Flow control method and device, storage medium and electronic equipment

Abstract

The invention discloses a flow control method and device, a storage medium and electronic equipment, and relates to the technical field of network flow management, and the method comprises the steps: determining a plurality of feature vectors matched with input network flow based on a plurality of feature extraction modules; performing feature fusion processing on the plurality of feature vectors through a gating attention mechanism to obtain distribution strategy features, and determining a traffic distribution strategy matched with the input network traffic according to the distribution strategy features; determining network resources respectively allocated to the network traffic of the plurality of traffic categories in the input network traffic according to a traffic allocation strategy; updating a flow distribution strategy according to the network transmission performance information; the decision model updating strategy is adjusted in real time based on the strategy execution result, learning and optimization of the adaptive network state in the operation process are achieved, it is ensured that the long-term service quality index is kept at the optimal level, the technical problems that the flow distribution strategy is poor in performance and cannot meet the requirement are solved, and the technical effect of improving the network resource utilization rate is achieved.

Description

Technical Field

[0001] The present application relates to the technical field of network traffic management, and particularly to a traffic control method, a device, a storage medium, and an electronic device. Background Art

[0002] Traditional traffic control methods usually rely on a pre-set rule table to allocate network resources and execute traffic policies. Especially when faced with a surge in encrypted traffic such as HTTPS video streams, static bandwidth allocation policies are often slow to respond, resulting in serious network congestion. With the advent of the 5G era, ultra-reliable low-latency communication has become a new industry standard, requiring the control latency to be maintained at the 1ms level. And when existing technical solutions process encrypted traffic, they often only focus on a single dimension of features, such as protocol fields or payload content, while ignoring the internal relationship between them and time behavior features. This isolated feature analysis leads to a significant decline in the recognition accuracy of the model for encrypted traffic and the formulation accuracy of control policies.

[0003] That is, there are technical problems in the prior art that due to the limitations of static rule configuration and the singularity of feature analysis, the performance of the used traffic allocation strategy is poor and cannot meet the requirements. Summary of the Invention

[0004] The present application provides a traffic control method, a device, a storage medium, and an electronic device, so as to at least solve the technical problems in the related art that due to the limitations of static rule configuration and the singularity of feature analysis, the performance of the used traffic allocation strategy is poor and cannot meet the requirements.

[0005] The present application provides a traffic control method, including: determining a plurality of feature vectors matching the input network traffic based on a plurality of feature extraction modules, where the plurality of feature extraction modules are trained based on the TensorFlow architecture, and the plurality of feature vectors respectively correspond to an attribute dimension of the input network traffic; performing feature fusion processing on the plurality of feature vectors through a gated attention mechanism to obtain an allocation policy feature, and determining a traffic allocation policy matching the input network traffic according to the allocation policy feature; determining the network resources allocated to the network traffic of a plurality of traffic classes in the input network traffic according to the traffic allocation policy; and updating the traffic allocation policy according to the network transmission performance information when the network transmission performance information matching each of the plurality of traffic classes is obtained, where the network transmission performance information is determined during the transmission according to the allocated network resources, and the updated traffic allocation policy is used for the next traffic allocation operation.

[0006] The present application also provides a traffic control device, including: a feature extraction module, configured to determine multiple feature vectors matching the input network traffic based on multiple feature extraction modules, wherein the multiple feature extraction modules are trained based on the TensorFlow architecture, and the multiple feature vectors respectively correspond to an attribute dimension of the input network traffic; a decision-making module, configured to perform feature fusion processing on the multiple feature vectors through a gated attention mechanism to obtain an allocation policy feature, and determine a traffic allocation policy matching the input network traffic according to the allocation policy feature; a traffic control execution module, configured to determine the network resources respectively allocated to the network traffic of multiple traffic classes in the input network traffic according to the traffic allocation policy; an update module, configured to update the traffic allocation policy according to the network transmission performance information in the case of obtaining the network transmission performance information respectively matching the multiple traffic classes, wherein the network transmission performance information is determined during the transmission according to the allocated network resources, and the updated traffic allocation policy is used for the next traffic allocation operation.

[0007] The present application also provides an electronic device, including: a memory, configured to store a computer program; a processor, configured to implement the steps of any one of the above traffic control methods when executing the computer program.

[0008] The present application also provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program implements the steps of any one of the above traffic control methods when executed by a processor.

[0009] The present application also provides a computer program product, including a computer program, wherein the computer program implements the steps of any one of the above traffic control methods when executed by a processor.

[0010] Through this application, multiple feature vectors matching the input network traffic are determined based on multiple feature extraction modules to adapt to the complex and changeable network environment; through the gated attention mechanism, the multiple feature vectors are subjected to feature fusion processing to obtain the allocation strategy features, and the traffic allocation strategy matching the input network traffic is determined according to the allocation strategy features, which can dynamically adjust the importance of the feature of each attribute dimension, so as to ensure that the traffic allocation strategy can comprehensively consider all key factors; the network resources allocated to the network traffic of multiple traffic categories in the input network traffic are determined according to the traffic allocation strategy; in the case of obtaining the network transmission performance information matching each of the multiple traffic categories, the traffic allocation strategy is updated according to the network transmission performance information; the decision model update strategy is adjusted in real time based on the result of policy execution, realizing continuous learning and optimization during operation to adapt to the network state and ensure that the long-term service quality index remains at the optimal level. Therefore, the technical problem in the prior art that the performance of the traffic allocation strategy used is poor and cannot meet the requirements due to the limitations of static rule configuration and the singularity of feature analysis can be solved, and the technical effect of improving the utilization rate of network resources can be achieved. Description of the Drawings

[0011] In order to more clearly illustrate the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0012] Figure 1 Hardware structure block diagram of a server device for a traffic control method provided by an embodiment of the present application;

[0013] Figure 2 Flowchart of a traffic control method provided by an embodiment of the present application;

[0014] Figure 3 Flowchart of another traffic control method provided by an embodiment of the present application;

[0015] Figure 4 Schematic diagram of a traffic control strategy provided by an embodiment of the present application;

[0016] Figure 5 Structural schematic diagram of a traffic control device according to an embodiment of the present application;

[0017] Figure 6 Structural schematic diagram of a traffic control electronic device according to an embodiment of the present application. Detailed Embodiments

[0018] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the protection scope of the present application.

