Drive Endurance Explained: AFR, TBW and DWPD for the Server+ Exam

Vendors publish several reliability figures for storage, and they measure different things. Choosing a drive for a workload means reading them correctly — and the most common mistake is treating MTTF as a lifespan.

For the general definitions of MTBF, MTTF and availability, see the companion article linked at the end. This one is about the storage-specific ratings.

MTTF and AFR

MTTF — mean time to failure — applies to items that are replaced rather than repaired, which is what a drive is. A figure of 1,200,000 hours does not mean the drive lasts 137 years.

It is a fleet statistic describing the failure rate during the drive's useful life. Across 1,000 such drives you would expect roughly one failure every 1,200 hours — about once every seven weeks.

AFR — annualised failure rate — expresses the same thing more usefully: the percentage of drives expected to fail in a year. Convert with AFR ≈ 8,760 ÷ MTTF, so 1,200,000 hours is about 0.73%.

AFR is the figure to plan with. In a 500-drive array at 0.73% you expect three or four failures a year, which tells you how many spares to stock and whether single-parity RAID is adequate.

Useful life and the bathtub curve

Separate from MTTF, and frequently confused with it. Drives carry a useful life or warranty period, typically five years.

The bathtub curve explains the relationship: high early failure from manufacturing defects (infant mortality), then a long flat period of low random failure, then rising failure from wear-out at end of life. MTTF describes only the flat middle.

So a drive can be within its MTTF statistics and still be beyond its useful life, at which point the failure rate climbs regardless of what the MTTF figure says. Burn-in testing exists to catch the early failures before deployment.

SSD endurance: TBW and DWPD

Flash cells wear out with writes, so SSDs carry endurance ratings that hard drives do not.

TBW — terabytes written — the total data that can be written over the drive's life. A 1 TB consumer SSD might be rated at 600 TBW.

DWPD — drive writes per day — how many times the full capacity can be overwritten daily for the warranty period. A 1 TB drive at 1 DWPD over five years allows writing 1 TB every day for five years.

The two are convertible: TBW ≈ DWPD × capacity × 365 × warranty years.

The practical tiers to recognise: read-intensive drives around 0.5 to 1 DWPD for workloads that mostly read; mixed-use around 3 DWPD; write-intensive at 10 DWPD or more for database logs and caching. Consumer drives sit well below all of these.

Exceeding the endurance rating does not usually mean sudden death — most drives transition to read-only when the write budget is exhausted — but it means unplanned replacement.

Workload rate

The figure people miss on hard drives. Enterprise drives specify a workload rate limit, typically 550 TB read and written per year, and desktop drives often only 55 TB.

Exceeding it voids the warranty and raises the failure rate. This is the real reason desktop drives fail early in servers and NAS units — not that they are badly made, but that they are rated for a fraction of the workload.

Enterprise drives also differ in vibration tolerance, which matters in multi-bay enclosures, and in error recovery behaviour, covered next.

TLER and why it matters in RAID

A specific and testable difference. When a drive hits a bad sector, a desktop drive retries for a long time — up to two minutes — because in a single-drive system recovering the data is the priority.

In a RAID array that pause looks like a failed drive, and the controller drops it. The array degrades, rebuilds, and the cycle repeats with the next drive.

TLER — time-limited error recovery, also called ERC or CCTL depending on the vendor — caps the retry at around seven seconds, so the controller reconstructs the data from parity instead. This is why RAID-rated drives exist, and why using desktop drives in an array causes drives to drop out mysteriously.

Monitoring

SMART attributes give early warning. The ones worth watching are reallocated sector count, pending sector count, uncorrectable sector count and, on SSDs, percentage of endurance used and media wear indicator.

Rising reallocated sectors indicate a drive in decline. A non-zero pending sector count is more urgent. And note that a meaningful share of drives fail with no SMART warning at all, which is why SMART supplements redundancy rather than replacing it.

What to remember

MTTF is a fleet failure rate during useful life, not a lifespan; AFR expresses it usefully for planning. Useful life is a separate warranty boundary, and the bathtub curve explains why. SSDs add TBW and DWPD endurance ratings — match the tier to the write workload. Hard drives have annual workload rate limits, which is why desktop drives fail in servers. TLER is required for RAID.

Related reading: MTBF and MTTF · JBOD · M.2 SSDs