Routing control method, multi-service parallel routing method, electronic equipment, medium and program product

Abstract

The invention provides a routing control method, which is applied to network equipment with multi-link access capability, and comprises the following steps: carrying out service identification on a to-be-forwarded data flow of the network equipment to obtain service type information of the data flow; obtaining evaluation indexes of a plurality of candidate links of the network equipment; evaluating the plurality of candidate links based on the evaluation index and the service type information to obtain a comprehensive quality score; selecting a target link from the plurality of candidate links according to the service type information, the evaluation index and the comprehensive quality score; and controlling the network equipment to transmit the data flow through the target link. The invention further provides a multi-service parallel routing method, electronic equipment, a computer readable medium and a computer program product. According to the invention, differentiated routing control of multiple service types is realized, quality fluctuation caused by switching of operators is reduced, and the stability and the intelligent level of network transmission are remarkably improved.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of software-defined wide area network (SD-WAN), in particular to a routing control method, a multi-service parallel routing method, an electronic device, a computer readable medium, and a computer program product. BACKGROUND

[0002] In order to meet the demand for more reliable 4G / 5G networking, current SD-WAN CPE (Customer Premises Equipment) products usually support inserting multiple operator SIM cards, so as to be able to select between different operator links to carry customer services. In the existing scheme, link selection is mainly determined according to the signal strength of each operator. Although this method solves the problem of 4G / 5G networking to some extent, there are still problems of insufficient timeliness of link switching and instability of network quality after switching. SUMMARY

[0003] The present disclosure provides a routing control method, a multi-service parallel routing method, an electronic device, a computer readable medium, and a computer program product.

[0004] In a first aspect, the present disclosure provides a routing control method applied to a network device with multi-link access capability, including: performing service identification on a data flow to be forwarded by the network device to obtain service type information of the data flow; obtaining evaluation indexes of multiple candidate links of the network device; evaluating the multiple candidate links based on the evaluation indexes and the service type information to obtain comprehensive quality scores; selecting a target link from the multiple candidate links according to the service type information, the evaluation indexes, and the comprehensive quality scores; and controlling the network device to transmit the data flow through the target link.

[0005] In some embodiments, obtaining the evaluation indexes of the multiple candidate links of the network device includes: periodically collecting real-time quality indexes of the multiple candidate links of the network device, the real-time quality indexes at least including quality indexes of physical layer, network layer, and application layer; predicting, based on the real-time quality indexes, quality index prediction values of the multiple candidate links at a future preset time through a prediction model; and obtaining network feature data of the multiple candidate links.

[0006] In some embodiments, the evaluation of the plurality of candidate links based on the evaluation indicators and the service type information to obtain a comprehensive quality score comprises: fusing the real-time quality indicators, the predicted values of the quality indicators and the network feature data by using a fusion algorithm to generate a link evaluation vector; and generating the comprehensive quality score by using a three-dimensional quality evaluation model based on the link evaluation vector and the service type information.

[0007] In some embodiments, the generation of the comprehensive quality score by using a three-dimensional quality evaluation model based on the link evaluation vector and the service type information comprises: inputting the link evaluation vector into the three-dimensional quality evaluation model; evaluating the data in the corresponding dimensions in the vector by a physical layer evaluation module, a network layer evaluation module and an application layer evaluation module in the model respectively to obtain a physical layer evaluation score P, a network layer evaluation score N and an application layer evaluation score A; and calculating and outputting the comprehensive quality score Q by a weighted fusion module in the model according to the formula Q = a × P + β × N + γ × A, wherein a, β and γ are weight coefficients determined according to the service type information and satisfy a + β + γ = 1.

[0008] In some embodiments, the evaluation of the data in the corresponding dimensions in the vector by the physical layer evaluation module, the network layer evaluation module and the application layer evaluation module in the model to obtain the physical layer evaluation score P, the network layer evaluation score N and the application layer evaluation score A comprises: extracting quality indicators of the physical layer dimensions from the link evaluation vector by the physical layer evaluation module, wherein the quality indicators comprise a reference signal received power, a reference signal received quality, a signal to interference plus noise ratio and a cell switching frequency, and the physical layer evaluation score P is calculated based on the quality indicators by using a weighted normalization method; extracting quality indicators of the network layer dimensions from the link evaluation vector by the network layer evaluation module, wherein the quality indicators comprise a delay, a jitter and a packet loss rate, and the network layer evaluation score N is calculated based on the quality indicators by using a weighted normalization method; and extracting quality indicators of the application layer dimensions from the link evaluation vector by the application layer evaluation module, wherein the quality indicators comprise a throughput, a connection success rate and a protocol efficiency, and the application layer evaluation score A is calculated based on the quality indicators by using a weighted normalization method.

[0009] In some embodiments, the selection of a target link from the plurality of candidate links according to the service type information, the evaluation indicators and the comprehensive quality scores comprises: obtaining a routing strategy corresponding to the service type information, wherein the routing strategy defines target evaluation indicators related to the service type and constraint conditions thereof; and performing multi-objective optimization calculation on the target evaluation indicators and the comprehensive quality scores of the plurality of candidate links based on the routing strategy to determine the target link.

[0010] In some embodiments, the multi-objective optimization calculation is performed on the target evaluation indicators and the comprehensive quality scores of the plurality of candidate links based on the routing strategy to determine the target link, including: filtering the plurality of candidate links based on the constraint conditions in the routing strategy to obtain a candidate subset, wherein the constraint conditions at least include threshold requirements for the target evaluation indicators; performing multi-objective optimization calculation on the target evaluation indicators and the comprehensive quality scores of each link in the candidate subset to obtain a comprehensive score of each link; and selecting the link with the highest comprehensive score as the target link.

[0011] In some embodiments, the multi-objective optimization calculation is performed on the target evaluation indicators and the comprehensive quality scores of the plurality of candidate links based on the routing strategy to determine the target link, including: filtering the plurality of candidate links based on the constraint conditions in the routing strategy to obtain a candidate subset, wherein the constraint conditions at least include threshold requirements for the target evaluation indicators; and selecting the link with the highest comprehensive quality score in the candidate subset as the target link.

[0012] In some embodiments, the control of the network device to transmit the data flow through the target link includes: generating a switching instruction containing a forwarding configuration corresponding to the target link; updating the forwarding rule of the network device according to the switching instruction; and routing the data flow to the target link for transmission according to the updated forwarding rule.

[0013] In some embodiments, the routing control method further includes: performing millisecond-level heartbeat detection on the target link to monitor the link connectivity state; and when it is determined based on the heartbeat detection result that the target link has failed, re-determining a target link from the remaining candidate links to perform fast recovery.

[0014] In the second aspect, the embodiments of the present disclosure provide a multi-service parallel routing method. The multi-service parallel routing method performs the routing control method of the first aspect of the embodiments of the present disclosure in parallel on a plurality of data flows to be forwarded received via a network device, so as to determine target links corresponding to the data flows of different service types respectively.

[0015] In some embodiments, the multi-service parallel routing method further includes: performing unified traffic scheduling on the plurality of data flows forwarded via the different target links, wherein the traffic scheduling includes bandwidth allocation, priority sorting, and queue scheduling strategy optimization.

[0016] In a third aspect, an electronic device is provided, and the electronic device includes one or more processors, and a memory storing one or more programs, wherein the one or more programs, when executed by the one or more processors, cause the one or more processors to implement the routing control method according to the first aspect of the embodiments of the present disclosure or implement the multi-service parallel routing method according to the second aspect of the embodiments of the present disclosure.

[0017] In a fourth aspect, a computer-readable medium is provided, and the computer-readable medium stores a computer program, wherein the computer program, when executed by a processor, implements the routing control method according to the first aspect of the embodiments of the present disclosure or implements the multi-service parallel routing method according to the second aspect of the embodiments of the present disclosure.

