Application-Aware QoE Prediction for Dynamic Network Routing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The challenge of dynamically classifying and optimizing the quality of experience (QoE) for numerous online applications in enterprise networks is complicated by their diversity and dynamic nature, with existing methods relying on static SLA thresholds that fail to accurately predict user experience and often result in SLA failures and poor QoE.
Innovation Solution
A predictive application aware routing engine uses machine learning models trained on application telemetry to dynamically classify application behavior and predict QoE, enabling proactive network configuration changes to optimize user experience across various applications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional SLA thresholds are used to predict user experience, then the method is simple and easy to implement, but the accuracy of user experience prediction deteriorates because SLA thresholds cannot account for application-specific factors like resilient codecs
Solution Approach 1:
The patent segments the prediction problem by classifying applications into different categories (e.g., video streaming, voice, file transfer) and selecting appropriate prediction models for each category. This allows accurate prediction for specific application types while avoiding the complexity of creating a single universal model for all applications.
Solution Approach 2:
The patent changes the parameters used for prediction from generic SLA thresholds to application-specific parameters such as codec resilience, packet loss tolerance, and quality of service requirements. This enables accurate prediction by adapting the measurement criteria to match the specific characteristics of each application type.
2Measurement precision
If individual prediction models are built for each of the thousands of applications, then prediction accuracy improves, but the time and effort required for model building increases significantly
Solution Approach 1:
The patent creates a universal framework that can handle multiple application types through a single classification and selection mechanism. Instead of building separate models for each application, the system uses a common approach to classify applications and select from a set of pre-defined prediction models, achieving multi-functionality across different application scenarios.
Solution Approach 2:
The patent performs preliminary classification of applications into categories before prediction is needed. By pre-classifying applications and preparing corresponding prediction models in advance, the system avoids the time-consuming task of building new models for each application at runtime, instead simply selecting the appropriate pre-prepared model.
3Ease of manufacture
If a single prediction model is built to cover all application types, then the effort required is reduced, but the model becomes unfeasible due to the diversity and number of applications
Solution Approach 1:
The patent segments the diverse application landscape into manageable categories or clusters based on shared characteristics. This segmentation allows the creation of a limited number of targeted prediction models that collectively cover all application types, making the overall system feasible while maintaining comprehensive coverage.
Solution Approach 2:
The patent implements a dynamic model selection mechanism that adapts to the specific application being analyzed. The system dynamically classifies the application and selects the most appropriate prediction model from the available set, enabling the system to handle application diversity flexibly without requiring a single static model for all scenarios.
Data Source
AI summary
In one embodiment, a device determines a type of content being accessed by an endpoint client within an online application during a particular period of time. The device selects a quality of experience prediction model for the online application, based on the type of content being accessed during the particular period of time. The device uses the quality of experience prediction model to predict a quality of experience metric for the online application. The device causes a configuration change in a network based on the quality of experience metric.