Scheduling method for multi-dimensional compatibility and conflict repair facing TSN end-side jitter

Abstract

The invention provides a scheduling method for multi-dimensional compatibility and conflict repair facing TSN end side jitter, and relates to the technical field of data transmission. A multi-dimensional compatibility evaluation and phase guidance mechanism is introduced to an end system side; a comprehensive compatibility index is constructed through four dimensions of periodic compatibility, load occupancy, jitter interval risk and time window overlapping degree, and clustering scheduling and end-side jitter absorption shaping of time sensitive flow are realized. According to the method, a global scheduling problem is divided into a plurality of mutually independent intra-cluster sub-problems through queue division based on multi-dimensional compatibility, so that service flows in the same cluster have relatively high similarity in the aspects of periodicity, bandwidth occupation and time tolerance characteristics, and the conflict probability of a port and a queue level is remarkably reduced. According to the method, an anchor point flow mechanism is introduced, a uniform time grid point structure is established, other flows select phases on the grid point and are arranged in a staggered manner, and periodic alignment and ordered peak shifting in the queue are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data transmission, in particular to a scheduling method for multi-dimensional compatibility and conflict repair of TSN end-side jitter. BACKGROUND

[0002] Real-time data transmission is becoming increasingly critical in modern information systems, especially in the context of the continuous development of emerging technologies such as industrial Internet, unmanned driving, smart healthcare, and smart cities. Time-Sensitive Networking (TSN) is a new Ethernet enhancement technology proposed by the IEEE 802.1 working group, aiming to provide real-time, low-jitter, congestion-free, low packet loss rate, and high robustness of data transmission capabilities. This technology achieves deterministic transmission of critical data by ensuring timely response of the network to specific events within a specific time period. Among the multiple standards of TSN, IEEE 802.1Qbv defines a Time-Aware Shaper (TAS). TAS periodically opens or closes queues on the port through a Gate Control List (GCL), achieving precise scheduling control, and is one of the core mechanisms to ensure deterministic communication of TSN. TAPRIO is a qdisc in the Linux kernel, which flexibly implements the 802.1Qbv / TAS protocol on the software side.

[0003] In the IEEE 802.1Qbv port structure, each egress port contains multiple priority-based transmit queues, and each queue is equipped with a time gate. When data frames arrive at the port, they will be classified into corresponding priority queues according to the VLAN priority class (PCP value). The transmission permission of the queue is controlled by the state of its corresponding time gate. This scheduling mechanism has a high dependence on precise time synchronization to ensure that different priority queues can be opened and closed on time within the scheduled time slot. However, when there is a soft or hardware clock deviation or a receiving time jitter between the end-side device and the TSN switch, the actual arrival time of the data packet may not match the expected time in the scheduling table. In some cases, data packets that arrive early may be immediately forwarded within the current scheduling period, disrupting the original transmission order and possibly causing tasks to miss the deadline; data packets that arrive late may directly miss the current transmission time slot and be forced to wait until the next period, causing tasks to timeout or scheduling to fail. This type of arrival time offset caused by end-side asynchronization not only destroys the predictability of TAPRIO scheduling, but also may cause serious problems such as link conflicts and frame order errors.

[0004] However, most existing TSN scheduling research assumes that all nodes in the network (including switches and end devices) can achieve strict time synchronization. Under this assumption, the design of the schedule table only needs to consider factors such as link bandwidth, queue priority, and task deadline, without needing to handle uncertainties caused by clock drift. However, in actual deployment, even if high-precision synchronization protocols such as IEEE 802.1AS or PTP can be relied upon to achieve precise time synchronization between switches, end devices may still have local clocks that are offset from the global reference time due to differences in hardware and software, network delay jitter, and other reasons. This end-device asynchronization phenomenon can cause the actual arrival time of data packets to be inconsistent with the expected schedule, posing a potential threat to the correctness and stability of the schedule.

[0005] Currently, the academic and industrial communities mainly use global optimization scheduling models to generate TSN gate schedules, assuming that the transmission time of the end system is precisely controllable. Typical methods such as SMT / ILP-based optimization models: Craciunas et al. proposed in "Combining optimization approaches for scheduling of time-triggered traffic in switched Ethernet networks" (Real-Time Systems, 2016, 52(2): 221-253) to formalize the TSN scheduling problem as an SMT / ILP model, and obtain the optimal schedule by solving global constraints.

[0006] The paper "Practical Aspects of Cross-Vendor TSN Time Synchronization Using IEEE 802.1AS" indicates that different interpretations and implementations of IEEE 802.1AS / gPTP can reduce synchronization accuracy, thereby compromising the predictability of scheduling, highlighting the need to explicitly consider terminal system jitter and clock offset in TSN design.

[0007] To solve the scheduling conflict problem caused by end system deviation, some research attempts to introduce robustness constraints into the optimization model. For example, by adding an upper bound item of end system jitter in the SMT or ILP model, the schedule can still be feasible within a limited time deviation. This method can ensure that frame isolation constraints are strictly met, and theoretically enhances the anti-deviation ability of the schedule table, but it mainly relies on global optimization modeling and high-complexity solving, resulting in high computational cost in large-scale networks, making it difficult to generate a scheduling scheme in real time in large-scale networks or engineering practice, and therefore lacking a lightweight repair mechanism.