[0019] It should be noted that in the description of the present application, the terms "including", "comprising" or any other variant thereof are intended to cover a non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device. The terms "first", "second", etc. in the present application are used to distinguish similar objects, rather than to describe a specific order or sequence.

[0020] In order to enable those skilled in the art of the present technology to better understand the solution of the present application, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0021] The method embodiments provided in the embodiments of the present application can be executed on a server device or a similar computing device. Taking running on a server device as an example, Figure 1 is a hardware structure block diagram of a server device for a traffic control method according to an embodiment of the present application. As Figure 1 shown, the server device may include one or more ( Figure 1 only one is shown in the figure) processors 102 (the processors 102 may include, but are not limited to, processing devices such as a microprocessor MCU or a field programmable gate array FPGA) and a memory 104 for storing data. Among them, the above server device may further include a transmission device 106 for communication functions and an input / output device 108. Those of ordinary skill in the art can understand that Figure 1 the structure shown is only schematic and does not limit the structure of the above server device. For example, the server device may further include more or fewer components than those shown in Figure 1 the figure, or have a different configuration from that shown in Figure 1 the figure.

[0022] The memory 104 can be used to store computer programs, such as software programs and modules of application software, such as the computer program corresponding to the traffic control method in the embodiments of the present application. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, implements the above-mentioned method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some instances, the memory 104 may further include a memory remotely disposed relative to the processor 102, and these remote memories can be connected to the server device through a network. Examples of the above-mentioned network include but are not limited to the Internet, intranet, local area network, mobile communication network, and combinations thereof.

[0023] The transmission device 106 is used to receive or send data via a network. Specific examples of the above-mentioned network may include a wireless network provided by a communication provider of the server device. In one instance, the transmission device 106 includes a network adapter (abbreviated as NIC), which can be connected to other network devices through a base station and thus communicate with the Internet. In one instance, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0024] Embodiments of the present application provide a traffic control method, and the method is described in detail in combination with the execution flow of the traffic control method.

[0025] To more clearly understand the technical solutions provided by the embodiments of the present application, the key terms related to the embodiments of the present application are introduced herein first:

[0026] Differentiable Hashing refers to a technique in which the weight matrix of a hash function can be optimized through backpropagation of neural network gradients.

[0027] The Gated Attention Mechanism can dynamically calculate the weights of each channel with learnable parameters to achieve adaptive weighted fusion of features.

[0028] The Dynamic Policy Matrix is a set of traffic control rules generated according to the real-time network state.

[0029] The Hurst Exponent is an index that measures the long-term dependence of a time series.

[0030] The Neural Hash Table combines the data structures of differentiable hashing and LRU cache (Least Recently Used cache).

[0031] The Traffic Fingerprint is a unique identifier for traffic based on multi-channel hash values.

[0032] TensorFlow Architecture: TensorFlow is an open-source software library for the design, training, and deployment of machine learning and deep learning algorithms for various perception and cognitive tasks.

[0033] Dilated Convolutional Neural Network: A special convolutional network structure mainly used to process signals with local correlations.

[0034] Transformer: A deep learning model based on the self-attention mechanism.

[0035] TCN (Temporal Convolutional Network): A convolutional neural network architecture specifically designed for time series analysis.

[0036] As an optional implementation, as Figure 2 shown, the application method of the above-mentioned compatible and interconnected switch includes:

[0037] S202, determining multiple feature vectors matching the input network traffic based on multiple feature extraction modules, where the multiple feature extraction modules are trained based on the TensorFlow architecture, and the multiple feature vectors respectively correspond to an attribute dimension of the input network traffic;

[0038] In some optional implementation manners, the above-mentioned feature extraction modules are algorithm modules for extracting features for each dimension of network traffic, such as protocol channels, payload channels, time series channels, etc. These modules are trained based on the TensorFlow architecture, that is, the feature extraction modules are optimized through deep learning algorithms and can automatically learn and extract meaningful features from network data; the above-mentioned input network traffic can be a real-time data stream being monitored or managed, and can be a single data packet or a series of consecutive data packets; the above-mentioned feature vectors are, for example: protocol feature vectors (which may contain field information such as TTL, DSCP, etc.); payload feature vectors (which may contain byte histograms, entropy, and n-gram sequences); time series feature vectors (which may contain time series statistical information such as Hurst exponent, autocorrelation function, etc.).

[0039] Specifically, protocol channel extraction includes using the DPDK library for zero-copy packet capture, parsing Ethernet / IPv4 / IPv6 / TCP / UDP protocol headers, extracting key fields to generate protocol feature vectors (dimension = 20), and further performing field normalization; for TLS / SSL traffic in the payload channel, extract the n-gram distribution (n = 2, 3) of the first 100 payload bytes, and calculate the semantic entropy through the following formula:

[0040]

[0041] where HMM_score is the anomaly probability output by the hidden Markov model, and p(b) represents the probability density function of the byte value (b) in the traffic. Further merge the 256-dimensional byte histogram, semantic entropy, and Top50 3-gram frequencies, and construct a feature vector with a total dimension of 307; the timing channel extracts the packet arrival time interval (IAT) sequence T = {t1, t2,..., tn} in the traffic, calculates the Hurst exponent to detect self-similarity, and further extracts periodic features through discrete Fourier transform (DFT). It can also be a feature vector after feature fusion. Here, there is no specific limitation on the source of the features and the method of obtaining the feature vector by processing information.

[0042] S204, perform feature fusion processing on multiple feature vectors through a gated attention mechanism to obtain allocation policy features, and determine the traffic allocation policy matching the input network traffic according to the allocation policy features;

[0043] Optionally, the above gated attention mechanism is a deep learning strategy used to dynamically adjust and fuse the weights of feature vectors in different channels. In this step, according to the importance of each feature vector, the weights are automatically adjusted through learnable parameters to achieve adaptive fusion of multi-channel features. The following is an optional self-attention mechanism formula:

[0044]

[0045] In the self-attention mechanism, the input sequence is decomposed into three parts: Query, Key, and Value. Query is used to ask for information, Key is used as the index of the information, and Value is the part containing the actual information. M is a predefined n-gram mask matrix, and d k is a normalization factor used to alleviate the scale problem of the attention score. The above formula allows selective modification of the attention matrix by introducing the mask matrix M. That is, when processing network traffic, M can be used to mask out n-grams or byte sequences irrelevant to the target traffic type, so that the model focuses more on meaningful features, enhances the model's attention to specific sequence features, ignores parts irrelevant to the current analysis target, and improves the efficiency and accuracy of the model.

