CMR vs SMR: the recording method that bites in arrays

CMR (conventional magnetic recording) and SMR (shingled magnetic recording) are two ways of writing data tracks onto a hard drive’s platters. The difference is invisible in casual use and on most marketing material, but it changes how a drive behaves under sustained writes — and in a RAID array or NAS, that difference can turn a routine rebuild into a multi-day ordeal or an outright failure.

How each method writes

On a CMR drive, write tracks sit side by side with small gaps, each independently writable. To change data, the drive simply rewrites that track. It is straightforward and consistent — the write speed you get at the start is roughly the write speed you get throughout.

SMR squeezes more capacity from the same platters by overlapping write tracks like roof shingles, because the write head is wider than the read head. That packs data more densely and lowers cost, but with a catch: you can no longer rewrite a single track without disturbing the ones overlapping it. The drive must read a whole band of shingled tracks, modify it, and rewrite the band. To hide this, SMR drives use a fast CMR-style cache zone and reorganise data later during idle time.

The SMR write cliff

That caching trick works beautifully for short bursts. Light, bursty writes land in the cache and the drive looks as fast as any CMR model. But sustained writes — copying hundreds of gigabytes, or worse, a RAID rebuild — fill the cache, and then the drive must do the slow read-modify-write dance in real time. Throughput can collapse from normal speeds to a tiny fraction, and stay there until the workload stops and the drive catches up. This is the SMR ‘write cliff’, and it is why SMR is poison for write-heavy roles.

Why SMR stalls RAID and NAS rebuilds

When a drive in a RAID array fails and you replace it, the array reconstructs the missing data by writing the entire contents of the new drive — a sustained, full-drive write that is exactly the workload SMR handles worst. An SMR drive can slow a rebuild from hours to days, during which the array is degraded and vulnerable to a second failure. In some cases the controller times the drive out and ejects it, failing the rebuild entirely. This is the real-world reason every serious NAS guide, including ours on NAS hard drives, insists on CMR for array members.

How to tell CMR from SMR

Manufacturers have historically been coy about labelling SMR, and a few episodes of quietly shipping SMR drives in NAS-marketed lines caused real anger. So verify before you buy:

• Check the maker’s spec sheet — most now state ‘CMR’ or ‘SMR’ explicitly, sometimes only in a footnote or a separate technical PDF.
• Consult community CMR/SMR lists that crowd-source drive models, since the recording method can vary by capacity within the same product family.
• Beware the cheapest small drives — SMR is most common in lower-cost desktop and portable 2.5-inch drives, and at certain mid capacities.
• Cache and warranty hints — unusually large caches on a budget drive, or a model marketed for ‘archive’ use, can be a tell, though they are not definitive.

When SMR is perfectly fine

SMR is not bad technology — it is mismatched technology when used wrong. For write-once, read-many data it is genuinely cost-effective: an external backup drive you fill once and shelve, a cold archive disk, a single-drive media store that mostly serves reads. In those roles the write cliff never bites because you never sustain heavy writes, and you bank the lower cost per terabyte. The rule is simple: CMR for anything that lives in an array or sees regular writes; SMR is acceptable only for write-once, single-disk archival.