A high-efficiency network traffic optimization control method based on a switch

Abstract

This invention discloses a high-efficiency network traffic optimization and control method based on a switch, belonging to the field of network traffic optimization technology. By introducing a micro-time detection window on the output port side and identifying synchronous burst time intervals, the traditional judgment logic based on long-term congestion thresholds is refined to a micro-time scale. This allows the switch to identify the independent phenomenon of short-term concentrated arrivals, rather than relying solely on whether the queue has grown for a long time as a trigger condition. This identification mechanism clearly locates the instantaneous traffic aggregation behavior that was previously difficult to distinguish individually, providing a targeted processing basis for subsequent scheduling actions. By constructing a set of micro-time slices and combining them with the offset budget of the service flow, a micro-time slice offset table is generated in the control plane. This allows multiple synchronously arriving service flows to be actively split into different time slices for dequeueing in terms of transmission timing, changing the traditional method of directly releasing queues in order.

Description

Technical Field

[0001] This invention relates to the field of network traffic optimization technology, specifically to a high-efficiency network traffic optimization and control method based on switches. Background Technology

[0002] In data center networks, cloud computing environments, and enterprise core switching networks, switches serve as critical nodes for traffic aggregation and forwarding. Their output ports need to process service flows from multiple input directions within a unit of time. When multiple short-duration service flows arrive simultaneously at the same output port queue within a very short timescale, a momentary traffic superposition phenomenon occurs at the port buffer. This phenomenon belongs to synchronous burst entry behavior at a micro-timescale, and is a relatively hidden but frequently occurring specific problem in switch port transmission scheduling.

[0003] In existing switch flow control mechanisms, port buffer handling primarily relies on congestion assessment mechanisms based on queue length thresholds or priority-based queuing scheduling strategies. These mechanisms typically use persistent bandwidth occupancy or long-term queue growth as trigger conditions, only implementing rate limiting, dropping, or rescheduling when the port buffer occupancy rate exceeds the threshold. However, when multiple short-lifecycle service flows arrive simultaneously within milliseconds or even less, the existing congestion control logic often fails to be triggered because the overall bandwidth utilization may be within a normal range and the queue growth duration is short.

[0004] When the aforementioned synchronous burst entry occurs, although the port's physical bandwidth is not saturated, the rapid entry of multiple service flows into the queue within a very short time scale causes the queue length to jump rapidly, forcing continuous compression of the dequeueing time. This results in drastic fluctuations in queuing time between adjacent time slices. This phenomenon typically manifests as packet latency jitter, abnormal instantaneous queuing delays, and even unstable response times in upper-layer applications. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a highly efficient network traffic optimization and control method based on switches, which solves the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a high-efficiency network traffic optimization and control method based on a switch, comprising the following steps: S1. At the switch output port Pout, monitor the arrival time of the service flow fi entering the output queue Qout, and use the micro-time detection window w to count the packet entry behavior, generating a candidate misaligned flow set Gb and a synchronous burst time interval [Ts, Te]. S2. Based on the synchronous burst time interval [Ts, Te], analyze the historical transmission completion time sequence of the output port, construct the set of micro time slices S for the output port to execute transmission, and generate the allowable time offset range for each service flow fi in the candidate misaligned flow set Gb, forming the offset budget Ω. S3. Based on the candidate misaligned flow set Gb, the micro-time slice set S and the offset budget Ω, determine the time offset Δti of each service flow fi and the corresponding micro-time slice sequence, and generate the micro-time slice offset table Toff. S4. Based on the micro-time slice offset table Toff, implement micro-time slice gating scheduling on the output port Pout, so that the synchronously entering service flow fi is dequeued out in a staggered manner according to different micro-time slices, and record the scheduling execution data Se; S5. Evaluate the queue change status based on the scheduling execution data Se, update the micro-time slice set S and offset budget Ω, and feed it back to step S2.

[0007] Preferably, S1 includes S11; S11: On the enqueue side of the queue Qout corresponding to the output port Pout, establish an arrival event record for each service flow fi, and write the arrival event sequence at the granularity of the message; record the arrival time and message length for each enqueue message, and synchronously write the service flow fi identifier and the target queue identifier to which the message belongs. Window aggregation is performed on all arriving events using a micro-time detection window w: Discretize the time axis into continuous window segments w(u) = [u × w, (u + 1) × w). The window starts from u×w, including the left endpoint, and ends at (u+1)×w, excluding the right endpoint. The advantage of this is that adjacent windows will not overlap, and any point in time will fall into only one window. Boundary points will be assigned to the left end of the next window and will not be counted repeatedly. All arriving events falling within the same window segment w(u) are organized into a two-level "stream-event" structure to form a window arrival matrix Mu, which includes the enqueued event count Mu(i), the accumulated enqueued bytes MuL(i), and the window queue sampling sequence {q(u)}. Specifically, when the arrival time of a message falls within the window segment w(u), the message is assigned to the corresponding window statistical unit; within the same window segment, all messages belonging to the same service flow fi are subjected to aggregation and statistical processing. The number of packets that fall within the window segment w(u) and belong to the service flow fi is accumulated to obtain the enqueued event count Mu(i) of the service flow fi within the window. The message lengths corresponding to the messages are accumulated to obtain the cumulative number of enqueued bytes MuL(i) of the service flow fi within the window. Mu(i) and MuL(i) are both obtained by window aggregation operation based on the collected arrival time and message length, and are used to characterize the degree of concentrated entry of service flow fi within the micro-time detection window; Read the queue length sample value q(u) of queue Qout at the end of the window to form the window queue sampling sequence {q(u)}.

[0008] Preferably, S1 further includes S12; S12, based on the enqueue event count Mu(i), the enqueue byte accumulation MuL(i), and the window queue sampling sequence {q(u)}, calculates the synchronization burst determination quantity for each window segment w(u): specifically as follows: First, calculate the number of concurrent streams within the window: Nu = |{fi|Mu(i)>0}|; In the formula, fi represents the identifier of the i-th business flow fi, and Mu(i) represents the enqueue event count, specifically the enqueue event count of business flow fi within the u-th micro-time detection window w(u); Then, the queue length transition Δq(u) = q(u) - q(u-1) and the continuous transition discrimination Δq(u) > 0 and Δq(u+1) > 0 are calculated; where q(u) represents the queue length value obtained by sampling the output queue Qout once when u micro-time detection windows w(u) end, and q(u-1) represents the queue length value obtained by sampling the output queue Qout once when u-1 micro-time detection windows w(u-1) end; When the number of concurrent flows Nu in the window reaches the preset concurrent flow judgment threshold, and Δq(u) satisfies the continuous transition judgment, the window segment w(u) is marked as a synchronous burst candidate window, and boundary locking is performed on the continuous candidate windows: the starting point of the earliest candidate window that meets the conditions is taken as Ts, and the ending point of the last candidate window that still meets the conditions is taken as Te, to obtain the synchronous burst time interval [Ts, Te]; and set extraction is performed on the participating flows within the synchronous burst time interval [Ts, Te]: the flows that satisfy the condition that have enqueuing events within [Ts, Te] are uniformly collected into the candidate misaligned flow set Gb; Furthermore, an arrival compactness parameter JCi is calculated for each flow within the candidate misaligned flow set Gb, which characterizes the degree of concentrated arrival of the flow within the burst window.

[0009] The arrival compactness parameter JCi is determined by the number of enqueued events in the stream within [Ts, Te], the dispersion of the event interval, and the clustering of enqueued bytes, so that subsequent misaligned scheduling can distinguish between extremely short dense streams and relatively dispersed streams; The compactness parameter JCi is obtained as follows: The number of messages entering within a given interval is counted to obtain the number of enqueued events; a time interval sequence is constructed based on the arrival times of adjacent messages; the lengths of messages within the interval are accumulated to obtain the total sum. First, count the number of packets that enter the queue for a certain service flow fi within the synchronous burst time interval, and at the same time accumulate the total amount of data that enters the queue for service flow fi within the time period, which is used to represent the scale of service flow fi entering in a short period of time.

