performance.

[0046] When the gating closing time C g = 0, the GAC enables the Backup GCL (Backup Gating Control List) to reserve dedicated time slots for high-priority flows, ensuring that critical traffic is not interrupted. If a switch port failure causes C g = 0, GAC immediately switches to the Backup GCL, allocates a fixed 5ms time slot for the control command flow, and the remaining traffic is transmitted in the degraded mode.

[0047] During the entire scheduling optimization process, the Time-Aware Shaper (TAS), the Enhanced Stream Attribute Module (ESAM), and the Gating Adaptive Controller (GAC) work together to jointly ensure the real-time performance of the TSN network. As shown in the appendix Figure 3 TAS is responsible for detecting the transmission status and generating adjustment requests, ESAM is responsible for enhancing the calculation of stream attributes, and GAC is responsible for recalculating the gating configuration. The close cooperation among the three enables the scheduling optimization unit to quickly generate an optimized scheduling scheme and dynamically adjust the gating configuration in the face of large-scale traffic and complex network topologies, thus effectively ensuring the real-time performance and transmission determinism of the TSN network.

[0048] To further illustrate the actual application scenarios of the present invention, the following takes an industrial automation control system as an example for detailed description. In an industrial automation scenario, the TSN network needs to transmit multiple types of data streams simultaneously, including control commands, sensor data, and video surveillance data. These data streams have different time sensitivities and priority requirements. For example, control commands have extremely low tolerance for time delay, while video surveillance data is more sensitive to bandwidth occupancy. By adopting the scheduling optimization method provided by the present invention, the application end can generate flow feature description information including time window constraints based on historical traffic statistics information and real-time traffic demands, and transmit it to the scheduling optimization unit. After the enhanced flow feature analysis module (ESAM) enhances the calculation of the feature weights of each data stream, the scheduling optimization unit generates an optimized scheduling plan. During the actual operation process, the time-aware shaper (TAS) monitors the transmission status of each data stream in real time and dynamically adjusts the gating configuration according to the detection results. For example, when the actual transmission delay of a certain sensor data stream exceeds the preset threshold, the TAS will generate a gating adjustment request, and the GAC will recalculate the gating closing time and adjust the gating configuration to ensure that the data stream can be transmitted according to the predetermined rules. In this way, the present invention can significantly improve the real-time performance and transmission determinacy of the TSN network in the industrial automation control system. For example, if the traffic prediction module predicts based on historical data that the bandwidth demand of the video stream will increase by 30% in the next cycle, the GAC will allocate an additional 5 ms time slot for the video stream in advance to avoid queue congestion caused by burst traffic.

[0049] In summary, the present invention solves the problem of ensuring the real-time performance of the TSN network in a large-scale traffic and complex network topology environment by introducing a flow feature enhancement algorithm and a dynamic gating adjustment mechanism. Its specific implementation covers the complete process from the application end generating flow feature description information to the scheduling optimization unit dynamically adjusting the gating configuration, and realizes efficient scheduling optimization through the collaborative work of multiple key components. Whether in theoretical analysis or actual application scenarios, the present invention demonstrates significant technical advantages and application values.

Claims

1. A TSN scheduling optimization method for merging after flow feature enhancement, characterized in that Including: A scheduling optimization unit deployed in a Time-Sensitive Network (TSN) receives flow feature description information sent from an application side; The scheduling optimization unit dynamically quantifies the time sensitivity and priority of each data flow based on a flow feature enhancement algorithm; The scheduling optimization unit matches the enhanced flow feature information with a predefined Gating Control List (GCL) template to generate an optimized scheduling plan; When the TSN network detects fluctuations in actual traffic transmission, it dynamically adjusts the gating configuration within the current period according to the enhanced flow features; The scheduling optimization unit includes a Time-Aware Shaper (TAS), an Enhanced Flow Feature Analysis Module (ESAM), and a Gating Adaptive Controller (GAC); before the scheduling optimization unit generates an optimized scheduling plan, it also includes the Enhanced Flow Feature Analysis Module (ESAM), and the Enhanced Flow Feature Analysis Module (ESAM) is based on the formula W f = α·T s + β·P q + γ·F t The feature weight W of each data stream f is enhanced calculated, where T s represents the time sensitivity score, P q represents the priority queue allocation score, F t represents the traffic fluctuation tolerance score, and α, β, γ are the corresponding weight coefficients respectively, and satisfy α + β + γ = 1; Before the scheduling optimization unit dynamically adjusts the gating configuration within the current period, it also includes the Time-Aware Shaper (TAS) and the Gating Adaptive Controller (GAC). The Time-Aware Shaper (TAS) detects the actual transmission status of each data flow within the current period and generates a gating adjustment request according to the detection result; the Gating Adaptive Controller (GAC) receives the adjustment request and recalculates the gating configuration in combination with the enhanced flow feature information; the process of the Gating Adaptive Controller (GAC) recalculating the gating configuration includes: the GAC is based on the formula C g = max(T w - D f , 0) Calculate the gating off-time C g , where T w represents the current time window length, and D f represents the actual transmission delay of the data stream; if C g > 0, then continue to maintain the current gating on state, otherwise turn off the gating and switch to the next cycle.

2. The method according to claim 1, wherein The enhanced flow feature information includes a time sensitivity weight, a priority queue allocation weight, and a traffic fluctuation tolerance threshold.

3. The method according to claim 1, wherein Before the scheduling optimization unit receives the flow feature description information sent from the application side, it also includes: the application side generates flow feature description information including time window constraints according to historical traffic statistical information and real-time traffic requirements; the application side transfers the flow feature description information to the scheduling optimization unit through control plane signaling.

4. The method according to claim 3, wherein The process of the application side generating the flow feature description information includes: the application side collects the historical transmission delay distribution, bandwidth occupancy rate, and jitter range of each data flow; the application side calculates the initial weight of the flow feature based on the collected data and reports it to the scheduling optimization unit.

5. The method according to claim 1, wherein Also including: When the TSN network detects that the traffic returns to normal, the scheduling optimization unit restores the current gating configuration to the initial state or maintains the optimized gating configuration.

Images