S.M.A.R.T., short for Self-Monitoring, Analysis and Reporting Technology and often written simply as SMART, is a monitoring system built into the firmware of hard disk drives and solid-state drives that tracks health indicators and reports them to the host. In storage, its purpose is to predict likely drive failure early enough for the user or administrator to copy data and replace the drive.
S.M.A.R.T. emerged in the mid-1990s from earlier vendor-specific failure prediction schemes and became part of the ATA command set. NVMe drives provide an equivalent SMART / Health Information log with a different, more standardized structure. The drive collects the data itself; software on the computer only reads and interprets it.
How S.M.A.R.T. monitors a drive
The drive firmware continuously counts events such as remapped sectors, read errors, power-on hours and temperature. For SATA and older ATA drives, each monitored quantity is an attribute with a numeric ID, and each attribute exposes several fields.
| Field | Meaning | How to read it |
|---|---|---|
| Attribute ID | Number identifying the measured quantity | Meaning can differ between vendors |
| Normalized value | Health score, often starting at 100 or 200 | Decreases as condition worsens |
| Worst value | Lowest normalized value ever recorded | Shows past deterioration |
| Threshold | Vendor-defined failure limit | Value at or below it means failing |
| Raw value | Underlying counter or measurement | Most informative, but format varies |
The overall PASSED or FAILED status reported by many tools only changes when a normalized value falls to its threshold. Because thresholds are set conservatively, a drive can report PASSED while already showing serious warning signs in the raw counters.
The most important S.M.A.R.T. attributes
Of the dozens of possible attributes, a handful carry most of the predictive value for hard drives. Rising raw values in the attributes below are the clearest early warnings, especially when they grow over days or weeks.
| ID | Attribute | What a rising raw value means |
|---|---|---|
| 5 | Reallocated sectors count | Bad sectors have been replaced by spares |
| 187 | Reported uncorrectable errors | Data could not be read even with ECC |
| 188 | Command timeout | Operations failed to complete in time |
| 197 | Current pending sector count | Unstable sectors waiting for remapping |
| 198 | Offline uncorrectable | Sectors that failed during offline scans |
| 199 | UDMA CRC error count | Transfer errors, usually a cable or connector fault |
Attribute 199 is a useful reminder that not every warning points to the drive itself. A growing CRC error count usually means a faulty SATA cable, a loose connector or electrical noise, and often stops increasing once the cable is replaced.
S.M.A.R.T. on SSDs and NVMe drives
SSDs have no moving parts, so their health indicators focus on flash wear and spare capacity rather than mechanical faults. SATA SSDs reuse the attribute format, but vendors assign their own IDs for wear, so interpretation requires vendor knowledge.
NVMe drives report health through a fixed log page defined in the NVMe specification, which makes values comparable across manufacturers. The table lists the most useful fields.
| Field | Meaning | Warning sign |
|---|---|---|
| Critical warning | Bit flags for serious conditions | Any bit set |
| Available spare | Remaining spare blocks as a percentage | Falling below the spare threshold |
| Percentage used | Estimate of rated endurance consumed | Approaching or exceeding 100 percent |
| Data units written | Total host writes in units of 512,000 bytes | Nearing the rated TBW |
| Media and data integrity errors | Unrecovered data errors | Any nonzero value |
| Unsafe shutdowns | Power losses without proper shutdown | Rapid growth suggests power problems |
Percentage used relates directly to the drive’s TBW rating and wear leveling. A value above 100 does not mean immediate failure, but it shows the drive has passed the endurance its manufacturer guarantees.
Self-tests and practical monitoring
Besides passive counters, S.M.A.R.T. supports drive self-tests that the host can start. A short test takes about two minutes and checks the main electronics and a sample of the surface. An extended test reads the entire surface and can take several hours on a large hard drive.
Common tools include the command-line smartctl utility and graphical health monitors in operating systems and NAS software. For servers and storage arrays, the practical approach is automated, scheduled monitoring with alerts:
- Record a baseline of raw values when a drive is installed.
- Alert on change, not only on threshold failure, especially for attributes 5, 187, 197 and 198.
- Run short tests weekly and extended tests monthly on hard drives.
- Watch temperature, because sustained heat accelerates wear.
- Replace drives early in redundant arrays when reallocations keep rising.
Limitations of failure prediction
S.M.A.R.T. is valuable but far from a complete failure predictor. Studies of large drive fleets have found that a substantial share of failed drives showed no warning attributes at all before failing. Sudden electronic failures, firmware bugs and controller faults often happen without any gradual sign.
Attribute definitions also vary by vendor and even by model, and some raw values are packed or encoded in ways that generic tools misreport. The only reliable protection remains backups and redundancy; S.M.A.R.T. merely improves the odds of acting before data is lost.
A further limitation concerns hardware between the drive and the system. USB enclosures and some RAID controllers do not pass S.M.A.R.T. commands through, so monitoring software may show no data or misleading data. In such setups, administrators query the drives through the controller’s own tools or connect them directly when health must be verified, and they treat a missing reading as unknown health, never as good health.
Stripped to the essentials
S.M.A.R.T. is a built-in health diary that every modern disk keeps about itself. It notes things like damaged spots, errors, temperature and how long the disk has been running.
Software can read this diary and warn you when the numbers start to look bad. It works like the warning lights in a car: helpful, but not perfect, so important files still need a backup.
Where it is applied
A small accounting office stores client files on a network storage box with four hard drives in a RAID 5 array. The device checks S.M.A.R.T. data daily and emails the office manager when any key value changes.
One morning, the alert reports that drive 3 shows eight current pending sectors and twelve reallocated sectors, although its overall status is still PASSED. A week later the reallocated count has risen to forty.
The manager orders a replacement and swaps the drive the same afternoon, and the array rebuilds overnight. Because the weakening drive was replaced before it failed completely, the array never ran in a degraded state with a second failing disk, and no client data was lost.