QUIC multipath load balancing method and system based on path quality dynamic evaluation

Abstract

The invention relates to the technical field of data communication, in particular to a QUIC multi-path load balancing method and system based on path quality dynamic evaluation, and the method comprises the steps: forming a network transmission layout according to network communication, the network transmission layout comprises a plurality of network paths, and monitoring and collecting the performance index data of each network path; analyzing the performance index data, and determining a quality score of each network path; inputting the performance index data into an exponentially weighted moving average model, outputting prediction index data, and obtaining a prediction quality score of each network path; carrying out dynamic load distribution based on the predicted quality score, and balancing the load; and a preset evaluation threshold value is compared with the predicted quality score, a network path switching signal is correspondingly triggered, and network path recovery processing is carried out. The flow distribution is adjusted in advance by predicting the prediction quality score of the network path, so that the path switching delay is obviously reduced, and the transmission performance reduction caused by network abrupt change is avoided.

Description

Technical Field

[0001] This invention relates to the field of data communication technology, and specifically to a QUIC multipath load balancing method and system based on dynamic path quality assessment. Background Technology

[0002] In modern society, the internet permeates all sectors, and various applications rely on high-quality network connections. Network transmission interruptions or performance fluctuations can affect user experience and cause economic losses, such as in real-time video conferencing. Therefore, the reliability and efficiency of network transmission are crucial to ensuring the stability of internet services and user experience. With the development of mobile internet and real-time interactive applications, network communication faces greater challenges. Data transmission quality is affected by various dynamic factors, making it more complex to achieve efficient, low-latency multipath transmission in complex network environments. Improving multipath transmission efficiency can enhance bandwidth utilization, reduce latency, and strengthen the robustness and continuity of network services.

[0003] With the diversification of network access methods and the presence of multiple network interfaces on terminal devices, the collaborative utilization of multiple paths to improve transmission performance has become a research hotspot. The QUIC protocol (Quick UDP Internet Connections) provides natural support for multipath transmission, and its multipath transmission method utilizes multiple links collaboratively through mechanisms such as path management. Multipath transmission methods have evolved from the traditional MP-TCP (Multipath TCP) to MP-QUIC (Multipath QUIC). Early MP-TCP deployments were complex and had poor compatibility, while MP-QUIC, based on the QUIC protocol, is more adaptable to modern network environments and has received widespread attention. Among these, MP-QUIC, as an emerging multipath solution, can effectively utilize the bandwidth of multiple paths through parallel multi-stream transmission; however, a single path selection mechanism may not be effective when faced with significant fluctuations in network quality.

[0004] To overcome this problem, many studies have proposed dynamic scheduling methods based on path quality assessment. For example, load balancing strategies based on real-time RTT and packet loss rate can dynamically adjust traffic proportions according to network conditions. Furthermore, research shows that combining multi-indicator comprehensive evaluation with dynamic weight adjustment can accurately reflect the true transmission capacity of a path and improve scheduling performance. Therefore, in QUIC multipath transmission, quality assessment + dynamic scheduling is an effective architecture. The assessment module monitors path status, the scheduling module allocates traffic, and a predictive mechanism responds to network changes in advance, improving transmission efficiency and stability. Especially in wireless networks, it can adapt to link quality fluctuations and provide high-quality service.

[0005] In practical operation, the real-time performance and accuracy of path quality assessment and load allocation are critical to the performance of multipath transmission systems. Existing methods mostly rely on current or historical indicators and lack the ability to predict future network conditions, leading to lag in scheduling decisions. For example, when path signals attenuate, traditional methods wait until the quality deteriorates significantly before switching, resulting in increased latency and throughput fluctuations. In addition, static thresholds or fixed weight strategies are difficult to adapt to complex networks and are prone to problems such as uneven load distribution and switching oscillations. Summary of the Invention

[0006] To address the limitations of existing QUIC multipath transmission technologies in load balancing and path switching, particularly the technical problems of uneven path load distribution, high switching latency, and insufficient adaptability to dynamic network changes in traditional schemes, this invention aims to provide a QUIC multipath load balancing method based on dynamic path quality assessment. The specific technical solution adopted is as follows:

[0007] A network transmission layout is formed based on network communication. The network transmission layout includes multiple network paths. The performance index data of each network path is monitored and collected.

[0008] Analyze performance metrics data to determine the quality score for each network path;

[0009] Obtain the exponentially weighted moving average model, input the performance index data into the exponentially weighted moving average model, output the prediction index data, and obtain the prediction quality score for each network path.

[0010] Dynamic load distribution is performed based on the predicted quality score to balance the load; and a preset evaluation threshold is compared with the predicted quality score to trigger a network path switching signal and perform network path recovery processing.

[0011] Preferably, the performance metrics data include data information corresponding to round-trip time, packet loss rate, and bandwidth utilization.

[0012] Preferably, performance index data is analyzed to determine the quality score of each network path, including:

[0013] Based on the performance index data, normalization processing is performed to determine the sub-scores for each round-trip time, packet loss rate, and bandwidth utilization in each network path.

