Resource allocation and flow control method, device and equipment based on multi-container network

Abstract

The invention relates to a resource allocation and flow control method, device and equipment based on a multi-container network. The method comprises the following steps: acquiring network state information of a multi-container network; performing feature extraction processing and feature enhancement processing on the network state information in sequence by using a pre-constructed multi-information fusion model to obtain multi-information fusion features; and performing decoding processing and mapping processing on the multi-information fusion features in sequence by using a pre-constructed multi-information decoding model to obtain a resource allocation result and a flow control result of each container. By adopting the embodiment of the invention, the continuously changing network service quality requirements can be met.

Description

Technical Field

[0001] The present application relates to the field of cloud computing technology, and in particular to a resource allocation and traffic control method, device and equipment based on a multi-container network. Background Art

[0002] In today's digital age, with the rapid development of cloud computing, enterprises and service providers rely on multi-container networks to support complex applications and services. Multi-container network architecture can achieve higher flexibility, scalability and fault tolerance by decomposing applications into multiple containers.

[0003] However, with the increase in the number of containers and the expansion of application scale, traditional resource allocation and traffic control methods are difficult to meet the ever-changing network service quality requirements. Summary of the invention

[0004] Based on this, it is necessary to provide a resource allocation and traffic control method, device and equipment based on a multi-container network that can meet the ever-changing network service quality requirements in response to the above technical problems.

[0005] In a first aspect, the present application provides a resource allocation and traffic control method based on a multi-container network, comprising:

[0006] Get network status information of multi-container network;

[0007] The pre-built multi-information fusion model is used to perform feature extraction and feature enhancement processing on the network status information in turn to obtain multi-information fusion features;

[0008] The pre-built multi-information decoding model is used to decode and map the multi-information fusion features in sequence to obtain the resource allocation result and flow control result of each container.

[0009] In a second aspect, the present application also provides a resource allocation and flow control device based on a multi-container network, including:

[0010] An information acquisition module is used to obtain network status information of a multi-container network;

[0011] An enhancement processing module is used to perform feature extraction and feature enhancement processing on the network status information in sequence using a pre-built multi-information fusion model to obtain multi-information fusion features;

[0012] The mapping processing module is used to use the pre-built multi-information decoding model to sequentially decode and map the multi-information fusion features to obtain the resource allocation result and flow control result of each container.

[0013] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the method of any one of the first aspects is implemented. In a fourth aspect, the present application further provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by the processor, the method of any one of the first aspects is implemented.

[0014] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, which implements any one of the methods in the first aspect when executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0016] FIG1 is a diagram showing the internal structure of a computer device in one embodiment;

[0017] Figure 2 It is a flowchart of a resource allocation and flow control method based on a multi-container network in one embodiment;

[0018] FIG3 is a schematic diagram of a flow chart of a resource allocation and flow control method based on a multi-container network in another embodiment;

[0019] FIG4 is a structural diagram of a multi-information fusion model in one embodiment;

[0020] FIG5 is a schematic flow chart of a resource allocation and flow control method based on a multi-container network in another embodiment;

[0021] FIG6 is a structural diagram of a feature enhancement model in one embodiment;

[0022] Figure 7 is a flow chart of a resource allocation and flow control method based on a multi-container network in another embodiment;

[0023] Figure 8 is a flow chart of a resource allocation and flow control method based on a multi-container network in another embodiment;

[0024] Fig. 9 is a flow chart of a resource allocation and flow control method based on a multi-container network in another embodiment;

[0025] Fig.10The structure block diagram is a resource allocation and flow control device based on a multi-container network in one embodiment. DETAILED DESCRIPTION

[0026] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0027] In an exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be shown in FIG1. ​​The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data in the resource allocation and flow control process based on a multi-container network. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a resource allocation and flow control method based on a multi-container network is implemented.

[0028] Those skilled in the art will understand that the structure shown in FIG. 1 is merely a block diagram of a partial structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different arrangement of components.

[0029] In an exemplary embodiment, Figure 2 As shown, a resource allocation and flow control method based on a multi-container network is provided, and the method is applied to the computer device in FIG. 1 as an example for explanation, including the following steps 201 to 203. Among them:

[0030] Step 201: Obtain network status information of a multi-container network.

[0031] Among them, multi-container networking usually refers to deploying applications using multiple containers in a distributed system. These containers may span multiple hosts and communicate through the network. Containers are lightweight virtualization environments that encapsulate applications and their dependent environments. Multiple containers are usually managed by a container orchestration platform (such as Kubernetes) and work together to provide efficient services.

[0032] Network status information includes information such as communication status, bandwidth utilization, latency, and packet loss rate between containers in the network.

[0033] In an embodiment of the present application, a computer device monitors the health status of a multi-container network in real time through a container management platform and obtains network status information of the multi-container network.

[0034] In some embodiments, the resource usage of each container can be collected, such as the central processing unit (CPU), memory, storage, and network bandwidth, to ensure that the status of each container in the network is effectively monitored. At the same time, the communication traffic between containers is monitored in real time through network traffic analysis tools to obtain information such as traffic size, protocol type, bandwidth usage, delay, and packet loss rate.

[0035] Step 202, using a pre-built multi-information fusion model to perform feature extraction and feature enhancement processing on the network status information in sequence, to obtain multi-information fusion features.

[0036] Among them, feature extraction processing refers to extracting important and effective features (such as traffic size, response time, resource consumption, etc.) from network status information.

[0037] Feature enhancement processing is based on the extracted features and uses algorithms to enhance their expressiveness to better reflect the status and load of the container network.

[0038] In the embodiment of the present application, when the network status information is processed using the pre-built multi-information fusion model, the network status information needs to be pre-processed first, such as removing redundant information, filling missing values, standardizing the format, and performing normalization operations, to ensure that data from different sources can be input into the subsequent multi-information fusion model with consistent standards.

[0039] Next is the feature extraction process. Feature extraction is to identify key and meaningful data features from network status information. For example, relevant features can be extracted from the resource usage of the container (such as CPU, memory, bandwidth, etc.), network performance indicators (such as latency, packet loss rate, bandwidth utilization), and the health status of the container (such as error rate, load status). The core purpose of feature extraction is to convert raw and complex data into a feature set that can effectively describe the network status. These features will provide a basis for subsequent processing.

