Write-back cache
Write-back caching completes a write after updating the cache and defers updating its backing store until later.
Write completion and durable commit are separate boundaries. Linux buffers modified file data and writes it back later; applications use synchronization and must handle writeback errors. A cache read alone is not proof of persistence. Device-level protection does not replace the application's own commit protocol. Write-back describes an update policy, not whether the cache is volatile: a persistent cache can precede the backing store. Linux dm-cache marks cached write-back updates dirty relative to the origin. Flush and FUA requests carry separate persistence requirements through the block stack.
WOW illustrates a separate policy choice: which dirty write groups to destage. It combines address ordering with rewrite recency; a higher write-cache hit rate alone does not determine backing-store service time. The destage pointer is not a physical disk-head position. [7]
Flash-backed controller cache is one protection mechanism for this policy: backup energy transfers volatile cache contents to flash on power loss. It is not the definition of write-back caching itself. [8]
In the cited controller design, current virtual-disk write policy, configured default and disk-cache policy are separate fields. A Linux cache-query warning can reflect an unanswered query without changing controller settings. During RAID-level migration or online capacity expansion, the documented controller switches all its virtual disks to write-through until completion. Record the observation time and background operation as well as the policy. Neither observation proves durability. [9]
Relations
- related
- Clean-First LRU (CFLRU), Clean-First Dirty-Clustered (CFDC), Write-around cache, Idle Read After Write (IRAW), Write-through cache, Solid-state drive (SSD), fsync, Writeback error, Crash consistency, Direct I/O, Storage cache flush, Force Unit Access, Hybrid flash array, Wise Ordering for Writes, Flash-backed write cache, Rio file cache, ATA read look-ahead, FlashTier
Citations
- Gill and Modha. WOW: Wise Ordering for Writes. FAST 2005. Section III.D, limits of the disk-head analogy.
- Dell PERC and HBA Unified User's Guide: cache preservation with nonvolatile cache, H965 and H975 scope.
- Dell EMC PERC 10 User's Guide. September 2021, Rev. A09, pp. 25, 65 (table 12) and 86: reconfiguration, policy fields and Linux cache-query message.
Further reading
- Linux VFS: Handling errors during writeback
- SQLite: Atomic Commit
- Hesela: Why retrying a failed fsync does not prove durability
- Linux kernel documentation. Explicit volatile write back cache control: REQ_PREFLUSH and REQ_FUA.
- Linux kernel documentation. Device-mapper cache: operating modes and on-disk metadata.
- NFS stable writes: what COMMIT, flush and FUA actually promise
Sources & caveats
- Reconfiguration, policy fields and Linux warning: Dell PERC 10, September 2021 Rev. A09 guide, printed pp. 25, 65 and 86. These are model-specific documentation claims, not hardware measurements. [9]