[0014] Each sub-score is assigned a pre-defined weight, and the quality score of the corresponding network path is obtained by combining the weights and sub-scores.

[0015] Preferably, an exponentially weighted moving average model is obtained. Performance index data is input into the exponentially weighted moving average model, and prediction index data is output to obtain the prediction quality score for each network path, including:

[0016] Input the performance index data corresponding to the current moment of each network path into the exponentially weighted moving average model to predict the performance index data corresponding to the next adjacent moment as the predicted index data;

[0017] The prediction quality score is output based on the prediction index data combined with weights.

[0018] Preferably, dynamic load allocation is performed based on the predicted quality score to balance the load, specifically as follows:

[0019] The traffic allocation weight for each network path is determined based on the predicted quality score, and the total traffic demand in the network transmission layout is obtained. The traffic allocation weight and the total traffic demand are then integrated to obtain the traffic allocated to each network path.

[0020] Preferably, a preset evaluation threshold is compared with the predicted quality score, triggering a network path switching signal and performing network path recovery processing, including:

[0021] Define the network path corresponding to the predicted quality score of the current analysis as the target path, preset the evaluation threshold, and compare it with the predicted quality score. If the predicted quality score is less than the estimated threshold, it means that the quality of the target path has deteriorated, triggering the target path switching signal, resetting the predicted quality score, and dynamically redistributing the load accordingly.

[0022] Once the predicted quality score of the target path returns to normal, the target path will be reinstated into the load balancing process, and the traffic allocation weight for the target path will be gradually increased until the predicted quality score is greater than or equal to the estimated threshold.

[0023] To address the aforementioned issues, this invention also provides a QUIC multi-path load balancing system based on dynamic path quality assessment. The system includes a processor, a communication interface, a memory, and a communication bus. The processor, communication interface, and memory communicate with each other via the communication bus. The processor invokes logical instructions from the memory to execute the QUIC multi-path load balancing method based on dynamic path quality assessment described in any of the preceding embodiments.

[0024] The present invention has the following beneficial effects:

[0025] 1. For multiple network paths in the network transmission layout, performance index data is collected to determine the quality score of each network path. An exponentially weighted moving average model is introduced to dynamically predict the relevant performance index data of the network path, which facilitates proactive load switching and allocation based on the predicted quality score, reducing switching latency. Compared with traditional static evaluation methods, by predicting the quality score of the network path to assess the future trend of quality changes and adjust traffic allocation in advance, the path switching latency is significantly reduced, avoiding the transmission performance degradation caused by network mutations. It can not only accurately capture the dynamic characteristics of the path, but also respond flexibly to complex network fluctuations, greatly improving the continuity and low latency of transmission, and solving the problems of uneven path load distribution and high switching latency in existing technologies. By determining the quality score, predicting the quality score, and dynamically allocating the load, path utilization, transmission latency, and fault tolerance can be optimized simultaneously, making it particularly suitable for high-requirement business scenarios such as real-time video transmission.

[0026] Furthermore, this method significantly outperforms the traditional MP-QUIC scheme in terms of switching latency, throughput stability, and video stuttering rate. It is suitable for real-time service transmission tasks in the mobile Internet and can provide end users with a smoother and more stable network experience in fields such as live video streaming, online conferencing, and cloud gaming. It helps improve the intelligence level of multi-path transmission and has important theoretical significance and practical application value.

[0027] 2. The QUIC multi-path load balancing system based on dynamic path quality assessment provided by this invention has the same beneficial effects as the QUIC multi-path load balancing method based on dynamic path quality assessment provided by this invention, and will not be elaborated here. Attached Figure Description

[0028] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 A flowchart illustrating the steps of a QUIC multi-path load balancing method based on dynamic path quality assessment, provided in an embodiment of the present invention.

[0030] Figure 2 This is a flowchart illustrating the implementation of a QUIC multi-path load balancing method based on dynamic path quality assessment, as provided in one embodiment of the present invention. Detailed Implementation

[0031] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of a QUIC multi-path load balancing method and system based on dynamic path quality assessment proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0032] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0033] The following description, in conjunction with the accompanying drawings, details the specific scheme of the QUIC multi-path load balancing method and system based on dynamic path quality assessment provided by this invention.

[0034] To better illustrate, the QUIC protocol, or Quick UDP Internet Connections, is a next-generation transport layer network protocol developed under the leadership of Google. Based on UDP (User Datagram Protocol), it integrates TLS (Transport Layer Security) encryption to achieve data transmission security and privacy protection. It aims to reduce connection establishment latency, improve congestion control, and natively support multipath transmission, thereby enhancing the robustness and flexibility of network communication.

[0035] During network communication, the quality of data transmission is affected by a variety of dynamic factors, such as network congestion, signal strength variations, multipath interference, and uneven node processing capabilities. This makes achieving efficient and low-latency multipath transmission in complex network environments even more challenging. In modern society, the Internet has permeated all aspects of production and life, from remote work and online education to real-time video communication and cloud computing services, all of which rely on high-quality network connections. Therefore, multipath transmission technology is an important means to cope with network fluctuations and improve transmission reliability.

