MIDAS: Minimizing Write Amplification in Log-Structured Systems through Adaptive Group...
summary
MiDAS minimizes garbage-collection write amplification by organizing data into a chain of adaptive groups, automatically tuning number of groups and group sizes plus a dynamically sized HOT group using an Update Interval Distribution and a Markov-Chain-based Analytical Model to predict movements. Implementation on a flash SSD prototype and simulations show concrete gains: about 25% lower WAF and 54% higher throughput versus state-of-the-art GC schemes, while using less CPU and memory.
AI-written summary · about
chapters (11)
- 0:00MIDAS introduction
- 2:05Optimal data placement for write amp
- 4:14Limitations of current techniques
- 7:48MIDAS design goals
- 9:55MIDAS architecture overview
- 13:08Finding optimal group config
- 16:22Dynamic workload handling
- 17:05MIDAS evaluation setup
- 18:08Performance results analysis
- 19:50Overhead and throughput comparison
- 20:11Conclusion and future work
related talks
33mSSDFS: Flash-friendly File System with Highly Minimized GC Activity, Diff-on-write, & Deduplication
15mADOC: Automatically Harmonizing Dataflow Between Components in Log-Structured Key-Value
18mAlleviating Garbage Collection Interference Through Spatial Separation in All Flash Arrays
25mFlashield: a Hybrid Key-value Cache that Controls Flash Write Amplification
13mClosing the B+-tree vs. LSM-tree Write Amplification Gap on Modern Storage Hardware...
01h 07mEmerging Embdedded Memories and Storage
13mLiquid-State Drive: A Case for DNA Block Device for Enormous Data
15mMAC: Metadata Acceleration for Sustainable Performance in Big-Data Systems with CXL DRAM