White box device congestion control optimization method and system based on large model technology

Abstract

The invention relates to a white-box device congestion control optimization method and system based on a large model technology, and the method comprises the steps: collecting network performance data and flow data through a white-box device, and transmitting the data to a central intelligent congestion control system through a bidirectional communication mechanism; the central intelligent congestion control system uses the received data to train a large model capable of predicting the network traffic trend, and constructs a knowledge database for storing traffic feature vectors and corresponding congestion control parameters; a central intelligent congestion control system adjusts congestion control parameters based on real-time flow characteristics transmitted by white-box equipment, the congestion control parameters retrieved by a knowledge database and a flow trend predicted by a large model, and issues the congestion control parameters to the white-box equipment through a two-way communication mechanism so as to optimize the network congestion control performance of the white-box equipment. According to the invention, the adaptive adjustment of the congestion control parameters of the white-box device is realized by combining the traffic characteristic fuzzy matching of the vector database and the traffic trend prediction capability of the center large model.

Description

Technical Field

[0001] The present invention belongs to the technical field of information transmission, and in particular relates to a white box device congestion control optimization method and system based on large model technology. Background Art

[0002] Network congestion control is a key technology for ensuring stable and efficient data transmission. It aims to regulate incoming network traffic and prevent system overloads that lead to packet loss and latency. This is a complex, network-wide collaborative process involving hosts and various forwarding devices, with the shared goal of achieving a lossless, low-latency, and high-throughput network environment. However, current congestion control parameter tuning faces challenges. If the parameters are set too high, forwarding devices will utilize long queues and large buffers to meet the needs of throughput-sensitive large flows. This can introduce significant queuing delays during congestion, severely impacting the transmission efficiency of latency-sensitive small flows. Conversely, if the parameters are set too low, while shortening queues and reducing buffering can reduce latency and meet the needs of small flows, this can compromise overall network throughput and limit the performance of throughput-sensitive large flows. Given that network traffic characteristics change dynamically over time, the current method of statically setting congestion control parameters by network administrators is no longer able to effectively adapt to these real-time dynamics. This leads to uneven utilization of network resources and makes it difficult to balance the needs of different traffic types. Summary of the Invention

[0003] The purpose of the present invention is to provide a white box device congestion control optimization method and system based on large model technology to solve the problem that the existing technology easily leads to uneven utilization of network resources.

[0004] To achieve one of the above-mentioned objectives, an embodiment of the present invention provides a white-box device congestion control optimization method based on a large model technology, the method comprising:

[0005] Collect network performance and traffic data through white box devices and transmit them to the central intelligent congestion control system through a two-way communication mechanism;

[0006] The central intelligent congestion control system uses the received data to train a large model that can predict network traffic trends and build a knowledge database that stores traffic feature vectors and their corresponding congestion control parameters;

[0007] The central intelligent congestion control system adjusts the congestion control parameters based on the real-time traffic characteristics transmitted by the white-box devices, the congestion control parameters retrieved from the knowledge database, and the traffic trends predicted by the large model. The parameters are then sent to the white-box devices through a two-way communication mechanism to optimize their network congestion control performance.

[0008] As a further improvement of an embodiment of the present invention, the method further includes that the two-way communication mechanism includes a streaming long connection based on remote procedure call, specifically including:

[0009] The white box device is configured as a GRPC server, and the central intelligent congestion control system is configured as a GRPC client to establish a bidirectional streaming long connection;

[0010] The white box device regularly pushes queue buffer occupancy rate, throughput rate and large and small flow ratio data to the central intelligent congestion control system through the streaming long connection;

[0011] When the white box device detects a burst of traffic, it also actively pushes the queue buffer occupancy rate, throughput rate, and large and small flow ratio data to the central intelligent congestion control system through the streaming long connection;

[0012] The central intelligent congestion control system regularly subscribes to and receives the queue buffer occupancy rate, throughput rate and large and small flow ratio data of the white box device through the streaming long connection.

[0013] As a further improvement of an embodiment of the present invention, the method further includes that the training of a large model capable of predicting network traffic trends includes:

[0014] The central intelligent congestion control system receives the network performance data and traffic data sent by the white box device, extracts feature information such as timestamp, device name, queue buffer occupancy, throughput rate, and ratio of large and small flows, and forms a structured data set for fine-tuning the large model;

[0015] During the idle period of the central intelligent congestion control system, the structured data set is used to fine-tune the large model, and the training generates a large model that can predict the network traffic trend within a predetermined time range in the future based on the input timestamp and device name.