[0036] Among diverse network access methods, terminal devices typically possess multiple network interfaces, such as Wi-Fi (Wireless Fidelity) and 5G (5th Generation Mobile Networks), and collaboratively utilize multiple paths to improve transmission performance. However, existing multi-path scheduling strategies, such as polling and minimum RTT (Round-Trip Time First), are insufficient in terms of dynamic adaptability, predictive ability, and real-time decision-making, resulting in limited transmission performance optimization and difficulty in meeting the stringent requirements of real-time applications for low latency and high stability.

[0037] Therefore, in real-time video transmission scenarios, terminal devices and / or access devices simultaneously connect to cellular networks such as Wi-Fi and 5G. However, traditional MP-QUIC solutions suffer from high latency (greater than 200ms) during path switching, leading to video stuttering or quality degradation. Therefore, a QUIC multi-path load balancing method based on dynamic path quality assessment is proposed, specifically addressing the network switching problem for mobile devices in real-time video transmission scenarios such as live streaming or video conferencing. A path quality trend prediction model is introduced, using an exponentially weighted moving average model to dynamically predict the performance indicators corresponding to network paths, enabling proactive load switching and reducing switching latency.

[0038] Please combine Figure 1 and Figure 2 It illustrates a flowchart and implementation flowchart of a QUIC multi-path load balancing method based on dynamic path quality assessment provided in the first embodiment of the present invention. The method includes:

[0039] Step S1: Form a network transmission layout based on network communication. The network transmission layout includes multiple network paths. Monitor and collect performance index data for each network path.

[0040] Step S2: Analyze the performance index data and determine the quality score for each network path;

[0041] Step S3: Obtain the exponentially weighted moving average model, input the performance index data into the exponentially weighted moving average model, output the prediction index data, and obtain the prediction quality score for each network path.

[0042] Step S4: Perform dynamic load distribution based on the predicted quality score to balance the load; and set an evaluation threshold, compare it with the predicted quality score, trigger the network path switching signal accordingly, and perform network path recovery processing.

[0043] As an optional implementation, in this embodiment, the method is specifically applied in real-time video transmission scenarios. When the Wi-Fi signal of a mobile device is unstable, for example, when the signal is attenuated due to user movement, the device can switch to the 5G network in advance to ensure the continuity and low latency of video transmission.

[0044] It can be explained that in step S1, in the relevant scenario of the current study, a network transmission layout is formed according to network communication. That is, based on the actual needs and application scenarios of network communication, multiple network paths are established. Each path carries a specific data transmission task, which together support the efficient operation and stable communication of the entire network, so as to form a corresponding network transmission layout.

[0045] Optionally, network path refers to the data transmission channel used by a device when connecting to the Internet, such as Wi-Fi or 5G, or other wired or wireless access methods, which together form the basic architecture of network communication.

[0046] Furthermore, the performance metrics data include round-trip time, packet loss rate, and bandwidth utilization.

[0047] Specifically, each network path in the network transmission layout is monitored in real time, and key network indicators, i.e., performance indicator data, are collected. Preferably, the data collection frequency is set to once every 100 milliseconds to ensure real-time performance. All performance indicator data is stored in a local buffer for subsequent calculations.

[0048] Round-trip time (RTT) is the time difference between sending and receiving data. The device sends one or more QUIC probe packets to the target server and records the sending time. When the target server receives the probe packet, it immediately returns a response. When the client receives the response packet, it calculates the time difference between sending and receiving based on the timestamp information carried by the response packet, thus obtaining a single RTT value. This process is repeated multiple times within the collection period to obtain multiple RTT values, and the average RTT value is calculated as the round-trip time data.

[0049] Packet loss rate, or L, is the difference between the number of packets sent and the number of acknowledgments within the statistical collection period. This difference is divided by the number of packets sent to obtain the packet loss rate. Bandwidth utilization rate, or U, is the amount of data transmitted within the measurement collection period divided by the estimated maximum bandwidth of the network path. It is obtained through bandwidth probing tools such as iperf (Internet Performance) to obtain bandwidth utilization rate information.

[0050] Further, step S2 includes:

[0051] Step S21: Normalize the performance index data to determine the sub-scores for each round-trip time, packet loss rate, and bandwidth utilization in each network path.

[0052] Specifically, in this embodiment, the network transmission layout is taken as the first The network path is described in detail, with the current time denoted as... The obtained performance index data is normalized, and the corresponding calculation formula is as follows:

[0053]

[0054] in, Indicates the current moment At that time, the first A sub-score of the round-trip time for each network path; Indicates An exponential function with base π, i.e., the natural constant; Indicates the current moment At that time, the first The measured round-trip time of the network path; Indicates the base round-trip time.

