Rethinking the Request-to-IO Transformation Process of File Systems for Full...

on YouTube ↗16m

" summary

Experimental analysis shows file systems pay a ~9× penalty when fsync exposes page-cache work and small SSD reads cause up to an 80% latency increase for 1 MB writes; page-cache management and fsync assembly together consume ~40–60% of time. OrchFS introduces a small NVM tier to partition writes into alignment-based write-partition (SSD-aligned blocks + unaligned NVM pages), a unified HR tree mapping, and a parallel IO engine that issues direct-IO SSD writes concurrently, demonstrating up to 29.76× write speedup and 6.79× read gains versus ext4/F2FS/NOVA and hybrid baselines.

AI-written summary · about

" chapters (18)

1. 0:00High-bandwidth SSD challenges
2. 1:18SSD vs file system performance gap
3. 2:20Write inefficiency root causes
4. 3:44Direct IO aligned vs unaligned
5. 5:11Page cache and fsync overhead
6. 6:14IO concurrency impact on SSD
7. 6:57Key issues and solution motivation
8. 8:03ORFS design overview
9. 9:10File write partitioning in ORFS
10. 10:14HR tree mapping for storage
11. 11:40Parallel IO engine and consistency
12. 12:45Experimental setup and baselines
13. 13:27Single-threaded random write results
14. 13:49ORFS read performance and path
15. 14:10Multi-threaded random write results
16. 14:53Real-world app performance
17. 15:36ORFS benefits summary
18. 15:56Q&A and open-source info