[0008] In summary, the existing scheme has two deficiencies: one is to rely on global modeling solution, long calculation time, difficult to deploy online; two is scheduling optimization is mostly carried out on the switch side, rather than on the source of jitter-the end system side. Because the uncertainty of the end system sending is amplified in the network, if the sending phase can be adjusted on the end side, the jitter can be absorbed and the conflict can be reduced before entering the switch, so as to improve the overall schedulability. Therefore, a scheduling method is needed, which can be quickly executed on the end system level, has adaptive conflict repair capability, and does not need global synchronization solution, to absorb time deviation at the source and improve system real-time performance and robustness. SUMMARY

[0009] In view of the deficiencies of the prior art, the purpose of the present application is to provide a scheduling method for multi-dimensional compatibility and conflict repair of TSN end-side jitter, comprising:

[0010] On the end-side device, a plurality of time-sensitive flows and parameters of the time-sensitive flows, timing attributes, paths and constraint information of the time-sensitive flows are obtained;

[0011] Based on the parameters of the time-sensitive flows, a multi-dimensional compatibility index matrix is calculated;

[0012] The multi-dimensional compatibility index matrix is processed by a hierarchical clustering algorithm to obtain a plurality of queues, each queue containing a plurality of time-sensitive flows, and each queue is assigned a number;

[0013] The sending phase of the anchor flow in each queue and the final sending phase of the anchor flow are determined, and the sending phase of the anchor flow represents the sending time of the anchor flow;

[0014] The sending phase of the non-anchor flow in each queue is determined;

[0015] According to the final sending phase of all non-anchor flows and the final sending phase of all anchor flows, a gating time slot configuration table of each queue is generated, and the final sending phase of each time-sensitive flow, the queue to which each time-sensitive flow belongs, and the gating time slot configuration table of each queue form a global scheduling table;

[0016] The end-side device sends the time-sensitive flows to the TSN switch according to the global scheduling table.

[0017] Optionally, the time-sensitive flow parameters at least include period, frame transmission duration, jitter error upper bound, starting earliest sending time, latest sending time, path first hop link, data packet size, data packet deadline and priority; the timing attribute is the generation time of the time-sensitive flow, and the constraint information includes queue isolation constraint, port isolation constraint, jitter safety margin constraint and deadline constraint.

[0018] Optionally, based on the parameters of the time-sensitive flow, a multi-dimensional compatibility index matrix is calculated, including:

[0019] Calculate the periodic compatibility between two time-sensitive flows , which is realized by the following formula:

[0020] ;

[0021] Wherein, T i represents the period of time-sensitive flow i, T j represents the period of time-sensitive flow j, lcm represents the least common multiple, and gcd represents the greatest common divisor;

[0022] Calculate the load occupancy of two time-sensitive flows , which is realized by the following formula:

[0023] ;

[0024] Wherein, represents the frame transmission duration of time-sensitive flow i, represents the frame transmission duration of time-sensitive flow j, represents the period of time-sensitive flow p, represents the period of time-sensitive flow q, represents the frame transmission duration of time-sensitive flow p, represents the frame transmission duration of time-sensitive flow q, and time-sensitive flow p and time-sensitive flow q are any two time-sensitive flows of all time-sensitive flows;

[0025] Calculate the jitter interval risk of two time-sensitive flows , which is realized by the following formula:

[0026] ;

[0027] Wherein, gap is the safety interval;

[0028] Calculate the time window overlap of two time-sensitive flows , which is realized by the following formula:

[0029] ;

[0030] ;

[0031] Wherein, represents the latest sending time of time-sensitive flow i, represents the latest sending time of time-sensitive flow j, represents the starting earliest sending time of time-sensitive flow i, a start earliest transmission time of a time sensitive flow j, a time sensitive flow i and time sensitive flow j window overlap rate;

[0032] a periodicity compatibility , a load occupancy , a jitter interval risk and a time window overlap degree are weighted and summed to obtain a comprehensive compatibility index , which is specifically realized by the following formula:

[0033] ;

[0034] ;

[0035] wherein, , , , are preset weights;

[0036] The above steps are repeated to obtain a multi-dimensional compatibility index matrix.

[0037] Optionally, determining the anchor flow and the final transmission phase of the anchor flow in each queue comprises:

[0038] For each queue, obtaining the greatest common divisor of the periods of all time sensitive flows in the queue as a common scale G, and obtaining the least common multiple of the periods of all time sensitive flows in the queue as a super period H;

[0039] Among all the time sensitive flows in the queue, obtaining the time sensitive flow with the longest period as the anchor flow, in the case that the periods of the time sensitive flows are the same, obtaining the time sensitive flow with the largest load as the anchor flow, wherein the load is obtained by dividing the data packet size of the time sensitive flow by the period, in the case that the periods of the time sensitive flows are the same and the loads are the same, obtaining the time sensitive flow with the smallest data packet deadline as the anchor flow, and further obtaining the anchor flow of each queue;

[0040] Obtaining the middle value of the target interval of the anchor flow as the initial transmission phase of the anchor flow, to obtain the initial transmission phase of each anchor flow, and the target interval of the anchor flow is the time interval from the start earliest transmission time to the latest transmission time;

[0041] For each anchor flow, it is judged whether the initial sending phase of the anchor flow satisfies the constraint information of the anchor flow, and in the case that the initial sending phase of the anchor flow satisfies the constraint information of the anchor flow, the initial sending phase of the anchor flow is taken as the final sending phase; in the case that the initial sending phase of the anchor flow does not satisfy the constraint information of the anchor flow, the initial phase of the anchor flow is phase-tuned, specifically, in the target interval of the anchor flow, the initial sending phase is moved with the granularity of the public scale G to obtain the final sending phase of the anchor flow, and then the final sending phase of each anchor flow is obtained.

[0042] Optionally, the sending phase of the non-anchor flow in each queue is determined, including:

[0043] The first non-anchor flow in the queue is obtained, and the candidate sending phase is obtained by removing the determined phase in the interval from the earliest sending time to the latest sending time of the non-anchor flow.

[0044] For each candidate sending phase, the cost evaluation

[0045] In the cost evaluation of all candidate sending phases, the candidate sending phase with the minimum cost evaluation is obtained, and is taken as the first sending phase of the non-anchor flow.

[0046] It is judged whether the first sending phase of the non-anchor flow satisfies the constraint information of the non-anchor flow, and in the case that the first sending phase of the non-anchor flow satisfies the constraint information of the non-anchor flow, the first sending phase is taken as the final sending phase of the non-anchor flow, and the next anchor flow of the queue is obtained, and the operation is returned to: removing the determined phase in the interval from the earliest sending time to the latest sending time of the non-anchor flow.

