Cloud QoS Estimation Using Conditional Adversarial Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current stochastic models for estimating quality of service in cloud computing often lead to incorrect estimates due to incorrect assumptions, and identifying the best predictors of request allocation metrics is challenging.
Innovation Solution
The use of Conditional Adversarial Neural Networks (cGANs) to estimate response times by learning the underlying metric distributions and generating synthetic values, considering the status of the computing network, allowing for unsupervised direct learning of response time distributions without requiring distribution assumptions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If stochastic models with distribution assumptions are used to estimate QoS metrics, then the estimation process is simplified, but the accuracy of the estimates deteriorates due to incorrect assumptions
Solution Approach 1:
The patent replaces traditional stochastic modeling approaches (mathematical/mechanical system) with machine learning models (data-driven system). Instead of assuming probability distributions and using queuing theory formulas, the invention trains neural networks on historical telemetry data to directly predict QoS metrics, eliminating the need for distribution assumptions while improving accuracy
Solution Approach 2:
The patent changes the fundamental parameters of the estimation approach by transitioning from fixed distributional assumptions to adaptive, data-driven parameter learning. The machine learning models automatically learn the appropriate statistical characteristics from historical data, allowing the system to adapt to changing conditions without requiring manual specification of distribution parameters
2Measurement precision
If machine learning models are trained to predict QoS metrics from request allocation telemetries, then distribution assumptions are eliminated, but the challenge of identifying the best predictor variables increases complexity
Solution Approach 1:
The patent creates a universal framework that handles multiple QoS metrics (response time, throughput, latency) using the same machine learning architecture and process. The system is designed to estimate various metrics simultaneously by training models on comprehensive telemetry data that captures multiple aspects of system performance, eliminating the need for separate specialized models for each metric
Solution Approach 2:
The machine learning models automatically perform feature selection and optimization without requiring manual intervention. The training process self-adjusts to identify the most predictive variables from the available telemetry data, and the models continuously learn from new data to improve their predictions, making the system self-optimizing rather than requiring complex external configuration
Data Source
AI summary
Models to predict quality of service metrics are disclosed. A response time is predicted using an occupancy status of an infrastructure and models that have been trained to predict a response time. Estimating a metric, such as the response time, allows the infrastructure to adjust to issues such that requests better satisfy quality of service requirements.