SD-WAN-based AIoT device data priority scheduling method

Abstract

The invention provides an AIoT device data priority scheduling method based on an SD-WAN, and relates to the technical field of data processing, and the method comprises the steps: analyzing device data at an AIoT gateway or an SD-WAN edge device into data atoms containing a security domain identifier, a service type identifier, a data size, a deadline, a basic value, an attenuation coefficient and a reliability requirement; an instant value function is generated; the edge device collects network measurement data for the superposition tunnel path according to a preset measurement period, calculates a path time delay quantile and a path packet loss rate quantile, obtains a deliverable coefficient and an expected delivery value of a data atom on each path, and further calculates a unit byte scheduling index; selecting the maximum scheduling index to send under the constraint of credit balance; when the deliverable coefficient is lower than the reliability requirement, double-path copying or system code action is executed, and a receiving end carries out duplicate removal and expired discarding; and the edge device reports statistical information, and the controller updates the next window allocation to realize closed-loop scheduling and resource isolation.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of data processing. BACKGROUND

[0002] With the development of industrial internet, smart park, smart building and smart city applications, a large number of AIoT devices continuously generate control instructions, alarm information, telemetry data, video streams and batch logs and other types of data, and different types of data have significant differences in latency, reliability and bandwidth requirements. Control and alarm type data usually have more stringent timeliness and reliability requirements, while telemetry and batch type data can tolerate latency to some extent but are more sensitive to transmission efficiency and cost. In the existing enterprise wide area network, site-to-cloud or site interconnection usually has multiple bearing links such as Internet, dedicated line or cellular link, and the latency, packet loss and available bandwidth of the link fluctuate significantly over time, making it difficult to rely on static routing, fixed priority queue or simple policy routing to long-term and stable guarantee the experience of key services.

[0003] SD-WAN can provide multi-path routing and policy arrangement capabilities through superposition of tunnels and centralized control, but still faces the following problems in the AIoT scenario. First, existing scheduling is usually based on sessions / flows, which is difficult to distinguish the timeliness and value differences of different "messages / samples" within the same service type, and often results in low-value data occupying bandwidth, causing key alarm or control data to be queued and delayed. Second, path quality assessment often uses instantaneous measurement or average value, which is difficult to provide robust routing basis in jitter and burst packet loss scenarios, resulting in seemingly available paths failing to meet the deadline and reliability requirements at critical moments. SUMMARY

[0004] Technical problems to be solved

[0005] In view of the defects of the prior art, the application provides an AIoT device data priority scheduling method based on SD-WAN, which solves the problems of the prior art.

[0006] Technical scheme

[0007] To achieve the above purpose, the application is implemented by the following technical scheme; an AIoT device data priority scheduling method based on SD-WAN, the method comprising the following steps:

[0008] S1, at the AIoT gateway or SD-WAN edge device, data from the AIoT device is parsed into data atoms, each data atom comprising; security domain identifier , service type identifier, data size , deadline , basic value attenuation coefficient Reliability requirements ;

[0009] S2. Generate an instantaneous value function for the data atoms. , among which; when hour ;when hour ;

[0010] S3. At the SD-WAN edge device, each superimposed tunnel path is measured according to a preset measurement cycle. Collect network measurement data and calculate the path delay quantiles. Path packet loss rate quantile quantile parameter and The quantile parameter is set in advance; and the data atoms in the path are calculated. On the deliverability coefficient and expected delivery value ;

[0011] S4, for the data atoms in the path Calculate the unit byte scheduling index in For pre-defined non-negative weight parameters, For path The pre-defined non-negative cost parameter of the binding;

[0012] S5, controlled by the SD-WAN controller in a time window A credit token is issued to the SD-WAN edge device, the credit token including the total site credit. and by security domain identifier With service level Sub-credit division and satisfy ;

[0013] S6. The SD-WAN edge device atomically maps the data to be sent to the corresponding service level. And deduct credits equal to the number of bytes actually sent by the data atom from the corresponding sub-credit counter;

[0014] S7. When the balance of the corresponding sub-credit counter is not less than the number of bytes to be sent, the SD-WAN edge device selects from its queued data atom set... Take the maximum value of the "data atom-path" combination and send it on the selected path;

[0015] S8. When the selected "data atom-path" combination satisfies When the SD-WAN edge device performs a redundant transmission action on the data atom, the redundant transmission action is one of the following two: - dual-path replication action; sending the same data atom on two different paths respectively; - system code action; after sending the data atom for the first time, sending the same data atom again after a preset retransmission interval;

[0016] S9. The SD-WAN edge device reports the on-time delivery count, the expired and discarded count, and the count of each path to the SD-WAN controller. and The SD-WAN controller updates and issues the credit token for the next time window based on the reported results.

[0017] Preferably, the basic value described in S1 Deadline attenuation coefficient With reliability requirements Determined jointly by business template parameters and semantic correction;

[0018] Pre-set template parameters for the business template corresponding to the business type identifier. , , , And obtain abnormal confidence levels. Severity normalization value Equipment health ,in , , ;

[0019] Define gate function ,in For the pre-set positive parameters, For pre-set threshold parameters;

[0020] And it is obtained from the following deterministic relation;

[0021]

[0022]

[0023]

[0024]

[0025] in For pre-defined non-negative parameters, To pre-set thresholds, This is an indicator function.

[0026] Preferably, S3 and S8 also include a closed-loop correction of the reliability requirements based on network default risk;

[0027] When satisfied or At that time, the data atoms are marked as redundant candidate data atoms, and the reliability requirement is modified to... ,in A pre-set packet loss limit parameter bound to the data atom. The non-negative correction amount is preset; in S8, the redundant candidate data atoms are preferentially sent using the redundant transmission action.

[0028] Preferably, the credit token described in S5 is issued using at least two time windows, and the first time window is satisfied. Smaller than the second time window ;

[0029] The SD-WAN controller in the first time window Sub-credits for the first service level set are issued in the second time window. Sub-credits are issued for the second service level set, wherein the first service level set and the second service level set are pre-defined and different service level sets.

[0030] Preferably, the total site credit mentioned in S5 Based on robust availability prediction, the SD-WAN controller determines the availability of sites within a pre-set observation window. Each path Calculate the available bandwidth quantile With instantaneous available bandwidth quantile parameter For pre-set and meet ; and according to Calculate available capabilities, among which Enable parameters for the preset path; and press Determine the site's total credit.