[0040] After feature extraction, feature enhancement processing is performed to obtain multi-information fusion features. The purpose of feature enhancement is to improve the expressiveness of the extracted features, making them more conducive to subsequent analysis and modeling. Feature enhancement methods can include data smoothing and denoising to reduce random fluctuations in the data; feature crossover and combination, by combining multiple features into new composite features to capture the relationship between features; and nonlinear transformations, such as logarithmic transformations or power transformations, to cope with the differences in features of different magnitudes.

[0041] In some embodiments, information fusion processing can also be performed to organically combine information from different data sources to form a multimodal feature representation. There are many ways to fuse information, including weighted fusion, splicing fusion, and hierarchical fusion. Weighted fusion combines different features into a single feature by assigning weights to them; splicing fusion directly connects each feature vector to form a higher-dimensional feature vector; and hierarchical fusion processes local features first and then merges them with global features. Information fusion can not only make up for the shortcomings of a single data source, but also capture more dimensional network status information.

[0042] In the fused feature space, there are often redundant or irrelevant features, which will affect the effect and efficiency of the multi-information fusion model. Therefore, feature selection and optimization can also be performed. Feature selection removes redundant features through correlation analysis, evaluates the importance of each feature, and retains features that contribute greatly to the decision-making of the multi-information fusion model. In addition, the use of dimensionality reduction methods can further reduce the dimension of the features, thereby improving the computational efficiency of the multi-information fusion model and avoiding overfitting.

[0043] Step 203, using a pre-built multi-information decoding model to sequentially perform decoding processing and mapping processing on the multi-information fusion features, to obtain a resource allocation result and a flow control result for each container.

[0044] Among them, the resource allocation result refers to how to allocate multi-container network resources (such as CPU, memory, network bandwidth) to different containers based on factors such as the current load of the container, network bandwidth, computing power, etc., to ensure the efficient operation of multiple containers.

[0045] Traffic control results refer to how to manage and adjust different traffic according to the network status during network transmission. Traffic control includes traffic scheduling, bandwidth allocation, delay optimization and other aspects to avoid network congestion, improve throughput and reduce packet loss.

[0046] In the above-mentioned resource allocation and flow control method based on multi-container network, the network status information of the multi-container network is first obtained; then, the network status information is sequentially subjected to feature extraction and feature enhancement processing using a pre-built multi-information fusion model to obtain multi-information fusion features; finally, the multi-information fusion features are sequentially decoded and mapped using a pre-built multi-information decoding model to obtain resource allocation results and flow control results for each container; in this method, by obtaining the network status information of the multi-container network, the current network environment can be accurately monitored; the use of the multi-information fusion model enables the network status information to be processed comprehensively, avoiding the one-sidedness of single-dimensional information. By performing feature extraction and enhancement processing on the network status information, effective and high-quality features can be extracted from the network status information, so that the network management system can better understand the changing trend of the network status. By decoding and mapping the fusion features using the multi-information decoding model, the resource allocation and flow control results of each container can be obtained, so that the resource allocation strategy and flow control rules can be dynamically adjusted according to the real-time changes in network demand, ensuring that each container can obtain the required resources under different load conditions, reducing resource waste or the occurrence of bottleneck phenomena.

[0047] In an exemplary embodiment, as shown in FIG3 , the multi-information fusion model includes a multi-level sub-fusion model, on which the pre-built multi-information fusion model is used to sequentially perform feature extraction and feature enhancement processing on the network state information to obtain multi-information fusion features, including steps 301 to 303. Wherein:

[0048] Step 301, use the multi-level sub-fusion model to iteratively calculate the network status information. In the first iteration, the network status information is input into the first-level sub-fusion model, and the first-level sub-fusion model is used to perform feature extraction and feature enhancement on the network status information in sequence to obtain the first fusion feature.

[0049] In the embodiment of the present application, in the first iteration process, the network state information is input into the first-level sub-fusion model. As shown in Figure 4, Figure 4 is a structural diagram of the multi-information fusion model. In Figure 4, transformer Encoder and Encoder represent encoders, mlp represents multi-layer perceptron, LSTM represents long short-term memory network, Qos main feature enhancement module represents feature enhancement model; secondary input features represent network traffic information, primary input features represent resource demand information, and time series input features represent time series information.

[0050] The purpose of this process is to extract basic features from the network status information and enhance them in order to provide preliminary fusion features for subsequent iterations. In this process, the first-level sub-fusion model will first perform feature extraction processing on the input network status information, and can identify important basic features such as resource usage, network performance, container health status, etc. from the original data. Next, after feature enhancement processing, these features will be further transformed and optimized, such as through feature crossover, data smoothing, or nonlinear transformation, to enhance their expressiveness and make them more suitable for subsequent model processing. Finally, after processing by the first-level sub-fusion model, the first fusion feature is obtained, which will be passed as input to the next stage of processing.

[0051] Step 302, in a non-first iteration process, the previous fusion feature obtained by the previous level sub-fusion model is input into the current level sub-fusion model, and the current level sub-fusion model is used to perform feature extraction and feature enhancement processing on the previous fusion feature in sequence to obtain the current level fusion feature.

[0052] In an embodiment of the present application, referring to Figure 4, in a non-first iteration process, the sub-fusion model of this level will be processed layer by layer according to the previous iteration results. The output of the sub-fusion model of the previous level, that is, the previous fusion feature, will be input into the sub-fusion model of this level. The sub-fusion model of this level will continue to perform feature extraction and enhancement processing on the fusion features of these previous level outputs. The feature extraction at this stage focuses more on discovering deeper patterns from the already fused features, such as extracting more complex relationships and trends by combining fusion features of multiple dimensions. After the enhancement processing, the sub-fusion model of this level will output the fusion feature of this level, which will be passed as input to the sub-fusion model of the next layer.

[0053] Through this iterative process, each layer of the sub-fusion model is constantly deepening and refining features, gradually enhancing the expressiveness of features and the richness of information. Each iteration can extract more potential key information based on the previous level of fusion features, and further improve the effectiveness of features through enhancement processing. This gradually deepening feature extraction and enhancement process makes the final fusion features more comprehensive and can better describe the complexity of network status information.

[0054] Step 303: determine the fusion features output by the final sub-fusion model as multi-information fusion features.

[0055] In the embodiment of the present application, after all iterative levels of processing, the final sub-fusion model will output the final multi-information fusion feature. This multi-information fusion feature is the cumulative result of all iterative processes, which integrates the extraction and enhancement of features by each level of model. Therefore, the multi-information fusion feature finally obtained not only contains the key information in the original network state information, but also captures more complex and deep network state features through multiple iterations and enhancements, which can provide strong data support for subsequent decision-making and optimization tasks.

