hesela.devNameless write
A nameless write is a research SSD operation in which the device selects a physical destination and returns its address for the client to record.
The 2010 proposal keeps placement under device control while the filesystem records the returned address. Its preferred reclaim callback asks the filesystem to rewrite and trim, unlike the later prototype's notification after device-side migration. This is not a raw-flash interface for caller-chosen physical overwrites. [1] [2]
The FAST 2012 prototype retains virtual-address mappings and temporary migration mappings; de-indirection is not zero metadata. Associated metadata locates filesystem references; a nameless-data migration is acknowledged only after changed reference metadata is written to the device, allowing its temporary remapping to be removed. Its recovery assumes atomic writes of each page with its out-of-band metadata, followed by a startup scan and reference repair after a crash. The modified-ext3 SSD-emulator study uses ordered journaling and reports four crash-injection points, not exhaustive hardware power-loss testing. [2]
NVMe Zone Append returns a logical address within a selected zone, not a physical flash address. The interfaces should not be treated as interchangeable. [3]
Relations
Citations
- Arpaci-Dusseau, Arpaci-Dusseau and Prabhakaran. Removing The Costs Of Indirection in Flash-based SSDs with Nameless Writes. HotStorage 2010, sections 2-4.
- Zhang et al. De-indirection for Flash-based SSDs with Nameless Writes. FAST 2012, updated 7 February 2012, sections 3-6.
- Western Digital. Zoned Storage Devices Overview, Zone Append.