[0046] The above allocation strategy feature is the final feature representation after being fused and processed by the gated attention mechanism, which synthesizes information from multiple dimensions of the original network traffic, can more comprehensively reflect the nature and requirements of the traffic, and then determine the network resource allocation strategy; the above traffic allocation strategy enables the network resource allocation scheme formulated based on the allocation strategy feature, including but not limited to bandwidth allocation, priority scheduling, and security rule setting, to optimize the use of network resources according to the determined traffic allocation strategy and meet the quality of service requirements of different traffic classes.

[0047] S206. Determine the network resources allocated to the network traffic of multiple traffic classes in the input network traffic according to the traffic allocation strategy.

[0048] Optionally, the above traffic classes include but are not limited to; the above network resources include but are not limited to network bandwidth, queue priority, security protection resources, etc. In this step, the system allocates corresponding resources to different classes of network traffic based on the traffic allocation strategy obtained in the previous step to ensure the transmission efficiency of the traffic and the security and stability of the network.

[0049] S208. When the network transmission performance information matching each of the multiple traffic classes is obtained, update the traffic allocation strategy according to the network transmission performance information, where the network transmission performance information is determined during the process of transmission according to the allocated network resources, and the updated traffic allocation strategy is used for the next traffic allocation operation.

[0050] Optionally, the above network transmission performance information is the network performance metrics related to different traffic classes monitored after the traffic allocation strategy is executed, such as packet loss rate, transmission delay, CPU utilization, etc., which reflect the effect of the strategy execution and are an important basis for evaluating and optimizing the allocation strategy.

[0051] Based on the network transmission performance information in the above steps, the system adjusts the traffic allocation strategy in real time through a feedback loop to optimize the network resource allocation, adapt to the changes in the network state, and the updated strategy may involve fine-tuning of model parameters, dynamic adjustment of hash tables, or real-time optimization of control rules to ensure the intelligence and self-adaptability of network traffic control.

[0052] The following uses a flowchart Figure 3 to illustrate the above process:

[0053] S302-1, protocol parsing; S302-2, payload analysis; S302-3, timing statistics; each processes the original traffic; further, three hash compression modules respectively perform parallel compression processing on protocol, payload, and timing features. Each module uses an independent differentiable hash function to convert the high-dimensional feature vectors into hash values of a fixed length, that is, respectively execute S304-1, hash compression module 1; S304-2, hash compression module 2; S304-3, hash compression module 3;

[0054] Execute S306-1, dilated CNN; S306-2, Transformer; S306-3, TCN; that is, three neural network models: the dilated CNN targets the protocol channel, the Transformer is for the payload channel, and the TCN processes the timing channel. They respectively receive their own hash compression features as inputs and perform in-depth analysis and abstraction of the features;

[0055] Further execute S308, attention fusion; S310, policy matrix generation; specifically, S312-1, bandwidth allocation; S312-2, priority scheduling; S312-3, anomaly blocking; S314, traffic shaping, to reduce the phenomenon of burst congestion and improve the robustness of the entire network; S316, execute feedback, establish a feedback loop, and adjust the hash parameters and decision model in reverse to realize parameter update, optimize hash compression, and update the strategy of the model output.

[0056] Through the above-described implementation manners recorded in this application, multiple feature vectors matching the input network traffic are determined based on multiple feature extraction modules, adapting to the complex and changing network environment; through the gated attention mechanism, feature fusion processing is performed on the multiple feature vectors to obtain the allocation strategy features, and based on the allocation strategy features, a traffic allocation strategy matching the input network traffic is determined, which can dynamically adjust the importance of the features in each attribute dimension, so as to ensure that the traffic allocation strategy can comprehensively consider all key factors; determine the network resources respectively allocated to the network traffic of multiple traffic categories in the input network traffic according to the traffic allocation strategy; in the case of obtaining the network transmission performance information respectively matching the multiple traffic categories, update the traffic allocation strategy according to the network transmission performance information; based on the results of policy execution, adjust the decision model update strategy in real time, realizing continuous learning and optimization during operation to adapt to the network state and ensure that the long-term service quality indicators are maintained at the optimal level. Therefore, it can solve the technical problems in the prior art that the performance of the traffic allocation strategy used is poor and cannot meet the requirements due to the limitations of static rule configuration and the singularity of feature analysis, and achieve the technical effect of improving the utilization rate of network resources.

[0057] In an alternative implementation manner, performing feature fusion processing on the multiple feature vectors through the gated attention mechanism to obtain the allocation strategy features includes:

[0058] S1. Determine the attention weight values respectively matching the feature vectors of multiple attribute dimensions through a gated attention mechanism, where the attention weight values are used to indicate the influence degree of the feature vectors matching the attribute dimensions on determining the policy features;

[0059] S2. Determine the product between the attention weight values respectively matching the multiple attribute dimensions and the corresponding feature vectors, and determine the allocation policy features according to the sum of the products.

[0060] In the above step S1, the attention weight values α respectively matching the feature vectors of multiple attribute dimensions are determined through a gated attention mechanism i It can be through the following formula:

[0061]

[0062] where, W g is the gated weight matrix, which is used to transform the feature vector f i to the vector space matching the attention calculation; v is the attention weight vector, which is used to evaluate the importance of the transformed feature vector, and the calculated attention weight value α i is normalized to ensure that the sum of all weight values is 1, accurately reflecting the relative importance of each feature vector in the process of determining the final policy features.

[0063] Furthermore, in the above step S2, determining the product between the attention weight values respectively matching the multiple attribute dimensions and the corresponding feature vectors, and determining the allocation policy features according to the sum of the products can be through the following formula:

[0064] Policy vector This comprehensive policy feature vector contains information of multiple attribute dimensions. Through the dynamic weight adjustment of the gated attention mechanism, it ensures the prominence of key features and also reduces the influence of secondary features on the decision-making process.