[0018] In a fifth aspect, a computer program product is provided, and the computer program product includes a computer program or instructions, wherein the computer program or instructions, when executed by a processor, implements the routing control method according to the first aspect of the embodiments of the present disclosure or implements the multi-service parallel routing method according to the second aspect of the embodiments of the present disclosure.

[0019] The embodiments of the present disclosure achieve differentiated routing control of multiple service types, reduce quality fluctuations caused by operator switching, and significantly improve the stability and intelligent level of network transmission. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 is a flowchart of a routing control method according to an embodiment of the present disclosure.

[0021] Figure 2 is a schematic diagram of a three-dimensional quality evaluation model according to an embodiment of the present disclosure.

[0022] Figure 3 is a schematic diagram of a multi-service parallel routing mechanism according to an embodiment of the present disclosure.

[0023] Figure 4 shows a service policy library structure according to an embodiment of the present disclosure.

[0024] Figure 5 is a whole architecture diagram of a routing control system according to an embodiment of the present disclosure.

[0025] Figure 6 is a composition diagram of an electronic device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0026] In order for those skilled in the art to better understand the technical solutions of the present disclosure, the technical solutions of the present disclosure are described in detail below with reference to the drawings.

[0027] Example embodiments will be described more fully hereinafter with reference to the accompanying drawings; however, the example embodiments can be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0028] The various embodiments in the present disclosure and the various features in the embodiments can be combined with each other without conflict.

[0029] As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0030] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the present disclosure. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "comprises" and / or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0031] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and the present disclosure and will not be interpreted in an overly literal or overly formal sense unless expressly so defined herein.

[0032] In the present disclosure, the following technical terms should be understood as follows unless otherwise specified: Multi-link access capability refers to the capability of a network device to access a network through multiple different types of links simultaneously. These links can include, but are not limited to, dedicated line links, broadband links, cellular mobile communication links (such as 4G, 5G links), etc.

[0033] Service type refers to a service category identified according to the characteristics of a data flow (such as application port, protocol type, transmission direction, QoS identifier, etc.), for example: voice service, video conference service, financial transaction, file transfer service, web access service, etc.

[0034] The prediction model refers to a model for predicting the future quality of a link. The model can be implemented based on statistical learning, machine learning or deep learning algorithms, such as linear regression model, LSTM network model, time series model, etc.

[0035] The LSTM prediction model refers to a machine learning model using a long short-term memory network, the input of which is a historical time series quality indicator, and the output of which is a predicted value of each link quality indicator in a future period (such as 5 seconds), which is used to realize predictive routing.

[0036] Deep packet inspection (DPI) refers to a network packet analysis technology that accurately identifies the application type of a data flow by checking the application layer features in the data packet payload and matching them with a pre-set feature library.

[0037] Quality metrics refer to technical parameters used to quantitatively evaluate the transmission performance of candidate links.

[0038] Millisecond-level heartbeat detection refers to a link fault detection mechanism that sends probe packets to the operator gateway at a high frequency (such as sending a probe packet every 20 ms) and low latency (such as a timeout of 15 ms) to quickly perceive the connectivity of the link (such as identifying faults within 50 ms) and trigger a fast recovery process.

[0039] In order to make the technical solutions of the present disclosure clearer and more specific, the system architecture and workflow of the present disclosure are described in detail below with the customer premises equipment (CPE) as a typical embodiment. It should be understood by those skilled in the art that this embodiment is only used to describe the technical solutions of the present disclosure, and is not a limitation on the protection scope of the present disclosure.

[0040] The core of the present disclosure is to propose a general routing control method and system based on intelligent prediction and service perception. The basic idea is not limited to a specific network device form. The network device in the text can be broadly understood as any device with multi-link access capability and the ability to dynamically select between multiple wireless links (especially multiple operator links). Therefore, any network terminal, gateway or edge computing node with multi-link access and data processing capabilities should be considered to fall within the protection scope of the present disclosure as long as it implements the routing control method described in the present disclosure.

[0041] In existing SD-WAN CPE products, in order to improve the reliability of 4G / 5G networking, multiple operator SIM cards are usually inserted, and the link carrying service is selected according to the signal strength of different operators. However, this method has problems such as insufficient timeliness of link switching and inability to guarantee the stability of the network after switching in actual application.

[0042] Therefore, the technical problem to be solved by the present application is how to realize more timely and stable link switching in a multi-operator link environment, thereby improving overall service continuity and network service quality.

[0043] The present disclosure relates to the technical field of software-defined wide area network (SD-WAN), and specifically proposes a CPE routing control system and method based on LSTM prediction and service awareness. The scheme realizes dynamic optimal path selection by real-time analysis of signal quality of multi-operator (such as mobile, China Unicom, and China Telecom) links, combined with machine learning prediction and a three-dimensional quality evaluation model, thereby guaranteeing the quality requirements of existing network services to the greatest extent.

[0044] Figure 1 is a flowchart of the routing control method of the present disclosure embodiment.

[0045] In a first aspect, with reference to Figure 1 The present disclosure embodiment provides a routing control method applied to a network device with multi-link access capability, including: S11, identifying the service of a data flow to be forwarded by the network device to obtain service type information of the data flow; S12, obtaining evaluation indexes of multiple candidate links of the network device; S13, evaluating the multiple candidate links based on the evaluation indexes and the service type information to obtain a comprehensive quality score; S14, selecting a target link from the multiple candidate links according to the service type information, the evaluation indexes, and the comprehensive quality score; S15, controlling the network device to transmit the data flow through the target link.

[0046] In some embodiments, obtaining the service type information of the data flow to be forwarded by the network device includes: Real-time analysis of the data flow is performed by deep packet inspection technology to identify the corresponding service type information.

[0047] In the present disclosure embodiment, based on deep packet inspection (DPI) technology, a data flow received via the network device and to be forwarded is analyzed in real time to identify the service application type to which the data flow belongs, and the corresponding service type information is output. It should be noted that the data flow is generated by a user device, accessed to the network device through local area network technology, and finally transmitted through a 4G / 5G wireless link through a multi-SIM card interface of the network device.

[0048] It should be noted that, although deep packet inspection (DPI) is a preferred and efficient technical means for implementing the service type identification of the present disclosure, the protection scope of the present disclosure is not limited to this specific technology. Any technical means that can achieve the purpose of identifying the service type of a data flow falls within the concept of the present disclosure.

[0049] In the embodiments of the present disclosure, the service types include at least video conference, financial transaction and file transmission.

[0050] In some embodiments, the evaluation indexes of the plurality of candidate links of the network device are acquired, including: The real-time quality indexes of the plurality of candidate links of the network device are periodically collected, and the real-time quality indexes include at least quality indexes of a physical layer, a network layer and an application layer. Based on the real-time quality indexes, a prediction model is used to predict quality index prediction values of the plurality of candidate links in a future preset time. Network feature data of the plurality of candidate links are acquired.

[0051] In the embodiments of the present disclosure, the quality indexes of the plurality of candidate links are detected with differentiated periods to realize real-time perception of the link state. For example, the physical layer quality indexes can be detected every 200 ms, and the network layer and application layer quality indexes can be detected every 1 s. Through the above differentiated detection period setting, the system overhead and data accuracy can be considered while ensuring real-time performance.

[0052] In the embodiments of the present disclosure, a long short-term memory network (LSTM) based prediction model can be used to predict the change trend of the physical layer, network layer and application layer quality indexes of each candidate link in a future period of time (for example, 5 seconds).

[0053] In addition, an operator feature library can be provided to store network characteristic information of each candidate link, including but not limited to the operator, the access network type (for example, 4G, 5G or 5GA, etc.), the typical bandwidth, the delay, etc.

