Separating Data via Block Invalidation Time Inference for Write Amplification Reduction...
summary
Claims an algorithm, SepBIT, that infers each block’s block-invalidation time (BIT) from workload traces and groups blocks with similar estimated BITs to cut garbage-collection rewrites. Uses mathematical analysis, probability conditional tests on 186 Alibaba Cloud volumes, and a C++ prototype on emulated zone devices comparing against 11 prior schemes plus an oracle. Results: reduces WA by 9.1–20.2%, raises I/O throughput about 20% at median and 25th percentile, and is deployed at Alibaba Cloud ESSDs.
AI-written summary · about
chapters (14)
- 0:00Intro
- 0:28Log-Structured Storage
- 1:20Garbage Collection
- 3:43Contribution
- 4:18Ideal Data Placement and Limitations
- 5:22Observations
- 7:21SepBIT Design
- 8:28User-Written Block Separation
- 9:44GC-rewritten Block Separation
- 10:56SepBIT Implementation
- 12:23Evaluation
- 13:10Trace Analysis on WA
- 13:43Prototype Throughput
- 14:02Conclusion
related talks
14mBurstable Cloud Block Storage with Data Processing Units
17mShieldReduce: Fine-Grained Shielded Data Reduction
21mMIDAS: Minimizing Write Amplification in Log-Structured Systems through Adaptive Group...
45mBig Data and SSDs
18mAlleviating Garbage Collection Interference Through Spatial Separation in All Flash Arrays
12mRevolutionizing Data Management
01h 07mEmerging Embdedded Memories and Storage
27mMempool Analysis and Simulation