[0047] In the case that the first sending phase of the non-anchor flow does not satisfy the constraint information of the non-anchor flow, conflict repair is performed to obtain a repair result, and the repair result includes the final sending phase of the non-anchor flow or conflict information, and the conflict information includes the non-anchor flow, the link of the non-anchor flow and all phases of the non-anchor flow.

[0048] ​In the case that the repair result includes conflict information, the cost weight of all phases of the non-anchor flow in the conflict information is increased, and the following is returned to perform: removing the determined phases in the interval from the earliest sending time to the latest sending time of the start of the non-anchor flow, until the sending phase of all non-anchor flows meets the constraint information, to obtain the final sending phase of the non-anchor flow in each queue; if in the process of returning to perform multiple rounds, there is a non-anchor flow whose sending phase does not meet the constraint information, the following is returned to perform again: removing the determined phases in the interval from the earliest sending time to the latest sending time of the start of the non-anchor flow, judging whether the sending phase of the non-anchor flow with high priority meets the constraint information, and not judging whether the sending phase of the non-anchor flow with low priority meets the constraint information, to obtain the final sending phase of the non-anchor flow in each queue.

[0049] Optionally, for each candidate sending phase, the cost evaluation is calculated , and the calculation is implemented by the following formula:

[0050] ;

[0051] wherein, the candidate sending phase is represented by ; QueueGapRisk represents the reciprocal penalty of the minimum head distance or tail distance of the adjacent interval of the same queue; PortGapRisk represents the occupation density penalty of other queues on the same port in the time window; SlackLossPenalty represents the penalty of insufficient window slack; , , , and all are cost weights.

[0052] Optionally, in the case that the first sending phase of the non-anchor flow does not meet the constraint information of the non-anchor flow, conflict repair is performed to obtain a repair result, including:

[0053] Phase fine-tuning is performed on the first sending phase of the non-anchor flow to obtain a second sending phase, specifically, in the candidate phases of the non-anchor flow, the first sending phase is moved multiple times with a granularity of a common scale G, the cost evaluation of each moved sending phase is calculated each time, the moved sending phase with the minimum cost evaluation is selected as the second sending phase, and in the case that the second sending phase meets the constraint information, the second sending phase is taken as the final sending phase of the non-anchor flow.

[0054] In the case that the second sending phase does not satisfy the constraint information, intra-cluster rearrangement is performed, specifically, all non-anchor point flows belonging to the queue to which the non-anchor point flow belongs are obtained, all non-anchor point flows belonging to the queue to which the non-anchor point flow belongs are sorted in ascending order of data packet deadline, and according to the order, for each non-anchor point flow, the following is returned: removing the determined phase in the interval from the earliest sending time to the latest sending time of the non-anchor point flow until the sending phase is obtained by phase fine tuning, if the sending phase obtained by phase fine tuning satisfies the constraint information, it is used as the final phase of the non-anchor point flow;

[0055] If the sending phase obtained by phase fine tuning does not satisfy the constraint information, inter-cluster migration is performed, specifically, the non-anchor point flow is migrated to the queue adjacent to the queue to which it belongs and having the lowest queue load, wherein the queue load is the sum of the data packet sizes of all time-sensitive flows in the queue divided by the least common multiple of the periods of all time-sensitive flows, and the following is returned: removing the determined phase in the interval from the earliest sending time to the latest sending time of the non-anchor point flow until the sending phase is obtained by intra-cluster rearrangement, if the sending phase obtained by intra-cluster rearrangement does not satisfy the constraint information, inter-cluster migration is performed again until the last queue is migrated, if there is a sending phase satisfying the constraint information during the inter-cluster migration process, it is used as the final phase of the non-anchor point flow, if there is no sending phase satisfying the constraint information during the inter-cluster migration process, the non-anchor point flow, the link of the non-anchor point flow and all phases of the non-anchor point flow are recorded.

[0056] The beneficial effects produced by the above technical solution are:

[0057] The application introduces a multi-dimensional compatibility evaluation and phase guiding mechanism on the end system side, constructs a comprehensive compatibility index through four dimensions of periodic compatibility, load occupancy, jitter interval risk and time window overlap degree, and realizes the clustering scheduling of time-sensitive flows and the end-side jitter absorption shaping. Without relying on a global synchronization optimization solver, the method can realize quasi-periodic data transmission with high reliability and low delay, thereby significantly improving the schedulability and delay stability of the system.

[0058] By dividing the queues based on multi-dimensional compatibility, the application divides the global scheduling problem into multiple independent intra-cluster sub-problems, so that the service flows in the same cluster have high similarity in periodicity, bandwidth occupation and time tolerance characteristics, reduces the scheduling competition across clusters, and significantly reduces the conflict probability at the port and queue levels. Further, the anchor flow mechanism is introduced, the flow with longer period, higher load or tighter deadline is used as the time reference to establish a unified time grid structure, other flows select phases on the grid and are arranged in staggered manner, realizing the periodic alignment and orderly staggered peak of the queue. This mechanism not only ensures the regularity and predictability of the queue-level scheduling, but also avoids the problem of jitter amplification caused by multi-reference conflicts.

[0059] In the phase distribution stage of the non-anchor flow, the application designs a plurality of cost functions including inter-queue interval risk, port occupancy density, time window margin and downstream link relaxation, to select the optimal sending phase in a heuristic search manner, so that the flow is as far away from the boundary as possible within the feasible window and avoids port-level competition. This mechanism effectively prevents the critical overlap phenomenon caused by port multiplexing and can adaptively adjust in a local range to relieve conflicts, thereby obtaining a high-quality feasible solution within a limited calculation time. In addition, through the local conflict repair strategy, after detecting intra-cluster or port-level conflicts, only the affected time window and flow are fine-tuned or the order is exchanged, without re-performing global backtracking, thereby greatly reducing the algorithm complexity and solving time.

