StreamCache - Revisiting Page Cache for File Scanning on Fast Storage Devices

Explore a groundbreaking conference talk from USENIX ATC '24 that introduces StreamCache, a novel page cache management system designed to optimize file scanning performance on fast storage devices. Delve into the challenges of using traditional buffered I/O via page cache for file scanning, including limited I/O bandwidth and interference from background writeback operations. Discover the three innovative techniques employed by StreamCache: lightweight stream tracking, stream-based page reclaiming, and two-layer memory management. Learn how these methods work together to significantly enhance I/O performance for scientific applications and large language models. Gain insights into the implementation of StreamCache in XFS and its impressive experimental results, showcasing substantial improvements in I/O bandwidth and reduced checkpoint/restart times compared to existing methods.