Adaptive gradient congestion control method based on utility function driving and related equipment

Abstract

The invention discloses an adaptive gradient congestion control method based on utility function driving and related equipment. The method comprises the following steps: acquiring a monitoring parameter and a historical monitoring value of a target transmission link; judging whether the monitoring period ends or not through a timer; if the monitoring period is not finished, target windows of different emphasized targets are quantified based on historical monitoring values and real-time monitoring parameters; performing weighted dynamic fusion on each target window based on the fusion factor corresponding to the previous monitoring period to update the congestion window of the target transmission link; if the monitoring period is finished, based on the historical monitoring value and the monitoring parameters in the monitoring period, the utility value of the monitoring period is obtained through utility function quantification, the fusion factor is adaptively updated based on the increment trend of the utility value, the next monitoring period is triggered, and the congestion window of the target transmission link is continuously adjusted. The method can meet the actual deployment requirements of a high-performance network on low overhead and fast response, and can be widely applied to the technical field of data processing.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to an adaptive gradient congestion control method and related equipment based on utility function-driven approaches. Background Technology

[0002] In modern high-performance network communication, congestion control is crucial for ensuring transmission efficiency and stability. Current mainstream solutions primarily rely on a single signal and can be categorized into three types: First, loss-based algorithms (such as Cubic) adjust the window by monitoring packet loss events, which can improve bandwidth utilization but easily lead to buffer bloat and high latency; second, delay-based algorithms (such as Timely) utilize RTT gradients as early congestion signals, effectively controlling latency, but are overly sensitive to random packet loss and network jitter, often causing a sharp drop in throughput; and third, learning-based algorithms (such as PCC-RL), while possessing theoretical adaptability, suffer from slow convergence, high computational overhead, and instability in unknown scenarios. Summary of the Invention

[0003] The main objective of this invention is to propose an adaptive gradient congestion control method, device, electronic device, storage medium, and program product based on utility function-driven approach, aiming to solve at least one problem in the prior art.

[0004] To achieve the above objectives, one aspect of this invention proposes an adaptive gradient congestion control method based on utility function-driven approaches, the method comprising: Obtain the monitoring parameters and historical monitoring values ​​of the target transmission link; The monitoring cycle is determined by a timer. If the monitoring period has not ended, the target window for different focused targets is quantified based on historical monitoring values ​​and real-time monitoring parameters; Based on the fusion factor corresponding to the previous monitoring period, each target window is dynamically fused with weights to update the congestion window of the target transmission link. If the monitoring period ends, the utility value of the monitoring period is obtained by quantifying the historical monitoring values ​​and the monitoring parameters within the monitoring period using a utility function. The fusion factor is then adaptively updated based on the incremental trend of the utility value to trigger the next monitoring period. Return to the step of determining whether the monitoring period has ended by using a timer, and continuously adjust the congestion window of the target transmission link.

[0005] In some embodiments, the monitoring parameters include instantaneous round-trip time and moving average latency gradient; historical monitoring values ​​include the last congestion event window value, timestamp variable, and baseline round-trip time; and the target window includes a first window aimed at preempting bandwidth and a second window aimed at suppressing latency. Quantifying the target windows with different focuses based on historical monitoring values ​​and real-time monitoring parameters includes the following steps: The first window is obtained by quantizing based on the window value and timestamp variable of the previous congestion event using a packet loss algorithm. The second window is obtained by using a delay-based algorithm to quantize based on instantaneous round-trip delay, baseline round-trip delay, and moving average delay gradient.

[0006] In some embodiments, a first window is obtained by quantizing based on the previous congestion event window value and timestamp variable using a packet loss algorithm, including the following steps: Based on the window value and timestamp variable of the previous congestion event, the first window is obtained by quantization through a preset cubic function, combined with a preset scaling factor and a multiplicative reduction factor. The expression for the first window is: ; In the formula, Indicates the first window; Indicates the preset scaling factor; Represents a timestamp variable; Indicates the window value of the last congestion event; parameter , This represents the multiplicative decreasing factor.

[0007] In some embodiments, a second window is obtained by quantizing based on a delay algorithm using instantaneous round-trip delay, a reference round-trip delay, and a moving average delay gradient, including the following steps: The gradient factor is determined based on the ratio of the difference between the instantaneous round-trip delay and the reference round-trip delay to the instantaneous round-trip delay. When the gradient factor is greater than the preset threshold, a second window is calculated based on the current congestion window using the gradient factor and the preset sensitivity coefficient. The expression for the second window is: ; In the formula, Indicates a second window; Indicates the current congestion window; Indicates the sensitivity coefficient; Indicates the gradient factor; When the gradient factor is less than or equal to the preset threshold, a second window is obtained based on the current congestion window using a preset additive growth constant. The expression for the second window is: ; In the formula, Indicates a second window; Indicates the current congestion window; This represents the additive growth constant.

[0008] In some embodiments, the monitoring parameters include real-time link capacity and average round-trip time within the monitoring period, and the historical monitoring values ​​include historical maximum link capacity and baseline round-trip time. Based on the historical monitoring values ​​and the monitoring parameters within the monitoring period, the utility value for the monitoring period is quantified using a utility function, including the following steps: The throughput term is constructed based on the ratio of real-time link capacity to the historical maximum link capacity; The delay term is constructed based on the ratio of the difference between the average round-trip delay and the baseline round-trip delay to the baseline round-trip delay; The utility value of the monitoring period is calculated based on the throughput and latency terms, combined with a preset latency penalty coefficient. The expression for utility value is: ; In the formula, This represents the utility value during the t-th monitoring period; This represents the real-time link capacity during the t-th monitoring period; Indicates the historical maximum link capacity; Indicates the delay penalty coefficient; This represents the average round-trip time during the t-th monitoring period; This indicates the base round-trip time delay.

[0009] In some embodiments, the fusion factor is adaptively updated based on the incremental trend of utility values, including the following steps: The utility value increment is obtained by quantifying the utility value based on the utility value of the two most recent monitoring periods; The fusion factor increment is obtained based on the fusion factor quantification of the most recent two monitoring periods; The target step size is obtained by quantizing the utility value increment and the fusion factor increment, and the fusion factor is updated using the target step size.

[0010] In some embodiments, the target step size is obtained based on the utility value increment and the fusion factor increment quantization, and the fusion factor is updated using the target step size, including the following steps: The target step size is obtained by multiplying the utility value increment and the fusion factor increment using a sign function and a preset step size factor. The expression for the target step size is: ; In the formula, Indicates the target step size; Indicates the step size factor; Represents a symbolic function; Indicates the increment of utility value; Indicates the increment of the fusion factor; The fusion factor is updated based on the target step size and step size factor. The expression for updating the fusion factor is: ; In the formula, This represents the fusion factor for the (t+1)th monitoring period; This represents the fusion factor for the t-th monitoring period.