[0065] Through the gated attention mechanism, it is possible to dynamically adjust the weights of the feature vectors of each attribute dimension during the feature fusion stage, ensuring that the allocation policy features can comprehensively and accurately reflect the requirements and characteristics of network traffic. It is applicable to processing multi-dimensional and highly complex network data, and can significantly improve the intelligence and efficiency of network traffic control decisions.

[0066] In an alternative implementation manner, determining multiple feature vectors matching the input network traffic based on multiple feature extraction modules includes:

[0067] S1. When the feature extraction module is used to obtain data features in the protocol dimension, the feature extraction module parses the protocol field information in the packet protocol header that matches the input network traffic, and converts the protocol field information into a low-dimensional feature vector according to a compression function. Here, the compression function is called based on the interface of the TensorFlow architecture, and the protocol field information is used to indicate the service category of the packet that matches the input network traffic.

[0068] Optionally, when the feature extraction module is used to obtain data features in the protocol dimension, the feature extraction module (such as the Protocol Channel) first deeply analyzes the input network traffic and extracts the key protocol field information in the packet protocol header, including but not limited to IP protocol version, window size, TTL (Time to Live), protocol type (such as TCP, UDP), source port, destination port, source IP, destination IP, etc. It can be understood that the protocol field information reflects the service category of the packet. For example, the TCP protocol is usually associated with reliable services at the transport layer, while the UDP protocol is often used for services with high real-time requirements but low requirements for data integrity (such as video streams).

[0069] Furthermore, a specific function or model under the TensorFlow architecture can be called to efficiently compress the extracted protocol field information and generate a low-dimensional feature vector. The principle is as shown in the formula:

[0070] h i =SignHash(W i T x i ), where

[0071] W i is a trainable weight matrix, x i is the corresponding high-dimensional feature vector, μ z and σ z are the dynamic mean and standard deviation of the feature z respectively, used to standardize the feature distribution, and h i is the low-dimensional feature vector after dimensionality reduction. By compression, while retaining the key features of the protocol field information, the data volume is reduced, accelerating the subsequent processing steps.

[0072] It should be noted that hash functions usually involve discretization operations, such as binarization, in order to map continuous inputs to a finite hash code space. However, such discretization operations are non-differentiable mathematically, which means that gradient descent optimization cannot be directly applied to it. In deep learning, the update of weights depends on the gradient of the loss function with respect to the weights, and in non-differentiable components, this gradient does not exist. Therefore, during the process of compressing features based on the TensorFlow architecture, a gradient approximation strategy can be further utilized. Assume that for the discretization operation, the gradient is almost equal to 1. When the input is within a certain threshold range, the gradient is 1, otherwise it is 0.

[0073] Through the automatic differentiation function of TensorFlow, the weight matrix W of the differentiable hash function can be incorporated into the training process of the entire network and optimized through backpropagation together with other model parameters, solving the problem of non-differentiability of traditional hash algorithms. And in TensorFlow, a straight-through estimator can be implemented using custom gradient operations, enabling the calculation of approximate gradients for training even for non-differentiable operations such as SignHash.

[0074] And a dynamic conflict detection strategy is used to optimize the training of the compression process. For example: the hash table maintains a conflict counter matrix C ∈ Z m×k , where m is the number of hash table slots and K is the number of predefined traffic class types. If C[h(x)][y] > θ (θ is verified through experiments), the conflict resolution mechanism is triggered to recalculate the hash table capacity. C[h(x)][y] represents the number of times the traffic feature vector belonging to class (y) conflicts at the hash table location (h(x)).

[0075] S2, when the feature extraction module is used to obtain data features of the payload semantic dimension, the feature extraction module parses the application information in the packet payload matching the input network traffic and converts the application information into a low-dimensional feature vector according to the compression function, where the application information is used to indicate the application type of the packet matching the input network traffic;

[0076] Optionally, when the feature extraction module is used to obtain data features in the payload semantic dimension, the feature extraction module (such as the Payload Channel) further parses the packet payload in the input network traffic, identifies the application information therein, including preprocessing of TLS / SSL encrypted traffic, to extract semantic features such as n-gram distribution, entropy (Shannon entropy, semantic entropy), and high-frequency sequences. It can be understood that the application information helps to identify the application type of the packet, such as web browsing, mail service, file transfer, video conferencing, etc.; further, by calling a compression function based on TensorFlow, the high-dimensional application information is converted into a low-dimensional feature vector, which can be by using a deep learning model, such as a convolutional neural network (CNN) or Transformer, to capture important patterns and correlations in the payload semantics, thereby generating a representative feature representation.

[0077] S3, when the feature extraction module is used to obtain data features in the temporal dimension, the feature extraction module parses the temporal information matching the input network traffic, and converts the temporal information into a low-dimensional feature vector according to the compression function, where the temporal information is used to indicate the time behavior pattern of the packet matching the input network traffic transmitted in the network.

[0078] Optionally, when the feature extraction module is used to obtain data features in the temporal dimension, the feature extraction module (such as the Temporal Channel) analyzes the temporal data in the input network traffic, such as the inter-arrival time (IAT) of packets, the periodicity and self-similarity of the packet sequence (calculated by the Hurst exponent and the autocorrelation function (ACF)); it can be understood that the temporal information reveals the time behavior pattern of the packet transmitted in the network. If the IAT sequence of the packet arrival time interval shows an obvious periodicity or steady-state distribution, then this traffic may come from a continuous and stable source, such as a video stream or a heartbeat packet. If there are a large number of short intervals or long intervals in the IAT sequence, this often indicates that there is bursty traffic in the network, which may be an activity at the application layer (such as the start of a file download) or an event at the network layer (such as a routing change); further, through the temporal data compression function based on TensorFlow, the temporal information is converted into a low-dimensional feature vector for easy processing and analysis, which can be by using a temporal convolutional network (TCN), a recurrent neural network (RNN), or other deep learning models suitable for processing sequence data.

[0079] In an alternative embodiment, after converting the information parsed by the feature extraction module into a low-dimensional feature vector according to the compression function, it includes:

[0080] S1. Determine the storage hash table that matches the input network traffic, where each position in the storage hash table corresponds to storing a feature identifier that matches a low-dimensional feature vector.