[0016] As a further improvement of an embodiment of the present invention, the method further includes the step of constructing a knowledge database storing traffic feature vectors and their corresponding congestion control parameters, comprising:

[0017] The central intelligent congestion control system converts historical traffic characteristics and their corresponding congestion control parameters into high-dimensional vector representations through an embedding model and stores them in a vector database;

[0018] After receiving the real-time traffic features sent by the white box device, the central intelligent congestion control system converts the real-time traffic features into vectors through the same embedding model, performs similarity calculation in the vector database, and retrieves the congestion control parameters that are most similar to the real-time traffic feature vectors.

[0019] As a further improvement of an embodiment of the present invention, the method further includes: the step of adjusting the congestion control parameter includes:

[0020] The central intelligent congestion control system compares the original congestion control parameter with the retrieved congestion control parameter;

[0021] When the original congestion control parameter is greater than the retrieved congestion control parameter and the traffic trend predicted by the large model is an upward trend, retaining the original congestion control parameter;

[0022] When the original congestion control parameter is greater than the retrieved congestion control parameter and the traffic trend predicted by the large model is a downward trend, updating the congestion control parameter to the retrieved congestion control parameter;

[0023] When the original congestion control parameter is less than the retrieved congestion control parameter and the traffic trend predicted by the large model is a downward trend, retaining the original congestion control parameter;

[0024] When the original congestion control parameter is smaller than the retrieved congestion control parameter and the traffic trend predicted by the large model is an upward trend, the congestion control parameter is updated to the retrieved congestion control parameter.

[0025] As a further improvement of an embodiment of the present invention, the method further includes, the adjusting of the congestion control parameters further includes,

[0026] In the initial stage when the amount of historical traffic data used for large model training is less than a preset threshold, the central intelligent congestion control system directly sends the congestion control parameters retrieved from the knowledge database to the white box device;

[0027] In the mature stage when the amount of historical traffic data used for large-scale model training is greater than or equal to a preset threshold, the central intelligent congestion control system combines the traffic trend predicted by the large-scale model, adjusts the congestion control parameters based on the comparison results between the original congestion control parameters and the retrieved congestion control parameters and the rising or falling state of the traffic trend, and then sends them to the white-box device.

[0028] As a further improvement of an embodiment of the present invention, the method further includes that the configuration optimization by sending to the white box device includes:

[0029] The white box device adjusts the network buffer size to adapt to the current traffic characteristics according to the received adjusted congestion control parameters;

[0030] The white box device updates the congestion control threshold to optimize queue buffer occupancy and throughput.

[0031] To achieve one of the above-mentioned objects of the invention, an embodiment of the present invention further provides a white-box device congestion control optimization system based on large model technology, the system comprising a collection module, a decision module and an adjustment module;

[0032] The acquisition module is used to collect network performance data and flow data through white box devices and transmit them to the central intelligent congestion control system through a two-way communication mechanism;

[0033] The decision module is used to use the received data through the central intelligent congestion control system to train a large model that can predict network traffic trends and build a knowledge database that stores traffic feature vectors and their corresponding congestion control parameters;

[0034] The adjustment module is used to adjust the congestion control parameters based on the real-time traffic characteristics transmitted by the white-box device, the congestion control parameters retrieved from the knowledge database, and the traffic trends predicted by the large model through the central intelligent congestion control system, and send them to the white-box device through a two-way communication mechanism to optimize its network congestion control performance.

[0035] In order to achieve one of the above-mentioned purposes of the invention, an embodiment of the present invention also provides an electronic device, including a memory and a processor, characterized in that the memory stores a computer program that can be run on the processor, and when the program is executed on the processor, the steps in the white-box device congestion control optimization method based on large model technology as described above are implemented.

[0036] In order to achieve one of the above-mentioned purposes of the invention, an embodiment of the present invention also provides a storage medium, which stores a computer program, and is characterized in that when the computer program is executed by a processor, it implements the steps in the white-box device congestion control optimization method based on large model technology as described above.

[0037] Compared to existing technologies, this invention provides a congestion control optimization method and system for white-box devices based on large-scale model technology. By combining the fuzzy matching of traffic characteristics in a vector database with the traffic trend prediction capabilities of a central large-scale model, this method enables dynamic adaptive adjustment of congestion control parameters for white-box devices. This effectively avoids congestion caused by sudden traffic changes and significantly improves service quality stability. Furthermore, because the large-scale model is centrally deployed, even white-box devices with limited computing power can obtain advanced dynamic congestion control capabilities, greatly expanding the application scope and deployment flexibility of this technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is an overall flow chart of the white box device congestion control optimization method based on large model technology described in the present invention.

