hesela.devAutomated storage tiering
Automated storage tiering relocates data among storage classes according to a policy without requiring a person to perform each move.
Scope: Tiers can differ in performance, capacity or cost. Their existence does not imply automatic movement; automation adds policy-controlled relocation within or across systems, including a cloud tier. [2] [1]
Policy and placement are different states. In the cited volume-tiering implementation, changing auto to none does not itself recall blocks already in the capacity tier. Separate retrieval controls affect their return. Check actual placement after a policy change. [4]
Retrieval: From ONTAP 9.8, cloud-retrieval-policy can override tiering defaults. Under default retrieval, auto brings cold data back on random reads but not sequential reads; on-read brings it back for either pattern, while never leaves read blocks in the cloud. Reading only filenames, timestamps or ownership metadata does not heat the underlying data blocks. [4] [6]
Completion: Using none plus promote requests proactive retrieval of all tiered volume data. Using snapshot-only plus promote targets tiered active-filesystem data, not blocks used only by snapshots. Check promotion status with volume object-store tiering show; accepting a policy change is not proof that promotion has finished. [7]
Cache safety cannot be reduced to a move-versus-copy distinction. A dirty write-back block can be newer than its stale origin copy; losing that cache can lose updates. The cited cache also describes promotion and demotion as migration. [5]
A capacity tier is not an independent backup merely because it is remote. In the cited architecture, required metadata stays on the performance tier, so the capacity-tier blocks alone cannot reconstruct the system after that metadata is lost. [6]
Relations
Limitations
- Automatic placement does not establish immediate migration or a guarantee that every read promotes data. [4]
Citations
- SNIA Dictionary: automated storage tiering, policy-controlled movement within and across systems.
- SNIA Dictionary: tiered storage, distinct classes and optional dynamic movement.
- SNIA Dictionary: auto-tiering synonym.
- NetApp ONTAP: FabricPool volume tiering policies, policy changes and cloud retrieval controls.
- Linux 6.12: dm-cache glossary and cache operating modes, origin-copy and writeback semantics.
- NetApp: Oracle database FabricPool tiering overview, Architecture, Data and metadata, and Backups.
- NetApp ONTAP: Promote data to the performance tier, ONTAP 9.8 and later; requested scope and promotion-status check.
Sources & caveats
- SNIA supplies the terminology and auto-tiering alias; the broader storage tiering alias remains attached to hot-warm-cold. [1] [2] [3]
- FabricPool supplies the policy-change, retrieval and metadata examples; these are not universal rules for every tiering system. [4] [6]
- Linux 6.12 supplies a counterexample to treating every cache as a disposable current copy. [5]
- Documentation synthesis, not a device experiment, measured speedup or independent recovery test.
- The ONTAP promotion procedure documents requested data scope and a separate status check; this is not a completed recall or a tier-removal procedure tested by Hesela. [7]