[0010] Subsequently, the arrival time between adjacent packets in the business flow fi is calculated sequentially to obtain the time interval between each packet, and these time intervals are statistically analyzed to determine whether the packets are concentrated.

[0011] Next, the fluctuations of all time intervals are calculated to reflect whether the message arrival rhythm is stable. The smaller the time interval variation, the more concentrated the message arrivals.

[0012] Finally, the number of times the business flow fi enters and the total amount of data entering are combined with the degree of fluctuation in the time interval. When a large number of packets enter continuously in a short period of time, a larger compactness result is obtained; when the packets enter more dispersedly, the result is relatively smaller. The concurrent flow determination threshold is not a fixed empirical value, but is obtained based on statistics of the port's historical stable operation phase. For example, it is determined by statistically analyzing the average number of active flows within a unit micro-time window under normal conditions and taking its upper bound interval. The condition for continuous transition is: Δq(u) is greater than 0 and Δq(u+1) is greater than 0, that is, the queue length continues to increase within two adjacent micro-time windows; Preferably, S2 includes S21; S21 takes the switch output port Pout as the object, extracts the continuous dequeue completion records of the output port Pout before [Ts, Te], and forms the actual dequeue completion time sequence of the port; the difference between two adjacent dequeue completion times is obtained to obtain the adjacent transmission completion interval, and forms an interval sequence in chronological order; The interval sequence is filtered for stable segments: obvious abnormal intervals are removed, such as isolated points that are too different from neighboring intervals, and the remaining intervals are aggregated into several beat clusters according to their proximity. Each beat cluster represents the sustainable transmission rhythm of a port at that time. Based on the beat cluster results, the time axis is divided into continuous micro-time slices sj, and the boundary of each micro-time slice is aligned with the typical interval of the beat cluster to obtain the set of micro-time slices S that the output port can execute to send. Based on the set of micro-time slices S, two types of execution boundaries are determined for each micro-time slice sj: One is the transmittable capacity, which is estimated by the duration of the time slice, the port transmission rhythm, and the number of message services that can be accommodated within the slice, and is expressed as the cumulative transmittable capacity that can be scheduled within the slice; The second is the minimum scheduling interval, which is estimated by the minimum stable interval between adjacent transmission completion intervals within the corresponding beat cluster; it is used to limit the shortest allocable gap between two dequeue actions within the chip, preventing multiple dequeue actions from being squeezed into gaps that the port beat cannot cover.

[0013] Preferably, S2 further includes S22; S22: Based on the micro time slice set S, generate an adjustable time offset range for each service flow fi in the candidate misaligned flow set Gb, and aggregate all offset ranges to form an offset budget set Ω; The specific processing steps are as follows: First, read the arrival compactness parameter JCi output by S1, and simultaneously read the enqueue behavior summary of the service flow fi in [Ts, Te], such as the packet length distribution of the flow in the burst interval, the continuous arrival interval characteristics, and whether there is an identifier that must maintain a strict order; then, classify the flow according to the arrival compactness parameter JCi, and divide the candidate misaligned flow set Gb into several categories: flows with high compactness are characterized by concentrated entry in a short period of time, and flows with low compactness are characterized by more dispersed entry; within the micro time slice boundary, allocate different offset windows for flows of different categories; For highly compact business flows fi, a finer-grained offset range is provided, and the offset is limited to rolling amortization only within a small number of adjacent micro-time slices; so that the dequeueing is no longer concentrated on the same window boundary; For low-compact service flows fi, the offset range converges to a narrower on-chip fine-tuning; avoiding the introduction of additional misalignment diffusion into the originally dispersed entry behavior; For service flows fi with sequence-sensitive identifiers, the offset budget allows inter-chip delays but not inter-chip advances, and a sequence constraint flag is written in the time offset range to ensure that the relative order of packets within the flow does not reverse during subsequent scheduling phases. Finally, a consistency check is performed between the time offset range in each service flow fi and the micro-time slice set S: Check whether there is sufficient transmittable capacity within the micro-time slice set covered by the time offset range to carry the dequeue demand of service flow fi within the burst interval, and ensure that the offset step size is not less than the scheduling resolution allowed by the minimum scheduling interval; output Ω after the verification is passed.

[0014] Preferably, S3 includes S31 and S32; S31. After obtaining the candidate misaligned flow set Gb, the micro-time slice set S, and the offset budget Ω, For each service flow fi in the candidate misaligned flow set Gb, a summary of the load to be released is constructed within the synchronous burst time interval. The summary includes: the cumulative number of bytes that service flow fi enters the queue within [Ts, Te], the discreteness of the message length, and the distribution of the message arrival points on the micro time slice. Read the offset budget ωi corresponding to the stream to obtain the time range and allowable offset step size boundary of the stream, and generate a set of offset candidate list for each service stream fi: Within the allowable range of the offset budget ωi, several offset values ​​are discretized by step size, and each candidate offset value is mapped to the set of micro time slices S to obtain a draft of the slice sequence of which micro time slices the stream should occupy if the offset value is adopted. When generating the slice sequence draft, a sequence constraint anchor point is established simultaneously: based on the order of packets in the output queue Qout of the service flow fi, the packet sequence is segmented according to the order of arrival, and each packet segment can only be mapped to a set of micro time slices that are no earlier than the previous segment in time, and packets within a segment are only allowed to be dequeued in the same micro time slice in the order of the queue head; this anchor point transforms the "intra-flow order" into a "slice sequence monotonicity constraint", so that when adjusting the offset later, there will be no situation where packets that arrive later are arranged in an earlier micro time slice; S32. Based on the slice sequence draft, perform slice content volume verification for each micro-time slice sj: count the total streaming load demand mapped to micro-time slice sj in all drafts, compare it with the sendable capacity of the micro-time slice, and identify the set of conflicting micro-time slices that exceed the capacity. For conflicting micro-time slices, instead of the conventional methods of "priority replacement" or "simple shift," inter-slice staggered rearrangement is performed: First, the traffic flows fi in the conflicting micro-time slice are grouped according to the strength of constraints on transmittable capacity and sequence anchor points, identifying compact flow groups that must be released close to the current window and non-compact flow groups that are allowed to be released more dispersedly; then, within the allowable range of the offset budget ωi, the offset candidates are replaced one by one for the adjustable flows, so that the slice sequence draft is moved out of the conflicting micro-time slice and falls into the adjacent free micro-time slice. During the migration, two constraints are checked simultaneously: First, the draft sequence of the migrated slices still satisfies the monotonicity of the sequential anchor point; Secondly, the remaining capacity of the migration destination micro-time slice is sufficient to handle the load of the migrated service flow fi without disrupting the minimum scheduling interval within the slice. Once the conflict is resolved, a unique time offset Δti and a corresponding final micro-time slice sequence draft are determined for each service flow fi. The micro-time slice sequences and time offsets Δti of all service flows are then summarized to generate the micro-time slice offset table Toff.