[0054] The physical layer quality indexes can include at least one of reference signal received power (RSRP), reference signal received quality (RSRQ), signal to interference plus noise ratio (SINR) and cell switching frequency; the network layer quality indexes can include at least one of delay, jitter and packet loss rate; and the application layer quality indexes can include at least one of throughput, connection success rate and protocol efficiency.

[0055] The real-time quality indicators are used to reflect the running state of the candidate link at different levels, and can be used for multi-level quality detection of each operator access link through the multi-SIM card interface. The multi-level quality detection includes: physical layer quality parameter collection, in which parameters such as RSRP, RSRQ and SINR of each link are obtained at a sampling period of about 200 ms, and are used to represent the signal strength and transmission stability of the link; network layer performance collection, in which performance indicators such as delay, jitter and packet loss rate are measured by combining active detection and passive statistics; and application layer quality collection, in which data flow transmitted through each link is monitored, and indicators such as throughput, connection success rate and retransmission rate are calculated to reflect the service capability of the link to the upper layer service.

[0056] It should be noted that the real-time quality indicators, the predicted values and the network feature data are only preferred examples for illustrating the implementation mode of the present disclosure, and are not a limitation on the protection scope. In other embodiments, the evaluation indicators can also include any parameter capable of reflecting the link quality or performance state, such as but not limited to link cost, security level, etc., as long as the parameter is used for link quality evaluation in the routing control framework of the present disclosure, which should be considered to fall within the protection scope of the present disclosure.

[0057] In the embodiments of the present disclosure, the candidate link of the present disclosure can include wireless links (such as mobile links, Unicom links and telecom links) from different operators accessed by the network device. In more extensive embodiments, the candidate link can generally refer to a plurality of data transmission channels that the network device can access, including but not limited to different technical standard links (such as 5G links and 4G links) of the same operator, links of different frequency bands under the same technical standard, a mixed set of wireless and wired links (such as optical fibers and xDSL), etc. Any link participating in the routing control process based on service awareness and quality prediction described in the present disclosure can be considered as a candidate link.

[0058] In some embodiments, the plurality of candidate links are evaluated based on the evaluation indicators and the service type information to obtain a comprehensive quality score, including: The real-time quality indicators, the quality indicator predicted values and the network feature data are fused by using a fusion algorithm to generate a link evaluation vector; Based on the link evaluation vector and the service type information, a three-dimensional quality evaluation model is used to generate the comprehensive quality score.

[0059] In the embodiments of the present disclosure, a fusion algorithm is used to fuse data from different sources to generate a link evaluation vector and calculate a comprehensive quality score. The fusion algorithm can flexibly select different implementation manners according to actual application scenarios. For example, in a scenario requiring high calculation efficiency and strong real-time performance, a weighted fusion algorithm can be used; in a scenario where there is noise, uncertainty or data loss, a Bayesian network model can be used for probabilistic reasoning; in a scenario where the input feature dimension dynamically changes or the weight distribution needs to be adaptively adjusted, an attention mechanism can be introduced to realize dynamic fusion. Subsequently, based on the link evaluation vector and service type information, a three-dimensional quality evaluation model is used to calculate a comprehensive quality score. The three-dimensional quality evaluation model can be constructed based on the weighted combination of physical layer quality, network layer quality and application layer quality.

[0060] In other embodiments, the fusion algorithm can further receive other types of network state information or service feature data to adapt to different network environments or service requirements, thereby improving the accuracy and robustness of the comprehensive quality score.

[0061] In some embodiments, based on the link evaluation vector and the service type information, a three-dimensional quality evaluation model is used to generate the comprehensive quality score, including: inputting the link evaluation vector into the three-dimensional quality evaluation model; using a physical layer evaluation module, a network layer evaluation module and an application layer evaluation module in the model to respectively evaluate the data in the corresponding dimensions of the vector to obtain a physical layer evaluation score P, a network layer evaluation score N and an application layer evaluation score A; using a weighted fusion module in the model to calculate and output the comprehensive quality score Q according to the formula Q=α×P+β×N+γ×A, wherein α, β and γ are weight coefficients determined according to the service type information, and satisfy α+β+γ=1.

[0062] In some embodiments, using a physical layer evaluation module, a network layer evaluation module and an application layer evaluation module in the model to respectively evaluate the data in the corresponding dimensions of the vector to obtain a physical layer evaluation score P, a network layer evaluation score N and an application layer evaluation score A, includes: the physical layer evaluation module extracts quality indicators of the physical layer dimension from the link evaluation vector, the quality indicators including reference signal received power, reference signal received quality, signal to interference plus noise ratio and cell switching frequency, and calculates the physical layer evaluation score P based on the quality indicators using a weighted normalization method; The network layer evaluation module extracts quality indexes of network layer dimensions from the link evaluation vector, the quality indexes including latency, jitter and packet loss rate, and obtains the network layer evaluation score N by using weighted normalization based on the quality indexes; The application layer evaluation module extracts quality indexes of application layer dimensions from the link evaluation vector, the quality indexes including throughput, connection success rate and protocol efficiency, and obtains the application layer evaluation score A by using weighted normalization based on the quality indexes.

[0063] In the embodiments of the present disclosure, the calculation formula of the physical layer evaluation score P is: P = w1 x RSRP_norm + w2 x RSRQ_norm + w3 x SINR_norm + w4 x (1-Switch_freq_norm) wherein RSRP_norm, RSRQ_norm and SINR_norm are normalized values of reference signal received power, reference signal received quality and signal to interference plus noise ratio respectively, Switch_freq_norm is a normalized value of cell switching frequency, w1, w2, w3 and w4 are weights and w1+w2+w3+w4=1.

[0064] In the embodiments of the present disclosure, the calculation formula of the network layer evaluation score N is: N = v1 x Latency_norm + v2 x Jitter_norm + v3 x (1-Loss_norm) wherein Latency_norm, Jitter_norm and Loss_norm are normalized values of latency, jitter and packet loss rate respectively, v1, v2 and v3 are weights and v1+v2+v3=1.

[0065] In the embodiments of the present disclosure, the calculation formula of the application layer evaluation score A is: A = u1 x Throughput_norm + u2 x Connection_success_norm + u3 x (1-Protocol_inefficiency_norm) wherein Throughput_norm, Connection_success_norm and Protocol_inefficiency_norm are normalized values of throughput, connection success rate and protocol inefficiency respectively, u1, u2 and u3 are weights and u1+u2+u3=1.

[0066] In the embodiments of the present disclosure, the weight coefficients a, b, g are weight coefficients preset in the policy library according to the service type information and are supported to be adjusted by the user.

[0067] In the embodiments of the present disclosure, for the video conference service, a=0.2, b=0.5, g=0.3 are preset; for the financial transaction service, a=0.1, b=0.2, g=0.7 are preset; and for the file transmission service, a=0.3, b=0.2, g=0.5 are preset.

[0068] Figure 2 is a schematic diagram of a three-dimensional quality evaluation model of the present disclosure.

[0069] It should be particularly noted that, Figure 2 The service types (such as video conference, financial transaction, and file transmission) shown in the above table are only examples and are intended to intuitively show how the model dynamically adjusts the weight distribution of the three evaluation dimensions of the physical layer, the network layer, and the application layer according to the characteristic requirements of different services to output the comprehensive quality score (Q) for a specific service.

[0070] The service types of the present disclosure are not limited to the above table, and any service scenario (such as voice call, online game, web browsing, etc.) that can be identified and has differentiated network quality requirements can be applicable to the model of the present disclosure.

[0071] As shown in the above table, Figure 2 the working principle of the three-dimensional quality evaluation model and the application examples in several typical service scenarios are shown.