[0056] In the above embodiment, a multi-level sub-fusion model is used to iteratively calculate the network status information, and features are gradually mined and optimized from simple to complex through staged feature extraction and enhancement. Through the preliminary processing of the first-level sub-fusion model and the progression of each level of sub-fusion model, high-quality multi-information fusion features are finally obtained, which can effectively improve the perception ability of complex network status, thereby improving the accuracy and efficiency of subsequent decision-making.

[0057] In an exemplary embodiment, the network status information includes resource demand information, network traffic information and timing information. The sub-fusion model includes an encoder, a multi-layer perceptron and a long short-term memory network. On this basis, the pre-built multi-information fusion model is used to perform feature extraction and feature enhancement processing on the network status information in sequence to obtain multi-information fusion features, including:

[0058] In case 1, the resource requirement information is input into the encoder for feature extraction processing to obtain the first resource requirement feature output by the encoder.

[0059] The encoder is a feedforward neural network consisting of multiple layers of neurons that can effectively extract high-dimensional abstract features from the input raw data. For resource demand information, the encoder performs nonlinear transformations on it through a multi-layer structure to capture complex patterns and relationships.

[0060] In the embodiment of the present application, referring to FIG4 , the resource requirement information is input into the encoder for feature extraction processing, and meaningful features are extracted from the resource requirement data, such as the CPU, memory, and bandwidth requirements of each container. After being processed by the encoder, the first resource requirement feature is output.

[0061] In some embodiments, since the network status information includes resource demand information, network traffic information and timing information, in actual application, a resource demand information data set, a network traffic information data set and a timing information data set may be constructed first.

[0062] (1) Construction of resource demand information dataset

[0063] The embodiment of the present application considers that the resource demand for containers, such as CPU usage and memory usage, is the main feature of multi-container network management. A resource demand information dataset is constructed based on these features. Specifically, data can be collected in the following ways:

[0064] Use the monitoring interface provided by the container orchestration tool to periodically obtain the CPU and memory usage data of each container. For example, obtain the data every 5 seconds and record it.

[0065] Deploy monitoring agents inside containers to collect and report resource usage in real time.

[0066] (2) Construction of network traffic information dataset

[0067] The embodiment of the present application considers the packet size in the network traffic characteristics and the idle or fault state in the running state of the container as secondary characteristics of multi-container network management, and constructs a data set based on these characteristics. , data can be collected in the following ways:

[0068] For the packet size in network traffic characteristics, traffic monitoring tools can be deployed on network nodes to capture data packets and analyze their sizes.

[0069] For the idle or faulty state of the container's running state, the status information can be obtained and recorded through the container's health check mechanism.

[0070] (3) Construction of time series information dataset

[0071] In the embodiment of the present application, the data with time series is considered as the time series characteristics of multi-container network management, such as CPU usage, memory usage, network bandwidth requirements, periodicity of network traffic, and the duration of the container being in a busy state over time, and a data set is constructed based on these characteristics. , which can be obtained by:

[0072] Set timestamps for indicators such as CPU usage, memory usage, and network bandwidth requirements, and sample and record them at fixed time intervals (such as every minute) to form time series data.

[0073] For the periodicity of network traffic, the traffic data over a period of time (such as a day or a week) is analyzed to determine its periodic pattern and record the relevant information in a time series dataset.

[0074] For the duration of a container being in a busy state, by monitoring the activity of the container, calculating and recording the start time and end time of each busy state, the time series data of the duration is obtained.

[0075] When constructing a data set, you also need to pay attention to the accuracy, completeness, and consistency of the data. At the same time, you need to clean and preprocess the data to remove outliers and erroneous data to ensure the quality of the data set.

[0076] Check the entire data set to identify and remove duplicate rows. This can be done by using the database's deduplication function or using a data processing library.

[0077] For outliers in numerical features (such as CPU usage, memory usage, etc.), you can use box plots or the 3σ principle to identify outliers. If the CPU usage of a container exceeds 3 standard deviations of the mean, it can be considered an outlier. For outliers, you can delete, modify, or retain them according to the specific situation. If it is a reasonable anomaly caused by a temporary system peak, it can be retained; if it is a data collection error, it should be modified or deleted.

[0078] The data preprocessing process can be:

[0079] (1) Data standardization / normalization: Min-Max normalization is performed on feature data of different magnitudes and ranges to facilitate subsequent analysis and model training.

[0080] (2) Feature encoding: For the classification features in the data set (such as application type, container running status, etc.), one-hot encoding is used for encoding.

[0081] (3) Data smoothing: For data with noise, such as network traffic data, moving average, exponential smoothing and other methods can be used for smoothing to reduce the impact of noise.

[0082] In case 2, the network traffic information is input into the multi-layer perceptron for feature extraction and processing, and the first network traffic feature output by the multi-layer perceptron is obtained.

[0083] Among them, a multilayer perceptron usually refers to a processing module specifically used to parse and convert input data. It can analyze information such as patterns, behaviors, and trends of network traffic.

[0084] In the embodiment of the present application, referring to FIG4 , the network traffic information is input into the multi-layer perceptron for feature extraction. The multi-layer perceptron extracts key features such as the time-varying nature of traffic, peak time period, bandwidth utilization, etc. by analyzing the network traffic data. At this stage, the task of the multi-layer perceptron is to convert the network traffic information into high-dimensional data that can reflect the traffic characteristics, and obtain the first network traffic feature, which provides a basis for subsequent traffic control, bandwidth allocation and other operations.

[0085] In case three, the time series information is input into the long short-term memory network for feature extraction to obtain the first time series feature output by the long short-term memory network.

[0086] Among them, the long short-term memory network is a special type of recurrent neural network that can effectively process time series data and capture long-term dependencies in the data.

[0087] In the embodiment of the present application, as shown in Figure 4, the timing information is input into the long short-term memory network for feature extraction. At this stage, the timing information (such as historical data of network load, traffic changes, container resource changes, etc.) will be input into the long short-term memory network. The long short-term memory network analyzes the time dependency of the data through its memory unit and extracts the timing features therein, such as trend prediction, periodic changes, sudden events, etc. These timing features can help the system understand the evolution law and future trends of the network status, and the final output is the first timing feature.

[0088] In the above embodiment, by inputting resource demand information, network traffic information and timing information into different neural network models for processing, the core features of various types of information are gradually extracted. Through the collaborative work of modules such as multi-layer perceptron, compiler and long short-term memory network, the key features of network status can be mined from multiple dimensions, providing comprehensive data support for subsequent decision-making.

