Real-Time Network Traffic Classification via Spectral Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current network traffic classification methods rely on packet headers and protocols, which are unreliable for real-time quality of service (QoS) guarantees, especially for hybrid traffic flows carrying multiple types of traffic over the same connection, such as VoIP and file transfers, as they do not accurately differentiate between audio, video, and file traffic.
Innovation Solution
The method involves representing network traffic as a stochastic process model, extracting power spectral density (PSD) feature vectors, and using similarity metrics to classify traffic flows by decomposing subspaces and identifying bases that minimize coding length, allowing for real-time classification of audio, video, and file traffic within hybrid flows.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If DiffServ method is used to guarantee QoS by configuring TOS field in packet headers, then QoS guarantee for specific traffic types is improved, but reliability deteriorates when multiple protocols set DSCP to highest number
Solution Approach 1:
The patent replaces the protocol-based mechanical classification system (DiffServ TOS/DSCP fields) with a signal processing-based system. By modeling packet arrival processes as stochastic processes and applying spectral analysis, the system transforms the classification problem from protocol interpretation to mathematical signal analysis, which is more reliable for hybrid traffic flows.
Solution Approach 2:
The patent changes the classification parameters from static header fields (DSCP values) to dynamic temporal characteristics (power spectral density features). By analyzing the frequency domain characteristics of packet arrival processes, the system can distinguish traffic types based on their unique temporal patterns rather than relying on potentially misconfigured header fields.
2Ease of operation
If protocol-based classification is used to identify traffic types, then ease of operation is improved, but measurement precision deteriorates for hybrid traffic flows
Solution Approach 1:
The patent replaces protocol-based classification with a universal stochastic process model that treats all packet streams as signals to be analyzed. This substitution eliminates the need for protocol-specific parsing rules and enables consistent classification across different traffic types including hybrid flows, by focusing on universal temporal characteristics rather than protocol-specific fields.
3Productivity
If real-time classification is implemented using spectral analysis, then productivity is improved, but device complexity increases
Solution Approach 1:
The patent extracts only the essential spectral features (power spectral density) from the packet arrival process, rather than performing complete protocol analysis or maintaining complex classification rules. By focusing on extracting and analyzing only the relevant temporal characteristics, the system achieves real-time performance while managing complexity through selective feature extraction.
Solution Approach 2:
The patent transforms the classification problem into the frequency domain by computing power spectral density, which converts complex temporal patterns into simpler spectral signatures. This parameter transformation simplifies the comparison and classification process, enabling faster real-time decision-making despite the initial computational effort of spectral analysis.
Data Source
AI summary
A method is provided to classify network traffic flows in real-time using spectral analysis techniques to extract regularities inside the network traffic flows. In one embodiment of the invention, subspace decomposition on power spectral density feature vectors and minimum coding length criterion are utilized for training traffic flows of different classifications. Experimental results are shown to demonstrate the effectiveness and robustness of the invention.