[0081] S2. Adjust the capacity of the storage hash table according to the total number of feature identifiers stored in the current storage hash table.

[0082] Optionally, in the above step S1, an initialized storage hash table can be preset according to engineering experience. Each position (or bucket) of the storage hash table stores a feature identifier (such as a traffic fingerprint) that matches a low-dimensional feature vector.

[0083] Furthermore, in the above step S2, a threshold can be defined to determine whether the hash table needs to be expanded. For example, when the load factor of the hash table (the number of stored feature identifiers divided by the capacity of the hash table) reaches or exceeds 75%, the expansion operation is triggered. For example, the golden ratio (about 1.618) can be used for dynamic growth to maintain good load balancing. It should be noted that during the adjustment of the hash table, new conflict situations may occur, and methods such as chaining, quadratic hashing, or open addressing can be used to resolve conflicts to ensure that each feature identifier can find a suitable storage location.

[0084] Specifically, the probability P of a conflict occurring in the hash table can be obtained through the following formula collision :

[0085]

[0086] where m i is the capacity of the hash table after the i-th round of adjustment, and N active is the total number of network traffic feature identifiers that are currently active. That is to say, the goal of the hash table design is to make the probability of a conflict less than or equal to the reciprocal of the capacity of the hash table. That is, the conflict rate should be as low as possible, preferably lower than 0.8 is an empirical value representing the ideal load factor of the hash table, that is, the proportion of the data volume stored in the hash table to the total capacity.

[0087] Optionally, the hash table can be adjusted through the following formula:

[0088]

[0089] represents the capacity of the hash table at different time points, and ΔL hashLet $\Delta L$ be the change in loss of the differentiable hashing algorithm in the most recent training cycle, and $\eta$ be the learning rate. This dynamic hash table capacity adjustment algorithm adaptively adjusts the size of the hash table by monitoring the performance of the model in the traffic classification task, so as to improve the accuracy and efficiency of traffic processing while maintaining low resource consumption. And the system can effectively cope with the fluctuations of network traffic, reduce the collision rate, improve the efficiency of query and insertion operations, and avoid over-allocation of resources while ensuring performance.

[0090] In an alternative implementation, determining the network resources allocated to the network traffic of multiple traffic classes in the input network traffic according to the traffic allocation policy includes:

[0091] S1, determining the bandwidth allocation information and queue priority information respectively matching the multiple traffic classes indicated by the traffic allocation policy;

[0092] S2, determining the network bandwidth quotas allocated to the network traffic of multiple traffic classes in the input network traffic according to the bandwidth allocation information;

[0093] S3, determining the traffic transmission order allocated to the network traffic of multiple traffic classes in the input network traffic according to the queue priority information.

[0094] Optionally, the above bandwidth allocation information can be, for example, that video streams are allocated 40% of the total bandwidth, VoIP is allocated 30%, and P2P downloads are allocated 10%; the above queue priority information can be that traffic classes with lower priority values have higher queue scheduling priorities; it should be noted that this solution can preset a target traffic allocation policy such as Figure 4 Based on the cooperation of traffic classes, bandwidth allocation, priority, and control actions, a dynamic traffic control policy system is formed. By training the model and optimizing the model parameters based on the feedback results, the finally trained output traffic allocation policy approaches the target traffic allocation policy.

[0095] Taking an alternative implementation to illustrate the above process, the model identifies categories such as real-time video, critical business data, P2P downloads, and suspicious attack traffic, and reads from the policy matrix that real-time video is allocated 40% bandwidth with a priority of 0; critical business data is allocated 30% bandwidth with a priority of 1; P2P downloads are allocated 10% bandwidth with a priority of 3; suspicious attack traffic is allocated 0% bandwidth with a priority of 7.

[0096] Further, set bandwidth allocation rules on the router or switch. Real-time video occupies 40% of the bandwidth, critical business data takes up 30%, P2P downloads occupy 10%, and the remaining bandwidth is reserved or used for other unclassified traffic. In this step, the system also needs to have a monitoring and adjustment mechanism to ensure that the dynamic allocation of bandwidth meets the policy requirements. Queue priority setting: Configure the queue priorities as follows: real-time video is the highest (0), critical business data is the second (1), P2P downloads are lower (3), and suspicious attack traffic is the lowest (7). This determines the order of packet processing, thus affecting the transmission speed and latency.

[0097] The intelligent recognition of deep learning models, the dynamic generation of the policy matrix, and the real-time response of network devices to the policy jointly ensure the efficient utilization of network resources and the reasonable scheduling of various traffic types, thus significantly improving the overall performance and service quality of the network.

[0098] In an alternative implementation, when obtaining network transmission performance information that matches each of multiple traffic categories, update the traffic allocation policy according to the network transmission performance information, including:

[0099] S1, when detecting that the network transmission performance information that matches each of multiple traffic categories does not meet the policy allocation conditions, obtain network transmission performance loss information, where the network transmission performance loss information is used to indicate the pressure on the network service;

[0100] S2, update the traffic allocation policy according to the performance loss information.

[0101] Optionally, in the above step S1, the above network transmission performance information can be but is not limited to packet loss rate, latency, bandwidth utilization, etc. The above non-compliance with the policy allocation conditions means that the actual transmission performance of the network fails to meet the service quality objectives set for each traffic category in the dynamic policy matrix. For example, the packet loss rate of real-time video traffic may exceed the allowed maximum value, or the latency of critical business data exceeds the specified range; the above network transmission performance loss information can be but is not limited to the additional number of packet losses, the number of milliseconds of increased latency, etc.

[0102] Optionally, in the above step S2, updating the traffic allocation policy according to the performance loss information can be reallocating bandwidth, adjusting queue priorities, enabling a traffic shaping mechanism, rate-limiting or blocking certain traffic categories, etc. As an alternative implementation, based on control effects (such as queue delay, packet loss rate, CPU utilization), reverse-adjust the model parameters as shown in the following formula:

[0103] L total =L class +λL QoS

[0104]

[0105] L class The evaluation values of the queue delay and packet loss rate for the traffic category, L QoS The evaluation values of the queue delay and packet loss rate that meet the quality of service, where λ is an adjustment factor, and in L total When it is greater than a certain threshold, update the parameters in the model, such as:

[0106]

[0107] ΔW i Is the update amount of the i-th parameter in the model parameters, and then obtain the updated traffic allocation strategy output after the model parameters are updated.

