QoE Inference Model Selection via Bottleneck Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Previous attempts to infer Quality of Experience (QoE) in computer networking have been unsuccessful due to the 'single-model-fits-all' approach, which fails to account for different vantage points in the network, leading to discordant views of network impairments and inaccurate QoE inference.
Innovation Solution
The implementation of multiple QoE inference models trained on different vantage points in the network, allowing the network traffic distribution node to select the optimal model based on observed conditions to accurately infer QoE by identifying the location of bottlenecks and impairments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single QoE inference model is used for all network locations, then the device complexity is reduced, but the measurement precision of QoE inference deteriorates due to discordant views of network impairments from different vantage points
Solution Approach 1:
The patent segments the network monitoring function by deploying multiple QoE inference models at different network locations (vantage points). Each model is trained on local network metrics from its specific position, allowing the system to capture discordant views of network impairments from multiple perspectives rather than using a single centralized model
Solution Approach 2:
The patent applies local quality by training each QoE inference model on locally-collected network metrics from its specific vantage point. Each model develops specialized knowledge of network conditions at its location, enabling more accurate local QoE measurements that reflect the actual network experience at that position
2Measurement precision
If multiple QoE inference models are deployed at different vantage points, then the measurement precision of QoE inference is improved, but the device complexity increases due to multiple models and selection logic
Solution Approach 1:
The patent implements dynamics by making the QoE inference model selection adaptive rather than static. The system dynamically selects which QoE model to use based on real-time bottleneck detection and network conditions, allowing the most appropriate model for the current situation to be applied rather than using a fixed model deployment strategy
Solution Approach 2:
The patent introduces an intermediary bottleneck detection mechanism that mediates between multiple QoE inference models and the final QoE measurement. This intermediary component analyzes network metrics to identify bottlenecks and determines which QoE model's perspective is most relevant, simplifying the complexity by providing a systematic selection process
3Ease of operation
If QoE inference is performed without identifying bottleneck locations, then the ease of operation is improved, but the measurement precision of QoE inference deteriorates due to inability to account for network impairments
Solution Approach 1:
The patent applies preliminary action by performing bottleneck detection and analysis before selecting and executing the QoE inference process. The system proactively identifies network bottlenecks and uses this information to guide the subsequent QoE measurement, ensuring that the most relevant QoE model is selected based on the current network state rather than using a default approach
Data Source
AI summary
In one example, a location of a potential bottleneck of network traffic in a network is identified. Based on the location of the potential bottleneck, a first QoE inference model is selected from a plurality of respective QoE inference models. The respective QoE inference models are each trained to infer a respective QoE of the network traffic based on one or more respective network traffic metrics generated by monitoring the network traffic at a respective location in the network. One or more first network traffic metrics of the one or more respective network traffic metrics are generated by monitoring the network traffic at a first respective location. The one or more first network traffic metrics are provided to the first QoE inference model to infer a first respective QoE.