Application feature and network modal mapping method oriented to intellectual computing

Abstract

The invention relates to the technical field of intellectual computing networks, in particular to an intellectual computing-oriented application feature and network modal mapping method. Comprising the steps of network flow data preprocessing and feature extraction, network flow classification model pre-training, application flow feature and modal mapping knowledge base construction, demand resource analysis and configuration protocol generation, networking verification and mapping rule optimization and the like. According to the method, storage, calculation and transmission of three-dimensional resource requirements are combined, a mapping model from application traffic features to network modals is constructed through a comprehensive application feature analysis method, accurate distribution of multi-dimensional resources is achieved, and the utilization efficiency of the network resources is remarkably improved. The mapping rule between the application traffic characteristics and the network mode is dynamically updated, the method adapts to the changes of the network environment and the traffic mode, and the flexibility and long-term applicability of the system are improved. The method has an efficient configuration generation mechanism, accelerates the optimization deployment process, and is particularly suitable for complex network environments and dynamic service requirements.

Description

Technical Field

[0001] The present invention relates to the field of intelligent connected computing network technology, and in particular to an application feature and network modality mapping technology for intelligent connected computing. Background Art

[0002] Intelligent computing applications in vertical industries, such as scientific computing, industrial applications, and large AI models, are showing trends of diversification, storage-transfer-computing convergence, and resource-intensiveness. Network carrying capacity is no longer limited to traffic access, distribution, and control. Computing and storage resources must be integrated and scheduled as network elements. Building on pure network modes such as best-effort, deterministic, and high-bandwidth, a new network mode of storage-transfer-computing convergence has emerged. Simple scheduling mechanisms under traditional single modes cannot guarantee the completion rate, latency jitter, and continuous stability requirements of applications. To ensure the efficient operation of intelligent computing applications, in-depth network traffic feature analysis is required to map application features to network modes and allocate network resources. Failure to do so may lead to improper network resource scheduling, resulting in a series of problems such as network congestion, increased latency, and increased packet loss, seriously affecting user experience and service quality.

[0003] In recent years, the application of deep learning technology in network traffic feature analysis has made significant progress. Researchers use deep learning models to automatically extract features from the bytes of raw traffic packets, achieving more effective traffic classification and management. However, this method has limitations. If a packet is too long, its bytes may obscure important information from other packets. As a result, traditional deep learning-based methods can lead to unstable performance and require large amounts of labeled traffic data for training. However, the acquisition of labeled data is expensive and time-consuming, limiting the application of these methods in real-world environments. Furthermore, the models have poor generalization capabilities for new scenarios or unknown traffic types, making them difficult to adapt to rapidly changing network environments.

[0004] Most current traffic classification and feature analysis methods only focus on the characteristics of the network forwarding dimension, while ignoring the demand characteristics of storage and computing resources. Invention CN118282952A only analyzes characteristics such as bandwidth and latency. This one-sided feature extraction method cannot fully characterize the characteristics of different modal traffic in complex scenarios, resulting in irrational resource allocation and limited optimization effects. Existing technologies usually fail to effectively construct an association mapping between application network traffic characteristics and storage-transfer-computation fusion network modalities, and can only optimize for a single dimension. For complex application scenarios that require multi-modal collaboration, existing methods cannot achieve accurate resource allocation, resulting in an inability to break through performance bottlenecks. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to propose an application feature and network modality mapping method for intelligent connected computing, so as to overcome the shortcomings of the existing technology in dealing with complex and diverse network scenarios, such as single feature extraction, reliance on a large amount of labeled data, and lack of multi-dimensional feature modality mapping.

