KVIMR: Key-Value Store Aware Data Management Middleware for Interlaced Magnetic...
summary
KVIMR is a middleware that sits between LSM-tree KV stores (RocksDB and others) and emulated 100 GB IMR HDD to cut costly remote rewrites. It implements Compaction-aware Track Allocation to place SSTables by level and lifetime and a Merged RMW to persist multi-track SSTables with fewer remote-device cycles. Evaluation inserts 75 million KV pairs and shows up to 1.55× throughput on write-heavy workloads, 2.17× at high space usage, and ~5.5% lag versus CMR.
AI-written summary · about
chapters (9)
- 0:00Intro
- 0:23LSM-tree based KV Store
- 1:22Interlaced Magnetic Recording
- 2:25Existing Solutions to Reduce RMWs
- 6:01Compaction-aware Track Allocation
- 8:16Merged Read-Modify-Write
- 10:51Evaluation Setup
- 11:40Evaluation Results (1/3)
- 13:28Conclusion
related talks
21mMatrixKV: Reducing Write Stalls and Write Amplification in LSM-tree Based KV Stores
18mSplitKV: Splitting IO Paths for Different Sized Key-Value Items with Advanced
14mLoLKV: The Logless, Linearizable, RDMA-based Key-Value Storage System
19mBuilding KVStore for ML Workloads
50mImmutable Values in the Big Wide World
15mMitigating Resource Usage Dependency in Sorting-based KV Stores on Hybrid...
12mREMIX: Efficient Range Query for LSM-trees
18mPreview: KV-Stores and Databases