Adaptive Online Cache Capacity Optimization via Lightweight Working Set Size Estimation at Scale

Explore a 20-minute conference talk from USENIX ATC '23 that delves into Cuki, an innovative approach to adaptive online cache capacity optimization. Learn about this approximate data structure designed for efficient estimation of working set size (WSS) and item repetition ratio (IRR) in variable-size item access scenarios. Discover how Cuki offers a cache-friendly, thread-safe, and lightweight solution with proven accuracy guarantees. Understand the adaptive online cache capacity tuning mechanism built upon Cuki and its application in estimating cache miss ratio curves (MRC). Examine the implementation of Cuki as a plugin for the distributed file caching system Alluxio and its performance compared to state-of-the-art algorithms. Gain insights into real-world applications, including its deployment on a major social platform for Presto query workloads, and learn how it significantly reduces table querying latency and improves file reading throughput.