[0006] To solve the above technical problems, the present invention proposes a method for mapping application features and network modalities for intelligent connected computing, comprising the following steps: Step 1: Perform traffic analysis on the original traffic data, extract the hierarchical features of the traffic, and convert the original traffic data into a multi-level flow representation matrix; Step 2: Pre-train the network traffic classification model to learn the essential characteristics of the traffic and reduce dependence on labeled data; Step 3: Build a dedicated knowledge base for traffic feature and modality mapping, and map network traffic features to network modalities through classification and labeling. Step 4: Based on the traffic characteristics and the guidance of the modal mapping knowledge base, perform demand resource analysis and generate an optimized configuration protocol corresponding to the network modal requirements based on the current traffic characteristics.

[0007] Further optimization is that after step 4, step 5 is also included: network verification is performed based on the user's task characteristics and the network modality requirements of traffic analysis, and resource allocation strategies and configuration protocols are adjusted based on the analysis results of the verification problems.

[0008] Further improved, step 3 also includes: adjusting the mapping rules between application network traffic characteristics and network modes in real time by monitoring current network traffic and analyzing feature changes; adding new traffic categories to the knowledge base through online learning or manual labeling.

[0009] Preferably, step 1 includes the following sub-steps: Step 1-1: Divide the original network bit stream into flows according to source IP address, destination IP address, source port, destination port, and protocol type, ensuring that each flow has a clear five-tuple identifier; Step 1-2: Select M consecutive data packets from each flow, where M is a natural number from 5 to 10, and extract the header information and payload information of each data packet. Steps 1-3: Combine the processed packet-level information in sequence to generate a two-dimensional multi-level flow representation matrix. The rows of the matrix represent the data levels, and the columns represent the contents of different packets, forming a unified flow representation structure.

[0010] Preferably, step 2 includes the following sub-steps: Step 2-1: Collect large-scale raw traffic sets from multiple real network scenarios and apply the preprocessing method in step 1 to convert the raw traffic data into a fixed-size multi-level flow representation matrix as model input; Step 2-2: Design a model architecture based on a masked autoencoder and train it using the multi-level flow representation matrix as input; Step 2-3: Use a lightweight decoder to reconstruct the masked multi-level flow representation matrix region through the potential representation generated by the encoder and the information of the mask position; Step 2-4: Use the optimizer to continuously optimize and adjust the parameters of the encoder and decoder by minimizing the reconstruction loss; Step 2-5: Save the optimized encoder parameters as the initialization model for subsequent classification tasks.

[0011] Preferably, step 3 includes the following sub-steps: Step 3-1: Extract the features of the multi-level flow representation matrix from the pre-trained model and classify the traffic data by category. Label the traffic features of each category according to specific task requirements to form an initial traffic feature classification database. Step 3-2: Based on the classified and labeled traffic characteristics, analyze the three-dimensional storage, transfer, and computing requirements of each type of traffic; Step 3-3: Through rule matching or statistical learning-based methods, a mapping relationship between traffic characteristics and network modes is constructed and stored in the knowledge base to form a preliminary association model.

[0012] Preferably, step 4 includes the following sub-steps: Step 4-1: Obtain the characteristic data of the current network traffic and normalize the data so that the format of the characteristic data of the current network traffic is consistent with the characteristic data format in the knowledge base; Step 4-2: Build a deep learning-based classifier to match current traffic features with traffic feature classifications in the knowledge base. If the features match the knowledge base, extract the corresponding network modality requirements. If the features do not match the knowledge base, enter a new learning process to analyze the storage, conversion, and computation requirements of the unmatched traffic, determine the modality type, and generate the corresponding resource requirement template. Step 4-3: Based on performance requirements and current resource availability, develop an optimized resource allocation strategy and generate a configuration agreement based on the strategy.

[0013] Preferably, step 5 includes the following sub-steps: Step 5-1: Build a network environment based on the user's task characteristics and the network mode requirements of traffic analysis, initialize the environment configuration, and load the generated network configuration parameters; Step 5-2: Inject user service-related traffic and use a monitoring agent to collect network performance data and resource usage during actual operation. Step 5-3: Compare the actual collected performance data with the optimization target to verify whether the user requirements are met; Step 5-4: Analyze the problems found during the verification process and adjust the resource allocation strategy and configuration protocol; record the mapping relationship between the effective strategies, new features and modal requirements found during the verification process and update them to the knowledge base.