[0039] Figure 2 This is a schematic diagram of the architecture of the white-box device congestion control optimization system based on the large model technology described in the present invention. DETAILED DESCRIPTION

[0040] The present invention will be described in detail below with reference to the specific embodiments shown in the accompanying drawings. However, these embodiments do not limit the present invention, and any structural, methodological, or functional changes made by those skilled in the art based on these embodiments are all within the scope of protection of the present invention.

[0041] The embodiments of the present invention are described in detail below. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention and are not to be construed as limiting the present invention.

[0042] In the first embodiment of the present invention, the present invention provides a white box device congestion control optimization method based on large model technology, such as Figure 1 As shown, the method includes,

[0043] S1: Collects network performance and traffic data through white box devices and transmits them to the central intelligent congestion control system through a two-way communication mechanism;

[0044] S2: The central intelligent congestion control system uses the received data to train a large model that can predict network traffic trends and build a knowledge database that stores traffic feature vectors and their corresponding congestion control parameters;

[0045] S3: The central intelligent congestion control system adjusts the congestion control parameters based on the real-time traffic characteristics transmitted by the white-box devices, the congestion control parameters retrieved from the knowledge database, and the traffic trends predicted by the large model. The parameters are then sent to the white-box devices through a two-way communication mechanism to optimize their network congestion control performance.

[0046] In a specific embodiment of the present invention, the two-way communication mechanism includes a streaming long connection based on remote procedure call, specifically including:

[0047] The white box device is configured as a GRPC server, and the central intelligent congestion control system is configured as a GRPC client to establish a bidirectional streaming long connection;

[0048] The white box device regularly pushes queue buffer occupancy rate, throughput rate and large and small flow ratio data to the central intelligent congestion control system through the streaming long connection;

[0049] When the white box device detects a burst of traffic, it also actively pushes the queue buffer occupancy rate, throughput rate, and large and small flow ratio data to the central intelligent congestion control system through the streaming long connection;

[0050] The central intelligent congestion control system regularly subscribes to and receives the queue buffer occupancy rate, throughput rate and large and small flow ratio data of the white box device through the streaming long connection.

[0051] It should be noted that to achieve efficient, real-time data exchange between the white-box device and the central intelligent congestion control system, the present invention employs a streaming persistent connection based on remote procedure call (GRPC) as a bidirectional communication mechanism. Specifically, in this communication architecture, the white-box device is configured as a GRPC server, responsible for responding to requests from the client and actively pushing data; while the central intelligent congestion control system is configured as a GRPC client, responsible for initiating connections and data subscriptions. This configuration enables a persistent, bidirectional streaming persistent connection to be established between the two. The advantage of this persistent connection is that, once established, multiple data exchanges can be performed within the session lifecycle, avoiding the overhead of frequent connection establishment and disconnection, significantly improving communication efficiency and real-time performance.

[0052] Furthermore, in terms of data transmission, this mechanism supports two modes to meet the needs of different scenarios: Scheduled Push Mode: The white-box device periodically and at preset intervals proactively pushes real-time network performance data and traffic characteristics to the central intelligent congestion control system via this persistent streaming connection. This includes, but is not limited to, key metrics such as queue buffer occupancy, throughput, and the ratio of large and small flows. This scheduled push mechanism ensures that the central system continuously obtains the latest network status information, providing a stable data stream for large-scale model training and trend prediction. Event-Driven Active Push Mode: To address unpredictable network conditions, the white-box device also has the ability to detect traffic bursts. Upon detecting significant traffic fluctuations or anomalies (e.g., a rapid increase in queue buffer occupancy or a sharp change in throughput), it immediately triggers data push via the persistent streaming connection, promptly reporting the latest queue buffer occupancy, throughput, and the ratio of large and small flows to the central intelligent congestion control system. This immediate response mechanism is crucial for quickly detecting and addressing the onset of network congestion.

[0053] Furthermore, the central intelligent congestion control system, acting as a GRPC client, regularly subscribes to and receives queue buffer occupancy, throughput, and the ratio of large and small flows from all white-box devices via streaming persistent connections. This subscription mechanism, in conjunction with the device's push mechanism, ensures that the central intelligent congestion control system can comprehensively, accurately, and in real time aggregate performance and traffic information from white-box devices across the entire network, laying a solid data foundation for subsequent large-scale model training, traffic forecasting, and intelligent adjustment of congestion control parameters.