[0060] Compared with the traditional scheme relying on a static protection zone or global robustness modeling, the application introduces a jitter absorption layer on the end system side, so that the data flow is reshaped into a quasi-periodic flow before entering the first-hop switch, avoiding the cascading amplification of jitter in multi-hop transmission, and effectively guaranteeing the timing stability of subsequent nodes. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 A flowchart of a scheduling method for multi-dimensional compatibility and conflict repair of TSN end-side jitter according to an embodiment of the application. DETAILED DESCRIPTION

[0062] The specific embodiments of the application will be further described in detail below with reference to the accompanying drawings and examples. The following examples are used to illustrate the application, but are not used to limit the scope of the application.

[0063] To solve the problems in the prior art, the application provides a scheduling method for multi-dimensional compatibility and conflict repair of TSN end-side jitter. To solve the problems of high computational complexity, sensitivity to end system jitter, and difficulty in real-time deployment in engineering practice in existing time-sensitive network (TSN) scheduling methods, the application proposes a time-sensitive network end-side jitter absorption scheduling method based on multi-dimensional compatibility and conflict repair, which can be optimized at the end system side.

[0064] The application absorbs the arrival time uncertainty caused by the clock deviation at the end system by introducing a multi-dimensional compatibility index in the end system sending stage, and avoids the cascading propagation of scheduling conflicts. At the same time, by combining the local conflict detection and repair mechanism, the dynamic recovery of feasible solutions is realized in the phase distribution process, and the adaptability and robustness of scheduling are improved.

[0065] Through heuristic queue division, anchor flow selection and cost function driven phase optimization mechanism, the application can quickly generate a robust feasible schedule table without a global solver, effectively reduce the calculation complexity, improve the port bandwidth utilization, and ensure the satisfaction of frame isolation and timing constraints, thereby significantly enhancing the real-time performance, determinacy and scalability of the time sensitive network.

[0066] The application provides a scheduling method for multi-dimensional compatibility and conflict repair of TSN end-side jitter. Figure 1 The method can include the following steps:

[0067] Step 1: obtaining a plurality of time sensitive flows and parameters of the time sensitive flows, timing attributes, paths and constraint information of the time sensitive flows on an end-side device, wherein the parameters of the time sensitive flows can be obtained from an upper network controller or a configuration file.

[0068] The parameters of the time sensitive flows at least include a period, a frame transmission duration, an upper bound of jitter error, a starting earliest sending time, a latest sending time, a path first hop link, a packet size, a packet deadline and a priority; the timing attribute is a generation time of the time sensitive flow; the constraint information includes queue isolation constraint, port isolation constraint, jitter safety margin constraint and deadline constraint;

[0069] Meanwhile, the number of queues is initialized in step 1, and the total number of queues in step 3 is less than or equal to the initialized number of queues; a compatibility matrix is also initialized, which is used to store multi-dimensional compatibility evaluation values between flows; a conflict record table is used to record conflict information in the subsequent repair stage, and both tables are empty in this stage, only two tables are set.

[0070] Step 2: calculating a multi-dimensional compatibility index matrix based on the parameters of the time sensitive flows;

[0071] Step 2.1: calculating the period compatibility between two time sensitive flows , which is realized by the following formula:

[0072] ;

[0073] Wherein, T i represents the period of the time sensitive flow i, T j represents the period of the time sensitive flow j, lcm represents the least common multiple, and gcd represents the greatest common divisor; the period compatibility is used to describe the harmonic relationship between the periods of two flows, and the smaller the index value is, the more compatible the periods are, and it is easier to arrange a conflict-free periodic schedule.

[0074] Step 2.2: calculating the load occupancy of two time sensitive flows , specifically implemented by the following formula:

[0075] ;

[0076] wherein, denotes the frame transmission duration of the time-sensitive flow i, denotes the frame transmission duration of the time-sensitive flow j, denotes the period of the time-sensitive flow p, denotes the period of the time-sensitive flow q, denotes the frame transmission duration of the time-sensitive flow p, denotes the frame transmission duration of the time-sensitive flow q, the time-sensitive flow p and the time-sensitive flow q being any two time-sensitive flows of all time-sensitive flows; load occupancy describes the occupancy intensity of two flows on time resources, and is used to reflect the bandwidth tension degree of a port or a queue, the load occupancy is greater, the more time resources occupied by the two time-sensitive flows, and the more likely to generate competition when arranged.

[0077] Step 2.3: Calculate the jitter interval risk of two time-sensitive flows , specifically implemented by the following formula:

[0078] ;

[0079] wherein, gap is a safety interval, used to absorb end system jitter and clock error; jitter interval risk is used to describe the overlap risk of two time-sensitive flows in the presence of clock bias, if the total transmission duration of two time-sensitive flows and the safety interval is less than the greatest common divisor of their periods, then , the conflict risk is low; if the value is close to 1, it means that the two time-sensitive flows almost certainly overlap in a minimum common period, and it is difficult to share the same queue. is greater, the higher the time overlap risk between the two time-sensitive flows.

[0080] Step 2.4: Calculate the time window overlap degree of two time-sensitive flows , specifically implemented by the following formula:

[0081] ;

[0082] ;

[0083] wherein, denotes the latest sending time of the time-sensitive flow i, denotes the latest sending time of the time-sensitive flow j, denotes the starting earliest sending time of the time-sensitive flow i, the start earliest transmission time of the time sensitive flow j, the time sensitive flow i and time sensitive flow j window overlap rate;

[0084] time window overlap degree measures the degree of overlap of the two time sensitive flow's transmittable time windows [est, lst], to reflect their schedulable time domain, if the time windows of the two time sensitive flows are completely overlapped, then , indicates strong time competition; if the time windows of the two time sensitive flows are completely not overlapped, then , indicates complete separation in the time domain. Therefore, can be directly used to evaluate whether the two time sensitive flows can be scheduled without interference in the time domain.

[0085] Step 2.5: Weighted sum of periodic compatibility , load occupancy , jitter interval risk and time window overlap degree to obtain the comprehensive compatibility index , which is specifically realized by the following formula:

[0086] ;

[0087] ;

[0088] wherein, , , , are all preset weights; The smaller the value is, the more compatible the two time sensitive flows are in multiple dimensions, and the more suitable they are for sharing the same sending queue.