[0014] More preferably, the step 5 further includes step 5-5: re-verifying the performance of the adjusted configuration in the actual environment to ensure that the optimization strategy continues to be effective.

[0015] The application feature and network modality mapping mechanism for intelligent connected computing proposed in the present invention breaks through the bottleneck of the single and solidified traditional network business model and network structure. The network traffic features are hierarchically modeled through a multi-level flow representation matrix, combined with the pre-training framework of the masked autoencoder, to achieve efficient and accurate traffic feature extraction and significantly reduce the dependence on labeled data. In addition, the present invention establishes an association mapping model between application traffic features and network modalities, and optimizes the network resource allocation strategy through real-time monitoring and dynamic update mechanisms to ensure that network performance can respond to dynamic needs in real time. Based on the optimization results, the present invention introduces an automatic generation process for standardized configuration files, providing a more comprehensive and efficient solution for modern network environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0017] Figure 1 This is a flowchart of the overall process of the application characteristics and network modality mapping method for intelligent connected computing of the present invention.

[0018] Figure 2 This is a structural diagram of the network traffic classification mechanism based on the large model.

[0019] Figure 3 Diagram of the feature network modal mapping mechanism for intelligent connected computing applications. DETAILED DESCRIPTION

[0020] The present invention provides an application feature network and modality mapping mechanism for intelligent connected computing, which aims to analyze network traffic characteristics and map them into network modalities, analyze and optimize network performance in an automated manner, and improve network resource utilization.

[0021] like Figure 1 As shown, the application feature and network modality mapping solution for intelligent connected computing provided by the present invention generally includes the following steps: Step 1: Network traffic data preprocessing and feature extraction. That is, the raw traffic data is converted into a multi-level flow representation matrix, and traffic analysis is performed to extract the hierarchical features of the traffic.

[0022] Step 2: Pre-training the network traffic classification model. That is, pre-train the network traffic classification model to learn the essential characteristics of the traffic and reduce the dependence on labeled data.

[0023] Step 3: Apply traffic characteristics and modality mapping knowledge base construction. That is, build a dedicated knowledge base for feature and modality mapping, map network traffic characteristics to network modalities through classification and labeling, and support dynamic updates to maintain real-time performance.

[0024] Step 4: Required resource analysis and configuration protocol generation. That is, based on the guidance of the feature and modality mapping knowledge base, the required resource analysis is performed, and the optimized configuration protocol corresponding to the network modality requirements is generated based on the current traffic characteristics.

[0025] Step 5: Network Verification and Mapping Rule Optimization. Specifically, network verification is performed based on the user's task characteristics and the network modality requirements from traffic analysis. The optimization strategy is continuously adjusted based on monitoring feedback to achieve the optimal steady-state mapping from task characteristics to network modality.

[0026] Combine Figure 2 As shown, step 1 specifically includes the following sub-steps: Step 1-1: Divide the original network bitstream into flows based on source IP address, destination IP address, source port, destination port, and protocol type, ensuring that each flow has a clear five-tuple identifier. Based on this division, remove the Ethernet header information from the flow, randomize the IP address, and reset the port number to zero to ensure data normalization.

[0027] Step 1-2: Group M consecutive packets (M is a natural number between 5 and 10) from each stream and extract the header and payload information for each packet. The header information includes the IP layer (20 bytes), the TCP / UDP layer (20 / 8 bytes), and the optional extended header information (80 bytes in total). Any header information less than 80 bytes is padded with zeros to a uniform length. The payload information consists of the payload portion of each packet (240 bytes in total). Any payload information exceeding 240 bytes is clipped, and any remaining payload information is padded with zeros to ensure a fixed matrix size (40 rows x 40 bytes).

