ScaleXFS: Getting scalability of XFS back on the ring

on YouTube ↗14m

" summary

ScaleXFS is a modified XFS that removes journaling bottlenecks by adding Double Committed Item List, Per-core In-memory Logging, and Strided Space Counting. Implemented in Linux and tested on a 112-core machine with Intel Optane SSD, ScaleXFS cuts unlink() latency from 0.59 ms to 0.09 ms (1/6th) and raises throughput to 1.5–2.2× over stock XFS across varmail, dbench and mdtest, showing the techniques eliminate lock contention and scale beyond XFS’s 20–40 core collapse.

AI-written summary · about

" chapters (17)

1. 0:00Intro
2. 0:21Hardware Trend
3. 0:52Journaling in Filesystem
4. 2:18Scalability Failure in XFS
5. 3:31Journaling in Ext4 vs. in XFS Ext4
6. 4:28Scalable Filesystem Design of Ext4
7. 5:14Latency of metadata operation
8. 5:52Lock contention in in-memory logging
9. 6:43Locks in In-memory Logging
10. 7:12Contentions
11. 7:27Contention between in-memory logging and on-disk logging
12. 7:39Contention among the multiple in-memory loggings
13. 8:00Double Log List
14. 8:48Per-core In-memory Logging
15. 11:05Evaluation Throughput
16. 12:05Evaluation Latency of unlink()
17. 12:55Conclusion