How SMR Drives Boost Disk Density and Challenge Storage Management

Shingled Magnetic Recording (SMR) Overview

SMR increases areal density by partially overlapping magnetic tracks on a platter, similar to roof shingles. Overlap allows more tracks per inch while keeping the write head size unchanged. Because a newly written track overwrites portions of the previous track, SMR disks cannot perform random in‑place updates; they support only sequential append writes. Random reads remain possible.

Physical Organization

Tracks are grouped into contiguous bands , which the host sees as zones . A typical zone is about 256 MiB. Within a zone, reads can be random, but writes must follow the zone’s write pointer (WP) , which always points to the next free LBA.

Zone Types (ZBC/ZAC standards)

Conventional Zone (C‑Zone) : fully random‑write capable; usually < 1 % of total capacity.

Sequential‑Preferred Zone (SP‑Zone) : located in the sequential‑write region; writes are expected at the WP for optimal performance, but out‑of‑order writes are allowed at a cost (additional data migration and garbage collection).

Sequential‑Required Zone (SR‑Zone) : the drive rejects any write that does not start at the current WP; strict sequential appends only.

When an SR‑Zone is full, the host must issue a Reset Write Pointer command to clear the zone and start writing from its beginning.

Management Models

Drive‑Managed SMR (DM‑SMR) : The firmware hides zones and WP behind a traditional block interface. Internally a Shingle Translation Layer (STL) converts random writes into sequential writes, performing background data migration and garbage collection.

Host‑Managed SMR (HM‑SMR) : Zones and WP are exposed via ZBC (SAS) or ZAC (SATA). The host must obey sequential rules; any violation returns an error.

Host‑Aware SMR (HA‑SMR) : Like HM‑SMR, zones and WP are exposed, but the drive may accept out‑of‑order writes, handling them internally at the expense of performance.

DM‑SMR offers drop‑in compatibility with existing software but can exhibit unpredictable latency under random‑write workloads because of the STL’s background work. HM‑SMR provides deterministic performance but requires software to be zone‑aware. HA‑SMR is a compromise, allowing limited out‑of‑order writes while still exposing zone information.

Standards and Commands

The T10 (SCSI) and T13 (ATA) committees defined the ZBC (Zoned Block Commands) and ZAC (Zoned‑device ATA Commands) specifications. Both abstract an SMR device as a collection of zones, each consisting of a contiguous LBA range, and define the three zone types above. Commands include: WRITE ZONE – writes at the current WP. READ ZONE – reads any LBA within the zone. RESET WRITE POINTER – clears a full SR‑Zone for reuse.

Design Implications

Effective use of SMR requires matching the application’s I/O pattern to the drive’s sequential‑write constraints:

Batch writes into large, zone‑aligned buffers (≥ zone size) to avoid frequent WP resets.

Prefer workloads with high sequential write volume (e.g., log aggregation, backup, archival).

For random‑write heavy workloads, consider using the small C‑Zone region or a DM‑SMR device with sufficient cache.

Choosing the appropriate management model and aligning software to respect WP semantics are essential to achieve the density benefits of SMR without incurring severe performance penalties.