[0055] It can be explained that the base round-trip time It is used to reflect typical network latency. It is used in an ideal environment where the network connection is stable, the transmission path is typical and the load is relatively balanced. That is, in the absence of obvious congestion or interference, and avoiding the impact of peak periods or abnormal routing, it can truly reflect the performance of the network under normal operating conditions, so as to measure a stable and reliable reference round-trip time.

[0056]

[0057] in, Indicates the current moment At that time, the first Sub-scores of packet loss rate for each network path; Indicates An exponential function with base 0; Indicates the current moment At that time, the first The measured packet loss rate of the network path; This represents the baseline packet loss rate.

[0058] What can be explained is the baseline packet loss rate. The baseline packet loss rate is determined by comparing the total number of packets sent and received. It is used to amplify the impact of packet loss and is measured in a highly standardized network environment with controlled conditions to minimize the influence of external variables on the test results.

[0059]

[0060] in, Indicates the current moment At that time, the first A sub-rating of network path bandwidth utilization; Indicates the current moment At that time, the first Measured bandwidth utilization of each network path; This indicates the maximum available bandwidth of the network path.

[0061] It can be stated that the maximum available bandwidth It is the maximum value determined by filtering based on the bandwidth utilization rate after the initial detection of all network paths in the network transmission layout, so as to accurately reflect the upper limit of network performance under the current conditions and to evaluate data transmission capacity and network load status.

[0062] Step S22: Preset weights for each sub-score, and combine the weights and sub-scores to obtain the quality score of the corresponding network path.

[0063] Specifically, still using the first Each network path is described, and after setting preset weights, the quality score of the network path is obtained by combining all sub-scores. The corresponding calculation formula is as follows:

[0064]

[0065] in, Indicates the current moment At that time, the first Quality score of each network path; , , All represent weighting coefficients; Indicates the current moment At that time, the first A sub-score of the round-trip time for each network path; Indicates the current moment At that time, the first Sub-scores of packet loss rate for each network path; Indicates the current moment At that time, the first A sub-rating of bandwidth utilization for each network path.

[0066] It can be explained that, in actual operation, Preferably, , , To emphasize the impact of latency and packet loss rate on network transmission scenarios such as video transmission, it can be adjusted according to the actual situation; according to the first Similarly, the quality score of all network paths in the network transmission layout is determined.

[0067] Furthermore, step S3 includes:

[0068] Step S31: Input the performance index data corresponding to the current moment of each network path into the exponentially weighted moving average model to predict the performance index data corresponding to the next adjacent moment as the predicted index data.

[0069] To better illustrate this, a path quality trend prediction model is adopted, which uses the Exponentially Weighted Moving Average (EWMA) algorithm to dynamically predict the performance indicators of network paths in the network transmission layout. This enables proactive judgment of the future state of network paths, avoiding the drawbacks of traditional solutions that rely solely on current static indicators. This allows the system to exhibit better foresight and adaptability when facing network fluctuations. In terms of load balancing, a comprehensive quality scoring mechanism is constructed, which comprehensively considers multiple dimensions such as round-trip latency, packet loss rate, and bandwidth utilization.

[0070] Preferably, in this embodiment, the EWMA prediction model is optimized by adopting a nonlinear smoothing factor adjustment strategy to dynamically adjust the prediction sensitivity based on the stability of the network state, thereby avoiding over-response of the model in a stable network environment.

[0071] Specifically, based on the current moment of the aforementioned steps To obtain the predicted index data for the next adjacent time step, the corresponding calculation formula is:

[0072]

[0073] in, , They represent the next adjacent time intervals. and the current moment At that time, the first Predicted round-trip time for each network path; Indicates the smoothing factor; Indicates the current moment At that time, the first The measured round-trip time of the network path.

[0074] It can be explained that the smoothing factor , refers to a key parameter used to control the weight of historical data in the current forecast. It directly affects the sensitivity and response speed of the EWMA forecast model to data changes. In other words, it can dynamically change according to the real-time fluctuation characteristics of the data, flexibly adapt to complex time series patterns, and improve the forecast performance in volatile environments.

[0075]

[0076] in, , They represent the next adjacent time intervals. and the current moment At that time, the first Predicted packet loss rate for each network path; Indicates the smoothing factor; Indicates the current moment At that time, the first The measured packet loss rate of the network path.

[0077]

[0078] in, , They represent the next adjacent time intervals. and the current moment At that time, the first Predicted bandwidth utilization of each network path; Indicates the smoothing factor; Indicates the current moment At that time, the first The measured bandwidth utilization of each network path.

[0079] It should be noted that, in actual operation, the relevant performance index data at the initial moment are as follows: , , .

[0080] Step S32: Output a prediction quality score based on the prediction index data and weights.

[0081] Specifically, based on the weights in step S22 and the prediction index data in step S31, the next adjacent time step is determined. The prediction quality score for each network path is calculated using the following formula:

[0082]

[0083] in, Indicates at time At that time, the first Predicted quality score for each network path; , , All represent weighting coefficients; Indicates An exponential function with base 0; Indicates at time At that time, the first Predicted round-trip time for each network path; Indicates the base round-trip time; Indicates at time At that time, the first Predicted packet loss rate for each network path; Indicates the baseline packet loss rate; Indicates at time At that time, the first Predicted bandwidth utilization of each network path; This indicates the maximum available bandwidth of the network path.