[0031] Preferably, the selection of redundant transmission actions in S8 satisfies the following deterministic rule; when the data atoms satisfy... And there are two different paths Make the correlation coefficient When selecting the dual-path copy action;

[0032] When the above conditions are not met and the conditions are met When selecting a system code action; among them For a pre-set data size threshold, For a pre-set correlation threshold, To determine the path within a pre-defined observation window With path The correlation coefficient is calculated from the time delay increment sequence.

[0033] Preferably, the system components corresponding to the method include an SD-WAN controller and an SD-WAN edge device; the SD-WAN edge device is configured to execute; a data atom generation module is used to parse AIoT device data into the data atoms described in claim 1S1; and a value calculation module is used to generate the instantaneous value function described in claim 1S2 and calculate the value described in claim 1S3. and The scheduling index calculation module is used to calculate the index as described in claim 1S4. The credit execution module is used to deduct the sub-credit issued in claim 1S5 and execute the transmission when the credit balance constraint is met; the redundancy execution module is used to execute the redundant transmission action when the condition of claim 1S8 is met; the reporting module is used to report the statistical information of claim 1S9 to the SD-WAN controller.

[0034] The SD-WAN controller is configured to execute a measurement fusion module for generating the measurement fusion module as described in claim 1S3. and The credit generation module is used to generate and issue the credit token as described in claim 1S5; the strategy update module is used to update and issue the credit token for the next time window based on the statistical information.

[0035] Preferably, the SD-WAN edge device further includes a credit lending module, which satisfies the following constraint: when the sub-credit balance of the first service level is insufficient to cover the number of bytes to be sent for the data atoms to be sent, the number of bytes is borrowed from the sub-credit balance of the second service level. satisfy ,in For pre-set and meet The loan-to-deposit ratio parameter, It is designated as the second service level; and will be compensated with a coefficient in the next time window. The number of bytes borrowed is deducted, where For pre-set and meet The replenishment coefficient.

[0036] Preferably, when the SD-WAN edge device performs sending and redundant sending operations, it adds an overlay tunnel extension field to each sent packet. The extension field includes atom_id and deadline_ts. The system also includes a receiver deduplication module, which deduplicates duplicate received packets based on atom_id and discards the corresponding packet when the current time is greater than deadline_ts and adds it to the expired discard count.

[0037] Preferably, the credit generation module of the SD-WAN controller is configured to generate credits in each time window. The sub-credits are generated and distributed according to the following rules. Obtain the identifier for each security domain from each SD-WAN edge device. With service level Reported queued bytes and value density statistics The value density statistic mentioned above For the SD-WAN edge device in the time window Internal scheduling metrics for the unit byte corresponding to the queued data atoms The pre-defined quantiles are calculated; when they meet the following conditions... Under the constraints, firstly for each Assign a pre-defined minimum sub-credit If the remaining credit exists after deducting the minimum sub-credit, the remaining credit is distributed according to... Distribute them in descending order to the corresponding... And for each The allocated sub-credit satisfies Thus determining the sub-credit And distribute it to the corresponding SD-WAN edge device.

[0038] Beneficial effects

[0039] This invention provides a data priority scheduling method for AIoT devices based on SD-WAN. It has the following beneficial effects;

[0040] 1. This invention parses AIoT service data at the AIoT gateway or SD-WAN edge device into data atoms containing security domain identifiers, service type identifiers, data size, deadlines, basic value, attenuation coefficients, and reliability requirements. Based on the real-time value function, it characterizes the attenuation of data atom value over time. Combining the latency quantiles and packet loss rate quantiles of the superimposed tunnel path, it calculates the deliverability coefficient and expected delivery value, and then forms a unit byte scheduling index to select the "data atom-path" combination for transmission from the queued data atom set. This improves the scheduling decision from the traditional static priority or fixed QoS queue to fine-grained adaptive scheduling based on deadlines and value density. It can more effectively ensure the priority transmission of high-timeliness and high-value data and reduce the occupation of link resources by low-value overdue data in scenarios with multiple concurrent services and fluctuating link status.

[0041] 2. This invention uses an SD-WAN controller to issue credit tokens containing total site credit and sub-credits categorized by security domain and service level to edge devices within a time window. The edge devices deduct credits based on the actual number of bytes sent, thus achieving controllable bandwidth allocation for security domain isolation and service level differentiation. Simultaneously, when the deliverability coefficient is lower than the reliability requirement, dual-path replication or system code actions are triggered, and the data atom identifier and expiration timestamp are carried in the overlay tunnel extension field. At the receiving end, deduplication and expired discarding are performed. The edge devices report the on-time delivery count, expired discard count, and path quantile measurement results to the controller to update the credit token for the next window, forming a closed-loop mechanism of "measurement—scheduling—redundancy—deduplication—reporting—update". This improves the effective delivery capability under unfavorable network conditions without changing the underlying bearer network and suppresses duplicate delivery and resource waste caused by redundancy. Attached Figure Description

[0042] Figure 1 This is a system composition diagram of the present invention;

[0043] Figure 2 This is a flowchart of the method of the present invention;

[0044] Figure 3 This is a diagram of the edge device scheduling structure of the present invention;

[0045] Figure 4 This is a structural diagram of the reporting and credit token issuance of the present invention;

[0046] Figure 5 This is a screenshot of the system in operation according to the present invention;

[0047] Figure 6 This is a screenshot showing the operation of the edge device and credit status of the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Specific Implementation Example 1:

[0050] like Figures 1 to 6 As shown, an AIoT device data priority scheduling method based on SD-WAN is implemented collaboratively by the control plane and the data plane. The control plane, executed by the SD-WAN controller, generates and distributes trust tokens, receives edge reports, and updates policies. The data plane, executed by the AIoT gateway or SD-WAN edge device, performs data atomization, value calculation, queue selection, path selection, and redundant transmission. A management channel is established between the SD-WAN edge device and the SD-WAN controller to maintain configuration consistency. This management channel is used to transmit trust tokens, measurement parameters, template parameters, and statistical reports. The method performs "data atom-level" scheduling decisions on the AIoT data entering each site, shifting the scheduling target from the traditional "stream / session" approach. The data atoms are refined to the "message / frame / sample level," enabling fine-grained sorting based on deadlines and value decay within the same service type. To ensure consistency in deadline calculations, SD-WAN edge devices and SD-WAN controllers are configured to use a unified time base, which is a reference clock obtained by calibrating the site's local clock via NTP or PTP. Timestamps under the same time base are used as input when calculating formulas involving time parameters. To ensure consistency of the "security domain identifier" in end-to-end forwarding, SD-WAN edge devices carry the security domain identifier or its mapping identifier when overlaying tunnel encapsulation and perform consistency verification at the receiving end. This ensures that the sub-credits of the credit token are isolated and effective according to the security domain and avoids resource encroachment between different security domains.