[0015] The micro-time slice offset table Toff not only records the offset of each flow, but also explicitly records the path occupied by the flow on the micro-time slice set S, so that subsequent gating scheduling can be executed slice by slice according to the micro-time slice offset table Toff, instead of relying on real-time preemptive queuing. Preferably, S4 includes S41; S41. After obtaining the micro time slice offset table Toff, establish a micro time slice gating rule table for the output port Pout, and expand the micro time slice sequence of each service flow fi in the time slice offset table Toff into slice-level authorization information: for each micro time slice sj, summarize all service flows that satisfy the micro time slice sj∈micro time slice sequence to form the set of allowed outgoing flows of the micro time slice. At the same time, based on the occupation arrangement of each service flow in the time slice in the micro time slice offset table Toff, the in-slice dequeue limit is generated and used to limit the cumulative number of bytes that service flow fi is allowed to send or the number of messages allowed to be dequeued in the micro time slice, so that the in-slice dequeue will not crowd out the range of the sendable capacity. During scheduling execution, the scheduler enters the in-slice gating state with the start boundary of the micro-time slice sj as the trigger point: it only grants dequeue permission to service flows that belong to the set of allowed dequeue flows, and during the dequeue process, it deducts the in-slice transmission of each service flow according to the in-slice dequeue limit. When a traffic flow reaches its intra-slice limit or its head packet has been sent, the dequeue permission for the traffic flow in the current micro-time slice is revoked. For traffic flows that do not belong to the allowed dequeue flow set, maintain gating and keep them in Qout waiting for the next authorized microtime slice.

[0016] To avoid "queue head switching jitter" during slice switching, a slice switching locking action is introduced at the micro-time slice boundary: new message initiation and transmission are prohibited within the minimum scheduling interval before the end of micro-time slice sj, and only the transmission action that has been initiated is allowed to be completed. At the end of micro-time slice sj, the in-slice metering value of each service flow is frozen to ensure that the gating authorization switching between micro-time slice sj and micro-time slice sj+1 has a clear boundary. Preferably, S4 includes S42; S42. Establish an execution record unit for each micro-time slice sj, record the actual dequeueing behavior that occurs within the micro-time slice, and form scheduling execution data Se; The scheduling execution data Se contains three types of information: One is the slice-level gating hit record, which records whether each service flow fi in the allowed dequeue flow set is actually dequeued within a micro-time slice, the start and end times of dequeueing, and the actual number of bytes dequeued; this is used to reflect whether the gating authorization is actually used. The second is the intra-slice upper limit consumption record, which records the proportion of the intra-slice outbound upper limit consumed for each business flow, and is used to determine whether the micro-time slice allocation is too tight or too loose. Thirdly, the queue state evolution record is recorded, which records the window queue sampling sequence {q(u)} and queue length transition amount Δq(u) at the micro-time slice boundary; this is used to characterize whether misaligned scheduling transforms the concentrated transition of the queue into a dispersed, gradually changing form. Preferably, S5 includes S51; S51. Perform time alignment processing on the acquired scheduling execution data Se: Using the synchronous burst time interval [Ts, Te] as the unified observation interval, merge the slice boundaries, actual dequeue start and end times, actual dequeue byte counts, and window queue sampling sequence {q(u)} of each micro-time slice sj in the scheduling execution data Se into the same sequence in chronological order. Two types of evaluation calculations are performed based on sequences: The first aspect is the calculation of queue stability assessment results: The changes in the window queue sampling sequence {q(u)} within the synchronous burst time interval [Ts, Te] are analyzed, and the criterion for whether concentrated transitions are broken up is extracted. That is, it is to identify whether there are still segments where the queue length transition amount Δq(u) is continuously positive and the growth rate is dense in a short period of time. Combined with the dispersion of the waiting time at the front of the queue at the slice boundary and the fluctuation of the upper limit consumption ratio of the dequeue within the slice, a comprehensive stability conclusion is formed. The queue stability assessment results are used to characterize whether the queue still exhibits burst accumulation characteristics after misaligned gating. Secondly, the calculation of execution deviation: using the micro time slice offset table Toff as the planning benchmark, the planned micro time slice sequence marked in the micro time slice offset table Toff for each service flow fi is compared to check whether the actual dequeue position of the service flow fi in the scheduling execution data Se is consistent with the micro time slice sequence, and further the difference between the planned release time point and the actual dequeue start time point is statistically analyzed to form the slice-level deviation of the service flow fi; Then, the deviations of all service flows within the candidate misaligned flow set Gb are summarized to obtain the overall execution deviation amount E; and the type of deviation source is marked in the execution deviation amount E, such as: gate authorization failure, insufficient slice content, delay caused by slice switching lock, and intra-slice crossing caused by queue head message length, etc.

[0017] Preferably, S5 also includes S52; S52. After obtaining the queue stability assessment results and execution deviation, determine the problem attribution—update object selection: When the queue stability assessment results show that there are still concentrated transitions, and the execution deviation shows that the plan and execution are highly consistent, it indicates that the micro time slice offset table Toff can be implemented but the misalignment force is insufficient. In this case, the offset budget Ω should be updated first. When the execution deviation is concentrated in slice switching, insufficient slice content, or low gating hit rate, it indicates that the division of the micro time slice set S does not match the actual clock cycle of the port. At this time, the micro time slice set S is updated first and the offset step size constraint of the offset budget Ω is corrected in conjunction with the adjustment. The update method for the offset budget Ω is based on the relationship between the slice-level deviation of each service flow in the execution deviation amount and the position of the concentrated transition residual segment in the queue stability assessment result; For traffic flows in the candidate misaligned flow set Gb that are relatively compact and still clustered and dequeued within the same window, expand their allowed offset range or refine their offset distance step size; so that they can be migrated to more distant adjacent micro time slices. For traffic flows with low compactness but excessive migration leading to a decrease in gating hit rate, narrow their allowable offset range to avoid new types of jitter caused by drifting between multiple micro-time slices. Update method for the set of micro-time slices S: When the execution deviation indicates that certain micro-time slices frequently fail to utilize their upper limit or frequently experience insufficient content, the micro-time slice boundaries are then further subdivided or merged: For segments with insufficient capacity, micro-time slices are subdivided to improve scheduling resolution and reduce intra-slice congestion; For segments with unused upper limits, adjacent micro-time slices are merged to reduce the number of slice switching and mitigate the delay effect caused by locking boundaries. At the same time, the corresponding transmittable capacity and minimum scheduling interval constraints are updated synchronously to fit the new clock division. After the update is completed, the updated micro-time slice set S and offset budget Ω are generated and written back to step S2 as input for the next round of synchronous burst processing. This enables the subsequent generation of the micro-time slice offset table Toff to directly perform misalignment solving based on the micro-time slice set S which is closer to the actual port behavior and the offset budget Ω which is more in line with the burst flow compactness distribution, thereby forming a self-calibrating misalignment scheduling closed loop driven by execution records.

[0018] This invention provides a high-efficiency network traffic optimization and control method based on switches, which has the following beneficial effects: (1) By establishing a message-level arrival event recording mechanism on the queuing side of the output port and combining it with micro-time detection windows to perform window aggregation processing on queuing behavior, the switch can perform structured observation of the entry behavior of multiple service flows at the micro-time scale, instead of relying solely on the traditional queue length threshold judgment. This method achieves a fine characterization of the phenomenon of "short-term concentrated entry", so that synchronous burst entry is no longer submerged in ordinary traffic fluctuations, and makes up for the lack of separate analysis methods for micro-time scale traffic aggregation in existing technologies from the identification level.

[0019] (2) By continuously analyzing the historical dequeue completion time of the output port and filtering the adjacent transmission intervals for stable segments, the division of micro-time slices no longer depends on a fixed time scale, but is constructed based on the actual transmission rhythm of the port. This method matches the time slice boundary with the physical transmission behavior of the port, avoiding scheduling deviations caused by the inconsistency between the manually set time granularity and the actual processing capacity of the port, thereby improving the feasibility of subsequent misaligned scheduling at the execution level.

[0020] (3) By introducing multi-layered processing actions such as load summary construction, offset candidate discretization, sequence constraint anchor point, and slice content volume verification in the misaligned scheduling generation stage, the time misaligned scheduling is transformed from a single time shift behavior into a structured scheduling process constrained by capacity, sequence, and tick. This approach enables the feasibility verification of the release path of each service flow to be completed in the generation stage, avoiding the random queuing fluctuations caused by traditional direct scheduling based on priority, and giving the scheduling behavior a clear time distribution structure.

