CN-122001818-A - Automatic parameter adjusting method, device, equipment and medium for controlling network congestion

CN122001818ACN 122001818 ACN122001818 ACN 122001818ACN-122001818-A

Abstract

The invention relates to the field of computer networks and discloses a method, a device, equipment and a medium for automatically adjusting congestion of a control network, wherein the method designs a sliding window sample maintenance strategy, combines random forest agent modeling and an acquisition function based on a prediction mean value and uncertainty, adopts a random exploration mechanism of global-local candidate generation and a certain probability, takes a retransmission rate as a main optimization target to automatically adjust congestion control parameters exposed by a core, and solves the problem that the traditional static experience parameter adjustment and single target optimization are difficult to ensure low retransmission rate and high throughput at the same time in a dynamic network environment, thereby improving transmission efficiency and system stability under changing network conditions.

Inventors

Xiao Mengbai
LI YIJUN
Bi Pengqiang
LU JIANGBIN
DU XIANCHANG
YU DONGXIAO
ZHENG YANWEI

Assignees

凌川峰(贵州)信息技术有限公司
山东大学

Dates

Publication Date: 20260508
Application Date: 20251226

Claims (12)

1. A method for controlling automatic parameter tuning of network congestion, which is operated on an electronic device having an operating system with a configurable congestion control algorithm, comprising the steps of: S1, selecting congestion control algorithm parameters to be adjusted, and firstly, initially sampling the parameters to obtain a preset number of candidate parameter sets, wherein the candidate parameter sets are values of each parameter to be applied to a system; s2, modifying the parameters of the operation kernel congestion control algorithm for each candidate parameter set to the parameter values corresponding to the candidate parameter sets so that the system operates under the parameter configuration; S3, constructing an objective function according to the performance index, and then training a proxy model based on the initial data set to fit the objective function; s4, performing iterative optimization on the objective function by using an iterative optimization method to obtain the optimal congestion control algorithm parameters.
2. The method of claim 1, wherein the congestion control algorithm parameters to be parameterized include an initial window, a slow start gain factor, a minimum bandwidth, a congestion window reduction factor after packet loss, or other congestion control algorithm related parameters exposed by the operating system kernel module, and wherein an upper and lower boundary needs to be defined for each parameter to constrain the search space; The initial sampling adopts Latin hypercube sampling method, so that the initial sample obtains uniform coverage in a high-dimensional parameter space.
3. The method for automatically adjusting parameters for controlling network congestion according to claim 1, wherein step S2 specifically comprises the following steps: S21, rounding and rounding the candidate parameter sets; S22, writing the candidate parameter group into a system kernel, and modifying corresponding parameters of a congestion control algorithm; S23, waiting for D seconds to ensure that the system state tends to be balanced, and calling a monitoring interface every t seconds in the subsequent K sampling cycles to acquire the downloading speed and the retransmission rate of each sampling, wherein K, D and t are configurable parameters; s24, calculating the average downloading speed And retransmission rate The calculation method is as follows: ; ; Wherein, the And The download speed and retransmission rate of the jth sample, respectively.
4. The automatic parameter adjusting method for controlling network congestion according to claim 3, wherein the monitoring interface can collect the request number, the downloading speed and the retransmission rate of each network card of the host machine in each sampling, and the downloading speed T and the retransmission rate R of each sampling are calculated as follows: ; Wherein, the , , The number of requests collected by the ith network card is the number of requests, the downloading speed and the retransmission rate are respectively, and n is the number of network cards of the machine.
5. The method for automatically adjusting parameters for controlling network congestion according to claim 3, wherein the objective function in step S3 is: ; Wherein x is a parameter set, And The average download speed and retransmission rate obtained for the corresponding samples of parameter set x, Penalty factor for retransmission rate.
6. The method of claim 1, wherein the agent regression model in the step S3 is a random forest regressor, the random forest regressor is composed of a plurality of independent decision trees, and can predict objective function values according to the parameter set x, and when predicting the candidate parameter set x, the random forest regressor respectively calls each decision tree to output predicted objective function values and predicted standard deviations, and calculates the arithmetic mean of the objective function values and the predicted standard deviations of all decision trees as the final predicted mean of the parameter set And prediction standard deviation 。
7. The method for automatically adjusting parameters for controlling network congestion according to claim 1, wherein the bayesian optimization method in step S4 comprises the following steps: s41, generating a candidate parameter set for next evaluation by using the agent regression model and the acquisition function; S42, applying the candidate parameter set to the system and adding the collected performance index to the data set; S43, maintaining a data set by using a sliding window, wherein the window length is W, and deleting earliest sample data from the head of the window when the number of samples exceeds W due to newly added data so as to ensure that the data set can reflect the dynamic change of the recent network environment; S44, updating the agent regression model by using the data set, and repeating the steps S41 to S44, wherein the candidate parameter set gradually approaches to the optimal value.
8. The method for automatically adjusting parameters for controlling network congestion according to claim 7, wherein the collection function in S41 is: ; Wherein, the And Is the mean and standard deviation of the objective function values predicted for the parameter set x proxy regression model, Is a configurable parameter.
9. The method for automatically adjusting parameters in network congestion according to claim 8, wherein the process of generating the candidate parameter set for the next evaluation is: Firstly, uniformly and randomly generating N_CAND parameter groups in defined parameter boundaries, calculating an acquisition function value U of each parameter group by using a proxy regression model, and selecting a parameter group corresponding to the largest U as a global preferred point x_best; Then, taking x_best as a center, applying Gaussian disturbance on each parameter dimension to generate n_noise local parameter sets, calculating acquisition function values again for the n_noise local parameter sets, and selecting a parameter set corresponding to the locally optimal acquisition function value as a candidate parameter set for next evaluation, wherein the n_cand and the n_noise are configurable parameters; the generation of the next estimated candidate parameter set also incorporates a random exploration strategy, performing a purely random candidate sampling with probability explore _p in each iteration to maintain global exploration ability, with probability A strategy based on acquisition function values is used.
10. The automatic parameter adjusting device for controlling network congestion is characterized by comprising a parameter sampling unit, a calculating unit, a proxy model unit and a candidate generating unit; The parameter sampling unit is used for writing the candidate parameter group into the system kernel, modifying the corresponding parameters of the congestion control algorithm through an interface allowed by the system, and sampling performance indexes in a plurality of periods after the system is stabilized to obtain the average downloading speed and the retransmission rate; the calculating unit is used for calculating a corresponding objective function value based on the downloading speed and the retransmission rate obtained by the parameter sampling unit, and establishing a corresponding relation between the parameter set and the objective function value and adding the corresponding relation into the data set; The agent model unit is used for training an agent regression model by using a data set based on a sliding window, wherein the agent regression model adopts a random forest regressor and can predict an objective function value and a standard deviation according to a parameter set; And the candidate generating unit is used for calculating an acquisition function value according to the prediction result of the agent model unit, generating a candidate parameter set for the next evaluation by combining a random exploration strategy with a certain probability so as to ensure the local optimization and maintain the global exploration capacity, and outputting the generated candidate parameter set to the parameter sampling unit for the next evaluation.
11. A computer device comprising at least one processor and a computer readable medium storing a computer program which, when read and executed by the processor, implements the method of any of claims 1-9.
12. A computer readable medium, characterized in that it stores a computer program, which, when read and run by a processor, implements the method according to any of claims 1-9.

