Kosmo: Efficient Online Miss Ratio Curve Generation for Eviction Policy Evaluation
summary
Kosmo generates online miss-ratio curves for non-inclusion eviction policies by keeping a single shared object copy and per-object eviction map of past evictions, reconstructing cache stacks on demand. Evaluated on 52 traces (≈126 billion accesses), Kosmo uses 3.6× less memory than Miniature Simulations on average (up to 36×) and 1.3× higher throughput, uses shard sampling to cut stored objects by >1000×, and pruning that yields about a 387x reduction in eviction-record size; 5–10 reconstructed stacks give <2% MAE.
AI-written summary · about
chapters (9)
- 0:00Kosmo and cache modeling
- 3:03Eviction policies trade-offs
- 5:29MiniSim MRC limitations
- 7:57Kosmo memory reduction
- 10:49Eviction maps for reconstruction
- 13:41Record pruning for efficiency
- 15:28Performance vs accuracy trade-offs
- 16:55Kosmo vs MiniSim memory
- 17:40Conclusion and future
related talks
22mOSCA: An Online-Model Based Cache Allocation Scheme in Cloud Block Storage Systems
11meMRC: Efficient Miss Ratio Approximation for Multi-Tier Caching
20mAdaptive Online Cache Capacity Optimization via Lightweight Working Set Size...
23mEfficient Miss Ratio Curve Computation for Heterogeneous Content Popularity
16mKVCache Cache in the Wild: Characterizing and Optimizing KVCache Cache at a...
36mCaching Strategies for an Erlang Based Web Stack
13mThe what, The from, and The to: The Migration Games in Deduplicated Systems
01h 00mCount-min Sketch to Infinity