[0028] Steps 1-3: Combine the processed packet-level information in sequence to generate a two-dimensional multi-level flow representation matrix. The rows of the matrix represent the data level (such as header bytes or payload bytes), and the columns represent the content of different packets, forming a unified flow representation structure.

[0029] Combine Figure 2 As shown, step 2 specifically includes the following sub-steps: Step 2-1: Collect large-scale raw traffic sets from multiple real network scenarios and use the preprocessing method in Step 1 to convert the raw traffic data into a fixed-size multi-level flow representation matrix (size H×W=40×40, where H is height and W is width) as model input.

[0030] Step 2-2: Design a model architecture based on masked autoencoder, taking the multi-level flow representation matrix as input. During training, randomly mask 90% of the patches in the multi-level flow representation matrix. The patch size is 2×2, so the matrix contains N=H×W / 4=400 patches. Only the remaining 10% visible patches are input to the encoder, represented as The encoder generates high-quality latent representation z = Encoder( ).

[0031] Step 2-3: Use a lightweight decoder to reconstruct the masked multi-level flow representation matrix area through the potential representation z generated by the encoder and the information Mask Token of the mask position = Decoder(z, Mask Token). The decoding process uses the mean square error (MSE) as the reconstruction loss, and the goal is to make the reconstructed pixel-level MFR matrix as close to the original matrix as possible: L rec = MSE( , x) = Ensure that the model can effectively capture the multi-scale traffic features at the byte level, packet level, and flow level in the multi-level flow representation matrix. Combined with the original traffic categories, it serves as the input parameter for training the large model.

[0032] Steps 2-4: Use an optimizer to continuously adjust the encoder and decoder parameters by minimizing the reconstruction loss. A random masking strategy is employed during training to further improve the model's generalization to diverse data distributions. Finally, the optimized encoder parameters are saved and used as the initialization model for subsequent classification tasks.

[0033] Combine Figure 3 As shown, step 3 includes the following sub-steps: Step 3-1: Obtain a series of raw traffic data to construct a knowledge base for mapping features and network modes. Group the raw traffic data and extract features to obtain multi-scale traffic features, which are then input into the pre-trained model generated in step 2. The model will classify the traffic data by category (such as video streaming, file transfer, online games, etc.). According to specific task requirements, the traffic features of each category are labeled to form an initial traffic feature classification database. Construct a traffic feature vector based on the multi-scale traffic features and classification results, expressed as F i = [f i1 , f i2 ,... f im ], f ij Represents the value of the i-th type of traffic on the j-th feature dimension.

[0034] Step 3-2: Use the monitoring system to monitor the application's storage and computing usage, analyze its storage and computing characteristics, and construct storage and computing feature vectors. Based on the storage, computing, and traffic characteristics, analyze the three-dimensional storage, conversion, and computing requirements of each type of traffic, expressed as Demand (M, N, C), where M represents storage, N represents network, and C represents computing. Storage is broken down into memory utilization, I / O rate, etc., network is broken down into bandwidth, throughput, latency, etc., and computing is broken down into CPU utilization, GPU utilization, and power consumption. Specifically: M = [Memory_Usage, IO_Rate: I / O] N = [Bandwidth, Throughput, Latency] C = [CPU_Usage, GPU_Usage, Power_Consumption] Through rule matching or statistical learning-based methods, the mapping relationship between traffic characteristics and network modes is constructed and stored in the knowledge base K to form a preliminary association model. The knowledge base is represented as a set K = {E1, E2, ... E n}, E i ={Modal_Type: modal type, Traffic_Feature_Type: traffic feature type (i.e. feature vector F extracted by pre-training model) i , each dimension is standardized), Demand (M, N, C)} Step 3-3: To adapt to changes in the network environment and traffic patterns, the knowledge base must support dynamic updates. By monitoring current network traffic and analyzing changes in characteristics, the mapping rules between application network traffic characteristics and network modalities can be adjusted in real time. Furthermore, new traffic categories can be added to the knowledge base through online learning or manual annotation to ensure continuous adaptability.