[0089] Step 2.6: Repeat steps 2.1 to 2.5 to calculate the multi-dimensional compatibility index matrix;

[0090] Step 3: Through the hierarchical clustering algorithm, the multi-dimensional compatibility index matrix is processed to obtain multiple queues, each queue containing multiple time sensitive flows, and each queue is assigned a number;

[0091] The goal of this stage is to ensure the feasibility of scheduling under the premise of making the service flows within the same cluster have as high compatibility as possible in periodicity, bandwidth occupancy, jitter risk and time window, so as to reduce the conflict probability of subsequent phase allocation. If the incompatibility in the current cluster exceeds the threshold, a new cluster (i.e. queue) is opened, and under the same conditions, the load of the cluster is balanced preferentially to make the utilization rate of each queue close.

[0092] To ensure the scheduling in the same cluster has a unified time reference, the concept of "anchor flow" is proposed. Anchor flow is a reference flow used to define the time beat structure of the cluster, and its period and phase are used as the reference for the phase allocation of other flows in the cluster.

[0093] Step 4: Determine the anchor flow and the final sending phase of the anchor flow in each queue, and the sending phase of the anchor flow represents the sending time of the anchor flow;

[0094] Step 4.1: For each queue, obtain the greatest common divisor of the periods of all time-sensitive flows in the queue as the common scale G, and obtain the least common multiple of the periods of all time-sensitive flows in the queue as the super period H;

[0095] Step 4.2: Among all time-sensitive flows in the queue, obtain the time-sensitive flow with the longest period as the anchor flow, in the case of the same period of time-sensitive flows, obtain the time-sensitive flow with the largest load as the anchor flow, wherein the load is obtained by dividing the data packet size of the time-sensitive flow by the period, in the case of the same period and the same load of the time-sensitive flow, obtain the time-sensitive flow with the smallest data packet deadline as the anchor flow: and thus obtain the anchor flow of each queue;

[0096] Step 4.3: Obtain the middle value of the target interval of the anchor flow as the initial sending phase of the anchor flow, and obtain the initial sending phase of each anchor flow, and the target interval of the anchor flow is the time interval from the earliest sending time to the latest sending time;

[0097] Step 4.4: For each anchor flow, determine whether the initial sending phase of the anchor flow meets the constraint information of the anchor flow, in the case that the initial sending phase of the anchor flow meets the constraint information of the anchor flow, the initial sending phase of the anchor flow is taken as the final sending phase; in the case that the initial sending phase of the anchor flow does not meet the constraint information of the anchor flow, the initial phase of the anchor flow is fine-tuned, specifically, in the target interval of the anchor flow, the initial sending phase is moved with the common scale G as the granularity, to obtain the final sending phase of the anchor flow, and thus the final sending phase of each anchor flow is obtained;

[0098] Other non-anchor flows are arranged at this time point, and the phase of the anchor flow is used as a reference for avoidance and alignment, so as to ensure the neatness and predictability of the periodic structure inside the queue. This mechanism ensures the order of scheduling while avoiding phase oscillation and rising computational complexity caused by multiple reference conflicts. For each candidate phase of the anchor flow, the following evaluation quantities are calculated: the boundary margin of the transmittable time slot, and the risk of port conflict between different queues.

[0099] Step 5: Determine the sending phase of the non-anchor flow in each queue;

[0100] Step 5.1: Obtain the first non-anchor flow in the queue, and remove the determined phase in the interval from the earliest sending time to the latest sending time of the non-anchor flow to obtain a candidate sending phase;

[0101] Step 5.2: For each candidate sending phase, calculate the cost evaluation , which is realized by the following formula:

[0102] ;

[0103] Wherein, represents the candidate sending phase; QueueGapRisk represents the reciprocal penalty of the minimum head distance or tail distance of the adjacent interval of the queue; PortGapRisk represents the occupation density penalty of other queues on the same port in the time window; SlackLossPenalty represents the penalty of insufficient window slack; , , are cost weights;

[0104] Step 5.3: In the cost evaluation of all candidate sending phases, obtain the candidate sending phase with the minimum cost evaluation, and take it as the first sending phase of the non-anchor flow;

[0105] Step 5.4: Judge whether the first sending phase of the non-anchor flow meets the constraint information of the non-anchor flow, and in the case that the first sending phase of the non-anchor flow meets the constraint information of the non-anchor flow, take the first sending phase as the final sending phase of the non-anchor flow, obtain the next anchor flow of the queue, and return to execute Step 5.1: remove the determined phase in the interval from the earliest sending time to the latest sending time of the non-anchor flow to obtain a candidate sending phase;

[0106] In the case that the first sending phase of the non-anchor flow does not meet the constraint information of the non-anchor flow, execute conflict repair (i.e. Figure 1 local repair) to obtain a repair result, i.e. the non-anchor flow cannot find a feasible phase that meets all the constraint conditions, or the system detects a conflict (such as frame overlap in the queue, simultaneous sending at the port level, etc.) in local arrangement, and enters the conflict repair stage.

[0107] The repair result includes the final sending phase of the non-anchor flow (corresponding to Figure 1 repair success), or conflict information (corresponding to Figure 1 repair unsuccessful), and the conflict information includes the non-anchor flow, the link of the non-anchor flow and all phases of the non-anchor flow;

[0108] Specifically, the first sending phase of the non-anchor flow is phase fine-tuned to obtain a second sending phase. Specifically, in the candidate phase of the non-anchor flow, the first sending phase is moved multiple times with a common scale G as a granularity. The cost evaluation of each moved sending phase is calculated each time. The moved sending phase with the minimum cost evaluation is selected as the second sending phase. In the case that the second sending phase meets the constraint information, the second sending phase is taken as the final sending phase of the non-anchor flow.