[0021] (4) By aligning the scheduling execution data with time and unifying the sequence, the scheduling plan and the actual execution results are compared and analyzed under the same time coordinate system, thereby avoiding misjudgment problems caused by the inconsistency of time scale between the planning layer and the execution layer. This processing method makes the observation of queue changes within the micro time scale more accurate, and provides a reliable basis for judging whether misaligned scheduling has really changed the queue concentration transition state.

[0022] By simultaneously introducing two analytical dimensions—queue stability assessment results and execution deviation—the nature of the problem can be differentiated and judged. On the one hand, by analyzing the queue change pattern, it is possible to identify whether the concentrated transition has been dispersed, and to determine whether the misaligned scheduling is effective from the overall structure. On the other hand, by comparing the planned micro-time slice sequence with the actual dequeue segments one by one, it is clear whether the deviation stems from the inadequacy of the scheduling strategy itself or from the mismatch of the execution phase. This dual-dimensional assessment mechanism avoids simply attributing queue fluctuations to a single factor, and improves the pertinence of subsequent adjustment decisions. Attached Figure Description

[0023] Figure 1 This is a schematic diagram illustrating the steps of a high-efficiency network traffic optimization and control method based on a switch according to the present invention. Figure 2 This is a schematic diagram of the updated micro-time slice set and offset budget acquisition process of the present invention; Figure 3 This is a schematic diagram illustrating the process of obtaining the micro-time slice set and offset budget according to the present invention. Detailed Implementation

[0024] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0025] Example 1 This invention provides a high-efficiency network traffic optimization and control method based on switches. Please refer to [link / reference]. Figure 1 This includes the following steps: S1. At the switch output port Pout, monitor the arrival time of the service flow fi entering the output queue Qout, and use the micro-time detection window w to count the packet entry behavior, generating a candidate misaligned flow set Gb and a synchronous burst time interval [Ts, Te]. S2. Based on the synchronous burst time interval [Ts, Te], analyze the historical transmission completion time sequence of the output port, construct the set of micro time slices S for the output port to execute transmission, and generate the allowable time offset range for each service flow fi in the candidate misaligned flow set Gb, forming the offset budget Ω. S3. Based on the candidate misaligned flow set Gb, the micro-time slice set S and the offset budget Ω, determine the time offset Δti of each service flow fi and the corresponding micro-time slice sequence, and generate the micro-time slice offset table Toff. S4. Based on the micro-time slice offset table Toff, implement micro-time slice gating scheduling on the output port Pout, so that the synchronously entering service flow fi is dequeued out in a staggered manner according to different micro-time slices, and record the scheduling execution data Se; S5. Evaluate the queue change status based on the scheduling execution data Se, update the micro-time slice set S and offset budget Ω, and feed it back to step S2.

[0026] In this embodiment, by introducing a micro-time detection window on the output port side and identifying synchronous burst time intervals, the traditional judgment logic based on long-term congestion thresholds is refined to a micro-time scale. This allows the switch to identify the independent phenomenon of "short-term concentrated entry," rather than relying solely on whether the queue has grown over a long period as a triggering condition. This identification mechanism clearly locates the instantaneous traffic aggregation behavior that was previously difficult to distinguish individually, providing a basis for targeted processing of subsequent scheduling actions.

[0027] By constructing a set of micro-time slices and combining them with the offset budget of the service flow, a micro-time slice offset table is generated in the control plane. This allows multiple synchronously arriving service flows to be proactively split into different time slices for dequeueing, changing the traditional method of directly releasing queues in sequence. This approach transforms the port transmission rhythm from "centralized release" to "staggered release." While maintaining the overall bandwidth, it distributes queue transitions within micro-time scales, thereby smoothing out short-term fluctuations in queue length and reducing drastic changes in packet waiting time between adjacent time slices.

[0028] By recording execution data and evaluating queue changes after scheduling, a self-calibration update mechanism for the micro-time slice set and offset budget is formed. This enables the system to dynamically correct the time slice division and offset range based on the actual transmission behavior of the ports. This closed-loop update method avoids the mismatch problem caused by fixed scheduling strategies under different load conditions, allowing subsequent synchronization burst scenarios to directly generate new misaligned scheduling schemes based on the corrected tick structure, thus enhancing the adaptability between the overall scheduling strategy and the actual transmission capacity of the ports.

[0029] This embodiment, while maintaining the physical bandwidth and hardware structure of the switch, achieves proactive dispersion processing of synchronous burst traffic at micro-timescales by reorganizing and finely organizing the transmission time structure. This suppresses short-term queuing jitter, reduces instantaneous delay fluctuations, and improves the continuity and stability of port output rhythm, thus making up for the lack of a dedicated identification and control mechanism for this type of problem in the prior art.

[0030] Example 2 Please refer to Figure 1 Specifically: S1 includes S11; S11. On the enqueue side of the queue Qout corresponding to the output port Pout, establish an arrival event record for each service flow fi, and write the arrival event sequence at the granularity of the message; record the arrival time and message length for each enqueue message, and synchronously write the service flow fi identifier and the target queue identifier to which the message belongs. Window aggregation is performed on all arriving events using a micro-time detection window w: Discretize the time axis into continuous window segments w(u) = [u × w, (u + 1) × w). All arriving events falling within the same window segment w(u) are organized into a two-level flow-event structure to form a window arrival matrix Mu, which includes the enqueued event count Mu(i), the enqueued byte accumulation MuL(i), and the window queue sampling sequence {q(u)}. Specifically, when the arrival time of a message falls within the window segment w(u), the message is assigned to the corresponding window statistical unit; within the same window segment, all messages belonging to the same service flow fi are subjected to aggregation and statistical processing. The number of packets that fall within the window segment w(u) and belong to the service flow fi is accumulated to obtain the enqueued event count Mu(i) of the service flow fi within the window. The message lengths corresponding to the messages are accumulated to obtain the cumulative number of enqueued bytes MuL(i) of the service flow fi within the window. Read the queue length sample value q(u) of queue Qout at the end of the window to form the window queue sampling sequence {q(u)}.

[0031] S1 also includes S12; S12, based on the enqueue event count Mu(i), the enqueue byte accumulation MuL(i), and the window queue sampling sequence {q(u)}, calculates the synchronization burst determination quantity for each window segment w(u): The specific method is as follows: First, calculate the number of concurrent streams within the window: Nu = |{fi|Mu(i)>0}|; In the formula, fi represents the identifier of the i-th business flow fi, and Mu(i) represents the enqueue event count, specifically the enqueue event count of business flow fi within the u-th micro-time detection window w(u); Then, the queue length transition Δq(u) = q(u) - q(u-1) and the continuous transition discrimination Δq(u) > 0 and Δq(u+1) > 0 are calculated; where q(u) represents the queue length value obtained by sampling the output queue Qout once when u micro-time detection windows w(u) end, and q(u-1) represents the queue length value obtained by sampling the output queue Qout once when u-1 micro-time detection windows w(u-1) end; When the number of concurrent flows Nu in the window reaches the preset concurrent flow judgment threshold, and Δq(u) satisfies the continuous transition judgment, the window segment w(u) is marked as a synchronous burst candidate window, and boundary locking is performed on the continuous candidate windows: the starting point of the earliest candidate window that meets the conditions is taken as Ts, and the ending point of the last candidate window that still meets the conditions is taken as Te, to obtain the synchronous burst time interval [Ts, Te]; and set extraction is performed on the participating flows within the synchronous burst time interval [Ts, Te]: the flows that satisfy the condition that have enqueuing events within [Ts, Te] are uniformly collected into the candidate misaligned flow set Gb; For each flow in the candidate misaligned flow set Gb, calculate the arrival compactness parameter JCi, which is used to characterize the degree of concentrated arrival of the flow within the burst window.

