Rebuilding a Faster OSD with Future
summary
Crimson rewrites the Ceph OSD IO path with Seastar futures to eliminate locks and pin one OSD thread per core, targeting NVMe-class devices (modern SSDs ~0.5M IOPS). A memory-backed, single-replica prototype runs random-read benchmarks that use about 3x fewer CPU cycles per IO and on one box even exceeds classic OSD IOPS; it uses the kernel socket stack today and plans BlueStore + RBD + SPDK support and CI perf gating for regressions.
AI-written summary · about
chapters (7)
- 0:00Introduction and background
- 7:10Current OSD data flow and bottlenecks
- 11:50Future technologies for OSD
- 13:19Crimson OSD redesign
- 15:51Performance results and goals
- 19:10Future goals and challenges
- 22:48Q&A
related talks
34mCrimson Project Update
01h 27mFile IO - Past, Present and Future
49mAn Engineering Approach to Optimising C++
46mInterview with – Vadim Tsesko
52mWriting fast C code for a modern CPU (and applying it to PostgreSQL)
19mStarMalloc: Verifying a Modern, Hardened Memory Allocator
55mAbusing Your Memory Model for Fun and Profit
47mOptimizing Multithreading Performance