[0089] In an exemplary embodiment, as shown in FIG5 , the sub-fusion model further includes a feature enhancement model, on which the first-level sub-fusion model is used to perform feature enhancement processing on the network state information to obtain the first fusion feature, including steps 401 to 403. Wherein:

[0090] Step 401, using the feature enhancement model to perform maximum pooling processing and average pooling processing on the first network traffic feature, and sequentially perform splicing processing, convolution processing and mapping processing on the processed first network traffic feature to obtain the first initial information feature.

[0091] Among them, the goal of the feature enhancement model is to enhance the expressiveness of network traffic features through different processing methods, so as to provide richer information for subsequent calculations.

[0092] In an embodiment of the present application, the feature enhancement model is used to perform maximum pooling and average pooling on the first network traffic feature. Maximum pooling can help extract the most significant information in the feature, thereby retaining the strongest network traffic pattern; while average pooling smoothes the feature by taking the average value, eliminating noise and retaining global features. These two pooling methods can help the feature enhancement model understand the details and overall trends of network traffic at different levels.

[0093] After the maximum pooling and average pooling processes, the pooling results will be sequentially processed by splicing, convolution, and mapping. The splicing process splices the feature vectors after the maximum pooling and average pooling together to generate a high-dimensional feature vector that combines the information from the two pooling methods. Next, the convolution process performs deep learning on the spliced ​​features through convolution operations to further extract meaningful spatial features. The convolution process can effectively capture local patterns and complex relationships in the features. Finally, the mapping process maps the convolution results to the target feature space through a fully connected layer to obtain the first initial information feature. This initial feature is an enhanced representation of the network traffic features, ready to support subsequent calculations.

[0094] In some embodiments, as shown in FIG6 , FIG6 is a diagram of a feature enhancement model structure. In FIG6 , the secondary feature Indicates the first network traffic feature, max pool indicates maximum pooling, avg pool indicates average pooling, and cat indicates concatenation. Represents the first network traffic feature after maximum pooling processing; represents the first network traffic feature after average pooling processing, represents the first network traffic feature after concatenation, BatchNorm represents batch normalization, represents the first network traffic feature after convolution processing, It represents the first network traffic feature after mapping, that is, the first initial information feature. represents the first resource demand feature, represents the first intermediate information feature, represents the first fusion feature, Represents the first time series feature.

[0095] This embodiment of the present application uses the secondary feature extraction attention mechanism, based on the inspiration of the channel attention mechanism, to perform effective feature extraction with the least amount of parameters, and uses the S-shaped sigmoid function to convert the features into a weight map. The features are obtained by using maximum pooling and average pooling respectively. and ,Will and Features obtained by splicing , will be passed The convolutional layer composed of 1*1 convolution, BatchNorm and 3*3 convolution is obtained , and then the secondary feature weight vector graph is obtained through the sigmoid function .

[0096] Step 402, performing a dot product process on the first initial information feature and the first resource demand feature to obtain a first intermediate information feature.

[0097] The dot product operation performs element-wise multiplication on two feature vectors to fuse their information.

[0098] In the embodiment of the present application, the first initial information feature and the first resource demand feature are subjected to dot product processing to obtain the first intermediate information feature. The dot product of the resource demand feature and the network traffic feature can capture the relationship between resource demand and traffic status, providing more accurate input for subsequent resource optimization and traffic control.

[0099] In some embodiments, referring to FIG. 6, the obtained and Point product to get the weighted features of the secondary features , and set the weight parameter to 0.3, and Add together to get , the specific formula is as follows:

[0100]

[0101] for , using the temporal attention mechanism for feature extraction, we get .

[0102] Step 403, the first intermediate information feature and the first time series feature are sequentially concatenated and convolved to obtain a first fused feature.

[0103] In an embodiment of the present application, the first intermediate information feature and the first time series feature are sequentially concatenated and convolved to obtain a first fused feature.

[0104] First, the intermediate information features and time series features are concatenated to fuse multi-dimensional features. Then, the concatenated features are further learned through convolution processing to extract the deep correlation between time series changes and network states. After these processes, the first fused feature is finally obtained.

[0105] In some embodiments, referring to FIG. 6 , and After concatenation, 1*1 convolution is performed to obtain The first fused feature of the same size .

[0106] In the above embodiment, the expression ability of network traffic features is enhanced through maximum pooling and average pooling, and resource requirements and traffic features are fused through point multiplication operations, and then the time series information is combined with the intermediate features through splicing and convolution processing, and finally a comprehensive and expressive first fusion feature is obtained. This feature not only integrates multi-dimensional information, but also captures the complex relationships between them through deep processing, providing a more accurate basis for subsequent optimization, prediction or decision-making.

[0107] In an exemplary embodiment, the last fused feature includes the last enhanced feature, the last network traffic feature, and the last time series feature, and the sub-fusion model of this level is used to perform feature extraction processing on the last fused feature, including:

[0108] In case 1, the previous enhanced feature is input into the encoder for feature extraction processing to obtain the enhanced feature of the current level output by the encoder.

[0109] In the embodiment of the present application, the last enhanced feature is input into the encoder for feature extraction processing. The encoder performs in-depth processing on the last enhanced feature to extract a more refined and meaningful feature representation. The encoder captures more structural relationships or potential patterns by performing nonlinear transformation on the features, thereby generating enhanced features at this level.

[0110] In case 2, the previous network traffic feature is input into the multi-layer perceptron for feature extraction and processing, and the network traffic feature of this level output by the multi-layer perceptron is obtained.

[0111] In the embodiment of the present application, the previous network traffic feature is input into a multi-layer perceptron for feature extraction processing. The multi-layer perceptron performs deep learning on the network traffic feature, extracts important rules and features in the traffic, and thus generates the current level network traffic feature.

[0112] In case three, the previous time series feature is input into the long short-term memory network for feature extraction to obtain the current level time series feature output by the long short-term memory network.

[0113] In the embodiment of the present application, the previous time series feature is input into the long short-term memory network for feature extraction. The long short-term memory network will analyze the dynamic changes of the time series information and extract information such as trends, periodic changes and emergencies related to time. After being processed by the long short-term memory network, the time series feature of this level is obtained.

[0114] In the above embodiment, the processing results of the previous level are input into the encoder, multi-layer perceptron and long short-term memory network, and the enhanced features, network traffic features and timing features are further extracted. These processing modules each play their own advantages. The encoder deeply learns and abstracts the complex relationship of enhanced features, the multi-layer perceptron focuses on the nonlinear mapping of traffic features, and the long short-term memory network extracts long-term trends and dynamic dependencies in time series data. The enhanced features, network traffic features and timing features finally obtained at this level provide richer and deeper feature expressions for subsequent decision-making and optimization tasks.