[0032] In this embodiment, by establishing a message-level arrival event recording mechanism on the queuing side of the output port and combining it with micro-time detection windows to perform window aggregation processing on queuing behavior, the switch can perform structured observation of multi-service flow inflow behavior at a micro-time scale, rather than relying solely on traditional queue length threshold judgment. This approach achieves a fine characterization of the phenomenon of "short-term concentrated inflows," ensuring that synchronous bursts are no longer submerged in ordinary traffic fluctuations, and compensating for the lack of separate analysis methods for micro-time scale traffic aggregation in existing technologies at the identification level.

[0033] A dual-dimensional identification mechanism was established by jointly determining the number of concurrent flows and the continuous transition state of queue length within a window: on the one hand, it confirms that multiple different business flows enter concurrently within the same micro-time window; on the other hand, it confirms that this concurrent entry has already created an actual backlog trend at the queue level. This combined determination method avoids misjudging single traffic amplification or occasional queue fluctuations as sudden events, improves the accuracy of synchronous sudden event identification, and provides a clear triggering basis for subsequent scheduling actions.

[0034] Example 3 Please refer to Figure 3 Specifically: S2 includes S21; S21. Taking the switch output port Pout as the object, extract the continuous dequeue completion records of the output port Pout before [Ts, Te] to form the actual dequeue completion time sequence of the port; calculate the difference between two adjacent dequeue completion times to obtain the adjacent transmission completion interval, and form an interval sequence in chronological order; The interval sequence is filtered for stable segments: obvious abnormal intervals are removed, and the remaining intervals are aggregated into several beat clusters according to their proximity. Each beat cluster represents the sustainable transmission rhythm of a port at that time. Based on the beat cluster results, the time axis is divided into continuous micro-time slices sj, and the boundary of each micro-time slice is aligned with the typical interval of the beat cluster to obtain the set of micro-time slices S that the output port can execute to send. Based on the set of micro-time slices S, two types of execution boundaries are determined for each micro-time slice sj: One is the transmittable capacity, which is estimated by the duration of the time slice, the port transmission rhythm, and the number of message services that can be accommodated within the slice, and is expressed as the cumulative transmittable capacity that can be scheduled within the slice; The second is the minimum scheduling interval, which is estimated by the minimum stable interval within the corresponding beat cluster when the adjacent transmission completion intervals are within the same timeframe.

[0035] S2 also includes S22; S22, based on the micro time slice set S, generates an allowable time offset range for each service flow fi in the candidate misaligned flow set Gb, and aggregates all offset ranges to form an offset budget set Ω; The specific processing steps are as follows: First, read the arrival compactness parameter JCi output by S1, and simultaneously read the enqueue behavior summary of the service flow fi in [Ts, Te]; then, classify the flow according to the arrival compactness parameter JCi, and divide the candidate misaligned flow set Gb into several categories: flows with high compactness are characterized by concentrated entry in a short period of time, while flows with low compactness are characterized by more dispersed entry; within the micro time slice boundary, allocate different offset windows for flows of different categories; For highly compact service flows fi, a finer-grained offset range is provided, and the offset is limited to rolling amortization only within a small number of adjacent micro-time slices; For low-compactness service flows fi, the offset range converges to a narrower in-chip fine-tuning; For service flow fi with a sequence-sensitive identifier, the offset budget allows inter-chip delay but not inter-chip advance, and a sequence constraint flag is written in the time offset range; Finally, a consistency check is performed between the time offset range in each service flow fi and the micro-time slice set S: Check whether there is sufficient transmittable capacity within the micro-time slice set covered by the time offset range to carry the dequeue demand of service flow fi within the burst interval, and ensure that the offset step size is not less than the scheduling resolution allowed by the minimum scheduling interval; after the verification is passed, output the offset budget set Ω.

[0036] In this embodiment, by continuously analyzing the historical dequeue completion time of the output port and filtering for stable segments between adjacent transmission intervals, the division of micro-time slices no longer depends on a fixed time scale, but is constructed based on the actual transmission rhythm of the port. This method matches the time slice boundaries with the physical transmission behavior of the port, avoiding scheduling deviations caused by the inconsistency between the manually set time granularity and the actual processing capacity of the port, thereby improving the feasibility of subsequent misaligned scheduling at the execution level.

[0037] By aggregating transmission intervals to form clock clusters, and estimating the transmittable capacity and minimum scheduling interval for each micro-time slice, each time slice not only has a time boundary but also a clear capacity boundary and scheduling resolution constraint. Compared to methods that only divide time periods at the logical level, this embodiment directly links time slices with the port's continuous transmission capability, enabling misaligned scheduling to be controlled simultaneously in both time and capacity dimensions, thereby reducing new jitter phenomena caused by intra-slice congestion or excessive scheduling at the source.

[0038] By incorporating arrival compactness parameters to tier candidate misaligned flows, a differentiated allocation strategy is applied to service flows with varying degrees of concentration across their time offset ranges. Flows with high compactness are distributed across adjacent micro-time slices for rolling release, while flows with low compactness undergo only intra-slice fine-tuning, thus avoiding excessive disturbance to the originally dispersed traffic structure. This tiering mechanism matches the time offset range with the flow arrival characteristics, enhancing the correlation between scheduling strategies and actual bursty behavior.

[0039] By performing consistency checks on the time offset range, micro-time slice capacity, and minimum scheduling interval, it is ensured that the offset arrangement for each service flow is executable within the port's capacity, avoiding the generation of theoretically feasible but unachievable scheduling plans under actual transmission cycles. This verification mechanism enables the offset budget set to complete feasibility verification during the generation phase, reducing the possibility of conflicts or deviations in subsequent execution phases.

[0040] Example 4 Please refer to Figure 1 Specifically: S3 includes S31 and S32; S31. After obtaining the candidate misaligned flow set Gb, the micro-time slice set S and the offset budget Ω, construct a summary of the load to be released for each service flow fi in the candidate misaligned flow set Gb within the synchronous burst time interval. The summary includes: the cumulative number of bytes of service flow fi entering the queue in [Ts, Te], the discreteness of the message length, and the distribution of the message arrival points on the micro-time slice. Read the offset budget ωi corresponding to the stream to obtain the time range and allowable offset step size boundary of the stream, and generate a set of offset candidate list for each service stream fi: Within the allowable range of the offset budget ωi, several offset values ​​are discretized by step size, and each candidate offset value is mapped to the set of micro-time slices S to obtain a slice sequence draft; When generating the slice sequence draft, establish the order constraint anchor point simultaneously: based on the order of packets in the output queue Qout of the service flow fi, segment the packet sequence according to the arrival order, and require that each packet segment can only be mapped to a set of micro time slices that are no earlier than the previous segment in time, and packets within the segment are only allowed to be dequeued in the same micro time slice in the order of the queue head. S32. Based on the slice sequence draft, perform slice content volume verification for each micro-time slice sj: count the total streaming load demand mapped to micro-time slice sj in all drafts, compare it with the sendable capacity of the micro-time slice, and identify the set of conflicting micro-time slices that exceed the capacity. For conflicting micro-time slices, inter-slice misalignment rearrangement is performed: First, the service flows fi in the conflicting micro-time slice are grouped according to the strength of constraints on transmittable capacity and sequence anchor points, identifying compact flow groups that must be released close to the current window and non-compact flow groups that are allowed to be released more dispersedly; then, within the allowable range of the offset budget ωi, the offset candidates are replaced one by one for the adjustable flows, so that the slice sequence draft is moved out of the conflicting micro-time slice and falls into the adjacent free micro-time slice. During the migration, two constraints are checked simultaneously: First, the draft sequence of the migrated slices still satisfies the monotonicity of the sequential anchor point; Secondly, the remaining capacity of the migration destination micro-time slice is sufficient to handle the load of the migrated service flow fi without disrupting the minimum scheduling interval within the slice. Once the conflict is resolved, a unique time offset Δti and a corresponding final micro-time slice sequence draft are determined for each service flow fi. The micro-time slice sequences and time offsets Δti of all service flows are then summarized to generate the micro-time slice offset table Toff.

