Powering a Petabyte-Scale Cache - Uber's Alluxio Implementation

Learn how Uber engineered a petabyte-scale distributed caching system using Alluxio to power their Presto analytics engine in this 23-minute conference talk. Discover the architectural enhancements Uber developed to scale their Alluxio deployment from 3 to 4 petabytes while maintaining 99.99% cache reliability for processing hundreds of thousands of daily queries across hundreds of petabytes of data. Explore the robust client fallback mechanisms and consistent hashing strategies that ensure efficiency during cluster scaling, along with the adaptive cache filter techniques that achieve over 80% cache hit rates. Gain insights into managing large-scale cache clusters, understand how this implementation delivers substantial egress bandwidth savings from underlying storage systems, and learn best practices for deploying high-performance caching solutions with Presto in high-throughput production environments.