5 facts filed under Endurance and qualification: all 5 were read in the original source. Each was checked against the source’s own words, so its figures appear there, and each names where it comes from. The idea is taught in Part 9, Performance & Endurance.
Primary sources
Facebook fleet study (nearly four years, six platforms): failure = at least one uncorrectable error; between 4.2% and 34.1% of drives per platform failed.
99.8% of SSDs that had an error in the first of two recent weeks also had one in the second.
Failure rate vs data written is not a bathtub: early detection (rises), early failure (falls), then useful life and wear-out (rise). Two-pool (weak/strong blocks) hypothesis. Early detection ends near 3TB for 720GB and 1.2TB drives and near 10TB for 3.2TB drives.
In Meza et al.’s field study, for every platform the top 10% of SSDs with the most errors have over 80% of all uncorrectable errors seen for that platform.
In Meza et al.’s field study, for Platforms B, C, E and F the 10% of SSDs with errors make up over 95% of all observed uncorrectable errors.
Sources
Primary is the original paper, standard or datasheet. Secondary is a review, a vendor summary or a news report. Measured is my own test data.
Show all 1 sources, 1 primary
- PrimaryA Large-Scale Study of Flash Memory Failures in the FieldACM SIGMETRICS 2015, 2015