[0041] S4 includes S41; S41, after obtaining the micro time slice offset table Toff, establish a micro time slice gating rule table for the output port Pout, and expand the micro time slice sequence of each service flow fi in the time slice offset table Toff into slice-level authorization information: for each micro time slice sj, summarize all service flows that satisfy the micro time slice sj∈micro time slice sequence to form the set of allowed outgoing flows of the micro time slice; At the same time, based on the occupation arrangement of each service flow in the time slice in the micro time slice offset table Toff, the in-slice dequeue limit is generated and used to limit the cumulative number of bytes that service flow fi is allowed to send or the number of messages allowed to be dequeued in the micro time slice, so that the in-slice dequeue will not crowd out the range of the sendable capacity. During scheduling execution, the scheduler enters the in-slice gating state with the start boundary of the micro-time slice sj as the trigger point: it only grants dequeue permission to service flows that belong to the set of allowed dequeue flows, and during the dequeue process, it deducts the in-slice transmission of each service flow according to the in-slice dequeue limit. When a traffic flow reaches its intra-slice limit or its head packet has been sent, the dequeue permission for the traffic flow in the current micro-time slice is revoked. Gating is maintained for service flows that do not belong to the allowed dequeue flow set.

[0042] S4 includes S42; S42 establishes an execution record unit for each micro-time slice sj, records the actual dequeueing behavior that occurs within the micro-time slice, and forms scheduling execution data Se; The scheduling execution data Se contains three types of information: One is the slice-level gating hit record, which records whether each service flow fi in the allowed dequeue flow set is actually dequeued within a micro time slice, the start and end times of dequeueing, and the actual number of bytes dequeued; The second is the intra-slice upper limit consumption record, which records the proportion of the intra-slice outbound upper limit consumed for each business flow, and is used to determine whether the micro-time slice allocation is too tight or too loose. Thirdly, the queue state is recorded as it evolves with the slice, including the window queue sampling sequence {q(u)} and the queue length transition Δq(u) at the micro-time slice boundary.

[0043] In this embodiment, by introducing multi-layered processing actions such as load summary construction, offset candidate discretization, sequence constraint anchor points, and slice content volume verification during the misaligned scheduling generation stage, the time misaligned scheduling is transformed from a single time shift behavior into a structured scheduling process constrained by capacity, sequence, and tick time. This approach ensures that the release path of each service flow is feasiblely verified during the generation stage, avoiding the random queuing fluctuations caused by traditional direct priority scheduling, and giving the scheduling behavior a clear time distribution structure.

[0044] During the conflict resolution phase, inter-slice staggered rearrangement is performed on conflict micro-time slices to identify and eliminate capacity overrun issues within the micro-time slices during the generation phase, rather than passively adjusting them during the execution phase. By distinguishing between compact and non-compact flows and grouping them for migration, highly concentrated incoming flows are preferentially dispersed, while dispersed flows are prevented from being over-migrated, thus achieving a more reasonable release distribution over the time dimension. This rearrangement method based on flow characteristic grouping and slice-level capacity matching helps reduce queuing concentration within micro-timescales.

[0045] By establishing an execution record unit in each micro-time slice and recording gating hit status, upper limit consumption ratio, and queue evolution status, the scheduling process becomes traceable and evaluable. This recording mechanism not only reflects whether misaligned releases have been completed, but also whether the time slice division is reasonable and whether the offset arrangement matches the actual port cycle time, providing a basis for subsequent adaptive updates.

[0046] Example 5 Please refer to Figure 2 Specifically: S5 includes S51; S51. Perform time alignment processing on the acquired scheduling execution data Se: Using the synchronous burst time interval [Ts, Te] as the unified observation interval, merge the slice boundaries, actual dequeue start and end times, actual dequeue byte counts, and window queue sampling sequence {q(u)} of each micro-time slice sj in the scheduling execution data Se into the same sequence in chronological order. Two types of evaluation calculations are performed based on sequences: The first is the calculation of the queue stability assessment results: the change pattern of the window queue sampling sequence {q(u)} is analyzed within the synchronous burst time interval [Ts, Te], and the criterion is whether the concentrated transition is broken up. That is, it is to identify whether there are still segments where the queue length transition amount Δq(u) is continuously positive and the growth rate is dense in a short period of time. Combined with the degree of dispersion of the waiting time of the queue head at the slice boundary and the degree of fluctuation of the upper limit consumption ratio of the dequeue within the slice, a comprehensive stability conclusion is formed. Secondly, the calculation of execution deviation: using the micro time slice offset table Toff as the planning benchmark, the planned micro time slice sequence marked in the micro time slice offset table Toff for each service flow fi is compared to check whether the actual dequeue position of the service flow fi in the scheduling execution data Se is consistent with the micro time slice sequence, and further the difference between the planned release time point and the actual dequeue start time point is statistically analyzed to form the slice-level deviation of the service flow fi; Then, the deviations of all business flows within the candidate misaligned flow set Gb are summarized to obtain the overall execution deviation E.

[0047] S5 also includes S52; S52, after obtaining the queue stability assessment results and execution deviation, performs the problem attribution—update object selection determination: When the queue stability assessment results show that there are still concentrated transitions, and the execution deviation shows that the plan and execution are highly consistent, it indicates that the micro time slice offset table Toff can be implemented but the misalignment force is insufficient. In this case, the offset budget Ω should be updated first. When the execution deviation is concentrated in slice switching, insufficient slice content, or low gating hit rate, it indicates that the division of the micro time slice set S does not match the actual clock cycle of the port. At this time, the micro time slice set S is updated first and the offset step size constraint of the offset budget Ω is corrected in conjunction with the adjustment. The update method for the offset budget Ω is based on the relationship between the slice-level deviation of each service flow in the execution deviation amount and the position of the concentrated transition residual segment in the queue stability assessment result; For traffic flows in the candidate misaligned flow set Gb that are highly compact and still clustered out of the queue within the same window, expand their allowed offset range or refine their offset step length. For service flows with low compactness but excessive migration leading to a decrease in gating hit rate, converge their allowed offset range; Update method for the set of micro-time slices S: When the execution deviation indicates that certain micro-time slices frequently fail to utilize their upper limit or frequently experience insufficient content, the micro-time slice boundaries are then further subdivided or merged: For segments with insufficient capacity, micro-time slices are subdivided to improve scheduling resolution and reduce intra-slice congestion; For segments with unused upper limits, adjacent micro-time slices are merged to reduce the number of slice switching and mitigate the delay effect caused by locking boundaries, while the corresponding transmittable capacity and minimum scheduling interval constraints are updated synchronously. After the update is completed, the updated set of micro-time slices S and the offset budget Ω are generated and written back to step S2 as input for the next round of synchronous burst processing.

[0048] In this embodiment, by aligning the scheduling execution data by time and unifying the sequence, the scheduling plan and the actual execution results are compared and analyzed under the same time coordinate system, thereby avoiding misjudgment problems caused by the inconsistency of time scales between the planning layer and the execution layer. This processing method makes the observation of queue changes within micro-time scales more accurate, providing a reliable basis for judging whether misaligned scheduling has truly changed the queue's centralized transition state.

[0049] By simultaneously introducing two analytical dimensions—queue stability assessment results and execution deviation—the nature of the problem can be differentiated and judged. On the one hand, by analyzing the queue change pattern, it is possible to identify whether the concentrated transition has been dispersed, and to determine whether the misaligned scheduling is effective from the overall structure. On the other hand, by comparing the planned micro-time slice sequence with the actual dequeue segments one by one, it is clear whether the deviation stems from the inadequacy of the scheduling strategy itself or from the mismatch of the execution phase. This dual-dimensional assessment mechanism avoids simply attributing queue fluctuations to a single factor, and improves the pertinence of subsequent adjustment decisions.