[0011] To achieve the above objectives, another aspect of the present invention proposes an adaptive gradient congestion control device based on utility function-driven methods, the device comprising: The first module is used to acquire the monitoring parameters and historical monitoring values ​​of the target transmission link; The second module is used to determine whether the monitoring period has ended by using a timer. The third module is used to quantify the target window for different focused targets based on historical monitoring values ​​and real-time monitoring parameters if the monitoring period has not ended. The fourth module is used to perform weighted dynamic fusion of each target window based on the fusion factor corresponding to the previous monitoring period in order to update the congestion window of the target transmission link. The fifth module is used to quantify the utility value of the monitoring period based on historical monitoring values ​​and monitoring parameters within the monitoring period when the monitoring period ends, and adaptively update the fusion factor based on the incremental trend of the utility value to trigger the next monitoring period. The sixth module is used to return to the operation performed in the second module and continuously adjust the congestion window of the target transmission link.

[0012] To achieve the above objectives, another aspect of the present invention provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the aforementioned method.

[0013] To achieve the above objectives, another aspect of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method.

[0014] To achieve the above objectives, another aspect of the present invention provides a computer program product, including a computer program that, when executed by a processor, implements the aforementioned method.

[0015] The embodiments of this invention include at least the following beneficial effects: This invention provides a utility function-driven adaptive gradient congestion control method, device, electronic device, storage medium, and program product. This scheme acquires monitoring parameters and historical monitoring values ​​of the target transmission link; determines whether the monitoring period has ended using a timer; if the monitoring period has not ended, it quantizes target windows for different focused targets based on historical monitoring values ​​and real-time monitoring parameters; it dynamically fuses each target window based on the fusion factor corresponding to the previous monitoring period to update the congestion window of the target transmission link; if the monitoring period has ended, it obtains the utility value of the monitoring period using utility function quantization based on historical monitoring values ​​and monitoring parameters within the monitoring period, adaptively updates the fusion factor based on the incremental trend of the utility value, and triggers the next monitoring period; it returns to the step of determining whether the monitoring period has ended using a timer, continuously adjusting the congestion window of the target transmission link. By introducing a utility function-driven adaptive gradient fusion mechanism, the embodiments of this invention can effectively overcome the shortcomings of existing technologies. Specifically, by constructing a hybrid window model and a dynamic fusion factor, this invention can achieve an autonomous balance between high throughput characteristics and low latency gradient characteristics, thereby solving the problem of "unbalanced" performance of single-signal algorithms. Furthermore, this invention utilizes a utility function to quantify throughput gains and latency costs in real time, enabling online optimization of the fusion factor. This allows the algorithm to accurately distinguish between congested packet loss and random packet loss, avoiding link starvation caused by misjudgment in weak network environments and significantly improving robustness. In addition, the entire adjustment process is based on lightweight periodic feedback and numerical gradient updates, eliminating the need for complex model training. This balances convergence speed and computational efficiency while adapting to the practical deployment requirements of high-performance networks for low overhead and fast response. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of an implementation environment for the utility function-driven adaptive gradient congestion control method provided in this embodiment of the invention. Figure 2 This is a flowchart illustrating an adaptive gradient congestion control method based on utility function driving provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the overall process of adaptive gradient congestion control based on utility function driven by an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating a comparison of experimental results provided in an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of an adaptive gradient congestion control device based on utility function driving provided in an embodiment of the present invention; Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the embodiments of this invention; they are merely examples of apparatuses and methods consistent with some aspects of the embodiments of this invention as detailed in the appended claims.

[0018] It is understood that the terms “first,” “second,” etc., used in this invention may be used herein to describe various concepts, but unless specifically stated otherwise, these concepts are not limited by these terms. These terms are used only to distinguish one concept from another. For example, first information may also be referred to as second information without departing from the scope of embodiments of the invention, and similarly, second information may also be referred to as first information. Depending on the context, the words “if,” “when,” or “in response to determination” as used herein may be interpreted as “when…” or “when…” or “in response to determination.”

[0019] The terms “at least one,” “multiple,” “each,” “any,” etc., used in this invention, “at least one” includes one, two, or more than two; “multiple” includes two or more than two; “each” refers to each of the corresponding multiple; and “any” refers to any one of the multiple.

[0020] 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. The terminology used herein is for the purpose of describing embodiments of the invention only and is not intended to limit the invention.

[0021] To facilitate understanding of the technical solutions of this invention, the technical terms of the proprietary technical means that may be involved in the embodiments of this invention will first be explained: AGC (Ada-Gradient Cubic): The name of the core congestion control algorithm proposed in this invention. It is an adaptive algorithm that integrates traditional cubic function logic (Cubic) with delay gradient control logic, aiming to balance ultra-high throughput and extremely low queuing latency in high-performance networks through utility function-driven approaches.

[0022] Hybrid Window Model: A dual-mode computational architecture for the algorithm. In this model, the system no longer generates a single congestion window, but instead simultaneously computes two target windows: a Cubic window aimed at preempting bandwidth, and a Gradient window aimed at suppressing latency. The final transmission window is a weighted fusion of these two windows.

[0023] Dynamic fusion factor (Alpha, α): A key parameter used to balance the two logical weights in the hybrid window model. The value of α ranges from 0 to 1: when α approaches 1, the protocol behaves more like a latency-based algorithm, focusing on reducing latency; when α approaches 0, the protocol behaves more like a packet loss-based algorithm, focusing on maximizing bandwidth utilization.

[0024] Utility Function: A mathematical evaluation system for quantifying transmission efficiency. It weights the positive throughput gains with the negative latency losses (penalties) to arrive at a numerical value reflecting the overall transmission efficiency under the current network conditions. It serves as the "guide" for algorithms to make adaptive decisions.

[0025] Gradient Ascent: An online optimization mathematical method for finding the maximum value of a utility function. At each decision point, the algorithm automatically determines the next adjustment direction (increase or decrease) of α by observing the positive or negative impact of small changes in the fusion factor α on the utility function value, thereby enabling the protocol strategy to evolve towards the optimal transmission performance point.

[0026] Monitoring Interval (MI): The smallest unit of time for algorithm execution feedback and decision-making. Within a monitoring interval, the system collects sufficient data samples (such as sampling RTT, statistical throughput), and at the end of the interval, calculates the utility value and performs a gradient ascent iteration to update the fusion factor.

[0027] RTT Gradient: The rate of change of Round-Trip Time over time. It is a core signal for detecting early network congestion: when the RTT gradient is positive and continues to increase, it indicates that data is starting to accumulate in the intermediate switch buffer; it is the main basis for the gradient pattern window shrinking in the AGC algorithm.

[0028] Buffer bloat: A phenomenon where excessively large buffers reserved by intermediate network nodes (such as routers or switches) cause packet loss-based congestion control algorithms to continuously fill with data, leading to a sharp increase in transmission latency and exacerbated jitter. AGC aims to eliminate this phenomenon by introducing RTT gradient awareness and proactively slowing down the buffers before they become full.

[0029] Among related technologies, existing technologies lack an autonomous adjustment mechanism that can dynamically balance throughput and latency based on real-time network conditions, making it difficult to maintain optimal performance in complex heterogeneous network environments.

