MTBF
Mean time between failures (MTBF) is the expected operating time between successive failures of a repairable system under specified conditions.
Under a constant repair rate model, MTBF is estimated as total operating time divided by the number of failures. The NIST example of 800 operating hours with 2 failures gives 400 hours, with a two-sided 90% confidence interval of 127 to 2251 hours. With no failures in the same 800 hours, only a 95% lower bound of 267 hours can be stated. [1]
For an exponential lifetime the mean time to failure is 1/λ, and the probability of failing by that time is 0.6321, so about 63% of units fail before reaching the mean. The mean is therefore not a calendar replacement age for an individual drive, and converting it into a failure probability requires a lifetime model. [2]
Schroeder and Gibson compared datasheet MTTF values of 1,000,000 to 1,500,000 hours, equivalent to annual failure rates of 0.58% to 0.88%, with replacement records for about 100,000 drives. Observed annual replacement rates ranged from 0.5% to 13.5% with a drive-weighted average of 3.01%, and replacement rates rose with age instead of staying constant. [4]
Relations
- related
- Annualized failure rate (AFR), Hard disk drive (HDD), S.M.A.R.T., Hazard rate, Annual replacement rate, Rate of occurrence of failures, Mean time to data loss (MTTDL)
Advantages
- Gives one exposure-based figure that can be bounded with confidence limits from observed operating time and failure counts. [1]
Limitations
- A drive datasheet usually states MTTF for a non-repairable item, so the quantity behind a quoted figure must be checked before comparing it with an MTBF. [4, 2]
- Short tests with few failures leave wide intervals, so they cannot confirm a large value. [1]
Citations
- NIST e-Handbook: Constant repair rate model and confidence bounds (8.4.5.1)
- NIST e-Handbook: Exponential distribution (8.1.6.1)
- Bianca Schroeder and Garth A. Gibson. Disk failures in the real world: What does an MTTF of 1,000,000 hours mean to you? FAST 2007.
Further reading
Sources & caveats
- nist-intervals is the NIST/SEMATECH e-Handbook section 8.4.5.1, re-read on 2026-10-22 including its worked example.
- The 800-hour, 2-failure, 400-hour, 127 to 2251 hour and 267-hour figures come from nist-intervals.
- nist-exponential is the NIST/SEMATECH e-Handbook section 8.1.6.1, re-read on 2026-10-22, and is the source of 0.6321.
- sg07 is the FAST 2007 paper by Schroeder and Gibson of Carnegie Mellon University, read in full on 2026-10-22.
- The 1,000,000 to 1,500,000 hours, 0.58% to 0.88%, about 100,000 drives, 0.5% to 13.5% and 3.01% figures come from sg07.