[0072] As shown in the above table, Figure 2 the three-dimensional quality evaluation model proposed by the present disclosure calculates the comprehensive quality score Q by weighting the evaluation scores of the physical layer (P), the network layer (N), and the application layer (A), which is used to represent the overall communication quality of the candidate link, and the calculation formula is as follows: Q=a×P+b×N+g×A Wherein, Q is the comprehensive quality score (the value range is 0-100 points), P, N, and A are the evaluation scores of the physical layer, the network layer, and the application layer respectively, a, b, and g are the weight coefficients of each dimension, which satisfy a+b+g=1, and are dynamically configured according to the service type.

[0073] It should be noted that the input of the three-dimensional quality evaluation model is the link evaluation vector generated after the fusion processing based on the real-time quality indicators, the quality indicator prediction values, and the network feature data. The vector is processed by the evaluation calculation modules of the physical layer, the network layer, and the application layer in turn to generate the evaluation scores P, N, and A respectively; then, the scores are weighted and calculated according to the preset comprehensive weights a, b, and g, and finally the comprehensive quality score Q is output.

[0074] To ensure the comparability between indicators, the disclosure normalizes all raw quality indicators and linearly maps them to the interval [0, 100]. For example, for reference signal received power (RSRP), its normalized value can be calculated as follows: RSRP_norm = (RSRP - RSRP_min) / (RSRP_max - RSRP_min) × 100 where RSRP_min and RSRP_max are the planning range boundaries (e.g., RSRP_min = -110 dBm and RSRP_max = -65 dBm). All parameters with “_norm” are normalized values.

[0075] Through the above normalization process, indicators of different dimensions are unified under the same scoring system, providing a basis for subsequent weighted calculation.

[0076] It should be noted that the normalization process used in each level of evaluation in the disclosure can be dynamically set according to the actual network environment and statistical data. The dynamic setting mechanism allows the system to periodically or adaptively update the minimum and maximum values of each indicator (i.e., the normalization boundary) based on real-time sampled indicator distribution (e.g., signal strength, latency, throughput, packet loss rate, etc.), thereby maintaining the adaptability and accuracy of the scoring model under different network environments, business loads, or time periods. Therefore, in addition to physical layer indicators (such as RSRP, RSRQ, SINR), network layer indicators (such as latency, jitter, packet loss rate) and application layer indicators (such as throughput, protocol efficiency, etc.) can also dynamically adjust the normalization reference interval according to the statistical characteristics of the actual observed data, to achieve a comprehensive dynamic normalization evaluation mechanism.

[0077] (1) Physical layer evaluation (P) Physical layer evaluation indicators include RSRP (reference signal received power), RSRQ (reference signal received quality), SINR (signal-to-noise ratio), and cell switching frequency.

[0078] The calculation formula of its evaluation score P is as follows: P = w1 × RSRP_norm + w2 × RSRQ_norm + w3 × SINR_norm + w4 × (1 - Switch_freq_norm) where RSRP_norm, RSRQ_norm, and SINR_norm are the normalized values of the corresponding indicators. Switch_freq_norm is the normalized value of cell switching frequency. Since frequent switching reflects link instability, it is converted to a positive indicator by (1-Switch_freq_norm); w1, w2, w3, w4 are weight coefficients of each indicator, satisfying w1+w2+w3+w4=1, and typical values are w1=0.3, w2=0.3, w3=0.3, w4=0.1.

[0079] (2) Network layer evaluation (N) The network layer evaluation indicators include: latency (Latency), jitter (Jitter), and packet loss rate (Loss), The network layer evaluation score (N) is calculated by weighting the normalized values of the above indicators, and the calculation formula is as follows: N=v1×Latency_norm+v2×Jitter_norm+v3×(1-Loss_norm) Where Latency_norm, Jitter_norm, Loss_norm are the normalized values of the corresponding indicators; v1, v2, v3 are weight coefficients of each indicator, satisfying v1+v2+v3=1, and typical values are v1=0.5, v2=0.3, v3=0.2.

[0080] After normalization, the values of latency and jitter satisfy the larger the score value, the better the quality (i.e. the smaller the original value, the larger the normalized value), so they are used directly. After normalization, the value of packet loss rate is still a negative indicator (the larger the value, the more serious the packet loss), so it is converted to a positive indicator by (1-Loss_norm).

[0081] (3) Application layer evaluation (A) The application layer evaluation indicators include: throughput (Throughput), connection success rate (Connection_success), and protocol efficiency (Protocol_efficiency, such as TCP retransmission rate), The application layer evaluation score (A) is calculated by weighting the normalized values of the above indicators, and the calculation formula is as follows: A=u1×Throughput_norm+u2×Connection_success_norm+u3×(1-Protocol_inefficiency_norm) Throughput_norm, Connection_success_norm, Protocol_inefficiency_norm are the normalized values of the corresponding indicators, respectively. u1, u2, u3 are the weight coefficients of each indicator, satisfying u1+u2+u3=1, and typical values are u1=0.5, u2=0.3, u3=0.2.

[0082] Throughput and Connection success rate are positive indicators; Protocol efficiency can be indirectly reflected by retransmission rate and other indicators, so (1-Protocol_inefficiency_norm) is used.

[0083] (4) Comprehensive weight (α, β, γ) The comprehensive weights α, β and γ in the three-dimensional quality evaluation model are used to reflect the focus of different business types on the quality of the physical layer, network layer and application layer. The weights provide a preset default configuration in the system and support user adjustment according to specific network environment or personalized needs. Typical business type weight configuration examples are as follows: Video conference: α=0.2, β=0.5, γ=0.3 (emphasis on network layer performance); Financial transaction: α=0.1, β=0.2, γ=0.7 (emphasis on application layer reliability); File transfer: α=0.3, β=0.2, γ=0.5 (emphasis on application layer throughput).

[0084] It should be noted that, Figure 2 The examples in the above formula appear a simplified variant of the model in a specific business scenario, and the general formula Q=α×P+β×N+γ×A should be used in actual deployment.

[0085] In some embodiments, according to the business type information, the evaluation indicators and the comprehensive quality score, a target link is selected from the plurality of candidate links, comprising: Obtaining a routing strategy corresponding to the business type information, the routing strategy defining target evaluation indicators related to business type and their constraint conditions; Based on the routing strategy, multi-objective optimization calculation is performed on the target evaluation indicators and the comprehensive quality score of the plurality of candidate links to determine the target link.

[0086] In some embodiments, based on the routing strategy, multi-objective optimization calculation is performed on the target evaluation indicators and the comprehensive quality score of the plurality of candidate links to determine the target link, comprising: filtering the plurality of candidate links based on the constraint condition in the routing strategy to obtain a candidate subset, wherein the constraint condition at least includes a threshold requirement for the target evaluation index; performing multi-objective optimization calculation on the target evaluation index and the comprehensive quality score of each link in the candidate subset to obtain a comprehensive score of each link; selecting a link with the highest comprehensive score as the target link.

[0087] In some embodiments, based on the routing strategy, performing multi-objective optimization calculation on the target evaluation index and the comprehensive quality score of the plurality of candidate links to determine the target link, comprising: filtering the plurality of candidate links based on the constraint condition in the routing strategy to obtain a candidate subset, wherein the constraint condition at least includes a threshold requirement for the target evaluation index; In the candidate subset, selecting a link with the highest comprehensive quality score as the target link.

[0088] In the embodiments of the present disclosure, the optimal path calculation based on predicted quality not only considers the current quality score, but also comprehensively considers multiple dimensional factors such as predicted quality, link cost, reliability, stability, etc. By introducing the predictive decision mechanism, the future quality trend output by the LSTM model can realize forward-looking path selection before network fluctuation; at the same time, according to the dynamic adjustment of the optimization target and the weight distribution according to the business type, a multi-objective optimization algorithm is used to find the optimal balance between conflicting targets, so as to select the truly "optimal" path for different business needs, not just the path with the highest quality.

