Two hard drives can sit side by side on a price list with identical capacity, the same spindle speed and near-identical interface specifications, yet behave in completely different ways once a sustained workload hits them. The reason is usually hidden one level deeper than the spec sheet: the magnetic recording technology used on the platters. Understanding the difference between conventional magnetic recording (CMR) and shingled magnetic recording (SMR) has become essential knowledge for anyone specifying storage for workstations, servers or network-attached storage, and it is one of the first questions procurement teams at DistriNode are trained to ask suppliers.

How conventional magnetic recording works

In a CMR drive, every data track is written as a separate, non-overlapping lane on the platter surface. The write element of the head is the same width as the track it creates, so each track has its own dedicated physical space. When the operating system needs to change a single sector, the head simply flies to that location and rewrites it in place. Neighbouring tracks are never touched.

This in-place rewrite capability is the property that makes CMR performance so predictable. Whether the drive is empty or 90 percent full, whether the workload is sequential or random, write throughput stays within a narrow, well-understood band. That consistency is why CMR remains the default choice for RAID arrays, database volumes, surveillance recorders and any system where write latency spikes are unacceptable.

What changes with shingled magnetic recording

SMR was developed to squeeze more capacity out of the same platter. Tracks are deliberately overlapped, with each new track partially covering the one written before it, which allows the readable portion of every track to be narrower than the write head itself. Areal density rises by roughly 10 to 20 percent at essentially zero manufacturing cost, which is why SMR dominates the value end of the market.

The trade-off is fundamental: because tracks overlap, the drive can no longer modify a sector in place. Changing even a few kilobytes inside a shingled zone forces the drive to read the entire band of overlapping tracks, apply the change in memory, and rewrite the whole band sequentially. A logical 4 KB update can trigger hundreds of megabytes of internal traffic.

The media cache: why short benchmarks mislead

Consumer SMR drives are almost always drive-managed, meaning the firmware hides the shingled layout from the host. Incoming writes land first in a persistent staging area recorded in conventional, non-overlapping format. While that cache has free space, the drive performs like any other disk, and a quick benchmark or a small file copy will show perfectly normal numbers.

The problem appears when the cache fills. The firmware must fold staged data into the shingled zones while new writes keep arriving, and sustained throughput can collapse from 180 MB/s to somewhere between 10 and 40 MB/s, occasionally pausing entirely for seconds at a time. The drive is not defective; it is doing exactly what its architecture requires.

Performance behaviour under real workloads

  • Light desktop duty: browsing, office documents and media playback rarely stress the staging cache, so an SMR drive feels indistinguishable from CMR in day-to-day use.
  • Large sequential ingest: copying a multi-terabyte backup set will exhaust the cache partway through, and transfer rates drop sharply for the remainder of the job.
  • Sustained random writes: this is the worst case for shingled media, since every scattered update multiplies into band rewrites; latency becomes erratic and queue depths climb.
  • Idle-time dependency: drive-managed SMR relies on quiet periods to clean its cache in the background, so systems that write around the clock never give the firmware room to recover.

Why RAID rebuilds expose the weakness

An array rebuild is the most punishing workload a hard drive can face: a continuous, days-long stream of writes covering the entire surface. On shingled media the staging cache saturates within minutes, throughput falls to a fraction of the nominal rate, and command latency can stretch beyond the timeout thresholds used by hardware controllers and software RAID layers alike. Some controllers respond by marking the disk as failed, turning a routine rebuild into a degraded-array emergency.

ZFS users have documented resilver operations on shingled disks taking a week or more where a conventional drive finished overnight. During that entire window the pool runs with reduced redundancy, which is precisely when a second failure does the most damage.

The financial argument is straightforward: the money saved on a shingled drive is rarely worth an extra five days of degraded redundancy during a rebuild.

For any multi-bay enclosure, specify drives that are explicitly validated for array duty. A curated selection is available in the NAS hard drive category, where recording technology is a primary selection criterion.

Where SMR still makes commercial sense

  • Cold archives: data written once and read occasionally never triggers band-rewrite penalties after the initial load.
  • Secondary backup targets: if the backup window is generous and jobs are sequential, the lower cost per terabyte is attractive.
  • Media libraries: a film or photo collection is a read-dominated workload that plays to SMR strengths.
  • Hyperscale object storage: host-managed SMR, where the operating system controls zone placement directly, delivers excellent economics in datacentres built around zoned block devices.

How to verify the technology before purchase

After considerable industry controversy in 2020, all three remaining hard drive manufacturers now publish recording-technology tables covering their product lines. Always cross-check the exact model number, not the product family name, because a single family frequently mixes both technologies across capacity points, and a 4 TB variant may be shingled while the 8 TB version is conventional.

Datasheets provide a second clue: shingled models often show unusually large cache figures (256 MB on an otherwise budget drive) and lower sustained transfer specifications. For standard workstation builds where the drive holds an operating system, applications and working files, the models listed in the desktop hard drive section can be filtered against manufacturer CMR listings in a few minutes.

Frequently asked questions

Does SMR affect read performance as well?

Largely no. Once data is written, reading it back proceeds at full platter speed regardless of recording technology. The exception is reads issued while the drive is busy folding its cache into shingled zones, when internal housekeeping competes with host requests and response times rise.

Can CMR and SMR drives be mixed in one array?

Technically the array will assemble, but it is poor practice. The array inherits the latency profile of its slowest member, and every rebuild or scrub will be gated by the shingled disk. Mixed arrays also complicate failure diagnosis because slow-response symptoms mimic a dying drive.

Do solid-state drives use shingling?

No. Shingling is specific to overlapping magnetic tracks on rotating platters. SSDs have their own asymmetric-write behaviour caused by flash erase blocks, which is managed through TRIM and garbage collection, but the mechanism is entirely different.

Is TRIM support on an SMR drive a complete fix?

It helps, but it is not a cure. TRIM tells the firmware which blocks no longer hold valid data, reducing unnecessary band rewrites during cleaning. The fundamental read-modify-write behaviour of overlapped tracks remains, and sustained-write collapse still occurs once the staging area is exhausted.

Which technology suits video surveillance recorders?

Conventional recording, without hesitation. An NVR writes continuously from multiple camera streams and never idles long enough for a shingled drive to clean its cache. Purpose-built surveillance drives are conventional for exactly this reason.

Key takeaways

Recording technology is not a footnote; it defines how a hard drive behaves under pressure. Choose conventional recording for arrays, surveillance, virtualisation and any write-intensive role. Reserve shingled drives for archives and read-mostly libraries where their cost advantage is real and their weaknesses stay dormant. Above all, verify the exact model number against the manufacturer documentation before the purchase order goes out, because the spec-sheet basics will not warn you.