[0051] S1. At the AIoT gateway or SD-WAN edge device, the data from the AIoT device is parsed into data atoms, each data atom including: security domain identifier. Business type identifier, data size Deadline Basic Value attenuation coefficient Reliability requirements "Parsing into data atoms" comprises two parts: protocol parsing and business semantic extraction. Protocol parsing extracts message boundaries from industrial protocols (including but not limited to MQTT, CoAP, OPCUA, Modbus, IEC104, RTSP, HTTP / HTTPS) or custom binary protocols and generates the smallest independently schedulable sending unit. Business semantic extraction determines the business type identifier based on the business topic, register / point table, URL path, message type field, or video frame type field. Data atoms are stored in the edge device's memory queue in the form of structured records, and each data atom is associated with a unique identifier, atom_id, for deduplication, statistics, and retransmission control. The atom_id is obtained by concatenating the site identifier, device identifier, generation time timestamp, and monotonically increasing sequence number. Security domain identifier. The domain label is a discrete value used to distinguish different tenants, different service segments, or different compliance domains. The service type is identified by a discrete value category label used to distinguish control commands, alarms, video, telemetry, and batch services. (Data size) This refers to the number of effective payload bytes of the data atom before overlay tunneling or the number of bytes after including encapsulation overhead according to preset rules. An index identifier for data atoms, used to distinguish different data atoms; deadline The non-negative time length represents the maximum end-to-end transmission delay budget or the maximum acceptable combined queuing and transmission delay budget allowed for this data atom from the moment it enters the scheduling queue; fundamental value. It is a non-negative scalar, used to characterize the contribution of this data atom to business objectives under the condition of on-time delivery; decay coefficient. It is a non-negative scalar used to characterize the rate at which value decreases after the deadline; reliability requirements. For interval The threshold within the range is used to characterize the minimum on-time delivery success probability threshold that the data atom must meet; to ensure that the parameters are calculable and do not produce inconsistencies, Business type identifier , , , All are determined when the data atoms are generated and remain unchanged as the data atoms move through the queue, unless a subsequent "closed-loop correction" step is performed. The updated value is then recorded in the data atom structure.

[0052] S2. Generate instantaneous value functions for data atoms. , among which; when hour ;when hour Instantaneous value function Used to quantify the "residual business value" of data atoms at the scheduling time. This is the time elapsed since the data atom entered the scheduling system, or the time difference between its generation and the current scheduling decision. Calculated using the clock difference with a consistent time base; when Time setting This indicates that the value remains at the base value during the deadline, and this setting ensures that the contribution of the value to the scheduling metric during the deadline does not change due to variations in waiting time; when Time setting This indicates that the value decays exponentially after the deadline, where... is the base of the natural constant. The length of the overdue period. Let be the decay rate parameter, and The larger the value, the faster it decreases after the expiration date; to avoid underflow or excessively small values ​​causing instability in the sorting, SD-WAN edge devices... Set lower bound truncation When the calculation is obtained When Set as ,in These are pre-defined non-negative constants used to ensure the stability of scheduling metrics in numerical calculations; and to ensure the consistency of business semantics at deadlines. The starting point is defined as the enqueue timestamp of the data atom entering the edge scheduling queue. Then it is used for calculation The time variable can be written as ,in The timestamp of the current scheduling decision, and , Use the same time unit.

[0053] S3. At the SD-WAN edge device, each overlay tunnel path is measured according to a preset measurement cycle. Collect network measurement data and calculate the path delay quantiles. Path packet loss rate quantile quantile parameter and The quantile parameters are set in advance; and the data atoms in the path are calculated. On the deliverability coefficient and expected delivery value Overlay tunnel path This is one of several bearer paths available for SD-WAN edge devices, which may include different physical links or different logical tunnels. The "preset measurement period" is used as a path index to distinguish different candidate paths; it is the measurement interval parameter. SD-WAN edge devices in each During the period, measurement probes are sent to each path or telemetry fields within the service packet are multiplexed to obtain one-way or round-trip delay samples and packet loss event samples, and the samples are written into the path measurement sliding window; path delay quantiles. This indicates the first time delay sample of the path within the preset observation window. quantile values, among which The pre-defined time delay quantile parameter is used to characterize the time delay at a confidence level. Conservative delay estimation; path packet loss rate quantiles This indicates the first instance of the packet loss rate statistics for this path within the preset observation window. quantile values, among which The pre-defined packet loss rate quantile parameter is used to characterize the packet loss rate at a confidence level. The conservative packet loss rate estimate is used; quantile calculation employs a specific quantile digest algorithm, either the Greenwald-Khanna quantile digest algorithm or the t-digest digest algorithm, and this algorithm is consistently chosen during system deployment with its parameters written into the configuration to ensure consistency across different devices; the deliverability coefficient... For interval The coefficients within the range are used to characterize the path selection. The lower bound of the probability that the data atom meets the deadline and successfully reaches it can be approximated, and its calculation logic can be explicitly expressed as a piecewise function; when Time to take This indicates that, under the condition that the conservative delay meets the deadline, the success probability is affected by the packet loss rate. Time to take This indicates that the path does not meet the on-time delivery requirement for the data atom when the conservative delay estimate exceeds the deadline; expected delivery value. Used to characterize the selected path And the expected value during transmission is estimated based on a conservative time delay, where , in the formula The instantaneous value function defined for S2, The current waiting time. For the conservative time delay estimate of this path, the two are added together to obtain the value input corresponding to the expected arrival time. The deliverability factor is used to discount the value to reflect delivery uncertainty caused by packet loss; to prevent path measurement from being severely affected by transient anomalies, SD-WAN edge devices... and Stable estimates are obtained by exponential smoothing. and and in calculation and When using , The exponential smoothing coefficient The parameters are preset and fixed in the configuration.