[0035] Combine Figure 3 As shown, step 4 includes the following sub-steps: Step 4-1: Obtain the current raw traffic data and normalize the data to ensure that its format is consistent with the feature data in the knowledge base for matching and analysis.

[0036] Step 4-2: Match the current features against the knowledge base and extract the corresponding network modality requirements based on the matching results. If the features are not matched in the knowledge base, the new learning process begins, analyzing the storage, conversion, and computation requirements of the unmatched traffic, determining the modality type, and generating the corresponding resource requirement template.

[0037] Step 4-3: Based on performance requirements and comprehensive consideration of current resource availability, formulate an optimized resource allocation strategy and generate a configuration agreement based on the strategy.

[0038] Step 5 includes the following sub-steps: Step 5-1: Build a network environment based on the user's task characteristics and the network modality requirements of traffic analysis, initialize the environment configuration, and load the generated network configuration parameters.

[0039] Step 5-2: Inject user service-related traffic. During actual operation, use the monitoring agent to collect network performance data and resource usage.

[0040] Step 5-3: Compare the actual collected performance data with the optimization target to verify whether user requirements are met.

[0041] Step 5-4: Analyze the problems found during the verification process and adjust the resource allocation strategy and configuration protocol; record the mapping relationship between the effective strategies, new features and modality requirements found during the verification process and update it to the knowledge base; Step 5-5: Verify the performance of the adjusted configuration in the actual environment again to ensure that the optimization strategy continues to be effective.

[0042] Compared to other existing mainstream technologies, this invention uses a multi-level flow representation matrix to hierarchically model traffic characteristics, fully capturing byte-level, packet-level, and flow-level feature information. Combined with a pre-trained masked autoencoder framework, this method can learn the essential characteristics of traffic without requiring extensive labeled data, thereby improving the efficiency and accuracy of feature extraction. Furthermore, this invention not only analyzes traffic characteristics at the network level but also integrates the three-dimensional resource requirements of storage, computing, and transmission, providing a comprehensive method for analyzing application characteristics. By mapping application traffic characteristics to network modalities, it achieves precise allocation of multi-dimensional resources, significantly improving network resource utilization efficiency. A dedicated knowledge base is designed to dynamically update the mapping rules between application traffic characteristics and network modalities to adapt to changes in network environments and traffic patterns. Real-time monitoring and online learning mechanisms ensure that the knowledge base always supports the latest optimization requirements, improving the system's flexibility and long-term applicability. This invention uses an automated process to complete network requirements analysis and generate standardized configuration files. This efficient configuration generation mechanism significantly reduces the need for manual intervention and accelerates the optimization deployment process, making it particularly suitable for complex network environments and dynamic business needs.

[0043] Finally, it should be noted that the above specific implementation methods are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for mapping application features and network modalities for intelligent connected computing, characterized in that: The following steps are involved: Step 1: Perform traffic analysis on the original traffic data, extract the hierarchical features of the traffic, and convert the original traffic data into a multi-level flow representation matrix; Step 2: Pre-train the network traffic classification model to learn the essential characteristics of the traffic and reduce dependence on labeled data; Step 3: Build a dedicated knowledge base for traffic feature and modality mapping, and map network traffic features to network modalities through classification and labeling. Step 4: Based on the traffic characteristics and the guidance of the modal mapping knowledge base, perform demand resource analysis and generate an optimized configuration protocol corresponding to the network modal requirements based on the current traffic characteristics.

2. The method for mapping application features and network modalities for intelligent connected computing according to claim 1, characterized in that: After step 4, the method further includes: Step 5: Perform network verification based on the user's task characteristics and the network mode requirements of traffic analysis. Adjust the resource allocation strategy and configuration protocol based on the analysis results of the verification problem.

