Technology card
ShippingDevice-Managed SMR
DMSMR · drive-managed SMR
How firmware hides shingled bands, why sustained updates differ from short bursts, and what the host can actually identify.
Glossary entry on hesela.dev · JSON record · Shingle translation layer · Persistent media cache · SMR stalls and error evidence
Related
On this page6 sections
In one minute
Device-managed SMR is shingled magnetic recording presented through a conventional block interface, with drive firmware managing data placement and rewriting.
Shingled tracks overlap, so updating one track can require preserving and rewriting neighboring data. Bands limit that interference. A drive-managed disk hides this organization: the host addresses logical blocks, while firmware decides where their current contents reside. Its ordinary block interface should not be confused with a host-visible zone map. [2] [1]
A shingle translation layer can stage updates in a persistent media cache before merging them into shingled bands. This is distinct from volatile DRAM buffering; the word cache alone does not establish power-loss behavior. [4]
Cache pressure is more than a byte counter. Skylight found separate journal-allocation and mapping-entry constraints: small updates could exhaust a resource before an equal volume of larger writes. Those sampled implementations do not define a universal cache capacity. [4]
Cleaning can overlap foreground requests. Evolving Ext4 for Shingled Disks evaluated four DM-SMR drives from two vendors and found different cleaning behavior. Its explanation of the WD40NMZW mapping strategy was an inference. A short write burst and a prolonged update workload can exercise different paths; neither a universal slowdown ratio nor an idle-only cleaning rule follows from that study. [5]
On Linux, queue/zoned reports the block-layer interface, not the recording technology. Linux v6.8 treats host-aware disks as conventional too, so none does not prove CMR or uniquely identify DM-SMR. The separate SCSI zoned_cap attribute reflects advertised capabilities when available; inspect the exact model and connection path rather than relying on one queue value. [6] [7]
For workload qualification, include sustained updates and the expected recovery or rebuild workload, not only a fresh-drive transfer. Record model, firmware, filesystem, request size, queue depth, write locality, preconditioning and idle time. This is an engineering implication of the observed workload sensitivity, not a new Hesela benchmark or a blanket NAS recommendation. [5]
Compared to neighbors
CMR and SMR describe track layout; drive-managed and host-managed describe how the shingled write constraints are handled. Interface compatibility and workload suitability require separate checks. [2] [3]
SMR interface contracts: the drive's model and the host kernel's treatment are distinct.
- Drive-managed
- What the host seesConventional rewritable logical blocks; bands hidden
- Who handles placement constraintsDrive firmware
- Host-aware
- What the host seesReported zones with sequential-write-preferred behavior
- Who handles placement constraintsHost may cooperate; nonsequential writes remain permitted
- Host-managed
- What the host seesSequential-write-required zones, with optional conventional zones
- Who handles placement constraintsHost must respect the required zones' write pointers
Uncertainty notes
- Source-based architecture guide, not a drive benchmark or a replication of the cited experiments.
- The 2015 and 2017 studies concern their sampled drives and firmware; internal mappings inferred from behavior are not universal specifications.
- Linux behavior is pinned to v6.8. Zoned Storage is maintained by Western Digital.
Sources
- 01Zoned Storage: Getting Started with SMR Hard Disks, interface models · accessed 2026-10-19
- 02Zoned Storage: Shingled Magnetic Recording Hard Disks, recording geometry and interface implementations · accessed 2026-10-19
- 03Zoned Storage: Zoned Storage Devices Overview, required and preferred write zones · accessed 2026-10-19
- 04Aghayev and Desnoyers. Skylight: a Window on Shingled Disk Operation. FAST 2015, sections 2 and 4.5-4.7 · accessed 2026-10-19
- 05Aghayev et al. Evolving Ext4 for Shingled Disks. FAST 2017, sections 2.1 and 4.3 · accessed 2026-10-19
- 06Linux v6.8: sd_read_block_characteristics and zoned_cap_show · accessed 2026-10-19
- 07Linux v6.8: queue_zoned_show · accessed 2026-10-19