[0030] In view of this, this invention provides an adaptive gradient congestion control method and related equipment based on utility function-driven approaches. This method acquires monitoring parameters and historical monitoring values ​​of the target transmission link; determines whether the monitoring period has ended using a timer; if the monitoring period has not ended, it quantizes target windows for different emphase targets based on historical monitoring values ​​and real-time monitoring parameters; it dynamically fuses each target window based on the fusion factor corresponding to the previous monitoring period to update the congestion window of the target transmission link; if the monitoring period has ended, it quantizes the utility value of the monitoring period using a utility function based on historical monitoring values ​​and monitoring parameters within the monitoring period, adaptively updates the fusion factor based on the incremental trend of the utility value, and triggers the next monitoring period; it then returns to the step of determining whether the monitoring period has ended using a timer, continuously adjusting the congestion window of the target transmission link. This invention, by introducing a utility function-driven adaptive gradient fusion mechanism, effectively overcomes the shortcomings of existing technologies. Specifically, by constructing a hybrid window model and a dynamic fusion factor, this invention can achieve an autonomous balance between high throughput and low latency gradient characteristics, thereby solving the problem of "unbalanced" performance of single-signal algorithms. Furthermore, this invention utilizes a utility function to quantify throughput gains and latency costs in real time, enabling online optimization of the fusion factor. This allows the algorithm to accurately distinguish between congested packet loss and random packet loss, avoiding link starvation caused by misjudgment in weak network environments and significantly improving robustness. In addition, the entire adjustment process is based on lightweight periodic feedback and numerical gradient updates, eliminating the need for complex model training. This balances convergence speed and computational efficiency while adapting to the practical deployment requirements of high-performance networks for low overhead and fast response.

[0031] It is understood that the utility function-driven adaptive gradient congestion control method provided by this invention can be applied to any computer device with data processing and computing capabilities, and this computer device can be various terminals or servers. When the computer device in the embodiment is a server, the server is an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. Optionally, the terminal can be a smartphone, tablet, laptop, or desktop computer, but it is not limited to these.

[0032] like Figure 1 The diagram shown is a schematic representation of an implementation environment provided by an embodiment of the present invention. (Refer to...) Figure 1 The implementation environment includes at least one terminal 102 and a server 101. The terminal 102 and the server 101 can be connected via a network, either wirelessly or via a wired connection, to complete data transmission and exchange.

[0033] Server 101 can be a standalone physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0034] Additionally, server 101 can also be a node server in a blockchain network. Blockchain is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms.

[0035] Terminal 102 can be a smartphone, tablet computer, laptop computer, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. Terminal 102 and server 101 can be directly or indirectly connected via wired or wireless communication, and this embodiment of the invention does not impose any limitations.

[0036] For example, based on Figure 1The implementation environment shown in this embodiment of the invention provides an adaptive gradient congestion control method based on utility function. The following description uses the application of this adaptive gradient congestion control method based on utility function in server 101 as an example. It can be understood that this adaptive gradient congestion control method based on utility function can also be applied to terminal 102.

[0037] Reference Figure 2 , Figure 2 This is an optional flowchart of the utility function-driven adaptive gradient congestion control method provided in the embodiments of the present invention. The execution subject of the utility function-driven adaptive gradient congestion control method can be any of the aforementioned computer devices (including servers or terminals). Figure 2 The method may include, but is not limited to, steps S100 to S600.

[0038] Step S100: Obtain the monitoring parameters and historical monitoring values ​​of the target transmission link; For example, in some specific implementations, the monitoring parameters include instantaneous round-trip time, moving average latency gradient, and real-time link capacity and average round-trip time in each monitoring period. Historical monitoring values ​​include the last congestion event window value, timestamp variable, baseline round-trip time, and historical maximum link capacity (or the network interface card rated bandwidth can be used directly).

[0039] Step S200: Determine whether the monitoring period has ended by using a timer; For example, in some specific implementations, the data transmission process of the target transmission link is continuously divided into several monitoring periods along the time sequence by a timer, so as to dynamically adjust the fusion factor of each monitoring period according to the data transmission situation to achieve adaptive gradient congestion control.

[0040] Step S300: If the monitoring period has not ended, quantify the target window for different focused targets based on historical monitoring values ​​and real-time monitoring parameters; It should be noted that the monitoring parameters include instantaneous round-trip time (RTT) and moving average delay gradient, historical monitoring values ​​include the last congestion event window value, timestamp variable, and baseline RTT, and target windows include a first window aimed at preempting bandwidth and a second window aimed at suppressing latency. In some embodiments, quantizing target windows with different focuses based on historical monitoring values ​​and real-time monitoring parameters may include the following steps: obtaining the first window based on the last congestion event window value and timestamp variable using a packet loss-based algorithm; and obtaining the second window based on the instantaneous RTT, baseline RTT, and moving average delay gradient using a latency-based algorithm.

[0041] For example, in some specific embodiments, the present invention is not limited to generating a single congestion window, but instead calculates two target windows simultaneously: one is a Cubic window aimed at preempting bandwidth (…). The first window is the gradient window, and the second is the gradient window, which aims to suppress latency. (i.e., the second window). The final sending window (congestion window) is formed by a weighted fusion of these two windows.

[0042] It should be noted that in some embodiments, the first window is obtained by quantization based on the previous congestion event window value and timestamp variable using a packet loss algorithm, which may include the following steps: quantizing the first window using a preset cubic function based on the previous congestion event window value and timestamp variable, combined with a preset scaling factor and a multiplicative decreasing factor.

[0043] For example, in some specific implementations, Primarily used to ensure high throughput performance in stable network environments, its data includes: Last congestion event window value ( ): Records the peak value of the window when packet loss occurs.

[0044] timestamp variable ( ): The physical time the system has been running since the last packet loss was detected and the window was reduced.

[0045] Preset scaling factor ( ): A constant used to control the slope of the growth curve.

[0046] The target value is determined by the following cubic function:

[0047] Among them, parameters (i.e., the window is restored to) The required time deviation is calculated by the following formula:

[0048] in This is a preset multiplicative decreasing factor, typically around 0.7.

[0049] The data used to achieve latency-sensitive predictive adjustments includes: Instantaneous round-trip delay ( ): The real-time RTT measured when the current ACK arrives.

[0050] Baseline round-trip delay ( ): The minimum RTT observed during the connection's lifetime, representing the physical propagation delay when there is no backlog in the link.

[0051] Moving average time delay gradient ( ): The trend of RTT change calculated by sliding window.

[0052] It should be noted that in some embodiments, the second window is obtained by quantizing based on a delay-based algorithm based on the instantaneous round-trip delay, the baseline round-trip delay, and the moving average delay gradient. This may include the following steps: determining a gradient factor based on the ratio of the difference between the instantaneous round-trip delay and the baseline round-trip delay to the instantaneous round-trip delay; when the gradient factor is greater than a preset threshold, calculating the second window based on the current congestion window using the gradient factor and a preset sensitivity coefficient; when the gradient factor is less than or equal to the preset threshold, calculating the second window based on the current congestion window using a preset additive growth constant.

[0053] For example, in some specific implementations, The calculation process consists of two steps: gradient determination and window mapping. Step A: Calculate the gradient factor :

[0054] Step B: Determine the target window based on the gradient interval: when (Positive gradient, buffer backlog):

[0055] Note: This is a sensitivity coefficient used to adjust the degree of window back-off triggered by increased latency; This indicates the current congestion window.

