Ceph and Flash
summary
Ceph proves flash benefits by put journal on flash for orders-of-magnitude lower write latency in hybrid nodes (typical 1 SSD serving 6–12 spindles); production clusters exceed 40 PB and 10^4 nodes. Years of tuning yield ~15× read improvement overall but writes lag, so BlueStore replaces FileStore to remove the file-system double-write, targeting ~2× write speed and better CPU use. BlueStore adds checksums on read, inline compression, and supports RocksDB or ZetaScale metadata engines; early results show substantial small-object CPU optimizations and mixed-media deployment working.
AI-written summary · about
chapters (12)
- 0:00Intro
- 1:45Conceptual Ceph System Model
- 4:05Terminology
- 4:34Inside the OSD
- 9:25Ceph Journal on Flash
- 17:05What's wrong with FileStore?
- 21:22BlueStore a rethink of ObjectStore
- 30:15BlueStore Architecture
- 31:08Two Write Path Options
- 32:07BlueStore vs FileStore
- 35:49KV Store Options
- 38:22BlueStore ZetaScale v RocksDB Performance
related talks
46mBlueStore V2 - an Evolutionary Step Forward
31mAI, ML, and the Ceph Advantage: Scalable Storage for Smarter Workflows
15mImproving Performance of Flash Based Key-Value Stores Using Storage Class Memory...
20mMaximizing the Benefit of Flash Storage for Aerospike
18mThe Solid State Storage Revolution
24mPerformance tuning in BlueStore & RocksDB
29mCeph Object Storage Overview, Capabilities and Future Plans
36mStorage Wars