[0115] In an exemplary embodiment, Figure 7 As shown, the sub-fusion model of this level is used to perform feature enhancement processing on the previous fusion feature to obtain the fusion feature of this level, including steps 501 to 503. Among them:

[0116] Step 501, use the feature enhancement model to perform maximum pooling and average pooling on the network traffic features of this level, and perform splicing, convolution and mapping on the processed network traffic features of this level in turn to obtain the initial information features of this level.

[0117] In an embodiment of the present application, the network traffic features at this level will be input into the feature enhancement model for processing. The goal of the feature enhancement model is to extract more expressive information from the input network traffic features. The feature enhancement model first performs maximum pooling and average pooling on the network traffic features. The maximum pooling operation can extract the most significant parts from the features and retain the high-intensity signals in the network traffic features, while average pooling helps to smooth the features, remove noise, and maintain global information. Through these two pooling processes, the feature enhancement model can understand the traffic features from different angles, thereby obtaining more comprehensive information.

[0118] Step 502, perform dot product processing on the initial information feature of this level and the resource demand feature of this level to obtain the intermediate information feature of this level.

[0119] In an embodiment of the present application, the pooled network traffic features are further transformed through splicing, convolution and mapping. The splicing process merges the feature vectors after maximum pooling and average pooling to generate a high-dimensional feature representation, which can combine the advantages of the two pooling processes. Then, the convolution process deeply extracts the spatial information and local patterns in the features through the convolution operation, further enhancing the expressive power of the features. Finally, the mapping process maps the convolved features to the target feature space to generate the initial information features at this level.

[0120] Step 503, sequentially perform concatenation processing and convolution processing on the intermediate information features of this level and the time series features of this level to obtain the fusion features of this level.

[0121] In the embodiment of the present application, the initial information features of this level and the resource demand features of this level will be processed by dot multiplication to obtain the fusion features of this level. The dot multiplication operation is to fuse these two types of features by multiplying the elements of the two vectors one by one and adding the results. The dot multiplication of resource demand features and network traffic features can capture the relationship between the two.

[0122] Finally, the intermediate information features and the time series features of this level will be concatenated and convolved in turn. The concatenation operation combines the time series features and the intermediate information features to form a multi-dimensional comprehensive feature. The time series features capture the time-related change trends, while the intermediate information features provide interactive information about network traffic and resource requirements, which can be integrated after concatenation. Next, the convolution process extracts the complex patterns between the time series and resource information by convolving the concatenated features, further enhancing the expressiveness of the features. Finally, after these processes, the fused features of this level are obtained.

[0123] In the above embodiment, the network traffic characteristics are enhanced by maximum pooling and average pooling, and different features are gradually fused by combining point multiplication processing and convolution processing, and finally a comprehensive current-level fusion feature is obtained. This feature integrates the deep correlation between traffic, resource requirements and timing information, and can provide an accurate basis for subsequent decision-making, resource optimization and traffic control.

[0124] In an exemplary embodiment, Figure 8 As shown, the multi-information decoding model includes a decoder, a first fully connected neural network and a second fully connected neural network. On this basis, the multi-information fusion features are sequentially decoded and mapped using the pre-built multi-information decoding model to obtain the resource allocation result and flow control result of each container, including steps 601 to 603. Among them:

[0125] Step 601, inputting the multi-information fusion features into the decoder to obtain the decoding features output by the decoder.

[0126] Among them, the decoded features are processed representations with higher semantic information, which can reflect the key patterns of network and resource status.

[0127] In the embodiment of the present application, the multi-information fusion feature is input to the decoder for processing. The role of the decoder is to extract more expressive decoding features from the fusion feature. The decoder extracts and maps the input multi-information fusion feature to generate a decoding feature.

[0128] Step 602: Input the decoded features into a first fully connected neural network, map the decoded features to a parameter space related to resource allocation through the first fully connected neural network, and obtain a resource allocation result for each container.

[0129] Among them, the first fully connected neural network is used in deep learning to map input features to a specific output space.

[0130] In an embodiment of the present application, the decoded features are input into a first fully connected neural network. The first fully connected neural network maps the decoded features into a parameter space related to resource allocation. The mapping process can calculate the resource allocation results required for each container based on the network and resource status information, such as the computing power, memory, bandwidth and other resources required for each container.

[0131] Step 603: Input the decoded features into the second fully connected neural network, map the decoded features into a parameter space related to flow control through the second fully connected neural network, and obtain the flow control result of each container.

[0132] In the embodiment of the present application, the decoded features will also be input into the second fully connected neural network. The task of the second fully connected neural network is to map the decoded features to the parameter space related to flow control. The second fully connected neural network calculates the flow control strategy required for each container, such as bandwidth allocation, delay control, etc., based on the flow pattern and demand in the decoded features.

[0133] Through the independent processing of these two neural networks, two key results can be obtained for each container: resource allocation result and flow control result. These two results can ensure that the system reasonably allocates resources among multiple containers while ensuring the efficiency and stability of flow control.

[0134] In some embodiments, the main decoding part of the decoder directly uses the decoder structure of the transformer to obtain the decoding features The decoded features are passed through two fully connected layers to output two results:

[0135] One of them The prediction result of how resources are allocated, that is, the resource allocation result, is obtained by mapping the decoding features to the relevant parameter space of resource allocation through the fully connected layer. The prediction result of resource allocation is mainly used to determine how to optimally allocate computing resources such as CPU, memory, storage resources and network bandwidth.

[0136] Another one is The predicted result of the flow control strategy, that is, the flow control result of each container, Another fully connected layer is used to obtain a specific flow control strategy. The strategy is mainly based on the decoding characteristics and the real-time status of the network to determine the data output flow of each container.

[0137] In the above embodiment, the decoder extracts key information from the fusion features, and then two fully connected neural networks are used to map this information into the parameter space of resource allocation and flow control, and finally an accurate resource allocation and flow control strategy is generated for each container.

[0138] In an exemplary embodiment, Fig. 9 As shown, the above method further includes step 701 to step 702. Among them:

[0139] Step 701: Determine a resource allocation loss value, a service quality loss value, and a flow control loss value based on a resource allocation result and a flow control result.

[0140] The resource allocation loss value is used to measure the accuracy of the model in resource allocation. The resource allocation loss is calculated by comparing the deviation between the resource allocation calculated by the model and the actual required resources.