[0084] It can be explained that by comprehensively predicting the data of the predictive indicators to obtain the predicted quality score of the corresponding network path, the entire network transmission layout can be prepared in advance to respond to network changes; according to the first Similarly, the prediction quality scores for all network paths are obtained.

[0085] Furthermore, in step S4, dynamic load allocation is performed based on the predicted quality score to balance the load, specifically as follows:

[0086] The traffic allocation weight for each network path is determined based on the predicted quality score, and the total traffic demand in the network transmission layout is obtained. The traffic allocation weight and the total traffic demand are then integrated to obtain the traffic allocated to each network path.

[0087] Specifically, firstly, based on the current moment The next adjacent time The corresponding predicted quality score determines the first The traffic allocation weights for each network path are calculated using the following formula:

[0088]

[0089] in, Indicates the current moment At that time, the first Traffic allocation weights for each network path; Indicates at time At that time, the first Predicted quality score for each network path; This indicates the number of network paths in the network transmission layout.

[0090] Next, in this embodiment, the total traffic requirement in the network transmission layout is defined as follows: For example, in live streaming, it refers to the data rate of the video stream; then, the traffic allocation weight and total traffic demand are combined to determine the... The formula for calculating the traffic that should be allocated to each network path is as follows:

[0091]

[0092] in, Indicates the current moment At that time, the first Traffic allocated to each network path; This indicates the total traffic demand.

[0093] It can be explained that during the actual operation of the network, traffic is allocated according to the network path. By dynamically adjusting the number of data packets sent on the corresponding network path, efficient traffic allocation and scheduling can be achieved, effectively avoiding the problem of some network paths being overloaded while others are idle, thus ensuring load balance and stable operation.

[0094] Further, in step S4, a preset evaluation threshold is compared with the predicted quality score, triggering a network path switching signal and performing network path recovery processing, including:

[0095] Step S41: Define the network path corresponding to the predicted quality score of the current analysis as the target path, preset the evaluation threshold, and compare it with the predicted quality score. If the predicted quality score is less than the estimated threshold, it indicates that the quality of the target path has deteriorated, triggering the target path switching signal, resetting the predicted quality score, and re-performing dynamic load redistribution accordingly.

[0096] As explained above, based on the description of the relevant steps, in this embodiment, the first... The network path will be denoted as the target path for further explanation.

[0097] Specifically, the evaluation threshold is denoted as This represents the threshold for network path quality degradation, which is set based on experimental experience. Taking the target path as an example, if the predicted quality score of the target path meets the evaluation threshold... , This represents the maximum value obtained from the predicted quality scores of all network paths; when this condition is met, it represents the target path, i.e., the i-th path. The quality of a network path is about to deteriorate, triggering an active switchover to send a corresponding switchover signal, reduce abrupt changes, gradually reduce the weight of the target path, and temporarily set the predicted quality score of the target path to... , Denotes a constant, and Based on the predicted quality score, the traffic allocation weight of the corresponding network path is recalculated to reduce the traffic allocation of the target path and increase the load of other network paths, so as to achieve smooth switching and ensure accurate adaptation to the fluctuation characteristics of mobile multi-network paths.

[0098] It can be explained that setting an evaluation threshold, which is based on the threshold adaptive mechanism of control theory and dynamically adjusts the evaluation threshold according to the historical handover success rate, improves the robustness of the entire network transmission layout in different network environments and ensures the stability of the load balancing strategy. Combining quality scoring, predicted quality scoring and dynamic load allocation to simultaneously handle the current network state and future changing trends significantly improves the performance of multi-path transmission. Especially in latency-sensitive scenarios such as video transmission on mobile devices, it can better adapt to rapid changes in network conditions. In addition, with the active handover mechanism, it not only reduces network path handover latency but also improves the continuity of data transmission, enabling flexible response to complex network fluctuations and significantly reducing video stuttering, providing important quality assurance for real-time video transmission applications.

[0099] Step S42: Once the predicted quality score of the target path returns to normal, the target path is reintegrated into the load distribution, and the traffic distribution weight of the target path is gradually increased until the predicted quality score is greater than or equal to the estimated threshold.

[0100] Specifically, based on step S41, it is known that the target path has deteriorated. During network operation, once the predicted quality score of the target path recovers, i.e., the predicted quality score meets the requirements... The target path is reintegrated into the load distribution, and to avoid oscillations, a gradual increase mechanism is adopted for the traffic distribution weight of the network path. That is, its traffic distribution weight is restored in a linear manner in subsequent running cycles, increasing the original traffic distribution weight by 10% every 100ms until the target path reaches a normal prediction quality score.