[0108] In addition, reinforcement learning (RL) can be introduced to optimize the long-term QoS metrics. The formula is as follows:

[0109]

[0110] Q(s,a) is the expected cumulative reward for taking action a in state s, E is the average expected value of future rewards, R t+τ Is the immediate reward obtained within τ time steps after the current time t, γ is the discount factor, used to evaluate the importance of future rewards. The smaller the value, the more importance is attached to recent rewards; the closer the value is to 1, the more long-term benefits are considered. β is a preset coefficient, used to balance the relationship between rewards and overheads. Overhead is the additional cost or overhead incurred by performing a specific action (such as a traffic control policy).

[0111] It should be noted that the dynamic policy matrix is predicted and generated by the deep learning model based on the historical data and real-time characteristics of network traffic. The output of the model directly determines the content of the policy. Therefore, each update of the model parameters will affect the generation of the policy. Conversely, the effect of the policy update can be fed back to the model to guide the further adjustment of its parameters. By adjusting the parameters, the model's understanding and analysis of network traffic characteristics can be optimized, so as to formulate traffic control policies more accurately. For example, through backpropagation and gradient descent, the weight matrix of the differentiable hash function can be optimized, making the identification of traffic characteristics more accurate, and thus improving the pertinence and effectiveness of policy updates.

[0112] In an optional implementation manner, convert the information parsed by the feature extraction module into a low-dimensional feature vector according to the compression function, including:

[0113] S1, determine the performance requirements for executing the compression function, and design a hardware acceleration architecture according to the performance requirements;

[0114] S2, design an algorithm pipelining structure on the hardware acceleration architecture;

[0115] In S3, in each clock cycle of the algorithm pipelined structure, the compression function converts an information into a low-dimensional feature vector.

[0116] Optionally, in the above step S1, the above performance requirements may include line speed processing capabilities, low latency, and high compression efficiency, etc. To meet the performance requirements, a dedicated hardware acceleration architecture needs to be designed, such as FPGA (Field Programmable Gate Array) or DPU (Data Processing Unit). These architectures have high parallel processing capabilities and are suitable for executing repetitive compute-intensive tasks, such as the hash calculation in the compression function.

[0117] In the above step S2, the pipeline is a common parallel processing mode in hardware accelerators. Its idea is to decompose a task into multiple interconnected stages or steps, and each stage uses hardware resources for parallel processing.

[0118] Specifically, the feature vector reception stage: receives the high-dimensional network traffic feature vector transmitted from the software layer. The hash function calculation stage: executes the hash calculation in parallel to convert the high-dimensional feature vector into a low-dimensional feature vector. The conflict handling stage: handles possible hash conflicts in the next stage of the pipeline to ensure the integrity and accuracy of the data. The feature vector output stage: outputs the compressed low-dimensional feature vector to the subsequent deep learning decision module or stores it in the cache. So that when a feature vector is in the hash calculation stage, the next feature vector can enter the feature vector reception stage, and so on, forming a continuous data processing flow to improve the processing efficiency.

[0119] In the above step S3, it can be understood that in hardware design, the clock cycle is the basic timing unit used to synchronize and control each part of the hardware. In the algorithm pipelined structure, the execution of the compression function is designed to process one feature vector per clock cycle, ensuring the continuity and high speed of the processing. Based on the designed pipelined structure, the compression function converts each high-dimensional network traffic feature vector into a low-dimensional feature vector, which can be achieved, for example, through a differentiable hash algorithm, mapping the vector to a smaller space that is convenient for storage and processing while trying to retain the discriminability of the original features.

[0120] Other hardware acceleration strategies can also be designed, such as using DPDK data plane integration. Specifically, the rte_flow API is used to compile the strategy into network card hardware rules (encapsulating the strategy information into an rte_flow object), and the constructed rte_flow rules are loaded into the DPDK data plane processing framework by calling the rte_flow_rule_add() function of DPDK. This function is responsible for translating the rules from the software level into an instruction set that the network card hardware can understand and storing them in the network card's rule table. Once the rule is added successfully, the network card will automatically start processing subsequent arriving data packets according to this strategy.

[0121] The Data Plane Development Kit (DPDK) is a high-performance data plane development kit designed specifically for network packet processing. It provides a set of libraries and drivers that allow applications to directly access network card hardware, bypassing the traditional operating system kernel network stack, thereby significantly improving packet processing speed and reducing latency. In this technical solution, DPDK is used to execute dynamic traffic control strategies, especially converting the strategies generated by the algorithm into hardware rules that the network card can directly understand through its rte_flow API, to achieve high-speed and low-latency network traffic control.

[0122] It can be understood that the above control of traffic can also be used in scenarios such as traffic scheduling between cross-regional data centers and Internet of Things device management to implement real-time policies with low latency, improve resource utilization, and resist malicious traffic attacks.

[0123] Through the description of the above implementation manners, those skilled in the art can clearly understand that the method according to the above embodiments can be implemented by means of software plus a necessary general hardware platform. Of course, it can also be implemented by hardware, but in many cases, the former is a better implementation manner.

[0124] According to another aspect of the embodiments of the present application, there is also provided a traffic control device for implementing the above traffic control method. As Figure 5 shown, the device includes:

[0125] A feature extraction module 502, configured to determine multiple feature vectors matching the input network traffic based on multiple feature extraction modules, where the multiple feature extraction modules are trained based on the TensorFlow architecture, and the multiple feature vectors respectively correspond to an attribute dimension of the input network traffic;

[0126] A decision-making module 504, configured to perform feature fusion processing on the multiple feature vectors through a gated attention mechanism to obtain an allocation strategy feature, and determine a traffic allocation strategy matching the input network traffic according to the allocation strategy feature;

[0127] A traffic control execution module 506, configured to determine the network resources allocated to the network traffic of multiple traffic classes in the input network traffic according to a traffic allocation policy;

[0128] An update module 508, configured to update the traffic allocation policy according to network transmission performance information when the network transmission performance information matching each of the multiple traffic classes is obtained, where the network transmission performance information is determined during the transmission according to the allocated network resources, and the updated traffic allocation policy is used for the next traffic allocation operation.