[0054] In a specific embodiment of the present invention, a large model capable of predicting network traffic trends is trained, specifically,

[0055] The central intelligent congestion control system receives the network performance data and traffic data sent by the white box device, extracts feature information such as timestamp, device name, queue buffer occupancy, throughput rate, and ratio of large and small flows, and forms a structured data set for fine-tuning the large model;

[0056] During the idle period of the central intelligent congestion control system, the structured data set is used to fine-tune the large model, and the training generates a large model that can predict the network traffic trend within a predetermined time range in the future based on the input timestamp and device name.

[0057] It should be noted that in order to enable the central intelligent congestion control system to accurately predict network traffic trends, the present invention deploys and trains a large model.

[0058] Furthermore, the central intelligent congestion control system continuously receives network performance and traffic data from all white box devices. This raw data contains a wealth of information about network operations. The system preprocesses and performs feature engineering on this data, extracting key features such as the timestamp of each data point, the device name to which the data belongs, and detailed traffic metrics, including queue buffer occupancy, throughput, and the ratio of large and small flows. This extracted feature information is then organized into a structured dataset. The format of this dataset is carefully designed, typically arranged in a time series format. For example, timestamps and device names serve as input features for the model, while the corresponding traffic metrics serve as the outputs or labels to be predicted by the model.

[0059] Furthermore, during idle periods, the central intelligent congestion control system fully utilizes these resources and fine-tunes the large model using the structured dataset. This large model can be a well-established deep learning architecture, such as the Transformer model or the LSTM model. These models are chosen for their excellent performance in processing time series data and capturing complex nonlinear relationships. By fine-tuning the previously constructed structured dataset, the large model can learn the inherent patterns, cyclical patterns, and burst characteristics of network traffic fluctuations from massive amounts of historical data. Once fine-tuned, the large model possesses powerful predictive capabilities, enabling it to be trained to accurately predict network traffic trends within a predetermined timeframe based on new inputs. This predictive capability is fundamental to dynamic congestion control, enabling the system to foresee potential congestion risks and make preventive adjustments, rather than simply reacting to them.

[0060] In a specific embodiment of the present invention, the step of constructing a knowledge database to store traffic feature vectors and their corresponding congestion control parameters is as follows:

[0061] The central intelligent congestion control system converts historical traffic characteristics and their corresponding congestion control parameters into high-dimensional vector representations through an embedding model and stores them in a vector database;

[0062] After receiving the real-time traffic features sent by the white box device, the central intelligent congestion control system converts the real-time traffic features into vectors through the same embedding model, performs similarity calculation in the vector database, and retrieves the congestion control parameters that are most similar to the real-time traffic feature vectors.

[0063] It should be noted that, in order to obtain the optimal congestion control parameters efficiently and intelligently, the present invention constructs and utilizes a knowledge database, the core of which is a retrieval enhancement technology based on a vector database.

[0064] Furthermore, the central intelligent congestion control system continuously accumulates a large amount of historical traffic characteristic data, such as various queue buffer occupancy rates, throughput rates, and combinations of large and small flows at different time points and network conditions. The system also records the corresponding congestion control parameters based on these specific traffic characteristics, verified through manual experience or optimized through simulation. These paired "traffic characteristic-congestion control parameter" data are the cornerstone of building a knowledge base.

[0065] Furthermore, to enable efficient retrieval and matching of this data, the system employs an embedding model. This model converts complex, multi-dimensional historical traffic features (such as numerical queue buffer occupancy and throughput) into high-dimensional vector representations. Each combination of traffic features is mapped to a unique vector point. These vectors are also associated with their corresponding congestion control parameters and stored in a vector database. The vector database is optimized for storing and querying high-dimensional vectors, enabling fast similarity searches.

[0066] Furthermore, when the central intelligent congestion control system receives real-time traffic features from the white-box device, it immediately initiates the parameter retrieval process. First, this real-time traffic feature is converted into a high-dimensional vector using the same embedding model. The system then uses this real-time traffic feature vector as a query point and performs a similarity calculation within the vast vector database. This means the system quickly locates those historical traffic feature vectors in the database that are "closest" or "most similar" to the real-time traffic feature vector in vector space. Through this fuzzy matching approach, the system can quickly retrieve the optimal congestion control parameters corresponding to the historical scenario that most closely resembles the current real-time traffic conditions. This vector similarity-based retrieval is far more accurate and efficient than traditional keyword matching or rule-based search, ensuring that even when faced with unprecedented traffic patterns, a verified, most appropriate optimization solution can be quickly found as the initial parameter recommendation.