[0056] In an optional embodiment of the present invention, the preset threshold involved in the steps This threshold is used to determine the current queuing backlog risk on the link. The value is set based on the physical characteristics of the target transmission link and the real-time disturbance tolerance. Specifically, for low-latency data center networks (DCNs), due to the short link distances and extreme sensitivity to buffer backlog, the preset threshold... The preferred value range is 1.1 to 1.2; for wide area networks (WANs) spanning multiple regions, considering path delay fluctuations and normal network jitter, the preset threshold... The preferred value range is 1.5 to 2.0; for wireless mobile networks with high-frequency random packet loss and physical layer jitter, the preset threshold... The preferred value range is 2.5 to 3.0. Through these differentiated threshold settings, this invention ensures accurate identification of congestion gradients under different network architectures, avoiding unnecessary frequent switching between packet loss-based and latency-based control modes.

[0057] when (Stable gradient, link saturation):

[0058] Note: It is an additive growth constant used to probe the remaining bandwidth without significantly increasing latency.

[0059] Step S400: Based on the fusion factor corresponding to the previous monitoring period, perform weighted dynamic fusion of each target window to update the congestion window of the target transmission link. For example, in some specific implementations, multiple target windows (such as those constructed based on the aforementioned steps) are used. and ), final (congestion) window It can be obtained through the following weighted dynamic fusion:

[0060] in, It is a dynamic fusion factor. The larger the value, the more the algorithm tends to control latency; The smaller the value, the more likely it is to increase throughput.

[0061] Step S500: If the monitoring period ends, the utility value of the monitoring period is obtained by quantifying the historical monitoring values ​​and the monitoring parameters within the monitoring period using the utility function. The fusion factor is adaptively updated based on the incremental trend of the utility value to trigger the next monitoring period. For example, the core decision-making aspect of the algorithm of this invention lies in using a utility function to quantify transmission performance in real time and dynamically driving the fusion factor through the gradient ascent method. The evolution of this mechanism allows the protocol to automatically find the optimal balance between "bandwidth probing" and "latency control".

[0062] It should be noted that, in some embodiments, the utility value of the monitoring period is quantified using a utility function based on historical monitoring values ​​and monitoring parameters within the monitoring period, which may include the following steps: The throughput term is constructed based on the ratio of real-time link capacity to the historical maximum link capacity; The delay term is constructed based on the ratio of the difference between the average round-trip delay and the baseline round-trip delay to the baseline round-trip delay; The utility value of the monitoring period is calculated based on the throughput and latency terms, combined with a preset latency penalty coefficient. The expression for utility value is: ; In the formula, This represents the utility value during the t-th monitoring period; This represents the real-time link capacity during the t-th monitoring period; Indicates the historical maximum link capacity; Indicates the delay penalty coefficient; This represents the average round-trip time during the t-th monitoring period; This indicates the base round-trip time delay.

[0063] For example, in some specific implementations, the normalized utility function model is as follows: To eliminate the difference between throughput (Gbps) and latency ( Due to the significant differences in dimensions and numerical magnitudes between the signals, this invention first normalizes the network feedback signal. The normalized utility function is defined as follows:

[0064] Explanation of the normalization formula: This represents the utility value during the t-th monitoring period; Throughput Item ( ): This represents the maximum historical link capacity (or network interface card rated bandwidth) known for the current path. After normalization, this item is mapped to... The interval represents the bandwidth utilization rate. This represents the real-time link capacity during the t-th monitoring period; This indicates the historical maximum link capacity.

[0065] Delayed term ( ): This represents the average round-trip time during the t-th monitoring period. This represents the link propagation delay (i.e., the baseline round-trip delay). This value calculates the percentage increase in additional queuing delay relative to the baseline delay. It is 0 when there is no queue backlog and 1 when queuing doubles the delay.

[0066] Delay penalty coefficient This is a core weight that adjusts the protocol's "personality," and its value determines the AGC algorithm's tolerance for latency fluctuations. Its value can be set according to the service's QoS requirements. For real-time interactive services (such as remote control), it is recommended to... To achieve extremely high latency sensitivity; for general data transmission, it is recommended This ensures that throughput gain and latency loss have equal weight.

[0067] It should be noted that, in some embodiments, the adaptive update of the fusion factor based on the incremental trend of the utility value may include the following steps: quantizing the utility value based on the utility value of the two most recent monitoring periods to obtain the utility value increment; quantizing the fusion factor based on the fusion factor of the two most recent monitoring periods to obtain the fusion factor increment; quantizing the utility value increment and the fusion factor increment to obtain the target step size, and updating the fusion factor using the target step size.

[0068] It should be noted that in some embodiments, the target step size is obtained based on the quantization of the utility value increment and the fusion factor increment, and the fusion factor is updated using the target step size, which may include the following steps: The target step size is obtained by multiplying the utility value increment and the fusion factor increment using a sign function and a preset step size factor; the fusion factor is then updated based on the target step size and the step size factor.

[0069] For example, in some specific implementations, adaptive evolution is based on gradient ascent: Within each monitoring interval (MI), the algorithm updates the fusion factor using the gradient ascent method based on the incremental trend of the utility value. :

[0070] in Step size factor The step size.

[0071] (1) Step size factor Specific values: Step size factor The sensitivity and stability of strategy evolution are determined and are usually set at... Between. Too large This will lead to It oscillates violently between Cubic and Gradient modes and fails to converge; too small This will cause the algorithm to react too slowly to sudden network conditions (such as a sudden drop in bandwidth). The recommended initial value in this invention is... .

[0072] (2) Step length Calculation method: To ensure The evolution can achieve both rapid response and smooth convergence. This invention uses the numerical gradient difference method to calculate the step size increment. At the end of each MI, the utility value increment is calculated. and the increase in fusion factor However, in a network environment, and The value is extremely small and full of noise. Directly... and The step size (original gradient) obtained by division will fluctuate drastically and may even produce a "division by zero" error. Therefore, a fixed step size control is used here, utilizing... The function removes the drastically fluctuating numerical magnitudes, retaining only the direction of the increment. This is combined with a step size factor. ,make Approaching the optimal solution at a constant and controlled pace fundamentally eliminates the possibility of drastic oscillations between Cubic and Gradient modes. The final step size is calculated as follows:

[0073] That is: if utility follows As it increases, it continues to increase. (Tilted towards lower latency); if utility follows If the increase is accompanied by a decrease, then the decrease is... (Regression to high throughput). In this way, A dynamic equilibrium will be achieved near the peak of the utility curve. Represents a symbolic function. .

[0074] Step S600: Return to the step of determining whether the monitoring period has ended by using a timer, and continuously adjust the congestion window of the target transmission link; For example, in some specific implementations, the data transmission process of the target transmission link is continuously divided into several monitoring periods along the time sequence, so as to continuously and dynamically adjust the fusion factor of each monitoring period according to the data transmission situation to achieve adaptive gradient congestion control.

[0075] To explain in detail the principle of the technical solution of the present invention, the overall process of the present invention will be described below with reference to some specific embodiments. It is easy to understand that the following is an explanation of the technical principle of the present invention and should not be regarded as a limitation of the present invention.

[0076] First, it's important to clarify that in modern high-performance network communication, especially in data centers, cloud computing, and ultra-large-scale wide area network scenarios, ensuring reliable data transmission while maintaining extremely high throughput and extremely low latency is a core requirement. As network bandwidth moves towards the 100Gbps level, the evolution of the underlying transmission architecture has laid the foundation for high-performance communication. However, achieving precise traffic scheduling in high-throughput environments still depends on the congestion control (CC) mechanisms of upper-layer protocols.