[0109] In the case that the second sending phase does not meet the constraint information, intra-cluster rearrangement is performed. Specifically, all non-anchor flows belonging to the queue to which the non-anchor flow belongs are obtained. All non-anchor flows belonging to the queue to which the non-anchor flow belongs are sorted in ascending order of data packet deadlines. According to the order, step 5.1 is performed for each non-anchor flow: removing the determined phase in the interval from the earliest sending time to the latest sending time of the non-anchor flow until the sending phase is obtained by phase fine-tuning. If the sending phase obtained by phase fine-tuning meets the constraint information, it is taken as the final phase of the non-anchor flow.

[0110] If the sending phase obtained by phase fine-tuning does not meet the constraint information, inter-cluster migration is performed. Specifically, the non-anchor flow is migrated to the queue adjacent to the queue to which it belongs and with the lowest queue load. The queue load is the sum of the data packet sizes of all time-sensitive flows in the queue divided by the least common multiple of the periods of all time-sensitive flows. Step 5.1 is performed again: removing the determined phase in the interval from the earliest sending time to the latest sending time of the non-anchor flow until the sending phase is obtained by intra-cluster rearrangement. If the sending phase obtained by intra-cluster rearrangement does not meet the constraint information, inter-cluster migration is performed again until the last queue is migrated. If there is a sending phase that meets the constraint information during the inter-cluster migration process, it is taken as the final phase of the non-anchor flow. If there is no sending phase that meets the constraint information during the inter-cluster migration process, the non-anchor flow, the link of the non-anchor flow, and all phases of the non-anchor flow are recorded.

[0111] Step 5.5: In the case that the repair result includes conflict information, the cost weight of all phases of non-anchor flows in the conflict information is increased, and step 5.1 is returned to execute: removing the determined phases in the interval from the starting earliest sending time to the latest sending time of the non-anchor flow to obtain candidate sending phases, until the sending phases of all non-anchor flows meet the constraint information, and the final sending phase of the non-anchor flow in each queue is obtained, if there is a non-anchor flow whose sending phase does not meet the constraint information in the process of returning to execute multiple rounds, step 5.1 is returned to execute again: removing the determined phases in the interval from the starting earliest sending time to the latest sending time of the non-anchor flow, for the sending phase of the non-anchor flow with high priority, it is judged whether the constraint information is met, and for the sending phase of the non-anchor flow with low priority, it is not judged whether the constraint information is met, and the final sending phase of the non-anchor flow in each queue is obtained;

[0112] The cost weight of all phases of non-anchor flows in the conflict information is increased, guiding the scheduler to avoid high-risk combinations. The conflict record participates in the next round of scheduling as a lightweight constraint, avoiding repeated conflicts while not permanently shielding possible feasible combinations. If the conflict cannot be resolved after multiple rounds of repair, the system triggers a degradation tolerance strategy, allowing a limited range of delay offset or delaying the transmission of non-critical flows (i.e., non-anchor flows with low priority) by one period to maintain global schedulability.

[0113] Step 6: According to the final sending phase of all non-anchor flows and the final sending phase of all anchor flows, a gating time slot configuration table of each queue is generated, and the final sending phase of each time sensitive flow, the queue to which each time sensitive flow belongs, and the gating time slot configuration table of each queue constitute a global scheduling table;

[0114] Step 7: The end-side device sends the time sensitive flow to the TSN switch according to the global scheduling table; the output scheduling table can be directly issued to the TAPRIO module in Linux or the TSN switch configuration interface to realize synchronous scheduling after absorbing the jitter on the end-side. Since the present application realizes jitter absorption on the end system side through a lightweight heuristic phase optimization, the flow entering the switch has been shaped into a quasi-periodic flow, thereby effectively improving the schedulability and robustness of the entire TSN network.

[0115] The present application proposes a time sensitive network end-side jitter absorption scheduling method based on multi-dimensional compatibility and conflict repair, which innovatively introduces a lightweight heuristic optimization mechanism on the end system side to realize integrated scheduling from "source jitter shaping" to "network synchronous transmission", aiming at the problem that existing TSN scheduling generally relies on global optimization solution, has complex calculation and ignores end-side jitter absorption.

[0116] The core innovation points of the application mainly embody three aspects: (1) a multi-dimensional compatibility index system is constructed, multiple factors such as periodicity, load occupation, jitter risk and time window overlap are comprehensively considered, a coexistence measurement model between business flows is established, and quantitative basis is provided for subsequent clustering and scheduling; (2) secondly, an anchor flow phase reference mechanism is proposed, by selecting a flow with longer period, higher load or tighter deadline as a time reference, a unified time grid is established, the uniformity and predictability of internal scheduling of the queue are ensured, and phase oscillation caused by multiple reference bases is avoided; (3) conflict repair and heuristic fine tuning strategies are introduced, when local conflicts are detected, only the phase or flow cluster in the affected window is adjusted, global backtracking is not needed, thereby significantly reducing the computational complexity. Through the above mechanisms, the application realizes fast and feasible scheduling and dynamic jitter absorption on the end system side, and provides a new type of end-side optimization framework with low complexity and high robustness for time sensitive network.

[0117] Compared with the existing time sensitive network scheduling method relying on SMT / ILP global modeling and solving, the application introduces a multi-dimensional compatibility evaluation and phase guiding mechanism on the end system side, constructs a comprehensive compatibility index through four dimensions of periodic compatibility, load occupation degree, jitter interval risk and time window overlap degree, realizes the clustering scheduling of time sensitive flow and end-side jitter absorption shaping. Without relying on global synchronous optimization solver, the method can realize quasi-periodic data transmission with high reliability and low delay, thereby significantly improving the schedulability and delay stability of the system.