[0067] In a specific embodiment of the present invention, the step of adjusting the congestion control parameters is specifically as follows:

[0068] The central intelligent congestion control system compares the original congestion control parameter with the retrieved congestion control parameter;

[0069] When the original congestion control parameter is greater than the retrieved congestion control parameter and the traffic trend predicted by the large model is an upward trend, retaining the original congestion control parameter;

[0070] When the original congestion control parameter is greater than the retrieved congestion control parameter and the traffic trend predicted by the large model is a downward trend, updating the congestion control parameter to the retrieved congestion control parameter;

[0071] When the original congestion control parameter is less than the retrieved congestion control parameter and the traffic trend predicted by the large model is a downward trend, retaining the original congestion control parameter;

[0072] When the original congestion control parameter is smaller than the retrieved congestion control parameter and the traffic trend predicted by the large model is an upward trend, the congestion control parameter is updated to the retrieved congestion control parameter.

[0073] It should be noted that in the central intelligent congestion control system, the step of dynamically adjusting congestion control parameters integrates real-time search results and future traffic predictions, ensuring that parameter adjustments can both respond to current network conditions and predict future trends.

[0074] Furthermore, after the system retrieves the "retrieved congestion control parameters" that are most similar to the current real-time traffic characteristics from the knowledge database, it first compares them with the "original congestion control parameters" currently in use by the white box device. The system also considers the future traffic trend (increase or decrease) predicted by the large model. Based on these two pieces of information, the system determines the final congestion control parameters according to the following logical rules:

[0075] Conservative optimization (congestion risk assessment): When the original congestion control parameters are greater than the retrieved congestion control parameters and the large-scale model predicts an upward traffic trend, this indicates that the current parameters may be relatively loose, while network traffic is still increasing. To avoid potential congestion, the system retains the original congestion control parameters. This strategy aims to maintain the stability of the existing parameters and avoid network congestion caused by overly aggressive parameter adjustments when traffic is rapidly increasing.

[0076] Moderate Adjustment (Congestion Relief or Optimization): When the original congestion control parameters are greater than the search congestion control parameters, but the large-scale model predicts a decreasing traffic trend, this indicates that network pressure is easing or that the current parameters may be too loose. In this case, the system identifies room for optimization and updates the congestion control parameters to the search congestion control parameters. This helps to release network resources more promptly or more precisely match the current decrease in traffic, thereby improving resource utilization.

[0077] Prioritizing stability (avoiding unnecessary tightening): When the original congestion control parameters are lower than the retrieved ones and the large-scale model predicts a downward traffic trend, this indicates that network pressure may be easing or that the current parameters are already conservative. To avoid unnecessary further tightening of parameters that could lead to a decrease in throughput, the system retains the original congestion control parameters. This strategy prioritizes the stable operation of existing services and avoids making adjustments that could negatively impact traffic when traffic is declining.

[0078] Aggressive optimization (performance improvement): When the original congestion control parameters are lower than the searched congestion control parameters and the large-scale model predicts an upward traffic trend, this indicates that the network may have underutilized bandwidth potential and that traffic is about to increase. To better handle the incoming traffic, the system updates the congestion control parameters to the searched congestion control parameters. This strategy aims to proactively improve network performance and ensure efficient use of network resources as traffic increases.

[0079] It should be noted that through the above-mentioned refined comparison and judgment logic, the system can realize intelligent decision-making of congestion control parameters, which can not only respond to the current network status, but also proactively consider future traffic changes, thereby maximizing resource utilization efficiency and service transmission quality while ensuring network stability.

[0080] In a specific embodiment of the present invention, adjusting the congestion control parameters further includes:

[0081] In the initial stage when the amount of historical traffic data used for large model training is less than a preset threshold, the central intelligent congestion control system directly sends the congestion control parameters retrieved from the knowledge database to the white box device;

[0082] In the mature stage when the amount of historical traffic data used for large-scale model training is greater than or equal to a preset threshold, the central intelligent congestion control system combines the traffic trend predicted by the large-scale model, adjusts the congestion control parameters based on the comparison results between the original congestion control parameters and the retrieved congestion control parameters and the rising or falling state of the traffic trend, and then sends them to the white-box device.

