BlueStore, Faster + Smaller = Better

on YouTube ↗33m

" summary

BlueStore gains concrete TCO wins: a write-path transformer now reclaims 10–15% lost space on heavily overwritten RBD images and an alternative scrub-time recompression is proposed; a new RocksDB WAL format moves to one IO WAL reduces latency 10–15% (bug found that triggers device repacking); multi-threaded discard and an allocator refactor enable smarter reuse of discarded ranges; segment-size tuning eliminates spanning blobs that double decode cost; a per-function PMU profiler proves tiny CPU wins, and recovery is now parallelizable 60x for much faster OSD restart.

AI-written summary · about

" chapters (12)

1. 0:00Introduction to BlueStore
2. 1:05Compression improvements
3. 6:55Write-ahead log optimizations
4. 8:22Discard multi-threading support
5. 11:13Turing cache concept
6. 13:48Processing speed optimizations
7. 17:22Metadata and cache efficiency
8. 21:49Memory and allocator challenges
9. 24:25Priority cache and fragmentation
10. 28:08Crash recovery improvements
11. 29:34CPU trace tool demo
12. 32:38Q&A