[0101] Understandably, for multiple network paths in a network transmission layout, performance index data is collected to determine the quality score of each network path. An exponentially weighted moving average model is introduced to dynamically predict the relevant performance index data of the network path, facilitating proactive load switching and allocation based on the predicted quality score, thus reducing switching latency. Compared to traditional static evaluation methods, predicting the quality score of the network path to assess future quality trends and adjust traffic allocation in advance significantly reduces path switching latency and avoids transmission performance degradation caused by network mutations. It can not only accurately capture the dynamic characteristics of the path but also flexibly respond to complex network fluctuations, greatly improving transmission continuity and low latency, and solving the problems of uneven path load distribution and high switching latency in existing technologies. By determining the quality score, predicting the quality score, and dynamically allocating the load, path utilization, transmission latency, and fault tolerance can be optimized simultaneously, making it particularly suitable for high-requirement business scenarios such as real-time video transmission.

[0102] Furthermore, this method significantly outperforms the traditional MP-QUIC scheme in terms of switching latency, throughput stability, and video stuttering rate. It is suitable for real-time service transmission tasks in the mobile Internet and can provide end users with a smoother and more stable network experience in fields such as live video streaming, online conferencing, and cloud gaming. It helps improve the intelligence level of multi-path transmission and has important theoretical significance and practical application value.

[0103] To better illustrate and verify the effectiveness of the QUIC multi-path load balancing method based on dynamic path quality assessment proposed in this invention, a specific explanation is given for a mobile high-definition live streaming scenario. In this network transmission scenario, the broadcaster uses a smartphone to conduct real-time video live streaming in indoor and outdoor mobile environments. The device is simultaneously connected to Wi-Fi 6 (2.4GHz / 5GHz dual-band) and 5G SA (standalone) networks, and transmits high-definition video streams (1080p@60fps, bitrate 8-12Mbps) to the cloud live streaming server through the QUIC multi-path protocol.

[0104] In actual operation, when the anchor moves from an indoor area with strong Wi-Fi signal to an outdoor corridor with weakened Wi-Fi signal but stable 5G signal, the traditional MP-QUIC solution will cause a switching delay of more than 200ms due to the lag in path quality assessment, resulting in video stuttering and image quality degradation. Therefore, the method proposed in this invention is used for specific operation.

[0105] Specifically, network parameters are set based on this network transmission scenario. When the network path is Wi-Fi 6, the interface identifier is wlan0; the maximum bandwidth is... The initial round-trip time is The initial packet loss rate was The bandwidth baseline value, i.e., the measured available bandwidth, is... When the network path is 5G SA, the interface identifier is rmnet0; the maximum bandwidth is... The initial round-trip time is The initial packet loss rate was The bandwidth baseline value is .

[0106] Then, define the key parameters for the relevant sub-scores of the network path performance metrics, with the round-trip time baseline value being... The baseline value for packet loss rate is In the weighting coefficients, , , In determining the predicted quality score, relevant key parameters are set, with the smoothing factor being... The evaluation threshold is The minimum score for the fault path is .

[0107] When the entire network transmission setup starts running, firstly, real-time monitoring and data collection are performed based on the quality of the dual network paths. The main monitoring period, i.e., the collection time, is set to... When the predicted quality score declines, activate the rapid monitoring mode and set it to [function name]. .

[0108] The network path performance metrics were measured sequentially, including round-trip time. A dedicated QUICPATH_CHALLENGE frame (16 bytes) was sent, and the sending time was recorded. and receive PATH_RESPONSE time The actual round-trip time is calculated using the following formula:

[0109]

[0110] in, Indicates the current moment At that time, the first The measured round-trip time of the network path; This indicates calibration via the device's local clock, which is set to a fixed value of 2ms in this embodiment.

[0111] Actual packet loss rate was measured using a sliding window method, with a window size of [missing information]. The calculation formula for each data packet is:

[0112]

[0113] in, Indicates the current moment At that time, the first The measured packet loss rate of the network path; Indicates the number of packets sent; This indicates the number of confirmed packages.

[0114] The actual bandwidth utilization rate is calculated using the following formula:

[0115]

[0116] in, Indicates the current moment At that time, the first Measured bandwidth utilization of each network path; Indicates the first The network path was obtained through the QUIC traffic statistics tool; Indicates the time difference, and .

[0117] Next, the performance index data is preprocessed by applying a 3-point moving average filter. The corresponding calculation formula is as follows:

[0118]

[0119] in, This indicates the measured round-trip time, packet loss rate, or bandwidth utilization rate after preprocessing.

[0120] Then, the performance index data is normalized, and the round-trip time sub-score, i.e., the exponential decay model, is calculated using the following formula:

[0121]

[0122] in, Indicates the current moment At that time, the first A sub-score of the round-trip time for each network path; Indicates the current moment At that time, the first Measured round-trip time of each network path after preprocessing; This represents the hyperbolic tangent function.

[0123] It can be explained that the hyperbolic tangent function This is used to introduce a non-linear adjustment that smooths out changes as the sub-routine score of round-trip time approaches the round-trip time baseline.