[0118] The application divides the global scheduling problem into multiple independent intra-cluster sub-problems through queue division based on multi-dimensional compatibility, so that the business flows in the same cluster have high similarity in periodicity, bandwidth occupation and time tolerance characteristics, reduces the scheduling competition between clusters, and significantly reduces the conflict probability at the port and queue level. Further, the anchor flow mechanism is introduced, a flow with longer period, higher load or tighter deadline is selected as a time reference, a unified time grid structure is established, other flows select phases on the grid and are arranged in staggered manner, the periodicity alignment and orderly staggered peak of the queue are realized. The mechanism not only ensures the uniformity and predictability of the queue level scheduling, but also avoids the jitter amplification problem caused by multiple reference conflicts.

[0119] In the phase assignment stage of non-anchor flows, the application designs a multi-item cost function including inter-queue interval risk, port occupancy density, time window margin and downstream link relaxation, to select the optimal sending phase in a heuristic search manner, so that the flow can be as far away from the boundary as possible within the feasible window and avoid port-level competition. This mechanism effectively prevents critical overlap caused by port multiplexing and can adaptively adjust in a local range to relieve conflicts, thereby obtaining a high-quality feasible solution within limited computation time. In addition, through a local conflict repair strategy, after detecting intra-cluster or port-level conflicts, the application only fine-tunes or exchanges the order of the affected time window and flow, without re-performing global backtracking, greatly reducing the algorithm complexity and solving time.

[0120] Compared with traditional schemes relying on static protection bands or global robustness modeling, the application introduces a jitter absorption layer on the end system side, so that the data flow is reshaped into a quasi-periodic flow before entering the first-hop switch, avoiding the cascading amplification of jitter in multi-hop transmission, and effectively ensuring the timing stability of subsequent nodes.

[0121] In summary, the application realizes the fusion of end-side jitter absorption and scalable scheduling of time-sensitive networks, ensuring the real-time and reliability of scheduling, and taking into account the feasibility of engineering landing. Its main advantages include: maintaining low-jitter deterministic communication without high-precision global synchronization; quickly generating a feasible schedule under limited computing resources; achieving lightweight deployment on the end side to improve the overall robustness and bandwidth utilization efficiency of the network, with good engineering promotion value and application prospect.

[0122] The above description is only the preferred embodiment of the present disclosure and the explanation of the applied technical principles. Those skilled in the art should understand that the scope of the application involved in the embodiments of the present disclosure is not limited to the technical solutions formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the above features are replaced with the technical features disclosed in the embodiments of the present disclosure (but not limited to) having similar functions to form technical solutions.

Claims

1. A scheduling method for multidimensional compatibility and conflict resolution oriented towards TSN end-side jitter, characterized in that, include: On the edge device, acquire multiple time-sensitive streams, as well as the parameters of the time-sensitive streams, their timing attributes, path, and constraint information; Based on the parameters of the time-sensitive flow, a multidimensional compatibility index matrix is ​​calculated; The multidimensional compatibility index matrix is ​​processed by a hierarchical clustering algorithm to obtain multiple queues, each containing multiple time-sensitive streams, and each queue is assigned a number. Determine the anchor stream and the final transmission phase of the anchor stream in each queue. The transmission phase of the anchor stream represents the transmission time of the anchor stream. Determine the transmission phase of non-anchor streams other than anchor streams in each queue; Based on the final transmission phase of all non-anchor streams and the final transmission phase of all anchor streams, a gating slot configuration table for each queue is generated. The final transmission phase of each time-sensitive stream, the queue to which each time-sensitive stream belongs, and the gating slot configuration table for each queue constitute a global scheduling table. End-side devices send time-sensitive flows to the TSN switch according to the global scheduling table.

2. The scheduling method for multi-dimensional compatibility and conflict repair oriented towards TSN end-side jitter according to claim 1, characterized in that, The time-sensitive flow parameters include at least the period, frame transmission duration, jitter error upper bound, earliest start time, latest start time, first hop link, data packet size, data packet deadline, and priority; the timing attribute is the generation time of the time-sensitive flow, and the constraint information includes queue isolation constraints, port isolation constraints, jitter safety margin constraints, and deadline constraints.

3. The scheduling method for multidimensional compatibility and conflict repair oriented towards TSN end-side jitter according to claim 1, characterized in that, Based on the parameters of the time-sensitive flow, a multidimensional compatibility index matrix is ​​calculated, including: Calculate the periodic compatibility between two time-sensitive streams Specifically, this is achieved through the following formula: ； Among them, T i T represents the period of time-sensitive flow i. j Let lcm represent the period of time-sensitive flow j, lcm represent the least common multiple, and gcd represent the greatest common divisor. Calculate the load occupancy of two time-sensitive streams Specifically, this is achieved through the following formula: ； in, Indicates the frame transmission duration of time-sensitive stream i. This represents the frame transmission duration of time-sensitive stream j. This represents the period of the time-sensitive flow p. This represents the period of the time-sensitive flow q. This indicates the frame transmission duration of the time-sensitive stream p. This represents the frame transmission duration of time-sensitive stream q, where time-sensitive stream p and time-sensitive stream q are any two time-sensitive streams from all time-sensitive streams. Calculate the jitter interval risk for two time-sensitive streams Specifically, this is achieved through the following formula: ； Where gap is the safety interval; Calculate the time window overlap between two time-sensitive streams Specifically, this is achieved through the following formula: ； ； in, Indicates the latest sending time of time-sensitive stream i. This represents the latest sending time of time-sensitive stream j. This indicates the earliest start time of time-sensitive stream i. This indicates the earliest start time of time-sensitive stream j. This represents the window overlap ratio between time-sensitive stream i and time-sensitive stream j; Periodic compatibility Load utilization jitter interval risk Overlap with time window We perform a weighted summation to obtain the comprehensive compatibility index. Specifically, this is achieved through the following formula: ； ； in, , , , All are preset weights; Repeat the above steps to calculate the multidimensional compatibility index matrix.