[0054] S4, for data atoms in the path Calculate the unit byte scheduling index in For pre-defined non-negative weight parameters, For path Bound pre-set non-negative cost parameters; scheduling metrics per byte This is used to unify "value, reliability, and cost" onto a single comparable metric, thereby enabling sortable scheduling across business operations and paths. middle, Expected value density per byte The expected delivery value obtained from S3. The size of the data obtained by S1 Pre-defined non-negative weight parameters are used to adjust the influence of cost items in scheduling. For path The pre-defined non-negative cost parameter is used to characterize the resource overhead of using the path; It is obtained by weighting path cost, encryption overhead, cross-domain compliance overhead, and energy consumption overhead with fixed weights, and the weight parameters of each overhead component are fixed as constants in the system configuration, so that... All data atoms must be kept consistent under the same configuration; this is to avoid differences in units of measurement. Parameter tuning is difficult for SD-WAN edge devices. Normalization is performed on all paths. Perform linear scaling to an interval The scaling factor is issued by the controller and displayed in the window. The internal structure remains unchanged; to ensure repeatable scheduling decisions, SD-WAN edge devices process the same data atom on different paths at the same scheduling moment. After parallel computation, the maximum value is selected, and when there are multiple combinations of "data atom-path"... When identical values ​​result in parallel results, a defined parallel resolution rule is applied, which is based on the cutoff date of the data atoms. If data is still tied, prioritize data from smallest to largest; if still tied, prioritize data from earliest to latest timestamps generated from atomic data; if still tied, prioritize data from path index. The fixed order takes precedence.

[0055] S5, controlled by the SD-WAN controller in a time window Issuing credit tokens to SD-WAN edge devices, the credit tokens include the total site credits. and by security domain identifier With service level Sub-credit division and satisfy Time window Generate credit tokens for the controller at discrete-time granularity. It can be represented by duration or by window number. For the site In the window The total number of bytes allowed to be sent within the specified range. Site indexes are used to distinguish different edge sites; sub-credits In the window Internal allocation to the site security domain With service level The available byte quota, of which The service level index is used to distinguish different scheduling levels, and the service level set is pre-set to a finite set during system initialization. ;constraint Used to ensure that the sum of sub-credits does not exceed the total credit of the site, thereby controlling the total amount of data sent by the edge device within the window; the credit token issuance content includes at least: window identifier, site identifier. Total site credit , sub-credit table The system includes sub-credit effective start and end timestamps, as well as version numbers for parallel resolution and lending parameters, enabling edge devices to atomically update credit counters during window switching and preventing erroneous deductions across windows. To ensure security domain isolation, the controller manages each security domain. Maintain independent minimum coverage and maximum usage limit, and in the calculation This constraint is applied to prevent a single security domain from crowding out all credits within a site; to ensure the interpretability of service levels, service levels... The mapping between the business type identifier and the service type identifier is fixed in the system configuration. The mapping rule is to map the service type identifier to the service level base value and to monotonically increase the service level by the anomaly confidence and severity, but the increase limit does not exceed the preset maximum service level.

[0056] S6 and SD-WAN edge devices atomically map the data to be sent to the corresponding service level. And deduct credits equal to the number of bytes actually sent from the corresponding sub-credit counter; "Map to the corresponding service level". "Use a deterministic mapping function" ,in For business type identification, For abnormal confidence levels, This is the severity normalization value. For device health, the mapping function consists of a configuration table and gating elevation rules. The configuration table specifies the basic service level corresponding to each service type identifier. Gating elevation rules are based on The comparison result with the preset threshold will Upgrade to a higher level and not exceeding The sub-credit counter is a counting variable maintained in memory by the edge device. Its initial value is in the window Initially set to be issued by the controller. The amount is deducted each time it is sent. ,in This represents the number of bytes actually sent by the data atom in the data plane, and The billing caliber is determined by "application payload bytes + overlay tunnel encapsulation bytes", and the billing caliber is fixed during system initialization. To ensure deduction consistency, if redundant sending actions are performed on the same data atomically, resulting in multiple transmissions, each transmission will independently deduct the sub-credit of the service level to which the corresponding path belongs, or deduct from the same sub-credit according to the configuration. This deduction rule is fixed as a single rule in the system configuration. To prevent credit from being consumed repeatedly due to race conditions, the deduction operation is implemented as an atomic operation within the edge device, and mutex locks or lock-free atomic instructions are used between concurrent scheduling threads to ensure consistency.

[0057] S7. When the balance of the corresponding sub-credit counter is not less than the number of bytes to be sent, the SD-WAN edge device selects the appropriate data atom from its queued data atom set. The data atom-path combination with the maximum value is selected and sent on the chosen path; the queued data atom set is determined by security domain. With service level The edge device is configured with a multi-queue structure, and each scheduling cycle... Trigger a dequeue selection. The scheduling period is pre-defined and can differ from the measurement period; in each dequeue selection, the edge device computes each candidate data atom on each available path within a candidate set that satisfies the credit balance constraint. The maximum value combination is selected, and to reduce computational complexity, the candidate set is limited to the head of each queue and the first few queues within a preset window. Data atoms, of which A pre-set positive integer; "Send" includes selecting a path. The corresponding overlay tunnel, data atoms are encapsulated and written with necessary extended fields, link layer or tunnel layer encryption is performed, and the encapsulated message is submitted to the corresponding path's sending queue; to ensure the deadline constraint is valid, the edge device performs an expiration check before sending, when... The data atom is discarded and added to the expired discard count, where To pre-define non-negative grace parameters to cover clock jitter and encapsulation overhead; to ensure fairness among different security domains, edge devices prioritize security domain queues with higher remaining credit ratios when queuing for cross-domain candidates. The remaining credit ratio is defined as... Furthermore, a deterministic comparison of this proportion is added to the parallel resolution rules to avoid long-term bias.