[0089] In the embodiments of the present disclosure, based on the business type information, a routing strategy corresponding thereto is obtained. The routing strategy defines the key target evaluation index related to a specific business type and the constraint condition that must be met. For example, for the "video conference" business, the constraint condition can be defined as "the jitter value of the link needs to be lower than 30 milliseconds". Subsequently, based on the obtained routing strategy, multi-objective optimization calculation is performed on the plurality of candidate links to determine the final target link.

[0090] The multi-objective optimization calculation can include: first, using the constraint condition defined in the routing strategy as a filtering threshold to filter all candidate links; by comparing the prediction index of each link with the constraint condition in the routing strategy, the candidate links that do not meet the constraint condition are excluded, thereby obtaining a candidate subset that meets the basic quality requirement of the business; then, in the candidate subset, performing multi-objective optimization calculation on the target evaluation index and the comprehensive quality score of each link to obtain a comprehensive score of each link, and selecting a link with the highest comprehensive score as the target link.

[0091] In the preferred embodiment, the multi-objective optimization calculation can be implemented by using a weighted linear model, analytic hierarchy process (AHP), Bayesian optimization model, genetic algorithm, or multi-objective linear programming algorithm. After normalizing each index, the objective function value is calculated to reflect the comprehensive performance of each link, and the optimal path is determined according to the calculation result.

[0092] In another embodiment, the comprehensive quality score can be directly used as the main optimization target. Under the premise of meeting the service policy constraints, the link with the highest comprehensive quality score is selected as the target link.

[0093] In some embodiments, the path calculation module can also evaluate the link performance trend within a future preset time window based on the predictive quality score, to support early link switching and priority path selection in network performance fluctuation or latency-sensitive service scenarios.

[0094] Through the above multi-objective optimization calculation process based on service policy constraints and comprehensive quality scores, the present disclosure can achieve differentiated and dynamic path decision-making in a multi-service parallel scenario, so that various types of services can meet the key performance indicator (KPI) requirements while achieving optimal overall network resource utilization.

[0095] It should be noted that the above multi-objective optimization calculation method is only a preferred embodiment. The specific optimization algorithm type, number of constraint conditions, and index weight can be flexibly adjusted according to different application scenarios. Any scheme that uses the service type information, evaluation indexes, and comprehensive quality scores as inputs, and performs multi-objective optimization decision-making based on routing strategies to determine the target link, should be considered within the protection scope of the present disclosure.

[0096] In some embodiments, the control of the network device to transmit the data flow through the target link includes: generating a switching instruction, the switching instruction containing a forwarding configuration corresponding to the target link; updating the forwarding rule of the network device according to the switching instruction; routing the data flow to the target link for transmission according to the updated forwarding rule.

[0097] In the embodiments of the present disclosure, the process of controlling the network device to transmit the data flow through the target link can include three stages of policy generation, rule updating, and data flow routing.

[0098] Firstly, in the policy generation phase, instructions for link switching can be generated based on the previously identified service types and real-time network state information. The switching instructions can be adaptively configured according to the performance parameters (such as bandwidth, latency, packet loss rate, jitter, etc.) of the target link, so as to determine the forwarding configuration corresponding to the target link. The forwarding configuration can include the interface identifier of the target link, the priority parameter, the quality of service (QoS) constraint condition, the queue scheduling strategy, and the rollback condition, etc., for guiding the subsequent update of the forwarding rules.

[0099] Subsequently, in the rule update phase, the forwarding rules of the network device are updated according to the generated switching instructions. The update can include operations such as modifying forwarding table entries, adjusting routing priorities, configuring traffic classification and marking, and reallocating bandwidth or scheduling queues, so that the forwarding path of the network device is consistent with the selected target link, and can meet the quality of service requirements of the service. In some embodiments, the update of the forwarding rules can specifically include adjustment of the routing table entries, queue scheduling strategies, flow table rules, or tunnel configurations of the device, to achieve forwarding of data flow via the target link.

[0100] In the data flow routing phase, the network device selects and forwards the subsequent data flow according to the updated forwarding rules, so that the data flow is transmitted via the target link. To ensure the smoothness of the switching process, the link state can be monitored in real time during the routing switching process, and when an abnormal situation is detected, the rollback or path reselection operation is performed according to the preset conditions, so as to ensure the continuity and stability of the data transmission process.

[0101] Further, the generation of the switching instructions can also combine the priority information of the service type to achieve differentiated policies for different types of services in link selection. For example, for service types with high real-time requirements (such as video conferencing, voice calls), links with lower latency and higher stability can be preferentially selected; while for service types with large bandwidth but tolerable latency (such as file transfer or cloud backup), links with lower cost or lighter load can be selected. In this way, the system can dynamically adjust the link allocation according to the service characteristics and network state, to achieve optimal utilization of overall network resources.

[0102] In addition, in the execution of the above control process, the running state information of the link, such as transmission rate, utilization rate, bit error rate, and latency fluctuation, etc. can be continuously obtained and analyzed, and the switching instructions or forwarding rules can be adjusted in real time accordingly, to achieve adaptive optimization control. In this way, the entire process not only enables accurate forwarding based on the target link, but also maintains the continuity and quality of service of the service transmission when the network environment changes.

[0103] In some embodiments, the routing control method further comprises: performing millisecond-level heartbeat detection on the target link to monitor link connectivity state; when it is determined that the target link fails based on the heartbeat detection result, re-determining the target link from the remaining candidate links to perform fast recovery.

[0104] In the embodiments of the present disclosure, the routing control method further comprises a failure detection and fast recovery mechanism. By performing high-frequency heartbeat detection on the target link, millisecond-level link state monitoring and fast recovery are achieved. Specifically, a probe packet can be periodically sent to the operator gateway, cloud control node or stable public IP address to detect link connectivity and delay jitter. In a preferred embodiment, a UDP Echo probe packet can be used as the heartbeat packet, the packet length is 64 bytes, the sending interval is 20 milliseconds, and the timeout time is 15 milliseconds; when a link does not receive a response for two consecutive heartbeat packets, it is determined that the link is in a failure state. When the target link is determined to be in a failure state, a new target link can be re-determined from the remaining candidate links based on the path calculation result or real-time link state, so as to ensure that the link after service recovery meets the key performance indicator (KPI) requirements.

[0105] In the embodiments of the present disclosure, the failure detection and link switching can be completed within 50 milliseconds, so that the upper-layer service has no awareness of the link interruption, and the communication can be restored to normal in a very short time. It should be noted that the implementation of the heartbeat detection is not limited to UDP Echo probe, and other methods such as ICMP probe, TCP KeepAlive, two-way handshake detection or SD-WAN control signaling heartbeat mechanism can also be used. The link recovery process is also not limited to a specific algorithm or path calculation strategy. Any solution that can trigger path recalculation after link failure detection and complete service recovery belongs to the protection scope of the present disclosure.

[0106] In some embodiments, the network device is a customer premises equipment.

[0107] In a second aspect, the present disclosure provides a multi-service parallel routing method, which performs the routing control method of any one of the above on a plurality of data streams to be forwarded received via a network device in parallel, so as to determine a target link corresponding to each data stream of different service types.

[0108] In some embodiments, the multi-service parallel routing method further comprises: performing unified traffic scheduling on a plurality of data streams forwarded via the different target links, wherein the traffic scheduling comprises bandwidth allocation, priority sorting and queue scheduling strategy optimization.

[0109] Figure 3 is a schematic diagram of the multi-service parallel routing mechanism of the embodiments of the present disclosure. As Figure 3As shown, the mechanism mainly includes core functions such as application recognition, policy matching, link selection, and CPE routing controller.

[0110] It should be understood that Figure 3 The functional units in the above table are logical divisions, which can be implemented by software, hardware, or a combination thereof in actual implementation, and do not necessarily constitute independent physical modules.

