Storage fundamentals
Shingled magnetic recording (SMR)
Shingled magnetic recording (SMR) is a hard drive technology that increases capacity by writing data tracks so that each partly overlaps the one before, like shingles on a roof. More tracks fit on each platter. The cost is that data can no longer be rewritten in place without disturbing the neighboring track, which changes how the drive handles writes.
How does SMR work?
A drive's write head is wider than its read head. Conventional recording leaves a gap between tracks so that writing one never touches the next. SMR removes the gap: each new track overlaps part of the previous one, leaving a narrower strip that the read head can still follow. Tracks are grouped into zones separated by guard bands. Within a zone, data has to be written in order from start to end. Changing something in the middle means rewriting everything after it in that zone.
How does SMR compare with conventional recording?
The two behave alike for reading and very differently for rewriting.
| Aspect | Conventional (CMR) | Shingled (SMR) |
|---|---|---|
| Capacity per platter | Baseline | Higher |
| Sequential write | Fast | Fast |
| Random overwrite | Fast, in place | Slow, requires rewriting part of a zone |
| Read performance | Normal | Normal |
| Best suited to | General use, mixed workloads | Write-once, sequential, archival data |
What are the types of SMR drive?
Drive-managed SMR hides the zones behind a conventional interface. The drive accepts random writes into a staging area and reorganizes in the background, which works until sustained writing fills the staging area and throughput collapses. Host-managed SMR exposes the zones, and the operating system or storage software must write each zone sequentially. That demands software written for it and gives predictable performance. Host-aware drives accept both styles. Large-scale storage systems that use SMR generally choose host-managed drives.
Where does SMR fit in storage systems?
SMR suits data that is written once, in large sequential pieces, and rarely changed: archives, backups, media, logs and the capacity tier of object storage. Object stores are a natural match because objects are written whole and never modified in place. It is a poor fit for databases, virtual machine disks and other workloads with random overwrites. One operational point deserves attention: rebuilding onto a drive-managed SMR drive can be very slow, so the drive type should be known and matched to software that supports it. Sequential vs. random I/O explains the access patterns involved.
Frequently asked questions
Are SMR drives less reliable?
Not inherently. The difference is in write behavior, not in how well data is retained.
How can an SMR drive be identified?
From the manufacturer's specification for the model. Drive-managed SMR is not always obvious from the drive's interface.
Can SMR drives be used in RAID?
Drive-managed SMR can cause very long rebuilds and timeouts in RAID arrays, and many vendors advise against it.
Related terms
- Flash storage vs HDD: the two main media compared.
- Sequential vs. random I/O: the access patterns that decide SMR suitability.
- Cold storage: the tier where SMR capacity is most useful.
- Data rebuild: an operation SMR can slow down.
- Write amplification: extra internal writes caused by rewriting.