[0058] S8. When the selected "data atom-path" combination satisfies At this time, the SD-WAN edge device performs a redundant transmission action on the data atom. The redundant transmission action is one of the following two: dual-path replication; transmitting the same data atom on two different paths; system code action; after the data atom is transmitted for the first time, the same data atom is transmitted again after a pre-set retransmission interval; condition middle, The deliverability factor calculated for S3. The reliability requirement threshold given by S1 is compared with that of S1 to determine whether the "lower bound of the probability of on-time delivery success under the current path's conservative measurement conditions" meets the business reliability threshold. When redundant transmission is triggered, the edge device first determines the type of redundant action. The implementation of the dual-path replication action is to select two different paths. And send the same data atoms respectively, where The primary path selected for S7 To make in the set of remaining paths The backup path with the maximum value is selected, and the security domain and compliance policy of the backup path must allow the service to be transmitted on that path; the system code action is implemented by immediately sending it once on the main path and setting a retransmission timer, with the retransmission interval... If the data atom is due and no delivery confirmation is received, or if telemetry indicates that the data atom has arrived on time, the same data atom is resent. To pre-set non-negative time parameters and The same time unit is used; to ensure that the determination of "whether it has arrived on time" does not rely on uncertain mechanisms, delivery confirmation adopts one of the deterministic acquisition methods: one is the application layer confirmation message (MQTTPUBACK, CoAPACK, or custom ACK frame) and the edge device maintains the mapping from atom_id to confirmation status; the other is the reception receipt returned by the receiving end's deduplication module and forwarded to the sending end by the controller or edge, and the confirmation channel is bound to the management channel or business channel and fixed during system initialization; to ensure that redundancy does not cause credit loss control, both transmissions of dual-path replication must meet the corresponding sub-credit balance constraints and deduct credit respectively. Before the retransmission of the system code is triggered, the sub-credit balance is checked again and the discard or degradation rule is executed if the balance is insufficient. The degradation rule is a deterministic strategy of performing a reduction in service level mapping or delaying the transmission to the next window for the data atomic, and the degradation strategy is fixed as a single strategy in the configuration; to ensure that duplicate arrivals do not affect business correctness, the receiving end deduplicates based on atom_id and delivers to the upper layer only once, while the duplicate reception is included in the redundancy overhead statistics for the controller to optimize credit allocation.

[0059] S9 and SD-WAN edge devices report on-time delivery counts, expired and discarded counts, and data for each path to the SD-WAN controller. and The SD-WAN controller updates and issues the credit token for the next time window based on the reported results. On-time delivery is counted as delivery within a preset statistical window if the arrival confirmation timestamp or the sending enqueue timestamp is not greater than the deadline. The count of data atoms or bytes is a statistical measure, while the count of expired data is a statistical measure of the number of data atoms or bytes discarded at the sending or receiving end due to exceeding the deadline. Both are based on security domains. With service level Maintain them separately to support fine-grained closed loops; each path's and For the quantile measurement results defined in S3, the edge device also carries the quantile parameters in the report. , The observation window length and sample size are used to enable the controller to perform consistency verification of measurement reliability; the logic for the controller to update the credit token based on the reported results includes; for each station Calculate the total site credit for the next window. Adjustments at the sub-credit level The adjustment criteria include at least the following: The system uses statistics on on-time delivery rate, delinquency rate, queue backlog, and value density to migrate credit from low-value, high-delinquency combinations to high-value, high-on-time delivery potential combinations. To avoid system instability caused by frequent fluctuations, the controller... Adjustment settings for the upper limit of the rate of change And adopt deterministic amplitude limiting rules ,in For the calculated adjustment amount, For the amplitude limiting function, and Pre-defined sub-credit lower and upper limits ensure the stability and predictability of closed-loop updates.

[0060] Basic value in S1 Deadline attenuation coefficient With reliability requirements The template parameters are determined jointly by the business template parameters and semantic correction; template parameters are pre-set for the business template corresponding to the business type identifier. , , , And obtain abnormal confidence levels. Severity normalization value Equipment health ,in , , Define the gate function ,in For the pre-set positive parameters, The threshold parameter is preset and obtained from the following deterministic relationship;

[0061]

[0062]

[0063]

[0064]

[0065] in For pre-defined non-negative parameters, To pre-set thresholds, This is an indicator function.

[0066] Business template parameters are used to provide baseline timeliness and value constraints for different business types. As the benchmark basic value, As the baseline cutoff date, As the reference attenuation coefficient, As a baseline reliability requirement, the aforementioned baseline parameters are stored in a template table on the edge device or controller using the service type identifier as the key and loaded during system initialization; anomaly confidence level. For interval The value within represents the confidence level of the anomaly detection model in that "the event corresponding to the current data atom is anomaly," and is the severity normalized value. For interval The value within represents a normalized representation of the event severity, indicating device health. For interval The values ​​within represent the health level of the equipment's operating status and A smaller value indicates lower health; gating function Let be a deterministic monotonic function, where For input variables, For the natural logarithm function, The base of the natural constant, The pre-set positive parameters are used to control the steepness of the gating curve. To pre-set threshold parameters to control the trigger position of the gating, thereby... , and The impact on value and deadline exhibits explainable monotonicity; parameters Non-negative weights are used to adjust the impact of outlier confidence, severity, and health on the base value. Gain amplitude, Non-negative weights are used to adjust the anomaly confidence and severity relative to the cutoff date. The compression range, Non-negative weights are used to adjust the attenuation coefficient based on anomaly confidence and severity. The gain amplitude; The more obvious the anomaly, the higher the severity, or the lower the health level, the higher the value of on-time delivery; Used to tighten deadlines in abnormal and high-severity scenarios to enable faster data scheduling; This is used to accelerate the decay of expired value in abnormal and high-severity scenarios to increase the tendency for expired data to be discarded. This is used to raise the reliability threshold when anomalies or severity exceed a certain threshold. , For a pre-set non-negative correction amount, , To pre-set thresholds, This is an indicator function that takes a value of 1 when the condition within the parentheses is true and 0 otherwise, thus making the triggering condition for improving reliability requirements clear and reproducible; to ensure the validity of the parameter range, the edge device calculates... , , , Then, it is subject to execution scope constraints, which include... , , , And values ​​that violate range constraints are truncated to the legal range according to preset rules.

[0067] The relationship between S3 and S8 also includes a closed-loop correction of reliability requirements based on network default risk; when the requirements are met... or At that time, data atoms are marked as redundant candidate data atoms, and the reliability requirements are modified to... ,in A pre-defined upper limit parameter for packet loss that is bound to data atoms. The pre-defined non-negative correction amount; in S8, redundant candidate data atoms are given priority in performing redundant transmission actions. and These are the path delay quantiles and path packet loss rate quantiles defined by S3, respectively. For data atoms The bound packet loss cap parameter is used to characterize the maximum acceptable packet loss rate threshold for this data atom; when a path exists... satisfy This indicates that delivery cannot be made within the deadline based on a conservative time delay estimate, when a path exists. satisfy This indicates that the conservative packet loss estimate cannot meet the reliability requirements of this data atom. Therefore, this data atom is marked as a redundant candidate data atom to trigger stronger delivery guarantees; the revised reliability requirements... middle, To meet the original reliability requirements, To pre-set a non-negative correction amount, used to raise the reliability threshold and trigger redundant actions when network default risk increases. To ensure that the corrected reliability requirement does not exceed 1; "priority execution" is achieved by adding a priority flag for redundant candidate rules in the parallel resolution rules of S7, that is, when two candidate combinations... The same or the difference is less than the preset threshold Redundant candidate data atoms are selected first, among which To pre-set a non-negative threshold for stable sorting; and to avoid resource consumption caused by long-term accumulation of redundant candidates, edge devices atomically set a maximum retention time for redundant candidate data. ,when Discard directly and include in the expired discard count, where To pre-set non-negative time parameters and satisfy .

