Fast and Synchronous Crash Consistency with Metadata Write-Once File System

on YouTube ↗15m

" summary

Build: a Linux kernel prototype called WOLVES implementing a metadata write-once file system that turns each file operation into a single checksum-protected package stored via per-core free lists and parsed by a package translation layer. Claim: this reduces many small ordered metadata I/Os to one ordered write per operation. Results: microbenchmarks (fio 1 GB, 4 KB IO) show existing PM approaches incur >75% sync overhead and >15% PM bandwidth waste, while WOLVES improves throughput by ~15% and can reach PM I/O bandwidth limits with stable write performance. Recovery and immediate package reclamation work in evaluation.

AI-written summary · about

" chapters (15)

1. 0:00Introduction
2. 0:19Crash consistency in PM
3. 2:45Synchronous crash consistency
4. 4:10High overhead causes
5. 5:17Design goals for WOLF
6. 6:01Metadata Write-Once File System
7. 7:05Package persistence
8. 8:53Package management
9. 10:20Package reclamation
10. 11:00WOLF evaluation setup
11. 12:06Performance by IO size
12. 12:51Metadata performance
13. 13:32Sync vs async consistency
14. 14:16Summary
15. 14:59Q&A