hesela.devRange write
A range write lets a storage device select a contiguous destination within client-approved address ranges and return the chosen starting address.
The OSDI 2008 research interface includes alignment and keeps blocks within one multi-block write in order. The filesystem supplies permissible free space and records the returned address; it does not surrender allocation policy. [1]
Overlapping outstanding ranges require temporary device-side tracking of occupied targets. Completion alone cannot retire that tracking: another request may still contain an outdated range. The proposal adds a barrier asserting that later requests account for earlier allocation decisions; pre-barrier writes must also complete before the device forgets them. [1]
The evaluation separates DiskSim/ext2 simulation from Bark, a software layer that skips journal positions on HDDs. Bark's recovery reconstruction was not fully implemented. Those performance experiments do not establish crash-safe deployment. [1]
Relations
Limitations
- Flexible write placement can harm later sequential reads; the filesystem must constrain layout accordingly. [1]
Citations
Further reading
Sources & caveats
- Historical research interface, not a claimed current NVMe feature or reproduced hardware benchmark.