[0068] In S5, credit tokens are issued using at least two time windows, and the first time window requirement must be met. Smaller than the second time window ; SD-WAN controller in the first time window Sub-credits for the first service level set are issued in the second time window. Sub-credits are issued for the second service level set, where the first and second service level sets are pre-defined and distinct service level sets. First time window. With the second time window For update cycles of different scales, and satisfying This allows for more frequent credit updates for high real-time services and smoother fluctuations in credit for low real-time services over a larger window; the first service level set and the second service level set are service level sets. Two disjoint subsets: the first service level set is used to carry services with short deadlines and high reliability requirements, and the second service level set is used to carry services with longer deadlines or those less sensitive to latency; the controller in each Only update the set corresponding to the first service level upon expiration. And maintain the credit of the second service level set within its The frequency of control plane issuance is kept constant within the cycle, thereby reducing the jitter caused by frequent credit resets at the edge; to ensure the determinism of window switching, the controller issues a window effective timestamp simultaneously when issuing the credit token. with expiration timestamp Edge devices Upon arrival, the credit counter is switched and the data atoms that have not been sent across windows are re-billed according to the preset rules. The re-billing rules are to keep the data atom in the queue unchanged, but deduct its subsequent transmission from the credit of the new window, thereby ensuring consistent behavior at the window boundary.

[0069] S5 site total credit Based on robust availability prediction, the SD-WAN controller determines the site availability within a pre-defined observation window. Each path Calculate the available bandwidth quantile With instantaneous available bandwidth quantile parameter For pre-set and meet ; and according to Calculate available capabilities, among which Enable parameters for the preset path; and press Determine the site's total credit.

[0070] Pre-set the observation window to a length of Within a time window, the controller collects bandwidth measurement samples for each path and calculates the available bandwidth quantiles. ,in Representing a path The first available bandwidth sample within the observation window quantile value To pre-set quantile parameters for selecting conservative available bandwidth; instantaneous available bandwidth. The instantaneous available bandwidth estimate measured at the start of the current window for the controller or edge; set Indicates site The set of candidate paths available; Enable parameters for the path when Time indicates path The current window is used to carry business operations and count the site's available capabilities. Time indicates path Those not involved in hosting or only used for backup in the current window are not included; Or in the original text Indicates the site is in the window Within the available capacity estimate, calculation formula middle, The smaller of the quantile bandwidth and the instantaneous bandwidth is used to form a robust conservative estimate, thereby avoiding excessive credit issuance due to inflated instantaneous bandwidth; total site credit. middle, The product is measured in bytes or bits, representing the time window length. In the implementation, bytes are consistently used as the billing unit, and bandwidth units are converted to bytes per second. To avoid inconsistencies in units, the controller will calculate... Convert to bytes per second. Converted uniformly to seconds, thus making Number of bytes.

[0071] The selection of redundant transmission actions in S8 satisfies the following deterministic rule: when the data atoms satisfy... And there are two different paths Make the correlation coefficient When the above conditions are not met, select the dual-path copy action; when the above conditions are not met and the conditions are met... When selecting a system code action; among them For a pre-set data size threshold, For a pre-set correlation threshold, To determine the path within a pre-defined observation window With path The correlation coefficient is calculated from the time delay increment sequence.

[0072] This is a data size threshold used to limit dual-path replication to only small data atoms to control redundancy costs. This indicates that the size of the data atom does not exceed a threshold; path , Two different candidate paths and satisfying Correlation coefficient The time delay increment sequence is a statistic calculated from the path delay increment sequence within the observation window. The time delay increment sequence is defined as... ,in For the first Delay samples for each measurement period, Calculated using Pearson correlation coefficient; ,in and These are the mean values ​​of the corresponding incremental sequences within the observation window; To pre-set a correlation threshold, when When the time delay fluctuations of the two paths are not highly correlated, the replication independence is stronger, thus improving redundancy benefits, so a dual-path replication action is chosen; when the replication condition is not met but still satisfies The system code action is selected in time, and the system code action achieves redundancy through time diversity and reduces dependence on the existence of low-relevance paths; to ensure that "the existence of two different paths" can be determined, the edge device will satisfy the requirements of security domain permission and path enablement. The set of paths for which the current path measurement is valid is denoted as . and only If the selection branch is selected, it is allowed to enter the copy selection branch; otherwise, it directly enters the system code branch, thus making the selection logic completely determined.

[0073] The system components corresponding to this method include an SD-WAN controller and SD-WAN edge devices; the SD-WAN edge devices are configured to execute; a data atom generation module is used to parse AIoT device data into S1 data atoms; and a value calculation module is used to generate the S2 real-time value function and calculate S3. and The scheduling index calculation module is used to calculate S4. The credit execution module is used to deduct the sub-credits issued by S5 and execute the transmission when the credit balance constraint is met; the redundancy execution module is used to execute redundant transmission actions when the S8 condition is met; the reporting module is used to report S9 statistical information to the SD-WAN controller.

[0074] The SD-WAN controller is configured to execute; the measurement fusion module is used to generate S3. and The credit generation module generates and issues S5 credit tokens; the policy update module updates and issues credit tokens for the next time window based on statistical information. The data atom generation module includes a protocol adaptation submodule and a field mapping submodule. The protocol adaptation submodule parses the boundaries according to the protocol and extracts business fields, while the field mapping submodule maps business fields to security domain identifiers. Business type identifier and template key are used to obtain , , , And perform semantic correction calculation to obtain , , , The value calculation module includes a value function calculation submodule and a path measurement input interface. The value function calculation submodule is used to calculate the value function input interface. The path measurement input interface is used to read the output of the measurement fusion module to calculate the instantaneous value. , And calculate , The scheduling index calculation module calculates for each candidate path. It outputs the sorting key to the scheduler; the credit execution module includes a credit counter maintenance submodule, a deduction atomic operation submodule, and a window switching submodule. The window switching submodule resets at the boundary of the credit token window. And ensure that the deduction is consistent within the window; the redundant execution module includes a copy submodule and a system code submodule. The copy submodule is used to send to two paths in parallel and ensure that the extended fields are consistent. The system code module is used to start the retransmission timer and execute retransmission when the retransmission condition is met; the reporting module reports according to a preset reporting cycle. The system summarizes and sends statistics to the controller, and the report includes data by security domain. With service level The statistical items are segmented to support the fine-grained allocation by the credit generation module; the measurement fusion module fuses the measurements reported from the edge with the controller's own detection measurements. The fusion rule is to weight the synthesis according to the sample size and confidence level and output quantile parameters that are consistent with the output. , Credit generation module estimates site capabilities. Generate total site credit And generate minimum guarantee and upper limit constraints at the sub-credit level. The strategy update module applies punitive credit reductions to combinations with abnormal on-time delivery rates and delinquency rates and transfers the released credits to combinations with higher value density, thereby forming a stable closed loop.

