Take Out the TraChe: Maximizing (Tra)nsactional Ca(che) Hit Rate
summary
Defines transactional hit rate and builds DeToX, a cache layer that groups keys by dependency (levels and interchangeable groups), scores groups by minimum frequency, and uses Redis plus Postgres/TiKV with two‑phase locking. Shows up to 130% higher transaction hit rate, 3.4× better cache efficiency, average latency down from 4.5 ms to 3 ms (≈30%), and 30% higher throughput. Demonstrates that cache contamination from popular but incomplete key sets explains why LRU/LFU fail and that group-aware eviction and prefetching fixes it.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:30Flaws in object hit rate
- 3:15Cash contamination impact
- 6:01Defining transaction length
- 7:29Transactional caching NP-hard
- 9:37Introducing Detox caching
- 11:02Experimental setup
- 12:24Results on hit rate
- 13:49Latency and throughput gains
- 14:10Summary
related talks
39mTransactor Performance Theory & Practice
24mLHD: Improving Cache Hit Rate by Maximizing Hit Density
26mCache Strategies with Best Practices
33mCache is King: Get the Most Bang for Your Buck From Ruby
28mPlanning next moves: improving performance when half your stack is someone else's problem
29mOracle TimesTen: extreme performance for latency-critical applications
16mKVCache Cache in the Wild: Characterizing and Optimizing KVCache Cache at a...
34mCache = Cash! 2.0