Kosmo: Efficient Online Miss Ratio Curve Generation for Eviction Policy Evaluation
Explore a 19-minute conference talk from USENIX FAST '24 that introduces Kosmo, a novel algorithm for generating Miss Ratio Curves (MRCs) for various cache eviction policies. Learn how Kosmo efficiently evaluates cache performance for policies like LFU, FIFO, 2Q, and LRFU, which don't adhere to the inclusion property. Discover how this algorithm outperforms existing methods, particularly Miniature Simulations, by significantly reducing memory overhead and increasing throughput. Gain insights into the importance of MRCs for optimizing in-memory cache configurations and their impact on backend storage server loads. Examine the evaluation results based on 52 publicly-accessible cache access traces, totaling approximately 126 billion accesses, and understand why Kosmo is more suitable for online MRC generation in real-world scenarios.