[0141] The service quality loss value reflects the model's ability to ensure service quality, and is usually calculated by evaluating indicators such as the system's response time, throughput, and latency.

[0142] The flow control loss value measures the accuracy of the system in traffic scheduling and bandwidth management. The flow control loss can be obtained by comparing the difference between the actual traffic distribution and the expected traffic control strategy.

[0143] In an embodiment of the present application, a resource allocation loss value, a service quality loss value, and a flow control loss value are determined based on the resource allocation result and the flow control result.

[0144] Step 702, adjusting the parameters of the multi-information fusion model based on the resource allocation loss value, the service quality loss value and the flow control loss value to optimize the multi-information fusion model.

[0145] In an embodiment of the present application, the multi-information fusion model is adjusted based on the resource allocation loss value, the service quality loss value and the flow control loss value to optimize the multi-information fusion model.

[0146] In some embodiments, the loss value may further specifically include:

[0147] Resource allocation loss value: For the resource allocation loss value, this embodiment of the application will consider the CPU, memory, storage resources and network bandwidth from four aspects:

[0148] CPU allocation loss:

[0149]

[0150] in Represents the actual value of CPU allocation, Represents the predicted value of CPU allocation; Indicates CPU allocation loss.

[0151] Memory allocation penalty:

[0152]

[0153] in represents the actual value of the memory allocation, Represents the predicted value of memory allocation; Represents memory allocation penalty.

[0154] Storage resource allocation loss:

[0155]

[0156] in represents the actual value of the memory allocation, Represents the predicted value of memory allocation, Indicates a loss in storage resource allocation.

[0157] Network bandwidth allocation loss:

[0158]

[0159] in represents the actual value of the memory allocation, Represents the predicted value of memory allocation, Indicates loss of network bandwidth allocation.

[0160] The resource allocation loss value is:

[0161]

[0162] The λ can be set according to the application preference. For example, if you pay more attention to the CPU, you can The coefficient is adjusted to be larger.

[0163] The service quality loss part is mainly based on the service quality requirements of the specific system, and is also an indicator for evaluating the overall network capabilities. It is also an evaluation indicator of the model. The embodiment of the present application will evaluate the predicted results based on the simulation tool. The evaluation does not depend on the actual label, but is more like an evaluation and loss value of a semi-supervised model. Specifically speaking, the following are some evaluation indicator rules specified for multi-container network management:

[0164] Resource Utilization:

[0165] CPU utilization: The average CPU utilization should be kept between 70% and 90% to avoid excessive idleness or overload. If the CPU utilization is lower than 70%, then loss= ; If the CPU utilization is greater than 90%, then loss= .

[0166] Memory utilization: The average memory utilization should be in the reasonable range of 60%-85%. If the memory utilization is lower than 60%, then loss= ; If the CPU utilization is greater than 85%, then loss= .

[0167] Storage utilization: Storage usage should not exceed 80% of the total capacity to reserve some room for expansion. If it exceeds 80%, loss= .

[0168] The above losses are concatenated with reasonable weights to obtain the resource utilization loss function .

[0169] Network performance:

[0170] Network delay: The average network delay should be less than 100 milliseconds, and the delay of key applications should be less than 50 milliseconds. Network delay loss function:

[0171]

[0172] Packet loss rate: The overall packet loss rate should be less than 1%. For applications with high real-time requirements, the packet loss rate should be less than 0.1%. Packet loss rate loss function:

[0173]

[0174] Bandwidth: Ensure that each container can obtain the predetermined minimum bandwidth, and the average bandwidth utilization is between 60% and 80%. If the bandwidth utilization is lower than 60%, loss= ; If the bandwidth utilization is greater than 80%, then loss= .

[0175] The above losses are concatenated with reasonable weights to obtain the network performance loss function .

[0176] Service Availability:

[0177] The service uptime should be above 99.9%, that is, the annual downtime should not exceed the threshold T hours. The service uptime is usually difficult to directly express with a loss function, but it can be indirectly evaluated by monitoring the gap between the actual uptime and the target time, then:

[0178] .

[0179] Response time:

[0180] For common requests, the average response time should be within 2 seconds. For urgent or critical requests, the response time should be within 500 milliseconds. , then the loss function is:

[0181]

[0182] Flow control accuracy:

[0183] The deviation between the actual flow rate and the set flow rate should be within 10%. , then the loss function is:

[0184] .

[0185] Fairness of resource allocation:

[0186] When resources are insufficient for containers of different priorities, resources are allocated according to priority, and low-priority containers should not be completely deprived of resources. In this embodiment of the application, the resource allocation difference between containers of the same type should be regarded as the loss value. Assume that the expected value of containers of the same priority is set to , each container obtains resources as , the loss function is:

[0187]

[0188] Fault recovery capability:

[0189] The target time from failure to restoration of normal service should be within 5 minutes. The amount of data loss during the failure should be controlled within an acceptable range, assuming the acceptable range is , the loss value calculation process is rounded up. Then the loss function is:

[0190]

[0191] Scalability:

[0192] When adding new containers or services, resource allocation and traffic control can be performed without significantly affecting the QoS of existing services. Let the loss function be:

[0193]

[0194] Energy efficiency:

[0195] Try to reduce the overall energy consumption, and set the target minimum energy consumption as . Then the loss function is:

[0196]

[0197] Cost-effectiveness:

[0198] Comprehensively consider resource usage costs and service quality to ensure the optimal QoS level within a certain budget. Assuming that the resource usage cost is C, the service quality score can be calculated based on the specific QoS indicators. The embodiment of this application uses principal component analysis to calculate Qos. By performing data analysis on multiple QoS indicators, the principal component that can explain the maximum variance in the data is found. These principal components are linear combinations of the original indicators and can be used as comprehensive indicators. The QoS indicator is calculated as Q, the budget is B, and the target service quality score threshold is . Then the loss function is:

[0199]

[0200] in, and is an adjustment coefficient used to weigh the losses caused by cost overruns and substandard service quality.

[0201] The above are relatively comprehensive service quality indicators formulated in the embodiment of the present application. The loss function is calculated based on each indicator to obtain arrive , set the parameters to arrive , the service quality loss function is:

[0202]

[0203] Considering that each deployment environment has different requirements for each service quality indicator, It can be fine-tuned according to the actual application scenario to better suit the application scenario.

[0204] The resource allocation loss, flow control loss and service quality loss are weighted and integrated to obtain the final loss function. This loss function can consider all aspects more comprehensively to train a model that is more suitable for the application environment, and adjust the model parameters based on this.