[0111] When inbound traffic arrives at the CPE device, the data flow is first identified by application recognition. Application recognition can be based on deep packet inspection (DPI) technology, using a protocol feature library to quickly match and analyze data packets. This module supports more than 1000 protocol feature fingerprints and allows users to customize the feature library; the first packet recognition time is less than 15ms, and the service type recognition accuracy can reach more than 99.2%. Through this identification process, different business types such as video conferencing, financial transactions, and file transfers can be accurately distinguished.

[0112] Based on the identified business type, policy matching is performed. Policy matching selects a corresponding routing policy for each business type based on a pre-established business policy library (such as Figure 4 As shown), which can include but is not limited to video conference policy, financial transaction policy, file transfer policy, industrial control policy, and default policy. The policy library supports extension and customization according to different user or business needs, so its content is not limited to the five example policies shown in Figure 4 .

[0113] For example, the policies for different business types can include the following configuration rules: Video conference policy rules: Select the lowest latency link for the primary path, preferably a telecom 5G SA dedicated line; Enable QoS accelerated forwarding mechanism; Set the switching threshold, automatically switch to the alternative link when the target path latency exceeds 100ms to ensure real-time performance.

[0114] Financial transaction policy rules: Use a primary-backup dual-path hot backup mechanism, with the primary path being a dedicated line and the backup path being a 5G SA; Enable AES-256 end-to-end encryption during transmission; Support dual-transmission single-reception synchronization mechanism to achieve zero-packet loss guarantee.

[0115] File transfer policy rules: Implement multi-link load balancing, jointly using Unicom and mobile links; Support dynamic bandwidth aggregation to improve overall throughput capacity; LZ4 real-time compression algorithm can be enabled to improve transmission efficiency.

[0116] Policy matching generates corresponding link selection results based on the selected service policy. For example, video conferencing services select the path with the lowest latency, financial transaction services select a high-reliability dual path, and file transfer services select the path with the highest throughput. The link selection results are received and executed by the CPE routing controller. The controller can update the forwarding rules of network devices according to policy instructions, thereby routing different types of service traffic to the optimal link for transmission.

[0117] In this embodiment, the CPE routing controller can further collaborate with the unified traffic scheduling module to achieve resource coordination and dynamic scheduling in multi-service parallel scenarios. By monitoring and analyzing the real-time status information of multiple links (including bandwidth utilization, latency fluctuations, bit error rate, etc.), the controller can dynamically adjust the link allocation and forwarding priority of each service flow, thereby ensuring that different services can obtain network resources that match their quality of service requirements when transmitting in parallel.

[0118] In addition, the unified traffic scheduling module can globally coordinate and allocate resources for multiple service flows forwarded through different target links. Through bandwidth allocation, priority sorting and queue scheduling strategy optimization, it can achieve the best overall performance and transmission stability of the system in multi-service parallel scenarios.

[0119] Figure 4 A schematic diagram of the business strategy library structure according to an embodiment of this disclosure is shown. Figure 4 As shown, each policy in the business policy library includes a primary path, alternative paths, and optimization parameters. Different policies can share some configuration rules, or they can independently define QoS parameters, encryption algorithms, compression methods, or synchronization mechanisms according to business needs. The above policy rules are general principles and can be flexibly adjusted according to network architecture and business requirements during actual deployment.

[0120] In summary, the multi-service parallel routing mechanism disclosed herein, through service type identification, differentiated policy matching, and intelligent link control, can achieve optimal dynamic routing decisions in scenarios where multiple services are transmitted simultaneously. This mechanism breaks the limitations of the unified routing strategy of traditional CPE devices, realizing parallel routing control and adaptive resource optimization oriented towards service characteristics.

[0121] Figure 5 This is an overall architecture diagram of the routing control system according to an embodiment of this disclosure. Figure 5As shown, the system mainly includes: a multi-SIM card interface (A), a signal detection module (B), a quality analysis engine (C), a service policy engine (D), a decision execution unit (E), a CPE routing controller (F), an application recognition module (G), an LSTM prediction model (H), an operator feature library (I), and multiple communication paths corresponding to mobile links (J), Unicom links (K) and telecom links (L) respectively. In the CPE routing control core part, it also includes a link state database (M), a policy routing engine (N), a path calculation module (O) and a fault detection unit (P).

[0122] The multi-SIM card interface (A) is used to support the parallel access of multiple operator SIM cards, which can simultaneously access the 4G / 5G networks of different operators (such as mobile, Unicom and telecom) and realize multi-band coverage to provide multi-link access capability.

[0123] The signal detection module (B) is connected to the multi-SIM card interface (A) and is used for quality detection of the links corresponding to each SIM card. This module detects the quality parameters of multiple communication protocol layers at different periods to obtain real-time data reflecting the link performance. Physical layer signal quality indicators (such as RSRP, RSRQ and SINR) are detected every 200ms, and network layer quality indicators (such as delay and packet loss rate) and application layer quality indicators (such as throughput, connection success rate and protocol efficiency) are detected every 1s to obtain real-time data reflecting the link performance.

[0124] The input of the LSTM prediction model (H) is the real-time measurement data output by the signal detection module (B), including multi-dimensional signal features of the physical layer and the application layer. This model uses a trained recurrent neural network structure to predict the signal change trend in the future (such as 5 seconds) and outputs the predicted value of the future signal quality for subsequent quality evaluation and path optimization.

[0125] The operator feature library (I) is used to store the inherent features and long-term statistical feature data of different operators, including operator identification, network access type (such as 4G and 5G), historical bandwidth, delay range and other information, which provides long-term policy guidance and reference basis for the system.

[0126] The quality analysis engine (C) is used to fuse the real-time quality parameters from the signal detection module (B), the prediction results from the LSTM prediction model (H), and the historical feature data from the operator feature library (I) to form a comprehensive link quality evaluation result. This evaluation uses Figure 2 The three-dimensional quality model outputs a unified three-dimensional quality score by considering the three dimensions of the physical layer, the network layer and the application layer, thereby providing a reliable basis for policy decision.

[0127] The application recognition module (G) performs real-time analysis on data streams passing through the network device based on deep packet inspection (DPI) technology, recognizes the service application type (such as video conference, web browsing, file downloading, etc.) to which the data stream belongs, and outputs the corresponding service type identifier.

[0128] The service policy engine (D) performs policy matching and policy generation according to the service type identifier output by the application recognition module (G) and the link state information of the quality analysis engine (C). It converts service requirements (such as real-time performance, bandwidth, cost constraints) into specific switching instructions, such as allocating a telecom 5G link, ensuring a 2Mbps bandwidth, and a latency <100ms, etc. The switching instructions include link switching parameters, execution conditions, monitoring requirements, and rollback policies, etc.

[0129] The decision execution unit (E) is connected to the service policy engine (D) and is used to receive the switching instructions and drive the lower-level components to complete the update and switching of forwarding rules based on the instructions. At the same time, it tracks the state according to the monitoring requirements during the execution of the instructions and triggers the rollback policy when the conditions are met.

[0130] The CPE routing controller (F) is the core intelligent decision unit of the system, which has the ability of global optimization decision, predictive planning, meta-decision and knowledge learning. It is responsible for coordinating the work of each module, and under the synergistic action of the policy routing engine (N), the path calculation module (O) and the fault detection unit (P), it completes the comprehensive management and optimal decision of multiple links.

[0131] The link state database (M) is used to store real-time and historical link performance data, including time-varying indicators such as latency, jitter, packet loss rate, throughput, and connection success rate, which are used to support real-time routing decisions and short-term trend analysis. At the same time, the link state data can be used to update the operator feature library (I) after long-term analysis.

[0132] The policy routing engine (N) is responsible for service priority mapping and routing policy management. It determines the priority level and corresponding policy parameters according to the service type output by the application recognition module (G) and the preset service priority table, so as to realize the policy execution and forwarding control of data streams in milliseconds.

