hesela.devWSBuffer
WSBuffer is a research I/O architecture that separates partial-write buffering from read caching and sends large aligned write regions directly to SSDs.
Its scrap buffer tracks valid byte ranges; a two-stage flusher fills missing data before writing complete regions. Only still-unfilled pages need a fill read; intervening writes can complete a page and bypass that read. The large-write alignment unit is the scrap-page data zone, not an application's direct-I/O alignment. [1]
The prototype modifies XFS on Linux 6.8. Evaluation used eight PCIe 4.0 SSDs in RAID0; ScaleCache comparisons used Linux 5.4. These are configuration-specific experiments, not a matched-kernel universal ranking. [1]
The flusher's Bcount tracks outstanding write bytes, not a measured latency or bandwidth rate; the paper uses it as a coarse busyness signal. [1]
Relations
- related
- Buffered I/O, Direct I/O, Page cache, fsync
Advantages
- Targets partial-write overhead while retaining read caching and the buffered application interface. [1]
Limitations
- Porting requires filesystem integration; the paper delegates durability and crash consistency to the underlying filesystem. [1]
Citations
Sources & caveats
- Source-reviewed research identity, not a shipping-kernel support assertion or an independent benchmark.
- The evaluation's SSD model is Samsung 990 PRO; no independent prototype reproduction was performed. [1]