[0075] SD-WAN edge devices also include a credit lending module, which satisfies the following constraint: when the sub-credit balance of the first service level is insufficient to cover the number of bytes to be sent for the data atoms, the number of bytes is borrowed from the sub-credit of the second service level. satisfy ,in For pre-set and meet The loan-to-deposit ratio parameter, It is designated as the second service level; and will be compensated with a coefficient in the next time window. The number of bytes borrowed is deducted, of which For pre-set and meet The replenishment coefficient.

[0076] Loan Bytes To switch from the second service level within the current window The corresponding sub-credit is temporarily transferred to the first service level sub-credit to meet the byte quota for sending high-value data. The loan-to-value ratio parameter is used to limit the loan ceiling to protect the basic resources of the second service level. For the site Security Domain Service Level In the window Sub-credit limit; replenishment factor Used to apply a "penalized" deduction to the loan in the next window. To ensure that borrowing incurs costs and thus prevent long-term overdrafts, the repayment method is through a window. When initializing credit, sub-credits of the first service level will be set according to... Reduce and return the corresponding amount to the second service level or credit it to the site's credit pool according to the configuration; to ensure auditability of lending activities, the edge device generates a loan record (loan_record) each time a loan occurs and carries it in the reporting module. The loan record must at least include a window identifier. Security Domain Identifier The service level of the loan Loan amount Triggering reason identifier (based on anomaly confidence level) With severity (The comparison results exceeding the threshold are determined) and the timestamp of the time when the lending occurred, so that the controller can perform quota structure adjustments on combinations of frequent lending and reducing the lending demand in the next window during policy updates.

[0077] When performing sending and redundant sending operations, the SD-WAN edge device adds a tunnel extension field to each sent packet. The extension field includes the data atom identifier (atom_id) and the deadline timestamp (deadline_ts). The system also includes a receiver deduplication module, which deduplicatizes duplicate received packets based on the atom_id and discards the corresponding packets when the current time is greater than the deadline_ts, and adds them to the expired discard count.

[0078] The overlay tunnel extension field is either the TLV field in the encapsulation header or a fixed-length field. `atom_id` is a unique identifier generated in S1 for the receiver to identify duplicate packets. `deadline_ts` is an absolute timestamp, representing the latest time the data atom is allowed to be submitted to the application layer by the receiver; its calculation method is `deadline_ts = t_{in,m} + d_m`, where... For the timestamp of joining the team, The deadline is set; the receiving end's deduplication module maintains a cache table for atom_id to status. Cache entries include the first reception timestamp, whether it has been submitted to the application layer, and an expiration / discard flag. The cache table uses a pre-defined maximum capacity and eviction policy. The eviction policy sorts and evicts entries based on a predetermined combination of deadline_ts and the most recent access time to avoid cache bloat. When the receiving end's current timestamp... When the value exceeds deadline_ts, the receiver discards the message and adds it to the expired discard count. The receiver uses its local clock to set the received time timestamp, and the receiver also uses NTP or PTP calibration to ensure a valid comparison with the transmitter's timestamp. To ensure end-to-end closed-loop operation, the receiver will categorize "on-time arrival / expired discard / repeated reception" statistics by security domain. With service level The data is aggregated and sent back to the sending edge or controller, thus supporting updates in S9.

[0079] The SD-WAN controller's credit generation module is configured to generate credits in each time window. Sub-credits are generated and distributed according to the following rules. Obtain the identifier for each security domain from each SD-WAN edge device. With service level Reported queued bytes and value density statistics Among them, the value density statistic For SD-WAN edge devices in the time window Internal scheduling metrics for the unit byte corresponding to the queued data atoms The pre-defined quantiles are calculated; when they meet the following conditions... Under the constraints, firstly for each Assign a pre-defined minimum sub-credit If the remaining credit exists after deducting the smallest sub-credit, the remaining credit will be distributed according to... Distribute them in descending order to the corresponding... And for each The allocated sub-credit satisfies Thus determining sub-credit And distribute it to the corresponding SD-WAN edge device.

[0080] Queued bytes For the site In the window Internal security domain Service Level The cumulative number of bytes queued in the corresponding queue or the number of bytes backlogged in the queue at the end of the window; the system uses a fixed statistical caliber during initialization to ensure consistency; value density statistics. In the window Internal scheduling metric per byte for all candidate "data atom-path" combinations within the queue Take the statistical value obtained from the pre-set quantiles, where the quantile parameter To ensure accuracy, parameters are pre-defined, and calculations are performed only on candidate combinations that meet the criteria of credit availability and path availability. Reflects true schedulable value; constraints middle, The total credit of the site, the smallest sub-credit This is a pre-defined protection limit used to ensure that each security domain and service level combination has the minimum available resources within the window. It can be generated by the controller according to the policy table and fixed in the configuration; "Remaining Credit" is defined as ,when Press at time Sort the data from largest to smallest and allocate the data accordingly, applying an upper limit to each combination during allocation. To avoid assigning invalid credits to queues with no backlog, "sequential allocation" is implemented using a defined water level filling rule, that is, filling the sorted list with fixed steps. Continuously increase the corresponding Until the remaining credits are exhausted or all combinations reach their limits, where To pre-set positive numbers, ensuring definite algorithm output and controllable implementation complexity; to guarantee smooth transitions across windows, the controller... Introducing a historical smoothing term and employing a deterministic update formula ,in The new value for this window obtained according to the above allocation rules. To pre-set a smoothing coefficient, thereby avoiding edge queue oscillations caused by large credit fluctuations.

[0081] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising a reference structure" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0082] 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 alterations 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 claims and their equivalents.