[0083] It should be noted that in addition to the parameter adjustment logic based on real-time comparison and trend prediction described above, this invention also considers the amount of historical traffic data accumulated by the system during the adjustment and issuance of congestion control parameters, adapting to the different stages of system startup and stable operation. This ensures that the solution operates effectively in both data-sparse and data-rich scenarios.

[0084] Furthermore, the parameter tuning strategy is divided into two main stages:

[0085] Fast response when data volume is sparse: In the initial stage when the amount of historical traffic data used for large-scale model training is less than the preset threshold, the large-scale model of the central intelligent congestion control system may not have been fully trained and fine-tuned, and its traffic trend prediction capabilities may not be mature enough. In this case, the system will not overly rely on the predictions of the large-scale model that is still learning. Instead, to ensure that white-box devices can quickly obtain reasonable congestion control parameters, the central intelligent congestion control system will directly send the congestion control parameters retrieved from the knowledge database to the white-box devices. At this time, the knowledge database, as a collection of empirical rules and historical best practices, can provide a reliable baseline parameter, allowing white-box devices to perform basic congestion control optimization even in the early stages of the system, avoiding a serious degradation of network performance.

[0086] Intelligent Decision-Making in the Face of Abundant Data: Over time, the central intelligent congestion control system continuously receives and accumulates large amounts of network performance and traffic data from white-box devices. When the amount of historical traffic data used to train the large model reaches or exceeds a preset threshold, the system enters the maturity stage. At this point, the large model has been fully trained and fine-tuned, and its traffic trend prediction capabilities have become highly reliable and accurate. During this stage, the central intelligent congestion control system fully utilizes its predictive capabilities. It combines the future traffic trends predicted by the large model with the rules described in claim 5 (i.e., a complex comparison and judgment logic based on the original parameters, the search parameters, and the predicted trends) to intelligently adjust the congestion control parameters. After the adjustment is completed, the final optimized parameters are immediately distributed to the corresponding white-box devices. This strategy ensures that when the system has sufficient data support, it can perform more refined and proactive congestion control parameter optimization, thereby maximizing network performance and ensuring service quality.

[0087] In a specific embodiment of the present invention, configuration optimization is performed by sending it to a white box device, specifically,

[0088] The white box device adjusts the network buffer size to adapt to the current traffic characteristics according to the received adjusted congestion control parameters;

[0089] The white box device updates the congestion control threshold to optimize queue buffer occupancy and throughput.

[0090] It should be noted that after the central intelligent congestion control system determines the optimal congestion control parameters, the next step is to send these parameters to the corresponding white box devices and guide them to perform local configuration optimization.

[0091] Furthermore, specific configuration optimization includes the following two main aspects:

[0092] Adjusting the size of the network buffer: White box devices dynamically adjust the size of their network buffers based on the latest congestion control parameters they receive. A network buffer is a temporary storage location for data packets before they are processed or forwarded. A reasonable buffer size is crucial for traffic management: if the buffer is too small, data packets are prone to overflow and discard during bursts or temporary link congestion, leading to transmission interruptions or retransmissions, which in turn reduces throughput and increases latency. If the buffer is too large, although it can accommodate more data packets, when congestion occurs, data packets will queue in the buffer for a longer time, resulting in significant queuing delays, which is unacceptable for latency-sensitive applications (such as real-time voice and video conferencing).

[0093] Therefore, through parameters dynamically issued by the central intelligent congestion control system based on predictions and real-time characteristics, white box devices can intelligently increase or decrease their buffers to better adapt to current network traffic characteristics. For example, when traffic is predicted to increase or the proportion of large flows increases, the buffer may be appropriately increased to absorb the burst and ensure large flow throughput; when traffic decreases or small flows predominate, the buffer may be reduced to reduce queue delay.

[0094] Updating the congestion control threshold: In addition to adjusting the buffer size, the white box device will also update its internal congestion control threshold according to the parameters sent down. The congestion control threshold is the key threshold for triggering specific congestion control behaviors (such as reducing the send window, starting the congestion avoidance algorithm, etc.): Queue buffer occupancy threshold: This threshold determines that when the data packets accumulate in the queue to a certain level, the system should start to take congestion control measures. By adjusting this threshold, congestion control can be triggered earlier or later, thereby affecting the speed and aggressiveness of the network's response to congestion. Throughput threshold: Some advanced congestion control algorithms may also refer to the throughput threshold. When the throughput is lower or higher than a certain expected value, the parameter adjustment is triggered.