Description

Automatic parameter adjusting method, device, equipment and medium for controlling network congestion Technical Field The invention relates to the field of computer networks and discloses a method, a device, equipment and a medium for controlling network congestion to automatically adjust parameters. Background Congestion control is a core mechanism in a network transport protocol, and the main objective of congestion control is to coordinate the data transmission rate of a sender under the condition of limited network resources and dynamic change so as to avoid network congestion and ensure end-to-end performance. Common congestion control algorithms all contain several configurable parameters such as initial congestion window, slow start gain, congestion window reduction factor, bandwidth estimation smoothing factor, etc. The values of these parameters directly affect the throughput, delay, retransmission rate and stability of the link, and under different network conditions (such as high delay, fluctuation of packet loss rate, bandwidth jitter, mobile network handover, etc.), there is often a significant difference in the optimal settings of the parameters. However, currently most congestion control algorithms employ static or quasi-static parameter configurations at deployment time, and these fixed parameters are often not applicable under different network conditions. In addition, conventional parameter adjustment usually depends on manual experience setting, and is difficult to match with fluctuating network states in real time, and the tuning process is complex, so that the network resource utilization rate is low. The network environment is commonly affected by various factors such as delay, packet loss rate, bandwidth fluctuation, link jitter, user traffic distribution, and mobility. The static parameters cannot be adaptively adjusted according to the change of the conditions, so that poor performance is caused in scenes such as high-delay links, packet loss sudden increases or bandwidth sharp fluctuation. If the parameters are not matched, the transmitting end still transmits according to a more aggressive strategy when the network condition is worsened, so that the retransmission rate is high, the bandwidth waste is serious, or the bandwidth is not fully utilized due to the conservative parameters when the network condition is good, so that the throughput is reduced. Disclosure of Invention Based on the defects of the prior art in facing a dynamic network environment and limited samples, the application provides a method, a device, equipment and a medium for automatically adjusting parameters for controlling network congestion, which have the following technical scheme: a method for controlling network congestion auto-tuning, running on an electronic device having an operating system with a configurable congestion control algorithm, comprising the steps of: S1, selecting congestion control algorithm parameters to be adjusted, and firstly, initially sampling the parameters to obtain a preset number of candidate parameter sets, wherein the candidate parameter sets are values of each parameter to be applied to a system; s2, modifying the parameters of the operation kernel congestion control algorithm for each candidate parameter set to the parameter values corresponding to the candidate parameter sets so that the system operates under the parameter configuration; S3, constructing an objective function according to the performance index, and then training a proxy regression model based on the initial data set to fit the objective function; s4, performing iterative optimization on the objective function by using an iterative optimization method to obtain the optimal congestion control algorithm parameters. Preferably, the congestion control algorithm parameters to be parameterized include an initial window, a slow start gain coefficient, a minimum bandwidth, a congestion window reduction coefficient after packet loss, or other congestion control algorithm related parameters exposed by the operating system kernel module, and an upper and lower boundary needs to be defined for each parameter to constrain the search space; The initial sampling adopts Latin hypercube sampling method, so that the initial sample obtains uniform coverage in a high-dimensional parameter space. Preferably, the step S2 specifically includes the following steps: S21, rounding and rounding the candidate parameter sets; S22, writing the candidate parameter group into a system kernel, and modifying corresponding parameters of a congestion control algorithm; S23, waiting for D seconds to ensure that the system state tends to be balanced, and calling a monitoring interface every t seconds in the subsequent K sampling cycles to acquire the downloading speed and the retransmission rate of each sampling, wherein K, D and t are configurable parameters; s24, calculating the average downloading speed And retransmission rateThe calculation