DJFS : Directory-Granularity Filesystem Journaling for CMM-H SSDs
summary
DJFS implements per-directory transactions for CXL.ME (CMH) SSDs, using path-based transaction selection, transaction coalescing, and transaction conflict resolution on an 88-core SPR machine with a CMH prototype (Ubuntu 22, Linux 5.18). It logs metadata at cache-line and 256‑byte granularities, allows parallel commits, and cuts lockup and conflicts. Compared to FastCommit it reports 4.5× higher throughput in Varmail, 2.5× in MDTest and 3.7× in Exim, shows far fewer transaction conflicts, shorter lockup intervals, and higher file-ops-per-transaction in evaluations.
AI-written summary · about
chapters (10)
- 0:00DJFS introduction and motivation
- 1:42Traditional journaling bottlenecks
- 4:53CMH SSD and CXL.ME overview
- 6:21File update patterns
- 7:47Directory-granularity journaling design
- 9:13Scalability in DJFS
- 10:40Per-directory transactions challenges
- 13:08Evaluation setup and benchmarks
- 14:56Performance results
- 15:17Conclusion and future work