[0124] The sub-score of packet loss rate, which uses double exponential decay, is calculated using the following formula:

[0125]

[0126] in, Indicates the current moment At that time, the first Sub-scores of packet loss rate for each network path; Indicates the current moment At that time, the first The measured packet loss rate after preprocessing the network path.

[0127] The sub-rating of bandwidth utilization, i.e., the S-shaped function with saturation characteristics, is calculated using the following formula:

[0128]

[0129] in, Indicates the current moment At that time, the first A sub-rating of network path bandwidth utilization; Indicates the current moment At that time, the first The measured bandwidth utilization rate after network path preprocessing.

[0130] It should be noted that score growth slows down when bandwidth utilization exceeds 60%, to avoid overuse of a single network path.

[0131] Secondly, weight settings are implemented. Based on the characteristics of the live streaming scenario, when a keyframe is detected (i.e., a video I-frame is transmitted), QUIC stream marking is used to increase the packet loss rate weight during the video I-frame transmission period (approximately 1 second). Instead, a P-frame transmission period is used to restore normal weights; the quality score of the network path is determined, and the corresponding calculation formula is:

[0132]

[0133] The range of quality scores is The higher the value, the better the path quality.

[0134] In this embodiment, double exponential smoothing of the trend term is introduced, along with a smoothing factor, to determine the predicted value of the performance index data. The trend term is denoted as... It is used in the exponentially weighted moving average model to enhance the ability to capture the direction and rate of change of network performance indicators, thereby improving the foresight of predictions. The corresponding calculation formula is as follows:

[0135]

[0136] in, This indicates the predicted round-trip time, predicted packet loss rate, or predicted bandwidth utilization. Indicates the smoothing factor; This represents the measured round-trip time, packet loss rate, or bandwidth utilization rate after preprocessing. Indicates the trend term, i.e. and Use trend terms to capture the direction of change.

[0137] The predicted quality score for the network path is determined by the following formula:

[0138]

[0139] Next, assess the prediction confidence level and calculate the root mean square of the prediction error. The corresponding formula is:

[0140]

[0141] in, Indicates the first The prediction confidence of a network path; when When this happens, the weight of the predicted path value for that network is reduced.

[0142] After obtaining the predicted quality score, dynamic load balancing is performed, assuming the total video bandwidth requirement, i.e., the live stream bitrate. The traffic allocation weights for network paths are determined using the following formula:

[0143]

[0144] Traffic is further segmented, with keyframes (i.e., video I-frames), accounting for approximately 20% of the traffic, prioritized for allocation to high-quality scoring network paths. The corresponding calculation formula is as follows:

[0145]

[0146] For video P-frames, the bandwidth is allocated according to a weighted ratio, and the corresponding calculation formula is as follows:

[0147]

[0148] The traffic adjustment adopts a smooth adjustment mechanism, that is, to avoid sudden changes in traffic, incremental adjustment is used, and the corresponding calculation formula is:

[0149]

[0150] in, This represents the adjustment rate factor, preferably, in this embodiment, .

[0151] The aforementioned predicted quality score With evaluation threshold A comparison is performed, and when the predicted quality score meets the requirements... When the prediction quality score is met, it indicates a decline in prediction quality; when the prediction quality score meets the requirements... and When this occurs, it indicates that the quality of the network path deteriorates rapidly; when the predicted quality score meets the following conditions... and When this occurs, it indicates that video I-frame protection is triggered, which checks the network path quality before transmission. This represents the traffic at the next moment corresponding to a video I-frame; This represents the current traffic volume of the network path; when the aforementioned conditions are met, an active handover signal is triggered, and a tiered handover strategy is adopted, where the traffic is reduced to a lower quality level. Perform a level-one switch, reducing traffic allocation weight by 20%, for When the quality is moderately degraded, that is... Perform a secondary switch, set the minimum weight, and then... When the quality is severely degraded, that is... A three-level switch will be implemented, with temporary removal adopted. It triggers a fast detection mode, with a cycle of 20ms.

[0152] During the switching of network transmission paths, the data packets to be transmitted on the faulty network path are marked, and the current flow is terminated by the FIN flag of the STREAM frame of QUIC; a new flow is established on the new network path, with the same flow ID and priority set, the data packets to be retransmitted are transmitted, and the status table corresponding to the network path is updated.

[0153] Finally, network path quality restoration and gradual restoration of traffic allocation weights are performed. Quality restoration is triggered when the predicted quality score of a network path meets all of the following conditions: first, prediction quality improves for three consecutive cycles. Secondly, absolute quality meets the standards, that is... Thirdly, stability verification, namely... , This indicates a function that runs stably.

[0154] The traffic allocation weights are gradually restored, i.e., an S-shaped recovery curve is used to avoid oscillations. The corresponding calculation formula is:

[0155]

[0156] in, Indicates the first Traffic allocation weights after network path recovery; Indicates the first The normal calculation of traffic allocation weights for each network path; Indicates the first The current traffic allocation weight for each network path; This indicates the recovery cycle calculation, specifically a count of 3 recovery cycles.