[0077] Currently, the most commonly used existing technology in high-performance network protocol stacks is the congestion control algorithm based on a single-dimensional signal-driven approach. Its representative approaches mainly include the following three: 1. Loss-based algorithms: represented by Cubic. These algorithms detect congestion by monitoring network packet loss events and use a cubic function to adjust the congestion window (cwnd) to achieve high bandwidth utilization.

[0078] 2. Delay-based algorithms: represented by Timely. These algorithms primarily utilize the gradient changes in round-trip time (RTT) as early congestion signals, minimizing queuing delays through precise rate control.

[0079] 3. Learning-based algorithms: Represented by PCC-RL (Performance-oriented Congestion Control with Reinforcement Learning). It utilizes a deep reinforcement learning (DRL) model to automatically generate rate adjustment decisions based on network observation characteristics (such as transmission rate, latency inflation, etc.).

[0080] Although the congestion algorithms mentioned above perform well in specific scenarios, they still have significant limitations when facing increasingly complex heterogeneous networks (such as environments with high bandwidth, high random packet loss, and long, fat pipes).

[0081] Limitation 1: Static, single feedback signal leads to compromises. Mainstream congestion control algorithms typically rely on a single feedback signal, making it difficult to balance throughput and latency in diverse network scenarios. Loss-based algorithms (such as Cubic) only reduce the sending rate when network packet loss is detected, tending to fill the buffer before packet loss occurs, resulting in severe buffer bloat and high latency. Conversely, delay-based algorithms (such as Timely) are extremely sensitive to round-trip time (RTT). While they can control latency, they are too conservative when faced with network jitter or random packet loss not caused by congestion, leading to a sharp drop in throughput.

[0082] Limitation 2: Inability to adapt to extremely weak network conditions and random packet loss. In weak network environments with low bandwidth and high packet loss (such as 5% random packet loss), existing technologies generally exhibit extremely poor robustness. Because they cannot effectively distinguish between "packet loss caused by network congestion" and "random packet loss inherent in the link itself," loss-based and delay-based algorithms often misjudge random packet loss as a severe congestion signal, thus incorrectly suppressing the congestion window to an extremely low level. This "misjudgment" leads to link starvation, causing actual transmission to collapse completely even if the physical link still has spare capacity under algorithms such as Cubic, Timely, or PCC.

[0083] Limitation 3: Learning-based algorithms suffer from performance and overhead bottlenecks. Although reinforcement learning (RL) algorithms (such as PCC-RL) have strong theoretical adaptability, they face many limitations in practical deployment. First, these algorithms typically require continuous inference and complex policy updates, resulting in high computational overhead and making it difficult to meet the low CPU usage requirements of high-performance user-space protocol stacks. Second, these algorithms have slow convergence speeds, especially when network conditions fluctuate drastically or when facing unseen scenarios; their performance is often uncontrollable and they struggle to quickly reach full bandwidth.

[0084] Limitation 4: Lack of a dynamically adjusting balancing mechanism. Existing technologies exhibit a "lopsided" design goal, lacking a navigation mechanism capable of automatically finding the optimal balance point based on real-time traffic conditions. For example, Cubic prioritizes high throughput at the expense of latency, while Timely prioritizes low latency but is susceptible to jitter. In multi-path transmission scenarios, if the algorithm cannot dynamically adjust its aggressiveness based on the real-time utility of each path, it cannot maintain optimal transmission performance across different network lifecycles (switching from high bandwidth to high packet loss).

[0085] In view of the shortcomings of existing technologies, this invention proposes an adaptive gradient congestion control strategy, Ada-Gradient Cubic (AGC), driven by a utility function. This algorithm combines the high bandwidth utilization of Cubic with the low latency of gradient-based algorithms, and introduces a dynamic adjustment mechanism based on a utility function to automatically find the optimal balance point under the current network environment.

[0086] Congestion control (CC) is a core mechanism for ensuring the efficient and stable operation of networks. However, current mainstream CC algorithms often fall short in addressing increasingly complex network environments (such as high-bandwidth, low-latency data centers, wireless networks with high random packet loss, and long, thick-pipe WANs). Loss-based algorithms (such as TCP Cubic) tend to fill the buffer when no packet loss occurs in the network, leading to severe buffer bloat and high latency.

[0087] Delay-based algorithms (such as TCP Vegas and Timely) can effectively control latency, but they are too conservative when faced with public network jitter or non-congestion packet loss, resulting in a significant drop in throughput.

[0088] While learning-based algorithms (such as PCC-RL and Aurora) are highly adaptable, they often face problems such as slow convergence speed, high computational cost, and uncontrollable performance in unseen scenarios.

[0089] To address the limitations of the aforementioned single mechanism, this invention proposes Ada-Gradient Cubic (AGC). This is a hybrid algorithm that combines the high bandwidth utilization of Cubic with the low latency of gradient-based algorithms, and automatically finds the optimal balance point in the current network environment through a dynamic adjustment mechanism based on a utility function.

[0090] like Figure 3 The diagram illustrates the complete processing logic of the AGC algorithm after receiving an ACK at the receiver: First, it updates real-time statistics such as RTT and throughput. Then, it determines whether the monitoring period (MI) has ended. If the period has ended, it enters the core adaptive adjustment module, calculates the current performance utility value and its gradient, and dynamically adjusts the direction and magnitude of the fusion factor α using a gradient ascent algorithm to optimize the strategy. Immediately afterwards (or directly if the period has not ended), the algorithm calculates two basic target windows, Cubic and Gradient, in parallel, and dynamically fuses them using the current α value, ultimately deriving the AGC target window to update the current congestion window (cwnd) and send new data packets. In some specific application scenarios, the utility function-driven adaptive gradient congestion control technical solution of this invention can be implemented as follows: 1. Design concept: AGC believes that there is no single static rule applicable to all networks. Algorithms should possess self-awareness and adjustment capabilities.

[0091] Skeleton: Based on the Cubic growth curve, it ensures that resources can be quickly preempted when bandwidth is abundant.

[0092] Sensor: Introduce RTT gradient as an early congestion signal to sense the trend of queue establishment.

[0093] Brain: Introduces utility functions and gradient ascent algorithms to periodically evaluate the effectiveness of the current strategy and dynamically adjust the dependence on the two signals mentioned above.

[0094] 2. Core Components: a) Mixed window calculation: AGC calculates two target windows each time an ACK (Acknowledgment, a control message sent back by the receiver to the sender) arrives. and And perform weighted fusion to obtain a hybrid window . It is the target window calculated by the standard Cubic algorithm. The target window is calculated based on the RTT gradient (if the gradient is positive, the window is reduced; if the gradient is stable, the window is kept or slightly increased).

[0095] Primarily used to ensure high throughput performance in stable network environments, its data includes: Last congestion event window value ( ): Records the peak value of the window when packet loss occurs.

[0096] timestamp variable ( ): The physical time the system has been running since the last packet loss was detected and the window was reduced.

[0097] Preset scaling factor ( ): A constant used to control the slope of the growth curve.

[0098] The target value is determined by the following cubic function:

[0099] Among them, parameters (i.e., the window is restored to) The required time deviation is calculated by the following formula:

[0100] in This is a preset multiplicative decreasing factor, typically around 0.7.