[0050] During the update phase, by associating the problem attribution results with the compactness characteristics of the business flow, the adjustment of the offset budget is no longer uniformly amplified or contracted, but rather differentiated for different types of business flows. For highly compact flows that are still concentrated in the queue, the offset range is appropriately expanded to further disperse the release; for low-compact flows that have been over-migrated, their offset range is converged to reduce unnecessary time drift. This hierarchical update approach based on flow characteristics helps maintain the balance of the time misalignment structure.

[0051] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended technical solutions and their equivalents.

Claims

1. A high-efficiency network traffic optimization and control method based on switches, characterized in that: Includes the following steps: S1. At the switch output port Pout, monitor the arrival time of the service flow fi entering the output queue Qout, and use the micro-time detection window w to count the packet entry behavior, generating a candidate misaligned flow set Gb and a synchronous burst time interval [Ts, Te]. S2. Based on the synchronous burst time interval [Ts, Te], analyze the historical transmission completion time sequence of the output port, construct the set of micro time slices S for the output port to execute transmission, and generate the allowable time offset range for each service flow fi in the candidate misaligned flow set Gb, forming the offset budget Ω. S3. Based on the candidate misaligned flow set Gb, the micro-time slice set S and the offset budget Ω, determine the time offset Δti of each service flow fi and the corresponding micro-time slice sequence, and generate the micro-time slice offset table Toff. S4. Based on the micro-time slice offset table Toff, implement micro-time slice gating scheduling on the output port Pout, so that the synchronously entering service flow fi is dequeued out in a staggered manner according to different micro-time slices, and record the scheduling execution data Se; S5. Evaluate the queue change status based on the scheduling execution data Se, update the micro-time slice set S and offset budget Ω, and feed it back to step S2.

2. The efficient network traffic optimization and control method based on a switch according to claim 1, characterized in that: S1 includes S11; S11. On the enqueue side of the queue Qout corresponding to the output port Pout, establish an arrival event record for each service flow fi, and write the arrival event sequence at the granularity of the message; record the arrival time and message length for each enqueue message, and synchronously write the service flow fi identifier and the target queue identifier to which the message belongs. Window aggregation is performed on all arriving events using a micro-time detection window w: Discretize the time axis into continuous window segments w(u) = [u × w, (u + 1) × w). All arriving events falling within the same window segment w(u) are organized into a two-level flow-event structure to form a window arrival matrix Mu, which includes the enqueued event count Mu(i), the enqueued byte accumulation MuL(i), and the window queue sampling sequence {q(u)}. Specifically, when the arrival time of a message falls within the window segment w(u), the message is assigned to the corresponding window statistical unit; within the same window segment, all messages belonging to the same service flow fi are subjected to aggregation and statistical processing. The number of packets that fall within the window segment w(u) and belong to the service flow fi is accumulated to obtain the enqueued event count Mu(i) of the service flow fi within the window. The message lengths corresponding to the messages are accumulated to obtain the cumulative number of enqueued bytes MuL(i) of the service flow fi within the window. Read the queue length sample value q(u) of queue Qout at the end of the window to form the window queue sampling sequence {q(u)}.

3. The efficient network traffic optimization and control method based on a switch according to claim 2, characterized in that: S1 also includes S12; S12. Based on the enqueue event count Mu(i), the enqueue byte accumulation MuL(i), and the window queue sampling sequence {q(u)}, calculate the synchronization burst determination quantity for each window segment w(u): The specific method is as follows: First, calculate the number of concurrent streams within the window: Nu = |{fi|Mu(i)>0}|; In the formula, fi represents the identifier of the i-th business flow fi, and Mu(i) represents the enqueue event count, specifically the enqueue event count of business flow fi within the u-th micro-time detection window w(u); Then, the queue length transition Δq(u) = q(u) - q(u-1) and the continuous transition discrimination Δq(u) > 0 and Δq(u+1) > 0 are calculated; where q(u) represents the queue length value obtained by sampling the output queue Qout once when u micro-time detection windows w(u) end, and q(u-1) represents the queue length value obtained by sampling the output queue Qout once when u-1 micro-time detection windows w(u-1) end; When the number of concurrent flows Nu in the window reaches the preset concurrent flow judgment threshold, and Δq(u) satisfies the continuous transition judgment, the window segment w(u) is marked as a synchronous burst candidate window, and boundary locking is performed on the continuous candidate windows: the starting point of the earliest candidate window that meets the conditions is taken as Ts, and the ending point of the last candidate window that still meets the conditions is taken as Te, to obtain the synchronous burst time interval [Ts, Te]. And within the synchronous burst time interval [Ts, Te], perform set extraction on the participating streams: streams that satisfy the condition that there are enqueuing events within [Ts, Te] are uniformly collected into a candidate misaligned stream set Gb; For each flow in the candidate misaligned flow set Gb, calculate the arrival compactness parameter JCi, which is used to characterize the degree of concentrated arrival of the flow within the burst window.

4. The efficient network traffic optimization and control method based on a switch according to claim 3, characterized in that: S2 includes S21; S21. Taking the switch output port Pout as the object, extract the continuous dequeue completion records of the output port Pout before [Ts, Te] to form the actual dequeue completion time sequence of the port; calculate the difference between two adjacent dequeue completion times to obtain the adjacent transmission completion interval, and form an interval sequence in chronological order; The interval sequence is filtered for stable segments: obvious abnormal intervals are removed, and the remaining intervals are aggregated into several beat clusters according to their proximity. Each beat cluster represents the sustainable transmission rhythm of a port at that time. Based on the beat cluster results, the time axis is divided into continuous micro-time slices sj, and the boundary of each micro-time slice is aligned with the typical interval of the beat cluster to obtain the set of micro-time slices S that the output port can execute to send. Based on the set of micro-time slices S, two types of execution boundaries are determined for each micro-time slice sj: One is the transmittable capacity, which is estimated by the duration of the time slice, the port transmission rhythm, and the number of message services that can be accommodated within the slice, and is expressed as the cumulative transmittable capacity that can be scheduled within the slice; The second is the minimum scheduling interval, which is estimated by the minimum stable interval within the corresponding beat cluster when the adjacent transmission completion intervals are within the same timeframe.

5. The efficient network traffic optimization and control method based on a switch according to claim 4, characterized in that: S2 also includes S22; S22. Based on the micro-time slice set S, generate an allowable time offset range for each service flow fi in the candidate misaligned flow set Gb, and aggregate all offset ranges to form an offset budget set Ω. The specific processing steps are as follows: First, read the arrival compactness parameter JCi output by S1, and simultaneously read the enqueue behavior summary of the service flow fi in [Ts, Te]; then, classify the flow according to the arrival compactness parameter JCi, and divide the candidate misaligned flow set Gb into several categories: flows with high compactness are characterized by concentrated entry in a short period of time, while flows with low compactness are characterized by more dispersed entry; within the micro time slice boundary, allocate different offset windows for flows of different categories; For highly compact service flows fi, a finer-grained offset range is provided, and the offset is limited to rolling amortization only within a small number of adjacent micro-time slices; For low-compactness service flows fi, the offset range converges to a narrower in-chip fine-tuning; For service flow fi with a sequence-sensitive identifier, the offset budget allows inter-chip delay but not inter-chip advance, and a sequence constraint flag is written in the time offset range; Finally, a consistency check is performed between the time offset range in each service flow fi and the micro-time slice set S: Check whether there is sufficient transmittable capacity within the micro-time slice set covered by the time offset range to carry the dequeue demand of service flow fi within the burst interval, and ensure that the offset step size is not less than the scheduling resolution allowed by the minimum scheduling interval; after the verification is passed, output the offset budget set Ω.