[0157] Once the predicted quality score of the network path recovers, a five-cycle observation period begins. During this period, the stability of the network path is assessed each cycle. If the quality deteriorates again, the system immediately reverts to the switching state. After the observation period ends, the calculation of normal traffic allocation weights is fully restored.

[0158] The study explains that an exponentially weighted moving average model is used to predict network path quality indicators in advance. Combined with dynamic weight allocation and active switching mechanisms, the performance of multi-path transmission is comprehensively optimized. In the design of the prediction model, a nonlinear smoothing factor and a dynamic evaluation threshold strategy are adopted to adjust the prediction accuracy in real time according to changes in network status, avoiding the performance degradation caused by the lag in traditional methods. At the same time, a traffic allocation weight gradually increasing mechanism is introduced for faulty paths to ensure smooth switching, helping to maintain transmission stability efficiently in complex network environments and ensuring low latency and high bandwidth utilization.

[0159] The second embodiment of the present invention provides a QUIC multi-path load balancing system based on dynamic path quality assessment. The system includes a processor, a communication interface, a memory, and a communication bus. The processor, the communication interface, and the memory communicate with each other through the communication bus. The processor calls logical instructions in the memory to execute the QUIC multi-path load balancing method based on dynamic path quality assessment as described in any of the preceding embodiments.

[0160] It can be noted that a QUIC multi-path load balancing system based on dynamic path quality assessment needs to utilize the QUIC multi-path load balancing method based on dynamic path quality assessment during operation. Therefore, whether the system and program data are integrated or different hardware is configured to produce functions with similar effects to those achieved by this invention, they all fall within the protection scope of this invention. Moreover, this system has the same beneficial effects as the aforementioned QUIC multi-path load balancing method based on dynamic path quality assessment, which will not be elaborated here.

[0161] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0162] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A QUIC multi-path load balancing method based on dynamic path quality assessment, characterized in that, The method includes: A network transmission layout is formed based on network communication. The network transmission layout includes multiple network paths. The performance index data of each network path is monitored and collected. Analyze performance metrics data to determine the quality score for each network path; Obtain the exponentially weighted moving average model, input the performance index data into the exponentially weighted moving average model, output the prediction index data, and obtain the prediction quality score for each network path. Dynamic load distribution is performed based on the predicted quality score to balance the load; and a preset evaluation threshold is compared with the predicted quality score to trigger a network path switching signal and perform network path recovery processing.

2. The QUIC multi-path load balancing method based on dynamic path quality assessment according to claim 1, characterized in that, The performance metrics data include round-trip time, packet loss rate, and bandwidth utilization.

3. The QUIC multi-path load balancing method based on dynamic path quality assessment according to claim 2, characterized in that, Analyze performance metrics data to determine the quality score for each network path, including: Based on the performance index data, normalization processing is performed to determine the sub-scores for each round-trip time, packet loss rate, and bandwidth utilization in each network path. Each sub-score is assigned a pre-defined weight, and the quality score of the corresponding network path is obtained by combining the weights and sub-scores.

4. The QUIC multi-path load balancing method based on dynamic path quality assessment according to claim 3, characterized in that, Obtain the exponentially weighted moving average model, input the performance index data into the exponentially weighted moving average model, and output the prediction index data to obtain the prediction quality score for each network path, including: Input the performance index data corresponding to the current moment of each network path into the exponentially weighted moving average model to predict the performance index data corresponding to the next adjacent moment as the predicted index data; The prediction quality score is output based on the prediction index data combined with weights.

5. The QUIC multi-path load balancing method based on dynamic path quality assessment according to claim 1, characterized in that, Dynamic load allocation and load balancing are performed based on predicted quality scores, specifically as follows: The traffic allocation weight for each network path is determined based on the predicted quality score, and the total traffic demand in the network transmission layout is obtained. The traffic allocation weight and the total traffic demand are then integrated to obtain the traffic allocated to each network path.

6. The QUIC multi-path load balancing method based on dynamic path quality assessment according to claim 5, characterized in that, A preset evaluation threshold is compared with the predicted quality score, triggering a network path switching signal and performing network path recovery processing, including: Define the network path corresponding to the predicted quality score of the current analysis as the target path, preset the evaluation threshold, and compare it with the predicted quality score. If the predicted quality score is less than the estimated threshold, it means that the quality of the target path has deteriorated, triggering the target path switching signal, resetting the predicted quality score, and dynamically redistributing the load accordingly. Once the predicted quality score of the target path returns to normal, the target path will be reinstated into the load balancing process, and the traffic allocation weight for the target path will be gradually increased until the predicted quality score is greater than or equal to the estimated threshold.

7. A QUIC multi-path load balancing system based on dynamic path quality assessment, characterized in that, The system includes a processor, a communication interface, a memory, and a communication bus. The processor, the communication interface, and the memory communicate with each other through the communication bus. The processor calls logical instructions in the memory to execute the QUIC multi-path load balancing method based on dynamic path quality assessment as described in any one of claims 1 to 6.

Images