[0205] In the above embodiment, parameter optimization based on resource allocation loss value, service quality loss value and flow control loss value is the core process of improving the performance of the multi-information fusion model. Through this process, the model can gradually reduce losses, improve the accuracy of resource allocation, optimize service quality, improve flow control strategy, and ultimately achieve more efficient and stable system performance.

[0206] According to some embodiments of the present application, a resource allocation and traffic control method based on a multi-container network is provided. The method is described by taking the application of the method to a computer device as an example, and the method may include the following steps:

[0207] Step 1: Obtain the network status information of the multi-container network. The network status information includes resource demand information, network traffic information, and timing information.

[0208] Step 2, use the multi-level sub-fusion model to iteratively calculate the network state information. In the first iteration, the network state information is input into the first-level sub-fusion model, and the first-level sub-fusion model is used to perform feature extraction processing on the network state information in turn. The sub-fusion model includes an encoder, a multi-layer perceptron, a long short-term memory network, and a feature enhancement model.

[0209] In some embodiments, the resource requirement information is input into an encoder for feature extraction processing to obtain a first resource requirement feature output by the encoder.

[0210] In some embodiments, the network traffic information is input into a multi-layer perceptron for feature extraction processing to obtain a first network traffic feature output by the multi-layer perceptron.

[0211] In some embodiments, the time series information is input into a long short-term memory network for feature extraction to obtain a first time series feature output by the long short-term memory network.

[0212] Step 3: Use the feature enhancement model to perform maximum pooling and average pooling on the first network traffic feature, and perform concatenation, convolution and mapping on the processed first network traffic feature in turn to obtain the first initial information feature.

[0213] Step 4: Perform dot multiplication on the first initial information feature and the first resource demand feature to obtain the first intermediate information feature.

[0214] Step 5: Concatenate and convolve the first intermediate information feature and the first time series feature in sequence to obtain the first fusion feature.

[0215] In the non-first iteration process, the previous fusion feature obtained by the previous sub-fusion model is input into the current sub-fusion model, and the current sub-fusion model is used to extract the previous fusion feature. The previous fusion feature includes the previous enhanced feature, the previous network traffic feature, and the previous time series feature.

[0216] In some embodiments, the previous enhanced feature is input into the encoder for feature extraction processing to obtain the enhanced feature of the current level output by the encoder.

[0217] In some embodiments, the previous network traffic feature is input into a multi-layer perceptron for feature extraction processing to obtain the current level network traffic feature output by the multi-layer perceptron.

[0218] In some embodiments, the previous time series feature is input into the long short-term memory network for feature extraction to obtain the current level time series feature output by the long short-term memory network.

[0219] Step 6: Use the feature enhancement model to perform maximum pooling and average pooling on the network traffic features of this level, and perform concatenation, convolution and mapping on the processed network traffic features of this level in turn to obtain the initial information features of this level.

[0220] Step 7, perform dot multiplication processing on the initial information features of this level and the resource demand features of this level to obtain the intermediate information features of this level.

[0221] Step 8: perform concatenation and convolution processing on the intermediate information features and the time series features of this level in sequence to obtain the fusion features of this level.

[0222] Step 9: determine the fusion features output by the final sub-fusion model as multi-information fusion features.

[0223] Step 10, input the multi-information fusion features into the decoder to obtain the decoding features output by the decoder.

[0224] Step 11: Input the decoded features into the first fully connected neural network, map the decoded features into a parameter space related to resource allocation through the first fully connected neural network, and obtain the resource allocation result of each container.

[0225] Step 12: Input the decoded features into a second fully connected neural network, map the decoded features into a parameter space related to flow control through the second fully connected neural network, and obtain the flow control result of each container.

[0226] Step 13: Determine a resource allocation loss value, a service quality loss value, and a flow control loss value based on the resource allocation result and the flow control result.

[0227] Step 14, adjusting the parameters of the multi-information fusion model based on the resource allocation loss value, the service quality loss value and the flow control loss value to optimize the multi-information fusion model.

[0228] It should be understood that, although the steps in the flowcharts involved in the above embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0229] Based on the same inventive concept, the embodiment of the present application also provides a resource allocation and flow control device based on a multi-container network for implementing the resource allocation and flow control method based on a multi-container network involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more embodiments of the resource allocation and flow control device based on a multi-container network provided below can be referred to the limitations of the resource allocation and flow control method based on a multi-container network above, and will not be repeated here.

[0230] In an exemplary embodiment, Fig.10 As shown, a resource allocation and flow control device based on a multi-container network is provided, including: an information acquisition module 801, an enhancement processing module 802 and a mapping processing module 803, wherein:

[0231] The information acquisition module 801 is used to obtain network status information of the multi-container network;

[0232] The enhancement processing module 802 is used to perform feature extraction and feature enhancement processing on the network status information in sequence using a pre-built multi-information fusion model to obtain a multi-information fusion feature;

[0233] The mapping processing module 803 is used to use the pre-built multi-information decoding model to sequentially perform decoding processing and mapping processing on the multi-information fusion features to obtain the resource allocation result and flow control result of each container.

[0234] In an exemplary embodiment, the above-mentioned enhancement processing module 802 is specifically used to iteratively calculate the network status information using a multi-level sub-fusion model. In the first iteration, the network status information is input into the first-level sub-fusion model, and the first-level sub-fusion model is used to perform feature extraction processing and feature enhancement processing on the network status information in sequence to obtain the first fusion feature; in the non-first iteration, the previous fusion feature obtained by the previous-level sub-fusion model is input into the current-level sub-fusion model, and the current-level sub-fusion model is used to perform feature extraction processing and feature enhancement processing on the previous fusion feature in sequence to obtain the current-level fusion feature; the fusion feature output by the last-level sub-fusion model is determined as the multi-information fusion feature.

[0235] In an exemplary embodiment, the above-mentioned enhanced processing module 802 is specifically used to input resource demand information into the encoder for feature extraction processing to obtain the first resource demand feature output by the encoder; input network traffic information into the multi-layer perceptron for feature extraction processing to obtain the first network traffic feature output by the multi-layer perceptron; input timing information into the long short-term memory network for feature extraction to obtain the first timing feature output by the long short-term memory network.

[0236] In an exemplary embodiment, the above-mentioned enhancement processing module 802 is specifically used to use the feature enhancement model to perform maximum pooling processing and average pooling processing on the first network traffic feature, and sequentially perform splicing processing, convolution processing and mapping processing on the processed first network traffic feature to obtain the first initial information feature; perform dot multiplication processing on the first initial information feature and the first resource demand feature to obtain the first intermediate information feature; perform splicing processing and convolution processing on the first intermediate information feature and the first time series feature in sequence to obtain the first fusion feature.