4. The scheduling method for multidimensional compatibility and conflict repair oriented towards TSN end-side jitter according to claim 1, characterized in that, Determine the anchor stream and the final transmission phase of the anchor stream in each queue, including: For each queue, obtain the greatest common divisor of the periods of all time-sensitive streams in the queue as the common scale G, and obtain the least common multiple of the periods of all time-sensitive streams in the queue as the supercycle H; Among all time-sensitive streams in the queue, the time-sensitive stream with the longest period is selected as the anchor stream. If the periods of the time-sensitive streams are the same, the time-sensitive stream with the highest load is selected as the anchor stream. The load is obtained by dividing the packet size of the time-sensitive stream by the period. If the periods and loads of the time-sensitive streams are the same, the time-sensitive stream with the shortest packet deadline is selected as the anchor stream. Thus, the anchor stream for each queue is obtained. The median value of the target interval of the anchor stream is obtained as the initial transmission phase of the anchor stream, and the initial transmission phase of each anchor stream is obtained. The target interval of the anchor stream is the time interval from the earliest transmission time to the latest transmission time. For each anchor stream, it is determined whether the initial transmission phase of the anchor stream satisfies the constraints of the anchor stream. If the initial transmission phase of the anchor stream satisfies the constraints, the initial transmission phase of the anchor stream is taken as the final transmission phase. If the initial transmission phase of the anchor stream does not satisfy the constraints, the initial phase of the anchor stream is fine-tuned. Specifically, within the target interval of the anchor stream, the initial transmission phase is moved at a granularity of a common scale G to obtain the final transmission phase of the anchor stream, and thus the final transmission phase of each anchor stream is obtained.

5. The scheduling method for multidimensional compatibility and conflict repair oriented towards TSN end-side jitter according to claim 1, characterized in that, Determine the transmission phase of non-anchor streams other than anchor streams in each queue, including: Get the first non-anchor stream in the queue, and remove the determined phases from the interval between the earliest and latest sending times of the non-anchor stream to obtain the candidate sending phases; For each candidate transmission phase, calculate the cost evaluation. ; Among all the cost evaluations of candidate transmission phases, the candidate transmission phase with the lowest cost evaluation is selected and used as the first transmission phase of the non-anchor stream. Determine whether the first transmission phase of the non-anchor stream satisfies the constraint information of the non-anchor stream. If the first transmission phase of the non-anchor stream satisfies the constraint information of the non-anchor stream, take the first transmission phase as the final transmission phase of the non-anchor stream, obtain the next anchor stream in the queue, and return to execute: remove the determined phase from the interval between the earliest and latest transmission times of the non-anchor stream. If the first transmission phase of a non-anchor stream does not satisfy the constraint information of the non-anchor stream, conflict repair is performed to obtain the repair result. The repair result includes the final transmission phase of the non-anchor stream or conflict information. The conflict information includes the non-anchor stream, the link of the non-anchor stream, and all phases of the non-anchor stream. If the repair result includes conflict information, increase the cost weight of all phases of non-anchor streams in the conflict information and return to the execution: remove the determined phases from the interval between the earliest and latest sending times of the non-anchor streams until all sending phases of the non-anchor streams satisfy the constraint information, and obtain the final sending phase of the non-anchor streams in each queue. If, during multiple rounds of execution, there are sending phases of non-anchor streams that do not satisfy the constraint information, return to the execution again: remove the determined phases from the interval between the earliest and latest sending times of the non-anchor streams; for sending phases of non-anchor streams with high priority, determine whether they satisfy the constraint information; for sending phases of non-anchor streams with low priority, do not determine whether they satisfy the constraint information, and obtain the final sending phase of the non-anchor streams in each queue.

6. A scheduling method for multidimensional compatibility and conflict repair oriented towards TSN end-side jitter according to claim 5, characterized in that, For each candidate transmission phase, calculate the cost evaluation. Specifically, this is achieved through the following formula: ； in, Indicates the candidate transmission phase; QueueGapRisk represents the inverse penalty of the minimum head-to-tail distance of adjacent intervals in the same queue; PortGapRisk represents the occupancy density penalty of other queues on the same port in this time window; SlackLossPenalty represents the penalty for insufficient window slack. , , All are cost weights.

7. A scheduling method for multidimensional compatibility and conflict repair oriented towards TSN end-side jitter according to claim 5, characterized in that, If the first transmission phase of a non-anchor stream does not satisfy the constraint information of the non-anchor stream, conflict repair is performed, and the repair result is obtained, including: The first transmission phase of the non-anchor flow is fine-tuned to obtain the second transmission phase. Specifically, among the candidate phases of the non-anchor flow, the first transmission phase is moved multiple times with a common scale G as the granularity. The cost evaluation of each moved transmission phase is calculated, and the transmission phase with the smallest cost evaluation is selected as the second transmission phase. If the second transmission phase satisfies the constraint information, the second transmission phase is taken as the final transmission phase of the non-anchor flow. If the second transmission phase does not meet the constraint information, intra-cluster rearrangement is performed. Specifically, all non-anchor flows in the queue to which the non-anchor flow belongs are obtained, and sorted in ascending order of data packet deadlines. For each non-anchor flow, the following steps are performed: remove the determined phase from the interval between the earliest and latest transmission times of the non-anchor flow until the transmission phase is obtained through phase fine-tuning. If the transmission phase obtained through phase fine-tuning meets the constraint information, it is taken as the final phase of the non-anchor flow. If the phase fine-tuning results in a transmission phase that does not meet the constraints, then inter-cluster migration is performed. Specifically, the non-anchor flow is migrated to the queue adjacent to its own queue and with the lowest queue load. The queue load is the sum of the packet sizes of all time-sensitive flows in the queue divided by the least common multiple of the periods of all time-sensitive flows. Then, the process returns to: removing the determined phases from the interval between the earliest and latest transmission times of the non-anchor flow until the transmission phase is obtained by intra-cluster rearrangement. If the transmission phase obtained by intra-cluster rearrangement does not meet the constraints, then inter-cluster migration is performed again until the last queue is reached. If a transmission phase meets the constraints during the inter-cluster migration process, it is taken as the final phase of the non-anchor flow. If no transmission phase meets the constraints during the inter-cluster migration process, then the non-anchor flow, the link of the non-anchor flow, and all phases of the non-anchor flow are recorded.

Images