[0095] By optimizing thresholds related to queue buffer occupancy and throughput, white box devices can more accurately determine the degree and trend of network congestion and adopt appropriate response strategies. For example, they can set more sensitive thresholds to quickly respond to minor congestion and prevent it from worsening; or, when network conditions are good, they can relax thresholds to allow higher throughput.

[0096] It should be noted that by dynamically adjusting the network buffer size and updating the congestion control threshold, the white box device can convert the intelligent decisions output by the central intelligent congestion control system into its local actual operations, thereby achieving continuous optimization of its own congestion control behavior and ensuring that network transmission maintains high performance and high stability in complex and changing environments.

[0097] In the second embodiment of the present invention, the present invention provides a white-box device congestion control optimization system based on large model technology, the system comprising a collection module 1, a decision module 2 and an adjustment module 3;

[0098] The acquisition module 1 is used to collect network performance data and flow data through white box devices and transmit them to the central intelligent congestion control system through a two-way communication mechanism;

[0099] The decision module 2 is used to use the received data through the central intelligent congestion control system to train a large model that can predict network traffic trends and build a knowledge database that stores traffic feature vectors and their corresponding congestion control parameters;

[0100] The adjustment module 3 is used to adjust the congestion control parameters based on the real-time traffic characteristics transmitted by the white-box device, the congestion control parameters retrieved from the knowledge database, and the traffic trends predicted by the large model through the central intelligent congestion control system, and send them to the white-box device through a two-way communication mechanism to optimize its network congestion control performance.

[0101] In embodiment three of the present invention, the present invention provides an electronic device, comprising a memory and a processor, characterized in that the memory stores a computer program that can be run on the processor, and when the program is executed on the processor, the steps in the white-box device congestion control optimization method based on the large model technology as described above are implemented.

[0102] In a fourth embodiment of the present invention, the present invention provides a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps of the white-box device congestion control optimization method based on the large model technology as described above.

[0103] In summary, the present invention provides a congestion control optimization method and system for white-box devices based on large-scale model technology. By combining the fuzzy matching of traffic characteristics in a vector database with the traffic trend prediction capabilities of a central large-scale model, this method enables dynamic adaptive adjustment of congestion control parameters for white-box devices. This effectively avoids congestion caused by sudden traffic fluctuations and significantly improves service quality stability. Furthermore, because the large-scale model is centrally deployed, even white-box devices with limited computing power can obtain advanced dynamic congestion control capabilities, greatly expanding the application scope and deployment flexibility of this technology.

[0104] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working process of the modules described above can refer to the corresponding process in the aforementioned method implementation, and will not be repeated here.

[0105] Modules described as separate components may or may not be physically separate, and components shown as modules may or may not be physical modules, i.e., they may be located in one place or distributed across multiple network modules. Some or all of these modules may be selected to achieve the purpose of this embodiment according to actual needs.

[0106] In addition, the functional modules in various embodiments of the present application may be integrated into a single processing module, or each module may exist physically separately, or two or more modules may be integrated into a single module. The aforementioned integrated modules may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0107] The above-mentioned integrated modules implemented in the form of software functional modules can be stored in a computer-readable storage medium. The above-mentioned software functional modules are stored in a storage medium and include a number of instructions for causing a computer system (which may be a personal computer, server, or network system, etc.) or a processor to execute some of the steps of the methods described in various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application.

Claims

1. A congestion control optimization method for white-box devices based on large model technology, characterized by: include, Collect network performance and traffic data through white box devices and transmit them to the central intelligent congestion control system through a two-way communication mechanism; The central intelligent congestion control system uses the received data to train a large model that can predict network traffic trends and build a knowledge database that stores traffic feature vectors and their corresponding congestion control parameters; The central intelligent congestion control system adjusts the congestion control parameters based on the real-time traffic characteristics transmitted by the white-box devices, the congestion control parameters retrieved from the knowledge database, and the traffic trends predicted by the large model. The parameters are then sent to the white-box devices through a two-way communication mechanism to optimize their network congestion control performance.

2. The white-box device congestion control optimization method based on large model technology according to claim 1 is characterized by: The two-way communication mechanism includes a streaming long connection based on remote procedure call, specifically including: The white box device is configured as a GRPC server, and the central intelligent congestion control system is configured as a GRPC client to establish a bidirectional streaming long connection; The white box device regularly pushes queue buffer occupancy rate, throughput rate and large and small flow ratio data to the central intelligent congestion control system through the streaming long connection; When the white box device detects a burst of traffic, it also actively pushes the queue buffer occupancy rate, throughput rate, and large and small flow ratio data to the central intelligent congestion control system through the streaming long connection; The central intelligent congestion control system regularly subscribes to and receives the queue buffer occupancy rate, throughput rate and large and small flow ratio data of the white box device through the streaming long connection.