[0133] The path calculation module (O) calculates the optimal communication path based on the three-dimensional quality score output by the quality analysis engine (C), considering factors such as predicted quality, cost, reliability, and service priority, through a multi-objective optimization algorithm, to realize dynamic and intelligent routing of service streams.

[0134] The fault detection unit (P) performs real-time state monitoring on each operator link through a high-frequency heartbeat detection mechanism. For example, a UDP Echo probe packet (packet length 64 bytes, sending interval 20 ms, timeout time 15 ms) can be used to achieve about 50 ms level link fault detection. When two consecutive probe packets are not responded, it is determined that the link is faulty, and the system automatically selects a backup link to ensure business continuity.

[0135] In the embodiment, the CPE routing controller (F) realizes intelligent decision control across operators and multi-link by integrating real-time data, prediction results and policy information from multiple modules. When the quality of a link decreases or a fault occurs, the system can automatically select the optimal path among the mobile link (J), the Unicom link (K) and the Telecom link (L) to realize high reliability and high performance communication guarantee.

[0136] Those skilled in the art should understand that, Figure 5 Each functional module of the system can be realized by software logic, hardware circuit or a combination thereof.

[0137] Figure 6 is a composition schematic diagram of an electronic device according to an embodiment of the present disclosure.

[0138] In a third aspect, referring to Figure 6 The present embodiment provides an electronic device, which comprises: one or more processors 601; a memory 602 having one or more programs stored thereon, when the one or more programs are executed by the one or more processors, the one or more processors implement the routing control method or the multi-service parallel routing method of any one of the above; one or more I / O interfaces 603 connected between the processor and the memory, configured to realize information interaction of the processor and the memory.

[0139] The processor 601 is a device with data processing capability, including but not limited to a central processing unit (CPU) and the like; the memory 602 is a device with data storage capability, including but not limited to a random access memory (RAM, more specifically SDRAM, DDR, etc.), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a flash memory (FLASH); the I / O interface (read-write interface) 603 is connected between the processor 601 and the memory 602, and can realize information interaction of the processor 601 and the memory 602, including but not limited to a data bus (Bus) and the like.

[0140] In some embodiments, the processor 601, the memory 602 and the I / O interface 603 are connected to each other through the bus 604, and further connected to other components of the computing device.

[0141] In a fourth aspect, the embodiments of the present disclosure provide a computer readable medium, which stores a computer program, and the program is executed by a processor to implement any one of the routing control methods or the multi-service parallel routing method.

[0142] In a fifth aspect, the embodiments of the present disclosure provide a computer program product, which includes a computer program or instructions, and the computer program or instructions are executed by a processor to implement any one of the routing control methods or the multi-service parallel routing method.

[0143] The present disclosure provides a routing control method, and the technical key points include: A three-dimensional quality evaluation model is constructed, which cooperatively scores the indicators of the physical layer, the network layer and the application layer, and maps the original indicators to the interval [0, 100] through a linear mapping mode to obtain a comprehensive quality score Q (0-100 points), wherein Q = α × P + β × N + γ × A, P is the physical layer evaluation score, N is the network layer evaluation score, A is the application layer evaluation score, α, β, γ are corresponding weight coefficients, and α + β + γ = 1.

[0144] A routing controller is set up to integrate link management, policy routing, path calculation and fault detection and other functions, so as to realize dynamic monitoring and scheduling of different links.

[0145] Through the cooperative action of the three-dimensional quality evaluation model and the routing controller, multi-service parallel routing is supported, and differentiated quality protection is provided for video services, financial services and file transfer services.

[0146] The present disclosure realizes multi-service parallel routing based on service types by introducing service identification, prediction evaluation and policy matching mechanisms on the CPE device side. Compared with the single routing strategy for all service traffic in the prior art, the present disclosure can dynamically load differentiated policy rules for different service types such as video conference, financial transaction and file transfer, and make path calculation and switching decisions in combination with the prediction model and the quality evaluation result. Therefore, the present disclosure can not only provide differentiated protection of low latency, high reliability or high throughput for key services, but also make more reasonable choices among multiple operator links, reduce the uncertainty of link quality caused by frequent switching of operators, and effectively improve the resource utilization and user experience of the overall network.

[0147] The present disclosure provides an intelligent customer premises equipment (CPE) routing control method, and constructs an automatic routing control architecture integrating prediction, evaluation, decision and execution. Through the cooperative action of the core modules and functional mechanisms, dynamic, accurate and stable link selection and control in a multi-link access environment are realized.

[0148] The prediction engine based on a long short-term memory (LSTM) network is introduced in the present disclosure, which can make high-precision prediction (the prediction accuracy can reach about 92.3%) on the signal trend in a future preset time window based on historical link quality data, and provide a forward-looking reference basis for path selection.

[0149] To accurately evaluate the carrying capacity of different links, the present disclosure proposes a three-dimensional quality evaluation model, which comprehensively considers multi-dimensional quality indicators of the physical layer, network layer and application layer. After the single-layer score is calculated for each level indicator, the weight of each dimension is dynamically adjusted according to the business type or preset scene, and a comprehensive quality score (Q value) is generated. The score can comprehensively reflect the comprehensive performance of the link and provide a quantitative basis for subsequent decision-making.

[0150] The routing controller designed in the present disclosure is the core decision-making unit, which integrates link state management, policy routing, path calculation and fault detection functions, and can coordinate resource scheduling and path planning from a global perspective to ensure the unity and optimality of routing decision-making. The routing controller also has fault tolerance and closed-loop feedback mechanisms, which can dynamically adjust policy parameters according to the execution results and changes in network state.

[0151] In addition, the present disclosure supports a multi-service parallel routing mechanism. Through deep packet inspection (DPI) technology, the business type to which the data flow belongs is identified, and combined with the pre-set business policy template, differentiated routing strategies and quality of service guarantee targets are matched for typical businesses such as video conferencing, financial transactions and file transfer, so as to realize the priority protection and performance optimization of key businesses at the link resource level.

[0152] Through the cooperative application of the above technical means, the present disclosure can significantly improve the overall performance of customer premises equipment in a multi-operator link environment. For example, in a video conference scenario, seamless link switching can be achieved within milliseconds; in a financial transaction scenario, high-reliability transmission close to zero packet loss can be achieved; in a file transfer scenario, the average throughput can be effectively improved to 12 Mbps, thereby comprehensively enhancing the stability of the network and the user experience.

[0153] In order for those skilled in the art to more clearly understand the technical solutions provided by the embodiments of the present disclosure, the technical solutions provided by the embodiments of the present disclosure are described in detail below through specific embodiments: Embodiment one This embodiment takes an enterprise video conference scenario as an example to illustrate the application process of the present disclosure in a low-latency business scenario.

[0154] In this embodiment, the service type of the data flow is identified as video conference service. According to the preset video conference policy rule in the service policy library, the system configures the corresponding link evaluation threshold condition for the service, which includes: the jitter threshold is 30 ms; the threshold of the comprehensive quality score Q is 70; and a low-latency link (such as a telecom 5G SA link) is preferentially selected.

[0155] It is monitored in real time that the jitter prediction value of the current Unicom link rises to 35 ms, which has exceeded the preset threshold 30 ms, and the comprehensive quality score Q calculated in real time is 68, which is lower than 70 defined in the policy rule, so the system determines that the service quality of the current primary link has a significant downward trend.

[0156] Subsequently, the quality trend of the link in the next 5 seconds is analyzed by an LSTM (Long Short-Term Memory) prediction model, and the prediction result shows that the comprehensive quality score will further decrease to 62. Based on the prediction result, the path re-evaluation process is started, and it is calculated that the real-time jitter of the telecom 5G link at the current time is 15 ms, and the predicted comprehensive quality score Q is 85, which is obviously better than the current link.