[0237] In an exemplary embodiment, the above device may further include:

[0238] A loss value determination module 804, configured to determine a resource allocation loss value, a service quality loss value, and a flow control loss value based on the resource allocation result and the flow control result;

[0239] The optimization module 805 is used to adjust the parameters of the multi-information fusion model based on the resource allocation loss value, the service quality loss value and the flow control loss value to optimize the multi-information fusion model.

[0240] Each module in the above-mentioned resource allocation and flow control device based on multi-container network can be implemented in whole or in part by software, hardware and their combination. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory in the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0241] According to some embodiments of the present application, there is also provided a non-transitory computer-readable storage medium including instructions, such as a memory including instructions, and the above instructions can be executed by a processor of an electronic device to complete the above method. For example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.

[0242] According to some embodiments of the present application, a computer program product is also provided, and when the computer program is executed by a processor, the above method can be implemented. The computer program product includes one or more computer instructions. When these computer instructions are loaded and executed on a computer, part or all of the above method can be implemented in whole or in part according to the process or function described in the embodiment of the present application.

[0243] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0244] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0245] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A resource allocation and flow control method based on a multi-container network, characterized in that: The method comprises: Get network status information of multi-container network; Using a pre-built multi-information fusion model, the network state information is sequentially subjected to feature extraction and feature enhancement processing to obtain multi-information fusion features; The multi-information fusion features are decoded and mapped in sequence using a pre-built multi-information decoding model to obtain resource allocation results and flow control results for each container.

2. The method according to claim 1, characterized in that The multi-information fusion model includes a multi-level sub-fusion model. The pre-built multi-information fusion model is used to sequentially perform feature extraction processing and feature enhancement processing on the network state information to obtain multi-information fusion features, including: Iteratively calculating the network state information using the multi-level sub-fusion model, inputting the network state information into the first-level sub-fusion model in the first iteration, and sequentially performing feature extraction processing and feature enhancement processing on the network state information using the first-level sub-fusion model to obtain a first fusion feature; In the non-first iteration process, the previous fusion feature obtained by the previous sub-fusion model is input into the current sub-fusion model, and the current sub-fusion model is used to perform feature extraction and feature enhancement processing on the previous fusion feature in sequence to obtain the current fusion feature; The fusion feature output by the final sub-fusion model is determined as the multi-information fusion feature.

3. The method according to claim 2, characterized in that The network status information includes resource demand information, network traffic information and timing information, the sub-fusion model includes an encoder, a multi-layer perceptron and a long short-term memory network, and the first-level sub-fusion model is used to perform feature extraction processing on the network status information, including: Inputting the resource demand information into the encoder for feature extraction processing to obtain a first resource demand feature output by the encoder; Inputting the network traffic information into the multi-layer perceptron for feature extraction processing to obtain the first network traffic feature output by the multi-layer perceptron; The time series information is input into the long short-term memory network for feature extraction to obtain the first time series feature output by the long short-term memory network.

4. The method according to claim 3, characterized in that The sub-fusion model also includes a feature enhancement model, and the first-level sub-fusion model is used to perform feature enhancement processing on the network state information to obtain a first fusion feature, including: Using the feature enhancement model, the first network traffic feature is subjected to maximum pooling processing and average pooling processing respectively, and the processed first network traffic feature is subjected to concatenation processing, convolution processing and mapping processing in sequence to obtain a first initial information feature; Performing a dot product process on the first initial information feature and the first resource demand feature to obtain a first intermediate information feature; The first intermediate information feature and the first time series feature are sequentially concatenated and convolved to obtain the first fusion feature.

5. The method according to claim 4, characterized in that The previous fusion feature includes the previous enhanced feature, the previous network traffic feature and the previous time series feature, and the using the sub-fusion model of the current level to perform feature extraction processing on the previous fusion feature includes: Inputting the previous enhanced feature into the encoder for feature extraction processing to obtain the current level enhanced feature output by the encoder; Inputting the previous network traffic feature into the multi-layer perceptron for feature extraction processing to obtain the current level network traffic feature output by the multi-layer perceptron; The previous time series feature is input into the long short-term memory network for feature extraction to obtain the current level time series feature output by the long short-term memory network.

6. The method according to claim 5, characterized in that The using the sub-fusion model of the current level to perform feature enhancement processing on the previous fusion feature to obtain the fusion feature of the current level includes: The feature enhancement model is used to perform maximum pooling processing and average pooling processing on the network traffic features of the current level, and the processed network traffic features of the current level are sequentially subjected to splicing processing, convolution processing and mapping processing to obtain initial information features of the current level; Performing a dot multiplication process on the initial information feature of the current level and the resource demand feature of the current level to obtain the intermediate information feature of the current level; The intermediate information features of this level and the time series features of this level are sequentially concatenated and convolved to obtain the fusion features of this level.

7. The method according to claim 1, characterized in that The multi-information decoding model includes a decoder, a first fully connected neural network and a second fully connected neural network. The multi-information fusion features are sequentially decoded and mapped using the pre-built multi-information decoding model to obtain resource allocation results and flow control results for each container, including: Inputting the multi-information fusion features into the decoder to obtain decoding features output by the decoder; Inputting the decoded features into the first fully connected neural network, mapping the decoded features to a parameter space related to resource allocation through the first fully connected neural network, and obtaining a resource allocation result for each of the containers; The decoded features are input into the second fully connected neural network, and the decoded features are mapped into a parameter space related to flow control through the second fully connected neural network to obtain a flow control result for each of the containers.

8. The method according to any one of claims 1 to 7, characterized in that: The method further comprises: Determining a resource allocation loss value, a service quality loss value, and a flow control loss value based on the resource allocation result and the flow control result; The multi-information fusion model is adjusted based on the resource allocation loss value, the service quality loss value and the flow control loss value to optimize the multi-information fusion model.

9. A resource allocation and flow control device based on a multi-container network, characterized in that: The device comprises: An information acquisition module is used to obtain network status information of a multi-container network; An enhancement processing module is used to perform feature extraction processing and feature enhancement processing on the network state information in sequence by using a pre-built multi-information fusion model to obtain a multi-information fusion feature; The mapping processing module is used to use a pre-built multi-information decoding model to sequentially perform decoding processing and mapping processing on the multi-information fusion features to obtain a resource allocation result and a flow control result for each container.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

Images