[0129] Optionally, the above decision module 504 includes: a first determination unit, configured to determine the attention weight values respectively matching the feature vectors of multiple attribute dimensions through a gated attention mechanism, where the attention weight values are used to indicate the influence degree of the feature vectors matching the attribute dimensions on determining the policy features; determine the product between the attention weight values respectively matching the multiple attribute dimensions and the corresponding feature vectors, and determine the allocation policy features according to the sum of the products.

[0130] Optionally, the above feature extraction module 502 includes: a compression unit, configured to, when the feature extraction module is used to obtain data features of a protocol dimension, parse the protocol field information in the packet protocol header matching the input network traffic through the feature extraction module, and convert the protocol field information into a low-dimensional feature vector according to a compression function, where the compression function is called based on an interface of the TensorFlow architecture, and the protocol field information is used to indicate the service category of the packet matching the input network traffic; when the feature extraction module is used to obtain data features of a payload semantic dimension, parse the application information in the packet payload matching the input network traffic through the feature extraction module, and convert the application information into a low-dimensional feature vector according to a compression function, where the application information is used to indicate the application type of the packet matching the input network traffic; when the feature extraction module is used to obtain data features of a timing dimension, parse the timing information matching the input network traffic through the feature extraction module, and convert the timing information into a low-dimensional feature vector according to a compression function, where the timing information is used to indicate the time behavior pattern of the packet matching the input network traffic during transmission in the network.

[0131] Optionally, the above compression unit includes a capacity adjustment unit, configured to determine a storage hash table matching the input network traffic, where each position in the storage hash table corresponds to storing a feature identifier matching a low-dimensional feature vector; adjust the capacity of the storage hash table according to the total number of feature identifiers stored in the current storage hash table.

[0132] Optionally, the above traffic control execution module 506 is further configured to determine bandwidth allocation information and queue priority information respectively matching multiple traffic classes indicated by the traffic allocation policy; determine the network bandwidth quotas respectively allocated to the network traffic of multiple traffic classes in the input network traffic according to the bandwidth allocation information; and determine the traffic transmission order respectively allocated to the network traffic of multiple traffic classes in the input network traffic according to the queue priority information.

[0133] Optionally, the above update module 508 is further configured to, when it is detected that the network transmission performance information respectively matching multiple traffic classes does not meet the policy allocation condition, obtain network transmission performance loss information, where the network transmission performance loss information is used to indicate the pressure of the network service; and update the traffic allocation policy according to the performance loss information.

[0134] Optionally, the above compression unit is further configured to determine the performance requirements for executing the compression function, and design a hardware acceleration architecture according to the performance requirements; design an algorithm pipelining structure on the hardware acceleration architecture; and in each clock cycle of the algorithm pipelining structure, convert one piece of information into a low-dimensional feature vector by the compression function.

[0135] For the descriptions of the features in the embodiments corresponding to the traffic control device, reference can be made to the relevant descriptions in the embodiments corresponding to the traffic control method, which will not be elaborated here one by one.

[0136] An embodiment of the present application further provides an electronic device, including a memory and a processor. A computer program is stored in the memory, and the processor is configured to run the computer program to execute the steps in any of the above traffic control method embodiments.

[0137] The electronic device to which the present application is applied may be a terminal device or a server. In this embodiment, the electronic device is taken as a mobile phone and a computer as an example for illustration. As Figure 6 shown, the electronic device includes a memory 602 and a processor 604. A computer program is stored in the memory 602, and the processor 604 is configured to execute the steps in any of the above method embodiments through the computer program.

[0138] Optionally, in this embodiment, the above electronic device may be at least one network device among multiple network devices in a computer network.

[0139] Optionally, those of ordinary skill in the art can understand that Figure 6 the structure shown is only schematic, and the electronic device may also be a terminal device such as a smart phone (such as an Android phone, an iOS phone, etc.), a tablet computer, a handheld computer, and a Mobile Internet Device (MID), a PAD, etc. Figure 6It does not limit the structure of the above electronic device. For example, the electronic device may further include more or fewer components (such as a network interface, etc.) than those shown in Figure 6 , or have a different configuration from that shown in Figure 6 .

[0140] Among them, the memory 602 can be used to store software programs and modules, such as the program instructions / modules corresponding to the application method and device of the compatible interconnected switch in the embodiments of the present application. The processor 604 executes various functional applications by running the software programs and modules stored in the memory 602, that is, implements the application method of the above compatible interconnected switch. The memory 602 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some instances, the memory 602 may further include a memory remotely disposed relative to the processor 604, and these remote memories can be connected to the terminal through a network. Examples of the above network include but are not limited to the Internet, enterprise intranets, local area networks, mobile communication networks, and combinations thereof. Among them, the memory 602 can specifically but not limitedly be used to store information such as signals and data. As an example, as shown in Figure 6 , the above memory 602 may include but are not limited to the feature extraction module 502, decision module 504, traffic control execution module 506, and update module 508 in the application device of the above compatible interconnected switch. In addition, it may further include but are not limited to other module units in the application device of the above compatible interconnected switch, which will not be elaborated in this example.

[0141] Optionally, the above transmission device 606 is used to receive or send data via a network. Specific examples of the above network may include a wired network and a wireless network. In one instance, the transmission device 606 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices and routers through a network cable, so as to communicate with the Internet or a local area network. In one instance, the transmission device 606 is a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0142] In addition, the above electronic device further includes: a display 608; and a connection bus 610, which is used to connect each module component in the above electronic device.

[0143] In other embodiments, the above terminal device or server may be a node in a distributed system. The distributed system may be a blockchain system, which may be a distributed system formed by connecting the multiple nodes in a form of network communication. Among them, the nodes may form a point-to-point network, and any form of computing device, such as electronic devices like servers and terminals, can become a node in the blockchain system by joining the point-to-point network.

[0144] An embodiment of the present application further provides a computer-readable storage medium, in which a computer program is stored. The computer program is configured to execute the steps in any of the above embodiments of the traffic control method when running.

[0145] In an exemplary embodiment, the above computer-readable storage medium may include, but is not limited to: USB flash drives, read-only memories (ROM), random access memories (RAM), mobile hard disks, magnetic disks, or optical discs and other various media that can store computer programs.

