eMRC: Efficient Miss Ratio Approximation for Multi-Tier Caching
summary
eMRC approximates multi-dimensional miss-ratio curves by partitioning cache space, applying a performance-cliff removal method and a convex-hull approximation so generating a two-tier surface needs 14× fewer samples and four-tier runs over 4,000× faster than brute force; prediction error stays under 2% MAE. The system handles LRU/ARC and stack or non-stack eviction, cuts MRC generation from 1h40m to ~3 minutes in one example, and ORCA uses eMRC to pick configurations that save about 36% cost while meeting ALPS SLOs.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:14MS Visual Curve
- 1:40Limitations
- 2:43Telos
- 3:39SingleTier
- 4:47MultiTier
- 6:06Evaluation
- 7:08Results
- 8:33Caching Framework
- 9:14Example
- 10:18Conclusion
related talks
22mOSCA: An Online-Model Based Cache Allocation Scheme in Cloud Block Storage Systems
23mEfficient Miss Ratio Curve Computation for Heterogeneous Content Popularity
36mCaching Strategies for an Erlang Based Web Stack
19mKosmo: Efficient Online Miss Ratio Curve Generation for Eviction Policy Evaluation
25mMemC3: Compact and Concurrent MemCache with Dumber Caching and Smarter Hashing
24mLHD: Improving Cache Hit Rate by Maximizing Hit Density
16mJohnny Cache: the End of DRAM Cache Conflicts (in Tiered Main Memory Systems)
20mAdaptive Online Cache Capacity Optimization via Lightweight Working Set Size...