[0101] The calculation process consists of two steps: gradient determination and window mapping. Step A: Calculate the gradient factor :

[0102] Step B: Determine the target window based on the gradient interval: when (Positive gradient, buffer backlog):

[0103] Note: This is a sensitivity coefficient used to adjust the degree of window back-off triggered by increased latency; This indicates the current congestion window.

[0104] when (Stable gradient, link saturation):

[0105] Note: It is an additive growth constant used to probe the remaining bandwidth without significantly increasing latency.

[0106] Final Window for:

[0107] in It is a dynamic fusion factor. The larger the value, the more the algorithm tends to control latency; The smaller the value, the more likely it is to increase throughput.

[0108] b) Utility-based adaptive adjustment: The core of this invention's algorithm lies in using a utility function to quantify transmission performance in real time and dynamically driving the fusion factor through a gradient ascent method. The evolution of this mechanism allows the protocol to automatically find the optimal balance between "bandwidth probing" and "latency control".

[0109] Normalized utility function model: To eliminate the difference between throughput (Gbps) and latency ( Due to the significant differences in dimensions and numerical magnitudes between the signals, this invention first normalizes the network feedback signal. The normalized utility function is defined as follows:

[0110] Explanation of the normalization formula: Throughput Item ( ): This represents the maximum historical link capacity (or network interface card rated bandwidth) known for the current path. After normalization, this item is mapped to... The interval represents the bandwidth utilization rate.

[0111] Delayed term ( ): This represents the link propagation delay (baseline round-trip time). This value calculates the percentage increase in additional queuing delay relative to the base delay. It is 0 when there is no queue backlog and 1 when queuing doubles the delay.

[0112] Delay penalty coefficient This is a core weight that adjusts the protocol's "personality," and its value determines the AGC algorithm's tolerance for latency fluctuations. Its value can be set according to the service's QoS requirements. For real-time interactive services (such as remote control), it is recommended to... To achieve extremely high latency sensitivity; for general data transmission, it is recommended This ensures that throughput gain and latency loss have equal weight.

[0113] Gradient-ascent-based adaptive evolution: Within each monitoring interval (MI), the algorithm updates the fusion factor using the gradient ascent method based on the incremental trend of the utility value. :

[0114] in Step size factor The step size.

[0115] (1) Step size factor Specific values: Step size factor The sensitivity and stability of strategy evolution are determined and are usually set at... Between. Too large This will lead to It oscillates violently between Cubic and Gradient modes and fails to converge; too small This will cause the algorithm to react too slowly to sudden network conditions (such as a sudden drop in bandwidth). The recommended initial value in this invention is... .

[0116] (2) Step length Calculation method: To ensure The evolution can achieve both rapid response and smooth convergence. This invention uses the numerical gradient difference method to calculate the step size increment. At the end of each MI, the utility value increment is calculated. and the increase in fusion factor However, in a network environment, and The value is extremely small and full of noise. Directly... and The step size (original gradient) obtained by division will fluctuate drastically and may even produce a "division by zero" error. Therefore, a fixed step size control is used here, utilizing... The function removes the drastically fluctuating numerical magnitudes, retaining only the direction of the increment. This is combined with a step size factor. ,make Approaching the optimal solution at a constant and controlled pace fundamentally eliminates the possibility of drastic oscillations between Cubic and Gradient modes. The final step size is calculated as follows:

[0117] That is: if utility follows As it increases, it continues to increase. (Tilted towards lower latency); if utility follows If the increase is accompanied by a decrease, then the decrease is... (Regression to high throughput). In this way, A dynamic equilibrium will be achieved near the peak of the utility curve. Represents a symbolic function. .

[0118] In some specific implementations, to verify the effectiveness of the proposed AGC congestion control strategy, experiments were conducted under different scenarios. The experiments employed three typical network environments: a high-bandwidth, low-packet-loss environment (data center / 5G), a low-bandwidth, high-packet-loss environment (weak network / edge network), and a medium-standard environment (standard WAN). In each environment, the AGC algorithm of this invention was compared with traditional algorithms such as TCP Cubic, Timely, and PCC-RL. Experimental results are as follows: Figure 4 The experimental results are shown.

[0119] In high-bandwidth, low-packet-loss scenarios, the AGC algorithm performs exceptionally well in both throughput and latency control. Experimental results show that AGC maintains near-maximum throughput even under network fluctuations, effectively avoiding the throughput drop issue that occurs with the Cubic algorithm after network fluctuations. Regarding latency, although AGC's throughput is higher than other algorithms, its median RTT remains within a reasonable range, outperforming the Cubic algorithm's latency increase due to packet loss, and significantly outperforming the Timely algorithm's low-bandwidth performance.

[0120] In low-bandwidth, high-packet-loss environments, the AGC algorithm demonstrates significant advantages. With a 5% packet loss rate, AGC can stably maintain a throughput of 200-500 pps, significantly higher than other algorithms. Especially when dealing with random packet loss, AGC adaptively adjusts its utility function, avoiding the performance crashes caused by Cubic and Timely algorithms misclassifying packet loss as congestion. This result proves the robustness and superiority of AGC in weak network environments.

[0121] In standard WAN environments, AGC also demonstrated superior throughput and better latency control. Compared to Cubic and Timely, AGC can quickly recover throughput, maintaining a high connection rate even with 1% packet loss, without sacrificing bandwidth utilization. Furthermore, AGC's window changes exhibit high-frequency and rapid adjustment characteristics, fully demonstrating its ability to effectively find the optimal operating point in dynamic environments.

[0122] In some optional implementations, the following experimental parameters can be used for different network environments: (1) High bandwidth, low packet loss environment (simulated data center or 5G core network scenario): In this environment, the focus is on verifying the invention's ability to maintain extremely low latency. Experimental parameters are set as follows: bottleneck link bandwidth is set to 10Gbps~100Gbps, and basic round-trip latency is set to 1ms~5ms. To simulate the typical shallow buffer characteristics of a data center, the router buffer size is set to 0.2 times the bandwidth-latency product. In this environment, a preset threshold is used. The value is set to 1.1 to 1.2, which is intended to trigger gradient adjustment when latency increases slightly, ensuring that high throughput and extremely low queuing latency coexist.

[0123] (2) Low bandwidth and high packet loss environment (simulating weak network or edge wireless network scenarios): In this environment, the robustness of the invention to random packet loss (non-congestion packet loss) is verified. Experimental parameters are set as follows: bottleneck link bandwidth is set to 10Mbps~50Mbps, and propagation delay is set to 50ms~150ms. Based on this, an independent random packet loss rate (BER) of 1%~5% is introduced into the bottleneck link. A preset threshold is also used. The value is set to 2.5 to 3.0 to tolerate physical jitter in the wireless link.

[0124] (3) Medium standard environment (simulating standard WAN scenario): In this environment, the general performance of this invention in long, thick internet pipelines was verified. The experimental parameters were set as follows: bottleneck link bandwidth was set to 100Mbps~1Gbps, and basic round-trip latency was set to 30ms~80ms. To simulate dynamic load, 20%~40% cross-traffic was injected into the link as background interference. A preset threshold was also set. The value ranges from 1.5 to 2.0.

