Rearchitecting Buffered I/O in the Era of High-Bandwidth SSDs
summary
WSBuffer is a rearchitecture of buffered I/O that separates small unaligned writes from page cache by introducing a scrap buffer, an opportunistic two-stage flush that uses idle SSD bandwidth, and an SX-array based concurrent manager (SX-array). Implemented in XFS on Linux 6.8 and evaluated on eight PCIe4 Samsung SSDs, it sends large aligned portions directly to SSDs, reduces page-cache churn, and reports up to 3.91X throughput, 82.80X latency improvement, 99.66% write-memory savings and 28.4% CPU reduction.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 2:05Buffered I/O limits with SSDs
- 4:56Existing approaches and flaws
- 5:58WS Buffer design
- 6:40WS Buffer components
- 8:25Buffer-minimized access and OT Flash
- 9:31Concurrent page management
- 10:34Evaluation setup
- 11:17Write performance results
- 11:58Direct vs hybrid I/O
- 13:00FileBench performance
- 13:44WS Buffer benefits
- 14:26Root causes and summary
- 14:48Conclusion
related talks
15mRearchitecting Linux Storage Stack for µs Latency and High Throughput
38mRethinking Writeback: Scaling Linux Filesystem and Memory Performance for the Next Decade
14mZNS+: Advanced Zoned Namespace Interface for Supporting In-Storage Zone Compaction
31mFrom Open-Channel SSDs to Zoned Namespaces
21mPreview: Operating Systems
17mHow advancements in storage are triggering innovations in open-source and vice-versa
38mMaximizing Heterogeneous Memory Bandwidth in Multi-Socket Systems
31mDesign and implementation of Autocaching for CXLSSD