[0157] According to the video conference policy rule, the corresponding switching instruction is generated and the forwarding configuration is updated. Through the MPTCP (Multi-Path Transmission Control Protocol) mechanism, seamless switching of the service flow from the Unicom link to the telecom 5G link is realized, and the whole switching process is completed within about 280 ms, which will not cause interruption or obvious lag of the video conference.

[0158] As can be seen from this embodiment, the routing control method of the present disclosure can dynamically optimize the link for real-time services such as video conferences under the synergistic effect of service quality prediction and policy matching. When the quality of the current link decreases, the system can trigger link switching in advance based on the prediction model, realize low-latency and seamless service guarantee, and thus effectively improve the service quality and user experience of real-time services.

[0159] Those of ordinary skill in the art will realize and understand that all or some of the steps in the methods disclosed above and the functional modules / units in the systems and devices can be implemented as software, firmware, hardware, and appropriate combinations thereof. In hardware implementation, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, one physical component can have multiple functions, or one function or step can be performed by several physical components in cooperation. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on computer-readable media, which can include computer storage media (or non-transitory media) and communication media (or transitory media). As is well known to those of ordinary skill in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by a computer. Further, it is common knowledge to those of ordinary skill in the art that communication media typically embodies computer readable instructions, data structures, program modules or other data in a modulated data signal such as a carrier wave or other transport mechanism and includes any information delivery media.

[0160] Example embodiments have been disclosed herein and, although the use of specific terms is expressly used herein, they are intended in a generic sense only and are not intended to limit the scope of the present disclosure. In some instances, it will be apparent to those skilled in the art that features, characteristics or / and elements described in connection with a particular embodiment can be used in conjunction with other embodiments unless otherwise explicitly stated. As such, those skilled in the art will appreciate that various changes can be made in form and detail without departing from the scope of the disclosure as set forth in the appended claims.

Claims

1. A routing control method, applied to a network device with multi-link access capability, comprising: The network device performs service identification on the data stream to be forwarded in order to obtain the service type information of the data stream; Obtain evaluation metrics for multiple candidate links of the network device; The multiple candidate links are evaluated based on the evaluation indicators and the business type information to obtain a comprehensive quality score; Based on the business type information, the evaluation indicators, and the comprehensive quality score, a target link is selected from the multiple candidate links; Control the network device to transmit the data stream through the target link.

2. The routing control method according to claim 1, wherein, The evaluation metrics for multiple candidate links of the network device are obtained, including: The real-time quality indicators of multiple candidate links of the network device are periodically collected, and the real-time quality indicators include at least the physical layer, network layer and application layer quality indicators. Based on the real-time quality indicators, the predicted quality indicator values ​​of the multiple candidate links are predicted at a future preset time using a prediction model. Obtain network feature data for the multiple candidate links.

3. The routing control method according to claim 2, wherein, The multiple candidate links are evaluated based on the evaluation metrics and the business type information to obtain a comprehensive quality score, including: The real-time quality index, the predicted value of the quality index, and the network feature data are fused using a fusion algorithm to generate a link evaluation vector; Based on the link evaluation vector and the service type information, the comprehensive quality score is generated using a three-dimensional quality evaluation model.

4. The routing control method according to claim 3, wherein, Based on the link evaluation vector and the service type information, a three-dimensional quality evaluation model is used to generate the comprehensive quality score, including: The link evaluation vector is input into the three-dimensional quality evaluation model; The physical layer evaluation module, network layer evaluation module, and application layer evaluation module in the model evaluate the data in the corresponding dimension of the vector respectively to obtain the physical layer evaluation score P, the network layer evaluation score N, and the application layer evaluation score A. The weighted fusion module in the model calculates and outputs the comprehensive quality score Q according to the formula Q=α×P+β×N+γ×A. Wherein, α, β, and γ are weighting coefficients determined based on the business type information, and satisfy α+β+γ=1.

5. The routing control method according to claim 4, wherein, The physical layer evaluation module, network layer evaluation module, and application layer evaluation module in the model evaluate the data in the corresponding dimensions of the vector, respectively, to obtain the physical layer evaluation score P, the network layer evaluation score N, and the application layer evaluation score A, including: The physical layer evaluation module extracts physical layer-level quality indicators from the link evaluation vector. The quality indicators include reference signal received power, reference signal received quality, signal-to-interference-plus-noise ratio, and cell handover frequency. Based on the quality indicators, the physical layer evaluation score P is calculated using a weighted normalization method. The network layer evaluation module extracts quality indicators of the network layer dimension from the link evaluation vector. The quality indicators include latency, jitter and packet loss rate. Based on the quality indicators, the network layer evaluation score N is calculated using a weighted normalization method. The application layer evaluation module extracts application layer-level quality indicators from the link evaluation vector. These quality indicators include throughput, connection success rate, and protocol efficiency. Based on these quality indicators, the application layer evaluation score A is calculated using a weighted normalization method.

6. The routing control method according to claim 5, wherein, Based on the business type information, the evaluation indicators, and the comprehensive quality score, a target link is selected from the multiple candidate links, including: Obtain the routing strategy corresponding to the business type information, wherein the routing strategy defines the target evaluation indicators and their constraints related to the business type; Based on the routing strategy, multi-objective optimization calculations are performed on the target evaluation indicators and comprehensive quality scores of the multiple candidate links to determine the target link.

7. The route selection control method according to claim 6, based on the route selection strategy, performs multi-objective optimization calculations on the target evaluation indicators and comprehensive quality scores of the multiple candidate links to determine the target link, including: Based on the constraints in the routing strategy, the multiple candidate links are filtered to obtain a candidate subset, wherein the constraints include at least a threshold requirement for the target evaluation index. Multi-objective optimization calculations are performed on the target evaluation indicators and comprehensive quality scores of each link in the candidate subset to obtain the comprehensive score of each link; Select the link with the highest overall score as the target link.

8. The routing control method according to claim 6, wherein, Based on the routing strategy, multi-objective optimization calculations are performed on the target evaluation indicators and comprehensive quality scores of the multiple candidate links to determine the target link, including: Based on the constraints in the routing strategy, the multiple candidate links are filtered to obtain a candidate subset, wherein the constraints include at least a threshold requirement for the target evaluation index. From the candidate subset, the link with the highest overall quality score is selected as the target link.

9. The routing control method according to any one of claims 1-8, wherein, Controlling the network device to transmit the data stream through the target link includes: Generate a handover instruction, the handover instruction containing the forwarding configuration corresponding to the target link; Update the forwarding rules of the network device according to the switching instruction; According to the updated forwarding rules, the data stream is routed to the target link for transmission.

10. The routing control method according to claim 9, wherein, Also includes: The target link is subjected to millisecond-level heartbeat detection to monitor the link connectivity status; When the target link is determined to have failed based on the heartbeat detection result, a new target link is determined from the remaining candidate links to perform fast recovery.

11. A multi-service parallel routing method, comprising: For multiple data streams to be forwarded received via network devices, the routing control method as described in any one of claims 1 to 10 is executed in parallel to determine the corresponding target link for each data stream of different service types.

12. The multi-service parallel routing method according to claim 11, wherein, Also includes: Unified traffic scheduling is performed on multiple data streams forwarded via different target links. The traffic scheduling includes bandwidth allocation, priority sorting, and queue scheduling strategy optimization.

13. An electronic device, comprising: One or more processors; A memory having stored one or more programs that, when executed by one or more processors, cause the one or more processors to implement the routing control method according to any one of claims 1 to 10 or the multi-service parallel routing method according to any one of claims 11 to 12.

14. A computer-readable medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the routing control method according to any one of claims 1 to 10 or the multi-service parallel routing method according to any one of claims 11 to 12.

15. A computer program product comprising a computer program or instructions, wherein the computer program or instructions, when executed by a processor, implement the routing control method according to any one of claims 1 to 10 or the multi-service parallel routing method according to any one of claims 11 to 12.

Images