Predictive Application Classification for Network Traffic Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
As the number of network applications increases, it becomes increasingly difficult for network traffic management systems to efficiently determine the application associated with a network flow, which is essential for applying appropriate management policies.
Innovation Solution
A packet traffic management device (PTMD) employs a prediction engine to generate prediction information based on subscriber behavior and historical data, which is then used by a classification engine to determine the applications associated with a network flow, thereby reducing the time and complexity of application identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional application identification methods are used, then all applications can be detected, but the time and complexity of identification increases significantly
Solution Approach 1:
The system performs preliminary classification by analyzing subscriber behavior patterns and historical data before detailed application identification. The prediction engine pre-processes subscriber information and generates predicted application lists, so when actual network traffic arrives, the classification engine can quickly match against pre-computed predictions rather than analyzing from scratch, significantly reducing identification time while maintaining accuracy
Solution Approach 2:
The application identification process is segmented into multiple stages: (1) subscriber behavior analysis and prediction generation, (2) predicted application list creation, (3) network traffic classification against predictions, and (4) fallback to detailed analysis if needed. This segmentation allows the system to handle most cases quickly through predictions while reserving comprehensive analysis for ambiguous cases, balancing speed and accuracy
2Measurement precision
If comprehensive application analysis is performed, then identification accuracy improves, but system complexity increases
Solution Approach 1:
The prediction engine acts as an intermediary between subscriber data and the classification engine. It transforms raw subscriber behavior data into structured predicted application lists that the classification engine can efficiently process. This intermediary layer simplifies the overall system architecture by pre-processing data and reducing the complexity of the main classification task
Solution Approach 2:
Subscriber behavior patterns and application predictions are computed in advance and stored for quick retrieval. This preliminary computation reduces the real-time processing complexity, as the classification engine only needs to match incoming traffic against pre-computed predictions rather than performing complex analysis on every traffic flow
3Productivity
If prediction information is used to assist classification, then identification speed increases, but accuracy may decrease without proper feedback
Solution Approach 1:
The system implements a feedback mechanism where classification results are fed back to update and refine the prediction engine. When the classification engine successfully identifies an application using predicted information, this confirmation feedback is used to reinforce and improve future predictions for similar subscriber behaviors. This closed-loop feedback ensures that prediction speed improvements do not compromise accuracy, as the system continuously learns from actual outcomes
Solution Approach 2:
The prediction accuracy and confidence levels are dynamic rather than static. The system adjusts the weight and reliability of predicted information based on historical accuracy, subscriber-specific patterns, and context. When predictions are highly confident, the system relies more on them for speed; when confidence is lower, it performs more comprehensive analysis, dynamically balancing speed and accuracy based on current conditions
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Embodiments are directed towards managing communication over a network assist application classification using predicted subscriber behavior. Subscriber information associated with a network flow may be determined. Prediction information that includes application identifiers may be generated based on the subscriber information. A classification engine may determine the applications associated with the network flow based on the prediction information and the network flow using a plurality of application classifiers that correspond to the applications. If an application identifier included in the prediction information is correspondent to an application classifier, the application may be determined based on the correspondent application classifier; otherwise the application is determined based on the remainder of the application classifiers. A policy for managing the network flow may be determined based on the determined application. Then the network traffic associated with the network flow may be communicated based on the at least one determined policy.