6. The efficient network traffic optimization and control method based on a switch according to claim 5, characterized in that: S3 includes S31 and S32; S31. After obtaining the candidate misaligned flow set Gb, the micro-time slice set S and the offset budget Ω, construct a summary of the load to be released for each service flow fi in the candidate misaligned flow set Gb within the synchronous burst time interval. The summary includes: the cumulative number of bytes of service flow fi entering the queue in [Ts, Te], the discreteness of the message length, and the distribution of the message arrival points on the micro-time slice. Read the offset budget ωi corresponding to the stream to obtain the time range and allowable offset step size boundary of the stream, and generate a set of offset candidate list for each service stream fi: Within the allowable range of the offset budget ωi, several offset values ​​are discretized by step size, and each candidate offset value is mapped to the set of micro-time slices S to obtain a slice sequence draft; When generating the slice sequence draft, establish the order constraint anchor point simultaneously: based on the order of packets in the output queue Qout of the service flow fi, segment the packet sequence according to the arrival order, and require that each packet segment can only be mapped to a set of micro time slices that are no earlier than the previous segment in time, and packets within the segment are only allowed to be dequeued in the same micro time slice in the order of the queue head. S32. Based on the slice sequence draft, perform slice content volume verification for each micro-time slice sj: count the total streaming load demand mapped to micro-time slice sj in all drafts, compare it with the sendable capacity of the micro-time slice, and identify the set of conflicting micro-time slices that exceed the capacity. For conflicting micro-time slices, inter-slice misalignment rearrangement is performed: First, the service flows fi in the conflicting micro-time slice are grouped according to the strength of constraints on transmittable capacity and sequence anchor points, identifying compact flow groups that must be released close to the current window and non-compact flow groups that are allowed to be released more dispersedly; then, within the allowable range of the offset budget ωi, the offset candidates are replaced one by one for the adjustable flows, so that the slice sequence draft is moved out of the conflicting micro-time slice and falls into the adjacent free micro-time slice. During the migration, two constraints are checked simultaneously: First, the draft sequence of the migrated slices still satisfies the monotonicity of the sequential anchor point; Secondly, the remaining capacity of the migration destination micro-time slice is sufficient to handle the load of the migrated service flow fi without disrupting the minimum scheduling interval within the slice. Once the conflict is resolved, a unique time offset Δti and a corresponding final micro-time slice sequence draft are determined for each service flow fi. The micro-time slice sequences and time offsets Δti of all service flows are then summarized to generate the micro-time slice offset table Toff.

7. The efficient network traffic optimization and control method based on a switch according to claim 6, characterized in that: S4 includes S41; S41. After obtaining the micro time slice offset table Toff, establish a micro time slice gating rule table for the output port Pout, and expand the micro time slice sequence of each service flow fi in the time slice offset table Toff into slice-level authorization information: for each micro time slice sj, summarize all service flows that satisfy the micro time slice sj∈micro time slice sequence to form the set of allowed outgoing flows of the micro time slice. At the same time, based on the occupation arrangement of each service flow in the time slice in the micro time slice offset table Toff, the in-slice dequeue limit is generated and used to limit the cumulative number of bytes that service flow fi is allowed to send or the number of messages allowed to be dequeued in the micro time slice, so that the in-slice dequeue will not crowd out the range of the sendable capacity. During scheduling execution, the scheduler enters the in-slice gating state with the start boundary of the micro-time slice sj as the trigger point: it only grants dequeue permission to service flows that belong to the set of allowed dequeue flows, and during the dequeue process, it deducts the in-slice transmission of each service flow according to the in-slice dequeue limit. When a traffic flow reaches its intra-slice limit or its head packet has been sent, the dequeue permission for the traffic flow in the current micro-time slice is revoked. Gating is maintained for service flows that do not belong to the allowed dequeue flow set.

8. The efficient network traffic optimization and control method based on a switch according to claim 7, characterized in that: S4 includes S42; S42. Establish an execution record unit for each micro-time slice sj, record the actual dequeueing behavior that occurs within the micro-time slice, and form scheduling execution data Se; The scheduling execution data Se contains three types of information: One is the slice-level gating hit record, which records whether each service flow fi in the allowed dequeue flow set is actually dequeued within a micro time slice, the start and end times of dequeueing, and the actual number of bytes dequeued; The second is the intra-slice upper limit consumption record, which records the proportion of the intra-slice outbound upper limit consumed for each business flow, and is used to determine whether the micro-time slice allocation is too tight or too loose. Thirdly, the queue state is recorded as it evolves with the slice, including the window queue sampling sequence {q(u)} and the queue length transition Δq(u) at the micro-time slice boundary.

9. The efficient network traffic optimization and control method based on a switch according to claim 8, characterized in that: S5 includes S51; S51. Perform time alignment processing on the acquired scheduling execution data Se: Using the synchronous burst time interval [Ts, Te] as the unified observation interval, merge the slice boundaries, actual dequeue start and end times, actual dequeue byte counts, and window queue sampling sequence {q(u)} of each micro-time slice sj in the scheduling execution data Se into the same sequence in chronological order. Two types of evaluation calculations are performed based on sequences: The first is the calculation of the queue stability assessment results: the change pattern of the window queue sampling sequence {q(u)} is analyzed within the synchronous burst time interval [Ts, Te], and the criterion is whether the concentrated transition is broken up. That is, it is to identify whether there are still segments where the queue length transition amount Δq(u) is continuously positive and the growth rate is dense in a short period of time. Combined with the degree of dispersion of the waiting time of the queue head at the slice boundary and the degree of fluctuation of the upper limit consumption ratio of the dequeue within the slice, a comprehensive stability conclusion is formed. Secondly, the calculation of execution deviation: using the micro time slice offset table Toff as the planning benchmark, the planned micro time slice sequence marked in the micro time slice offset table Toff for each service flow fi is compared to check whether the actual dequeue position of the service flow fi in the scheduling execution data Se is consistent with the micro time slice sequence, and further the difference between the planned release time point and the actual dequeue start time point is statistically analyzed to form the slice-level deviation of the service flow fi; Then, the deviations of all business flows within the candidate misaligned flow set Gb are summarized to obtain the overall execution deviation E.

10. The efficient network traffic optimization and control method based on a switch according to claim 9, characterized in that: S5 also includes S52; S52. After obtaining the queue stability assessment results and execution deviation, determine the problem attribution—update object selection: When the queue stability assessment results show that there are still concentrated transitions, and the execution deviation shows that the plan and execution are highly consistent, it indicates that the micro time slice offset table Toff can be implemented but the misalignment force is insufficient. In this case, the offset budget Ω should be updated first. When the execution deviation is concentrated in slice switching, insufficient slice content, or low gating hit rate, it indicates that the division of the micro time slice set S does not match the actual clock cycle of the port. At this time, the micro time slice set S is updated first and the offset step size constraint of the offset budget Ω is corrected in conjunction with the adjustment. The update method for the offset budget Ω is based on the relationship between the slice-level deviation of each service flow in the execution deviation amount and the position of the concentrated transition residual segment in the queue stability assessment result; For traffic flows in the candidate misaligned flow set Gb that are highly compact and still clustered out of the queue within the same window, expand their allowed offset range or refine their offset step length. For service flows with low compactness but excessive migration leading to a decrease in gating hit rate, converge their allowed offset range; Update method for the set of micro-time slices S: When the execution deviation indicates that certain micro-time slices frequently fail to utilize their upper limit or frequently experience insufficient content, the micro-time slice boundaries are then further subdivided or merged: For segments with insufficient capacity, micro-time slices are subdivided to improve scheduling resolution and reduce intra-slice congestion; For segments with unused upper limits, adjacent micro-time slices are merged to reduce the number of slice switching and mitigate the delay effect caused by locking boundaries, while the corresponding transmittable capacity and minimum scheduling interval constraints are updated synchronously. After the update is completed, the updated set of micro-time slices S and the offset budget Ω are generated and written back to step S2 as input for the next round of synchronous burst processing.

Images