3. The method for mapping application features and network modalities for intelligent connected computing according to claim 1, characterized in that: The step 3 further comprises: By monitoring current network traffic and analyzing feature changes, the mapping rules between application network traffic features and network modes are adjusted in real time; New traffic categories are added to the knowledge base through online learning or manual labeling.

4. The method for mapping application features and network modalities for intelligent connected computing according to claim 1, characterized in that: The step 1 includes the following sub-steps: Step 1-1: Divide the original network bit stream into flows according to source IP address, destination IP address, source port, destination port, and protocol type, ensuring that each flow has a clear five-tuple identifier; Step 1-2: Select M consecutive data packets from each flow, where M is a natural number from 5 to 10, and extract the header information and payload information of each data packet. Steps 1-3: Combine the processed packet-level information in sequence to generate a two-dimensional multi-level flow representation matrix. The rows of the matrix represent the data levels, and the columns represent the contents of different packets, forming a unified flow representation structure.

5. The method for mapping application features and network modalities for intelligent connected computing according to claim 4, characterized in that: The step 2 includes the following sub-steps: Step 2-1: Collect large-scale raw traffic sets from multiple real network scenarios and apply the preprocessing method in step 1 to convert the raw traffic data into a fixed-size multi-level flow representation matrix as model input; Step 2-2: Design a model architecture based on a masked autoencoder and train it using the multi-level flow representation matrix as input; Step 2-3: Use a lightweight decoder to reconstruct the masked multi-level flow representation matrix region through the potential representation generated by the encoder and the information of the mask position; Step 2-4: Use the optimizer to continuously optimize and adjust the parameters of the encoder and decoder by minimizing the reconstruction loss; Step 2-5: Save the optimized encoder parameters as the initialization model for subsequent classification tasks.

6. The method for mapping application features and network modalities for intelligent connected computing according to claim 1, characterized in that: Step 3 includes the following sub-steps: Step 3-1: Extract the features of the multi-level flow representation matrix from the pre-trained model and classify the traffic data by category. Label the traffic features of each category according to specific task requirements to form an initial traffic feature classification database. Step 3-2: Based on the classified and labeled traffic characteristics, analyze the three-dimensional storage, transfer, and computing requirements of each type of traffic; Step 3-3: Through rule matching or statistical learning-based methods, a mapping relationship between traffic characteristics and network modes is constructed and stored in the knowledge base to form a preliminary association model.

7. The method for mapping application features and network modalities for intelligent connected computing according to claim 1, characterized in that: The step 4 includes the following sub-steps: Step 4-1: Obtain the characteristic data of the current network traffic and normalize the data so that the format of the characteristic data of the current network traffic is consistent with the characteristic data format in the knowledge base; Step 4-2: Match the current features in the knowledge base and extract the corresponding network modality requirements based on the matching results. If the features are not matched in the knowledge base, enter the new learning process to analyze the storage, conversion, and computing requirements of the unmatched traffic, determine the modality type, and generate the corresponding resource requirement template. Step 4-3: Based on performance requirements and current resource availability, develop an optimized resource allocation strategy and generate a configuration agreement based on the strategy.

8. The method for mapping application features and network modalities for intelligent connected computing according to claim 2, characterized in that: The step 5 includes the following sub-steps: Step 5-1: Build a network environment based on the user's task characteristics and the network mode requirements of traffic analysis, initialize the environment configuration, and load the generated network configuration parameters; Step 5-2: Inject user service-related traffic and use a monitoring agent to collect network performance data and resource usage during actual operation. Step 5-3: Compare the actual collected performance data with the optimization target to verify whether the user requirements are met; Step 5-4: Analyze the problems found during the verification process and adjust the resource allocation strategy and configuration protocol; record the mapping relationship between the effective strategies, new features and modal requirements found during the verification process and update them to the knowledge base.

9. The method for mapping application features and network modalities for intelligent connected computing according to claim 8, characterized in that: The step 5 also includes Step 5-5: Verify the performance of the adjusted configuration in the actual environment again to ensure that the optimization strategy continues to be effective.

Images