[0146] An embodiment of the present application further provides a computer program product. The above computer program product includes a computer program, and when the computer program is executed by a processor, it implements the steps in any of the above embodiments of the traffic control method.

[0147] An embodiment of the present application further provides another computer program product, including a non-volatile computer-readable storage medium. The non-volatile computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the steps in any of the above embodiments of the traffic control method.

[0148] Those skilled in the art can further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Skilled professionals can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0149] The above has introduced in detail a traffic control method, device, storage medium and electronic device provided by the present application. Specific examples are used in this article to elaborate on the principle and implementation manner of the present application. The description of the above embodiments is only used to help understand the method and its core idea of the present application. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present application, several improvements and modifications can be made to the present application, and these improvements and modifications also fall within the protection scope of the claims of the present application.

Claims

1. A flow control method, characterized in that, Including: Determining a plurality of feature vectors matching the input network traffic based on a plurality of feature extraction modules, wherein the plurality of feature extraction modules are trained based on the TensorFlow architecture, and the plurality of feature vectors respectively correspond to an attribute dimension of the input network traffic; Performing feature fusion processing on the plurality of feature vectors through a gated attention mechanism to obtain an allocation policy feature, and determining a traffic allocation policy matching the input network traffic according to the allocation policy feature; Determining the network resources respectively allocated to the network traffic of multiple traffic categories in the input network traffic according to the traffic allocation policy; When obtaining network transmission performance information respectively matching the multiple traffic categories, updating the traffic allocation policy according to the network transmission performance information, wherein the network transmission performance information is determined during the transmission according to the allocated network resources, and the updated traffic allocation policy is used for the next traffic allocation operation.

2. The traffic control method according to claim 1, wherein: Performing feature fusion processing on the plurality of feature vectors through a gated attention mechanism to obtain an allocation policy feature, including: Determining attention weight values respectively matching the feature vectors of the multiple attribute dimensions through the gated attention mechanism, wherein the attention weight values are used to indicate the influence degree of the feature vectors matching the attribute dimensions on determining the policy feature; Determining the product between the attention weight values respectively matching the multiple attribute dimensions and the corresponding feature vectors, and determining the allocation policy feature according to the sum of the products.

3. The traffic control method according to claim 1, wherein: Determining a plurality of feature vectors matching the input network traffic based on a plurality of feature extraction modules, including: When the feature extraction module is used to obtain data features in the protocol dimension, parsing protocol field information in the packet protocol header matching the input network traffic through the feature extraction module, and converting the protocol field information into a low-dimensional feature vector according to a compression function, wherein the compression function is called based on an interface of the TensorFlow architecture, and the protocol field information is used to indicate the service category of the packet matching the input network traffic; When the feature extraction module is used to obtain data features in the payload semantic dimension, parsing application program information in the packet payload matching the input network traffic through the feature extraction module, and converting the application program information into a low-dimensional feature vector according to the compression function, wherein the application program information is used to indicate the application type of the packet matching the input network traffic; When the feature extraction module is used to obtain the data features in the time series dimension, the time series information matching the input network traffic is parsed by the feature extraction module, and the time series information is converted into the low-dimensional feature vector according to the compression function, where the time series information is used to indicate the time behavior pattern of the data packet matching the input network traffic during transmission in the network.

4. The traffic control method according to claim 3, wherein After converting the information parsed based on the feature extraction module into the low-dimensional feature vector according to the compression function, it includes: Determine the storage hash table matching the input network traffic, where each position in the storage hash table correspondingly stores a feature identifier matching the low-dimensional feature vector; Adjust the capacity of the storage hash table according to the total number of the feature identifiers stored in the current storage hash table.

5. The traffic control method according to claim 4, wherein Determining the network resources respectively allocated to the network traffic of multiple traffic classes in the input network traffic according to the traffic allocation policy includes: Determine the bandwidth allocation information and queue priority information respectively matching multiple traffic classes indicated by the traffic allocation policy; Determine the network bandwidth quotas respectively allocated to the network traffic of multiple traffic classes in the input network traffic according to the bandwidth allocation information; Determine the traffic transmission order respectively allocated to the network traffic of multiple traffic classes in the input network traffic according to the queue priority information.

6. The traffic control method according to claim 5, wherein When obtaining the network transmission performance information respectively matching multiple traffic classes, updating the traffic allocation policy according to the network transmission performance information includes: When it is detected that the network transmission performance information respectively matching multiple traffic classes does not meet the policy allocation conditions, obtain the network transmission performance loss information, where the network transmission performance loss information is used to indicate the pressure of the network service; Update the traffic allocation policy according to the performance loss information.

7. The traffic control method according to claim 3, wherein Converting the information parsed based on the feature extraction module into the low-dimensional feature vector according to the compression function includes: Determine the performance requirements for executing the compression function, and design a hardware acceleration architecture according to the performance requirements; Design an algorithm pipelining structure on the hardware acceleration architecture; In each clock cycle of the algorithm pipelining structure, the compression function converts one piece of information into one low-dimensional feature vector.

8. A flow control device, characterized in that, It includes: A feature extraction module, configured to determine multiple feature vectors matching the input network traffic based on multiple feature extraction modules, where the multiple feature extraction modules are trained based on the TensorFlow architecture, and the multiple feature vectors respectively correspond to one attribute dimension of the input network traffic; A decision-making module, configured to perform feature fusion processing on the multiple feature vectors through a gated attention mechanism to obtain allocation policy features, and determine a traffic allocation policy matching the input network traffic according to the allocation policy features; A traffic control execution module, configured to determine the network resources respectively allocated to the network traffic of multiple traffic classes in the input network traffic according to the traffic allocation policy; An update module, configured to update the traffic allocation policy according to the network transmission performance information when the network transmission performance information respectively matching the multiple traffic classes is obtained, wherein the network transmission performance information is determined during the transmission according to the allocated network resources, and the updated traffic allocation policy is used for the next traffic allocation operation.

9. An electronic device, characterized in that, Comprising: A memory, configured to store a computer program; A processor, configured to implement the steps of the traffic control method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that, A computer program is stored in the computer-readable storage medium, wherein the computer program implements the steps of the traffic control method according to any one of claims 1 to 7 when being executed by a processor.

Images