A Learned Cache Eviction Framework with Minimal Overhead — mentioned by benlivengood

Mon, 08 Jan 2024 16:02:47 GMT

https://arxiv.org/abs/2301.11886 for storage and CDN caches achieves pretty remarkable improvements over standard heuristics by using common heuristics as a filter for ML decisions. I imagine even TLB caches might achieve some benefit from ML models with batched pre-prediction as done in MAT, but L1-3 are almost certainly too high throughout.

Paper and discussion history. A mention is not an endorsement.

A Learned Cache Eviction Framework with Minimal Overhead · HackerPapers discussions

A Learned Cache Eviction Framework with Minimal Overhead — mentioned by benlivengood