RocksDB
RocksDB is an LSM-tree key-value store that serves in this corpus as both a deployable storage engine and a baseline for studying compaction, caching, I/O paths, and device-aware optimization.
是什么
RocksDB turns an LSM-tree design into a configurable engine with write-ahead logging, memtables, sorted files, compaction, block cache, and pluggable storage behavior. Its configuration and workload mix materially affect observed performance, so it is useful as a systems baseline but not a single fixed point of comparison.
关键观察 / 隐含假设
- 观察：compaction and device behavior jointly determine write and tail-latency costs. DOGI-FAST26 and RASK-FAST26 use RocksDB-related workloads to expose this interaction.
- 观察：data and metadata placement can change the bottleneck without changing the application interface. DecouKV-ATC25 and MlsDisk-FAST26 study such storage-path choices.
- 假设：a benchmark configuration represents deployment behavior. HotRAP-ATC25 and UnICom-FAST26 show why workload distribution and I/O mode need explicit boundaries.
演进时间线
- 2025 ATC：DecouKV-ATC25 — explores decoupled KV-storage paths.
- 2025 ATC：HotRAP-ATC25 — examines storage behavior under application workload constraints.
- 2026 FAST：DOGI-FAST26 — connects LSM/KV behavior with storage-device mechanisms.
相关概念
相关论文
- DOGI-FAST26 — device-aware storage evaluation.
- RASK-FAST26 — storage-system mechanism under KV workloads.
- DecouKV-ATC25 — decoupled data and metadata paths.
- MlsDisk-FAST26 — storage placement and engine behavior.