Claims

1. A data priority scheduling method for AIoT devices based on SD-WAN, characterized in that: The method includes the following steps; S1. At the AIoT gateway or SD-WAN edge device, data from the AIoT device is parsed into data atoms, each of which includes: a security domain identifier. Business type identifier, data size Deadline Basic Value attenuation coefficient Reliability requirements ; S2. Generate an instantaneous value function for the data atoms. ; S3. At the SD-WAN edge device, each superimposed tunnel path is measured according to a preset measurement cycle. Collect network measurement data and calculate the path delay quantiles. Path packet loss rate quantile quantile parameter and The quantile parameter is set in advance; and the data atoms in the path are calculated. On the deliverability coefficient and expected delivery value ; S4, for the data atoms in the path Calculate the unit byte scheduling index ; S5, controlled by the SD-WAN controller in a time window A credit token is issued to the SD-WAN edge device, the credit token including the total site credit. and by security domain identifier With service level Sub-credit division ; S6. The SD-WAN edge device atomically maps the data to be sent to the corresponding service level. And deduct credits equal to the number of bytes actually sent by the data atom from the corresponding sub-credit counter; S7. When the balance of the corresponding sub-credit counter is not less than the number of bytes to be sent, the SD-WAN edge device selects from its queued data atom set... Take the maximum value of the "data atom-path" combination and send it on the selected path; S8, when satisfied At that time, the SD-WAN edge device atomically performs redundant transmission of the data; S9. The SD-WAN edge device reports the on-time delivery count, the expired and discarded count, and the count of each path to the SD-WAN controller. and The SD-WAN controller updates and issues the credit token for the next time window based on the reported results.

2. The AIoT device data priority scheduling method based on SD-WAN according to claim 1, characterized in that: The fundamental value described in S1 Deadline attenuation coefficient With reliability requirements Determined jointly by business template parameters and semantic correction; Pre-set template parameters for the business template corresponding to the business type identifier. , , , ; And obtain abnormal confidence level Severity normalization value Equipment health .

3. The AIoT device data priority scheduling method based on SD-WAN according to claim 1, characterized in that: Between S3 and S8, there is also a closed-loop correction of the reliability requirements based on network default risk; When satisfied or At that time, the data atoms are marked as redundant candidate data atoms, and the reliability requirement is modified to... In S8, the redundant candidate data atoms are preferentially sent using a redundant transmission action.

4. The AIoT device data priority scheduling method based on SD-WAN according to claim 1, characterized in that: The credit token described in S5 is issued using at least two time windows, and the first time window is satisfied. Smaller than the second time window ; The SD-WAN controller in the first time window Sub-credits for the first service level set are issued in the second time window. Sub-credits are issued for the second service level set, wherein the first service level set and the second service level set are pre-defined and different service level sets.

5. The AIoT device data priority scheduling method based on SD-WAN according to claim 1, characterized in that: Total site credit as described in S5 Based on robust availability prediction, the SD-WAN controller performs site monitoring within a pre-defined observation window. Each path Calculate the available bandwidth quantile With instantaneous available bandwidth quantile parameter For pre-set and meet ; And calculate the available capacity.

6. The AIoT device data priority scheduling method based on SD-WAN according to claim 1, characterized in that: The selection of redundant transmission actions in S8 satisfies the following deterministic rule: when the data atoms satisfy... And there are two different paths Make the correlation coefficient When the above conditions are not met, select the dual-path copy action; when the above conditions are not met and the conditions are met... When selecting a system code action; among them For a pre-set data size threshold, For a pre-set correlation threshold, To determine the path within a pre-defined observation window With path The correlation coefficient is calculated from the time delay increment sequence.

7. A data priority scheduling method for AIoT devices based on SD-WAN according to any one of claims 1-6, characterized in that: The system components corresponding to the method include an SD-WAN controller and SD-WAN edge devices; The SD-WAN edge device is configured as follows: A data atom generation module is used to parse AIoT device data into the data atoms described in S1; The value calculation module is used to generate the instantaneous value function of S2 and calculate the value of S3. and ; The scheduling index calculation module is used to calculate the index described in S4. ; The credit execution module is used to deduct sub-credits issued by S5 and execute the transmission when the credit balance constraint is met; The redundant execution module is used to perform redundant transmission actions when the S8 conditions are met. The reporting module is used to report the statistical information mentioned in S9 to the SD-WAN controller; The SD-WAN controller is configured as follows: The measurement fusion module is used to generate the measurement fusion module described in S3. and ; The credit generation module is used to generate and issue the credit token described in S5; The strategy update module is used to update and issue the credit token for the next time window based on the statistical information.

8. The system corresponding to the AIoT device data priority scheduling method based on SD-WAN according to claim 7, characterized in that: The SD-WAN edge device also includes a credit lending module, which satisfies the following constraint: when the sub-credit balance of the first service level is insufficient to cover the number of bytes to be sent for the data atoms to be sent, the number of bytes is borrowed from the sub-credit balance of the second service level. satisfy ,in For pre-set and meet The loan-to-deposit ratio parameter, It is designated as the second service level; and will be compensated with a coefficient in the next time window. The number of bytes borrowed is deducted, where For pre-set and meet The replenishment coefficient.

9. The system corresponding to the AIoT device data priority scheduling method based on SD-WAN according to claim 7, characterized in that: When performing sending and redundant sending operations, the SD-WAN edge device adds a tunnel extension field to each sent packet. The extension field includes atom_id and deadline_ts. The system also includes a receiver deduplication module, which deduplicates duplicate received packets based on atom_id and discards the corresponding packet when the current time is greater than deadline_ts, adding it to the expired discard count.

10. The system corresponding to the AIoT device data priority scheduling method based on SD-WAN according to claim 7, characterized in that: The credit generation module of the SD-WAN controller is configured to generate credits in each time window. The sub-credits are generated and distributed according to the following rules. ; Obtain the identifier for each security domain from each SD-WAN edge device. With service level Reported queued bytes and value density statistics The value density statistic mentioned above For the SD-WAN edge device in the time window Internal scheduling metrics for the unit byte corresponding to the queued data atoms The pre-defined quantiles are calculated; when they meet the following conditions... Under the constraints, firstly for each Assign a pre-defined minimum sub-credit If the remaining credit exists after deducting the minimum sub-credit, the remaining credit is distributed according to... Distribute them in descending order to the corresponding... And for each The allocated sub-credit satisfies Thus determining the sub-credit And distribute it to the corresponding SD-WAN edge device.

Images