Skip to content

Technology card

Limited

Host-Managed SMR

HM-SMR · host-managed zoned SMR

Write rules, zone lifecycle and software compatibility for host-managed shingled disks.

Glossary entry on hesela.dev · JSON record · Host-aware SMR · ZBC · SMR error evidence

On this page6 sections

In one minute

Host-managed SMR is a hard-disk interface model that requires host software to write sequential-write-required zones at their reported write pointers.

The host-managed contract applies to sequential-write-required zones, not necessarily every sector on the disk. Each such zone has a write pointer identifying the next allowed write position. Optional conventional zones accept ordinary random writes. Check the reported zone type and boundaries instead of treating the entire device as one append-only stream. [7]

Reusing a sequential zone requires a reset, after preserving any data still needed. A successful reset of an eligible sequential zone returns its write pointer to the start and invalidates previous contents. Closing a zone retains its write position for later writes; finishing makes it full even if it had unwritten capacity. Finish does not turn an unwritten tail into meaningful data. These operations do not establish application durability. [7]

Host-aware is a different model, not another name for host-managed. Its sequential-write-preferred zones permit non-sequential writes, allowing use as a regular disk. Drive-managed SMR goes further and hides zoning from the host. The three models can share shingled recording physics while exposing materially different software interfaces. [9]

Host-aware also needs historical qualification. Upstream Linux 6.8 removed the host-aware block-layer model: these drives remain usable as ordinary block devices. The old queue/zoned value host-aware disappeared; none therefore does not prove that a disk is CMR. Distinguish the drive interface, kernel block-layer support and controller command path when interpreting an older setup guide. [5]

Wu and colleagues tested Seagate ST8000AS0022 drives with prototype firmware ZN03 in 2016. Their burst workload showed a sequential-write throughput drop around the reported 128-open-zone recommendation; sustained non-sequential writes exposed media-cache cleaning delays. This is evidence about those prototypes and workloads, not a universal threshold or a benchmark of host-managed disks. [6]

Do not copy every NVMe ZNS rule onto an SMR disk. The ZBC/ZAC model can impose an open-zone limit, but does not define the ZNS-style active-zone resource limit. It also has no native Zone Append command; host software can emulate append semantics. A host-aware disk's optimal open-zone count is guidance rather than an enforced maximum. [8]

Compatibility begins below the filesystem. A SAS HBA must recognize the host-managed device type; attaching an ATA ZAC disk through it additionally requires suitable SCSI-to-ATA translation of zone commands. A working cable or ordinary disk support does not establish this capability. Confirm the drive model, controller firmware and reported zones before selecting a storage layout. [9]

Linux block-layer zoned support arrived in 4.10, but that is not a minimum-version promise for every zoned filesystem: zonefs arrived in 5.6 and Btrfs zoned mode in 5.12. A general-purpose zoned-aware filesystem, or a suitable filesystem over dm-zoned, can handle placement for unmodified applications. Raw zoned access leaves the sequential-write obligation with the application. [10]

zonefs presents zones as files without turning them into unrestricted writable files. Writes to sequential-zone files must be append-only and use direct I/O; a write failure can leave only part of the data written. Applications must reconcile file size and their own record boundaries before continuing. Use synchronization and recovery appropriate to the application; a legal zone write alone does not establish a complete transaction. [11]

Btrfs zoned mode is a specific on-disk format and allocation strategy, not a switch that exposes hidden zones in drive-managed SMR. All member devices of the zoned filesystem must share the same zone size, and its supported features differ from ordinary mode. Check the documented restrictions for the deployed version before committing a pool to this layout. [12]

SMORE takes an application-level route: an object spans zone sets across disks, while a flash-resident index locates its segments. Recovery combines index snapshots with embedded layout markers, tombstones and zone-set digests. Its six-drive experiment mostly sampled every 60th zone; reported recovery extrapolations are not full-capacity measurements. Optional NVRAM buffering also needs its own failure policy: disk parity cannot reconstruct data that never reached the disks. [1]

SMRSTORE connects Pangu chunk operations directly to HM-SMR zones. It combines stream-type separation, chunk-size limits matched to zones and round-robin allocation from preopened pools. A large chunk can then be deleted without copying unrelated live data from its exclusive zone. Small chunks can still share zones, and the paper's ablation found stream separation alone insufficient. [2]

The placement trade-off appears before cleaning starts: SMRSTORE's round-robin policy reduces small-write throughput in its clean-state microbenchmark by spreading writes across zones. Its eventual reclamation benefit must be assessed after deletions and with the actual workload. The study used an OSS/Pangu stack, SSD caches and specific disks, with 6 TB baseline comparisons and a separate full-capacity SMRSTORE variant; it does not establish a universal SMR advantage. [2]

Our reading: compare placement and recovery together. An allocation policy determines both how much live data must move during reclamation and which records recovery must inspect. Check whether published write amplification includes parity, whether recovery starts from surviving metadata, and which buffering failures the design covers. The structured records retain these qualifications for SMORE and SMRSTORE. [3] [4]

SMR interface models

Choose the interface model by the software stack's write contract. Capacity and latency still depend on the specific drive, workload and data-management policy.

SMR interface models, including historical host-aware designs: device contracts, not a promise of current Linux zoned support or measured speed.

Drive-managed
Host-visible zonesNot exposed
Random writesAccepted; firmware handles placement
Host-aware
Host-visible zonesSequential-write-preferred; optional conventional zones
Random writesAccepted; sequential placement remains preferable
Host-managed
Host-visible zonesSequential-write-required; optional conventional zones
Random writesRejected away from the write pointer in required zones

Uncertainty notes

  • This is a source-based interface guide, not a drive benchmark or a capacity comparison.
  • Zoned Storage is maintained by Western Digital; Linux and Btrfs documentation describe their own software behavior.
  • Kernel feature-introduction dates do not replace checking current fixes and filesystem restrictions.

Sources

  1. 01Macko et al. SMORE (extended version), 2017: recovery, buffering and evaluation scope. · accessed 2026-11-14
  2. 02Zhou et al. SMRSTORE, FAST 2023: placement, recovery and evaluation, sections 3-7. · accessed 2026-11-14
  3. 03Hesela corpus: SMORE, parity accounting and recovery limits. · accessed 2026-11-14
  4. 04Hesela corpus: SMRSTORE, placement and workload scope. · accessed 2026-11-14
  5. 05Linux commit 7437bb73f087: remove support for the host-aware zone model (included in v6.8) · accessed 2026-10-18
  6. 06Wu et al., Evaluating Host Aware SMR Drives, HotStorage 2016, sections 3.1-3.3 · accessed 2026-10-18
  7. 07Zoned Storage: zone models, write pointers and zone-management commands · accessed 2026-10-18
  8. 08Zoned Storage: SMR zone types, resource limits and Zone Append · accessed 2026-10-18
  9. 09Zoned Storage: SMR identification and host bus adapter compatibility · accessed 2026-10-18
  10. 10Zoned Storage: Linux access paths and kernel feature history · accessed 2026-10-18
  11. 11Linux kernel documentation: zonefs file operations and I/O error handling · accessed 2026-10-18
  12. 12Btrfs documentation: zoned mode requirements and limitations · accessed 2026-10-18

Published by Hesela team

Report a correction · Editorial standards · Reuse guidance

Cite this page
Hesela team. Host-Managed SMR. Hesela. https://hesela.com/tech/hm-smr/