Understanding the cloud through data analysis with ripe atlas

Abstract

Cloud service performance remains challenging to characterize due to the distributed and dynamic nature of modern cloud infrastructure. This presentation demonstrates how RIPE Atlas measurement data can be systematically analyzed to develop comprehensive models of cloud latency behavior. We leverage traceroute data combined with extensive metadata from the RIPE Atlas platform to extract actionable insights about cloud performance patterns across major service providers including AWS, Azure, and Google Cloud.

Our methodology encompasses multiple analytical approaches: characterization and modeling of latency patterns, predictive modeling through time series forecasting, and geospatial regression techniques to capture location-dependent performance variations. We employ post-hoc explainable machine learning methods to identify the most influential features and variables affecting latency performance. Through real-world measurement campaigns targeting the three major cloud providers, we demonstrate how data-driven analysis can reveal underlying performance characteristics and provide predictive capabilities for cloud service latency, offering practical tools for network optimization and service quality assessment.