Predictive Scaling and Capacity Planning With Machine Learning at Amazon.com

Learn how Amazon.com handles high-velocity events like Prime Day, Black Friday, and Diwali through predictive scaling and machine learning-based capacity planning in this keynote presentation from KubeCon + CloudNativeCon. Discover the limitations of traditional reactive horizontal scaling approaches that respond to increased traffic, CPU utilization, or other metrics, and understand why reaction times may be insufficient for high-velocity events, leading to customer-facing errors such as high latency and checkout failures. Explore Amazon's proactive scaling methodology that uses machine learning algorithms and historical event data to anticipate demand and scale thousands of services running on AWS infrastructure before traffic spikes occur. Gain insights into the technical strategies for maintaining service availability during peak demand periods and learn how to implement de-scaling mechanisms that return systems to forecasted loads for normal business operations, ensuring optimal resource utilization while maintaining exceptional customer experience.