No More LRU: Simple Scalable Caching with FIFO Queues

Watch a technical conference talk exploring an innovative approach to cache eviction algorithms that challenges the traditional Least-Recently-Used (LRU) method. Dive deep into the limitations of LRU, including its scalability issues with multi-core CPUs and complex implementation requirements. Learn about S3-FIFO, a novel algorithm using three static FIFO queues that demonstrates superior performance across 6594 cache traces from 14 datasets. Discover how this simpler approach achieves 6× higher throughput compared to optimized LRU at 16 threads, while maintaining lower miss ratios than state-of-the-art algorithms. Master the principles behind efficient cache eviction, understand how to implement and troubleshoot caching systems, and gain insights into designing scalable storage solutions for modern computing environments. Presented by Juncheng Yang from Harvard University's School of Engineering and Applied Science, this 49-minute presentation equips developers and system architects with practical knowledge to improve storage system performance through better caching strategies.