[0125] Based on the comprehensive experimental results, the AGC algorithm of this invention exhibits excellent performance in various network environments, especially demonstrating significant advantages in scenarios with high packet loss and low bandwidth. Experiments verify the effectiveness of the utility function-based adaptive adjustment mechanism, proving that the algorithm can achieve a balance between throughput and latency under different network conditions, exhibiting strong adaptability and robustness.

[0126] In summary, the core of this invention lies in the integration of a utility function-driven adaptive gradient congestion control (AGC) algorithm. By dynamically adjusting the hybrid window ratio using the gradient ascent method, the protocol can automatically identify and combat random packet loss. While ensuring nanosecond-level low latency, it achieves throughput performance several times that of traditional algorithms in complex weak network environments.

[0127] This invention constructs a dual-mode driven hybrid window computing model, breaking the limitations of a single feedback signal. The algorithm introduces a dynamic fusion factor α to weightedly fuse the Cubic mode, which possesses high bandwidth preemption capabilities, with the gradient-based mode, which senses early congestion. This design allows the sending window to simultaneously absorb the advantages of both logics: rapidly increasing throughput using the cubic function curve of Cubic when the link is idle, and suppressing delay using gradient logic when the round-trip time (RTT) gradient is detected, thus fundamentally resolving the contradiction between bandwidth utilization and queuing delay in high-performance networks.

[0128] This invention integrates an online gradient optimization mechanism based on utility functions, which serves as the "brain" for adaptive decision-making. The system no longer relies on preset static parameters but instead calculates a utility function in real time, incorporating both throughput gains and latency losses. Within each monitoring cycle, the algorithm dynamically evolves the fusion factor α using gradient ascent: when increasing α yields higher overall utility, the algorithm tilts towards lower latency; conversely, it reverts to higher throughput. This dynamic evolution capability allows the protocol to automatically find the optimal balance point based on real-time traffic conditions, avoiding the blindness of manual tuning.

[0129] Specifically, this invention aims to solve the following core technical problems through a utility function-driven adaptive gradient congestion control algorithm (Ada-Gradient Cubic, AGC): 1. Addressing the performance imbalance and poor environmental adaptability caused by a single congestion signal: Existing algorithms such as Cubic are prone to buffer bloat, while Timely suffers throughput collapse under random packet loss conditions. The technical problem this invention aims to solve is how to integrate packet loss events, round-trip time gradients (RTGradient), and performance utility values ​​from multiple dimensions. By designing a dynamic adjustment mechanism, the algorithm can automatically find the optimal balance between pursuing high bandwidth utilization (Cubic characteristics) and maintaining low latency (gradient control characteristics) based on real-time network conditions.

[0130] 2. Solving the problems of misjudgment and link starvation caused by non-congestion packet loss in weak network environments: In extreme weak network environments such as 5% high packet loss, existing technologies often misjudge random packet loss as congestion, leading to the collapse of the entire transmission line. The technical problem to be solved by this invention is to introduce a utility function evaluation mechanism to quantify the gains in throughput and latency in real time. When the algorithm detects that excessive window restriction causes a decrease in utility value, it can automatically identify non-congestion packet loss and adjust the strategy, thereby maintaining stable throughput even in complex packet loss environments. Its throughput performance can reach more than 5 times that of traditional algorithms.

[0131] 3. During network operation, the state undergoes drastic shifts in real time, from "high bandwidth and low latency" to "low bandwidth and high packet loss." Existing static algorithms struggle to automatically adjust their strategy preferences across different lifecycles. The technical problem this invention aims to solve is by introducing a gradient ascent optimization mechanism, dynamically evolving the fusion factor α through online closed-loop feedback. Gradient control is strengthened when latency gains outweigh throughput losses; otherwise, it reverts to a Cubic mode, enabling the protocol to possess self-healing capabilities and rapid convergence in unknown and variable scenarios.

[0132] like Figure 5 As shown, this embodiment of the invention also provides a utility function-driven adaptive gradient congestion control device 900, which can implement the above-described method. This device may include: The first module 910 is used to acquire the monitoring parameters and historical monitoring values ​​of the target transmission link; The second module 920 is used to determine whether the monitoring period has ended by using a timer. The third module 930 is used to quantify the target window for different focused targets based on historical monitoring values ​​and real-time monitoring parameters if the monitoring period has not ended. The fourth module 940 is used to perform weighted dynamic fusion of each target window based on the fusion factor corresponding to the previous monitoring period in order to update the congestion window of the target transmission link. The fifth module 950 is used to quantify the utility value of the monitoring period based on historical monitoring values ​​and monitoring parameters within the monitoring period when the monitoring period ends, and adaptively update the fusion factor based on the incremental trend of the utility value to trigger the next monitoring period. Module 6, 960, is used to return to the operation performed in Module 2 and continuously adjust the congestion window of the target transmission link.

[0133] It is understood that the content of the above method embodiments is applicable to the present device embodiments. The specific functions implemented by the present device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0134] This invention also provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described above. This electronic device can be any smart terminal, including tablet computers, in-vehicle computers, etc.

[0135] It is understood that the content of the above method embodiments is applicable to this device embodiment. The specific functions implemented by this device embodiment are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0136] like Figure 6 As shown, Figure 6 The hardware structure of an electronic device 1000 according to another embodiment is illustrated. The electronic device 1000 includes: The processor 1001 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (aSIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention. The memory 1002 can be implemented as a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RaM). The memory 1002 can store the operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 1002 and is called and executed by the processor 1001. Input / output interface 1003 is used to implement information input and output; The communication interface 1004 is used to enable communication and interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.). Bus 1005 transmits information between various components of the device (e.g., processor 1001, memory 1002, input / output interface 1003, and communication interface 1004); The processor 1001, memory 1002, input / output interface 1003 and communication interface 1004 are connected to each other within the device via bus 1005.

[0137] The electronic device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0138] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method.

[0139] It is understood that the content of the above method embodiments is applicable to this storage medium embodiment. The specific functions implemented in this storage medium embodiment are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those achieved in the above method embodiments.

[0140] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0141] It is understood that the content of the above method embodiments is applicable to the embodiments of this program product. The specific functions implemented by the embodiments of this program product are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0142] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs. Furthermore, memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, memory may optionally include memory remotely located relative to the processor, and these remote memories can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0143] The present invention provides a utility function-driven adaptive gradient congestion control method, apparatus, electronic device, storage medium, and program product. This method acquires monitoring parameters and historical monitoring values ​​of the target transmission link; determines whether the monitoring period has ended using a timer; if the monitoring period has not ended, it quantizes target windows for different focuses based on historical monitoring values ​​and real-time monitoring parameters; it dynamically fuses each target window based on the fusion factor corresponding to the previous monitoring period to update the congestion window of the target transmission link; if the monitoring period has ended, it obtains the utility value of the monitoring period using utility function quantization based on historical monitoring values ​​and monitoring parameters within the monitoring period; it adaptively updates the fusion factor based on the incremental trend of the utility value, triggering the next monitoring period; and it returns to the step of determining whether the monitoring period has ended using a timer, continuously adjusting the congestion window of the target transmission link. The present invention, by introducing a utility function-driven adaptive gradient fusion mechanism, effectively overcomes the shortcomings of existing technologies. Specifically, by constructing a hybrid window model and a dynamic fusion factor, the present invention can achieve an autonomous balance between high throughput characteristics and low latency gradient characteristics, thereby solving the problem of "unbalanced" performance of single-signal algorithms. Furthermore, this invention utilizes a utility function to quantify throughput gains and latency costs in real time, enabling online optimization of the fusion factor. This allows the algorithm to accurately distinguish between congested packet loss and random packet loss, avoiding link starvation caused by misjudgment in weak network environments and significantly improving robustness. In addition, the entire adjustment process is based on lightweight periodic feedback and numerical gradient updates, eliminating the need for complex model training. This balances convergence speed and computational efficiency while adapting to the practical deployment requirements of high-performance networks for low overhead and fast response.

