Accelerating Filesystem Checking and Repair with pFSCK

on YouTube ↗29m

" summary

pFSCK parallelizes fsck at inode-granularity parallelism, giving each worker a per-thread context (buffers, cache, problem state) to avoid global locking, then hides remaining contention by running pipeline parallelism across passes and delaying dependent checks as intermediate tasks. a dynamic thread scheduler samples queue lengths and migrates threads to match file-system metadata skew. evaluation on NVMe shows >3.72× speedups over vanilla fsck and ~1.71× over XFS repair while keeping correctness and without changing on-disk layout.

AI-written summary · about

" chapters (25)

1. 0:00Intro
2. 0:51File System Checking (cont.)
3. 1:19Storage Evolution and Modern Consistency Mechanisms
4. 3:02Challenges on Parallelizing File System Checking
5. 3:30Traditional EXT Disk Layout
6. 3:53File System Checker and Repair
7. 4:26File System Checker Passes
8. 6:24Evaluating Current FSCK Performance
9. 6:58File Count Sensitivity
10. 7:40Directory Count Sensitivity
11. 8:23Take Aways and Research Questions
12. 9:32Serial Execution in FSCK
13. 10:29Challenge: Buffers and Intermediate Data Structures
14. 11:54PFSCK Pipeline Parallelism
15. 13:10Pipeline Parallelism Dependent Check Problem
16. 13:52Delaying Dependent Checks
17. 14:32Pipeline Parallelism Work Imbalance Problem
18. 15:22Solution Dynamic Thread Scheduling
19. 17:13PFSCK with file system errors
20. 17:47Evaluation Goals
21. 18:22Evaluation Data Parallelism
22. 20:56Evaluation: Pipeline Parallelism
23. 22:06Evaluation: Dynamic Thread Scheduler
24. 23:19Summary
25. 23:45Conclusion

" related talks

• 15mParaSync: Exploiting Fine-Grained Parallelism for Efficient File SynchronizationHuiba Li, Yue Yu, Zhihao Zhang, Jiwu Shu, Yiming Zhang, Guangtao Xue, Lu Tang · FAST '26
• 15mInfiniFS: An Efficient Metadata Service for Large-Scale Distributed FilesystemsJiwu Shu, Yiming Zhang, Youyou Lu, Wenhao Lv, Peile Duan · FAST '22
• 17mParaRC: Embracing Sub-Packetization for Repair Parallelization in MSR-Coded StorageJie Li, Patrick P. C. Lee, Dan Feng, Yuchong Hu, Xiaolu Li, Keyun Cheng, Kaicheng Tang, Ting-Yi Wu · FAST '23: 21st USENIX Conference on File and Storage Technologies
• 16mBoosting Full-Node Repair in Erasure-Coded StoragePatrick P. C. Lee, Jiwu Shu, Zhirong Shen, Shiyao Lin, Guowen Gong · USENIX ATC '21
• 15mLODIC: Logical Distributed Counting for Scalable File AccessChangwoo Min, Youjip Won, Jeoungahn Park, Taeho Hwang, Jongmoo Choi · USENIX ATC '21
• 20mscoutfs: Large Scale POSIX ArchivingZach Brown · Vault '19
• 17mP99 Pursuit: 8 Years of Battling P99Dor Laor · DevOpsDays Tel Aviv 2024
• 22mEveryone Loves File: File Storage Service (FSS) in Oracle Cloud InfrastructureBradley C. Kuszmaul, Matteo Frigo, Justin Mazzola Paluska · USENIX ATC '19