3. The white-box device congestion control optimization method based on large model technology according to claim 2 is characterized in that: The training of large models capable of predicting network traffic trends includes: The central intelligent congestion control system receives the network performance data and traffic data sent by the white box device, extracts feature information such as timestamp, device name, queue buffer occupancy, throughput rate, and ratio of large and small flows, and forms a structured data set for fine-tuning the large model; During the idle period of the central intelligent congestion control system, the structured data set is used to fine-tune the large model, and the training generates a large model that can predict the network traffic trend within a predetermined time range in the future based on the input timestamp and device name.

4. The white-box device congestion control optimization method based on large model technology according to claim 3 is characterized by: The step of constructing a knowledge database for storing traffic feature vectors and their corresponding congestion control parameters includes: The central intelligent congestion control system converts historical traffic characteristics and their corresponding congestion control parameters into high-dimensional vector representations through an embedding model and stores them in a vector database; After receiving the real-time traffic features sent by the white box device, the central intelligent congestion control system converts the real-time traffic features into vectors through the same embedding model, performs similarity calculation in the vector database, and retrieves the congestion control parameters that are most similar to the real-time traffic feature vectors.

5. The white-box device congestion control optimization method based on large model technology according to claim 4 is characterized in that: The step of adjusting the congestion control parameters includes: The central intelligent congestion control system compares the original congestion control parameter with the retrieved congestion control parameter; When the original congestion control parameter is greater than the retrieved congestion control parameter and the traffic trend predicted by the large model is an upward trend, retaining the original congestion control parameter; When the original congestion control parameter is greater than the retrieved congestion control parameter and the traffic trend predicted by the large model is a downward trend, updating the congestion control parameter to the retrieved congestion control parameter; When the original congestion control parameter is less than the retrieved congestion control parameter and the traffic trend predicted by the large model is a downward trend, retaining the original congestion control parameter; When the original congestion control parameter is smaller than the retrieved congestion control parameter and the traffic trend predicted by the large model is an upward trend, the congestion control parameter is updated to the retrieved congestion control parameter.

6. The white-box device congestion control optimization method based on large model technology according to claim 5 is characterized by: The adjusting of the congestion control parameters further includes: In the initial stage when the amount of historical traffic data used for large model training is less than a preset threshold, the central intelligent congestion control system directly sends the congestion control parameters retrieved from the knowledge database to the white box device; In the mature stage when the amount of historical traffic data used for large-scale model training is greater than or equal to a preset threshold, the central intelligent congestion control system combines the traffic trend predicted by the large-scale model, adjusts the congestion control parameters based on the comparison results between the original congestion control parameters and the retrieved congestion control parameters and the rising or falling state of the traffic trend, and then sends them to the white-box device.

7. The white-box device congestion control optimization method based on large model technology according to claim 6 is characterized by: The configuration optimization by sending the configuration to the white box device includes: The white box device adjusts the network buffer size to adapt to the current traffic characteristics according to the received adjusted congestion control parameters; The white box device updates the congestion control threshold to optimize queue buffer occupancy and throughput.

8. A white-box device congestion control optimization system based on large model technology, characterized by: Including acquisition module, decision module and adjustment module; The acquisition module is used to collect network performance data and flow data through white box devices and transmit them to the central intelligent congestion control system through a two-way communication mechanism; The decision module is used to use the received data through the central intelligent congestion control system to train a large model that can predict network traffic trends and build a knowledge database that stores traffic feature vectors and their corresponding congestion control parameters; The adjustment module is used to adjust the congestion control parameters based on the real-time traffic characteristics transmitted by the white-box device, the congestion control parameters retrieved from the knowledge database, and the traffic trends predicted by the large model through the central intelligent congestion control system, and send them to the white-box device through a two-way communication mechanism to optimize its network congestion control performance.

9. An electronic device comprising a memory and a processor, characterized in that: The memory stores a computer program that can be run on the processor, and when the program is executed on the processor, the steps of the white-box device congestion control optimization method based on large model technology as described in any one of claims 1 to 7 are implemented.

10. A storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the white-box device congestion control optimization method based on large model technology as described in any one of claims 1 to 7 are implemented.

Images