[0144] The embodiments described in this invention are for the purpose of more clearly illustrating the technical solutions of the embodiments of this invention, and do not constitute a limitation on the technical solutions provided by the embodiments of this invention. As those skilled in the art will know, with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of this invention are also applicable to similar technical problems.

[0145] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present invention, and may include more or fewer steps than shown, or combine certain steps, or different steps.

[0146] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0147] Those skilled in the art will understand that all or some of the steps in the methods disclosed above, as well as the functional modules / units in the systems and devices, can be implemented as software, firmware, hardware, or suitable combinations thereof.

[0148] The preferred embodiments of the present invention have been described above with reference to the accompanying drawings, but this does not limit the scope of the claims of the present invention. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and spirit of the present invention should be within the scope of the claims of the present invention.

Claims

1. A utility function-driven adaptive gradient congestion control method, characterized in that, The method includes the following steps: Obtain the monitoring parameters and historical monitoring values ​​of the target transmission link; The monitoring cycle is determined by a timer. If the monitoring period has not ended, the target window for different focused targets is quantified based on the historical monitoring values ​​and the real-time monitoring parameters; Based on the fusion factor corresponding to the monitoring period mentioned above, each target window is dynamically fused with weights to update the congestion window of the target transmission link; If the monitoring period ends, the utility value of the monitoring period is obtained by quantifying the historical monitoring values ​​and the monitoring parameters within the monitoring period using a utility function. The fusion factor is then adaptively updated based on the incremental trend of the utility value to trigger the next monitoring period. Return to the step of determining whether the monitoring period has ended by using a timer, and continuously adjust the congestion window of the target transmission link.

2. The method according to claim 1, characterized in that, The monitoring parameters include instantaneous round-trip time and moving average delay gradient; the historical monitoring values ​​include the last congestion event window value, timestamp variable, and baseline round-trip time; the target window includes a first window aimed at preempting bandwidth and a second window aimed at suppressing latency; quantifying the target window with different focuses based on the historical monitoring values ​​and the real-time monitoring parameters includes the following steps: Based on the previous congestion event window value and the timestamp variable, the first window is obtained by quantization using a packet loss algorithm. The second window is obtained by using a delay-based algorithm quantization based on the instantaneous round-trip delay, the baseline round-trip delay, and the moving average delay gradient.

3. The method according to claim 2, characterized in that, The process of obtaining the first window based on the previous congestion event window value and the timestamp variable using a packet loss algorithm includes the following steps: Based on the previous congestion event window value and the timestamp variable, the first window is obtained by quantization through a preset cubic function, combined with a preset scaling factor and a multiplicative reduction factor. The expression for the first window is: ； In the formula, Indicates the first window; Indicates the preset scaling factor; Represents a timestamp variable; Indicates the window value of the last congestion event; parameter , This represents the multiplicative decreasing factor.

4. The method according to claim 2, characterized in that, The process of obtaining the second window based on the instantaneous round-trip delay, the baseline round-trip delay, and the moving average delay gradient using a delay-based quantization algorithm includes the following steps: The gradient factor is determined based on the ratio of the difference between the instantaneous round-trip delay and the reference round-trip delay to the instantaneous round-trip delay. When the gradient factor is greater than a preset threshold, the second window is obtained by calculating the gradient factor and the preset sensitivity coefficient based on the current congestion window. The expression for the second window is: ； In the formula, Indicates a second window; Indicates the current congestion window; Indicates the sensitivity coefficient; Indicates the gradient factor; When the gradient factor is less than or equal to a preset threshold, the second window is obtained based on the current congestion window using a preset additive growth constant. The expression for the second window is: ； In the formula, Indicates a second window; Indicates the current congestion window; This represents the additive growth constant.

5. The method according to claim 1, characterized in that, The monitoring parameters include real-time link capacity and average round-trip time within the monitoring period, and the historical monitoring values ​​include historical maximum link capacity and baseline round-trip time. The process of quantifying the utility value of the monitoring period using a utility function based on the historical monitoring values ​​and the monitoring parameters within the monitoring period includes the following steps: The throughput term is constructed based on the ratio of the real-time link capacity to the historical maximum link capacity; A delay term is constructed based on the ratio of the difference between the average round-trip time and the baseline round-trip time to the baseline round-trip time. The utility value of the monitoring period is calculated based on the throughput term and the latency term, combined with a preset latency penalty coefficient. The expression for the utility value is as follows: ； In the formula, This represents the utility value during the t-th monitoring period; This represents the real-time link capacity during the t-th monitoring period; Indicates the historical maximum link capacity; Indicates the delay penalty coefficient; This represents the average round-trip time during the t-th monitoring period; This indicates the base round-trip time delay.

6. The method according to claim 1, characterized in that, The adaptive update of the fusion factor based on the incremental trend of the utility value includes the following steps: The utility value increment is obtained by quantifying the utility value based on the utility value of the two most recent monitoring periods; The fusion factor increment is obtained based on the quantification of the fusion factor in the two most recent monitoring periods; The target step size is obtained by quantizing the utility value increment and the fusion factor increment, and the fusion factor is updated using the target step size.

7. The method according to claim 6, characterized in that, The process of quantizing the target step size based on the utility value increment and the fusion factor increment, and updating the fusion factor using the target step size, includes the following steps: The target step size is obtained by multiplying the utility value increment and the fusion factor increment using a sign function and a preset step size factor. The expression for the target step size is: ； In the formula, Indicates the target step size; Indicates the step size factor; Represents a symbolic function; Indicates the increment of utility value; Indicates the increment of the fusion factor; The fusion factor is updated based on the target step size and the step size factor. The expression for updating the fusion factor is: ； In the formula, This represents the fusion factor for the (t+1)th monitoring period; This represents the fusion factor for the t-th monitoring period.

8. A utility function-driven adaptive gradient congestion control device, characterized in that, The device includes: The first module is used to acquire the monitoring parameters and historical monitoring values ​​of the target transmission link; The second module is used to determine whether the monitoring period has ended by using a timer. The third module is used to quantify the target window for different focused targets based on the historical monitoring values ​​and the real-time monitoring parameters if the monitoring period has not ended. The fourth module is used to perform weighted dynamic fusion of each target window based on the fusion factor corresponding to the previous monitoring period, so as to update the congestion window of the target transmission link. The fifth module is used to, if the monitoring period ends, quantify the utility value of the monitoring period using a utility function based on the historical monitoring values ​​and the monitoring parameters within the monitoring period, adaptively update the fusion factor based on the incremental trend of the utility value, and trigger the next monitoring period; The sixth module is used to return to the operation performed in the second module and continuously adjust the congestion window of the target transmission link.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method according to any one of claims 1 to 7.

10. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the method described in any one of claims 1 to 7.

Images