mean-latent-error-time · research · also MLET, Latent error time, LET

Mean latent error time

Mean latent error time (MLET) is the expected fraction of a drive's operating horizon during which at least one latent sector error remains undetected.

Related Observatory article: Mean latent error time on hesela.com

Oprea and Juels define LET for one error history and MLET as its mean over the error-model distribution; both are dimensionless. [1]

Overlapping undetected intervals contribute once to exposed time, while an event-mean detection delay counts each event separately. In Hesela's ten-hour example, the same two onsets can yield LET of 40%, 50% or 60%; the respective mean delays are 3, 2.5 and 3 hours. These are invented complete histories, not a fitted MLET or a comparison at equal I/O cost. Detection ends the undetected state, not necessarily the need for repair. [3]

The trace-based FAST 2010 comparison uses MTTED and addresses uncertainty in fault-onset timestamps. Preserve the population, horizon, detection rules and censoring when comparing either metric. [2]

Relations

related

Mean time to detection, Staggered scrubbing, Latent sector error, Data scrubbing

Citations

1. Oprea and Juels. A Clean-Slate Look at Disk Scrubbing. FAST 2010, section 5. (accessed 2026-10-20)
2. Schroeder, Damouras and Gill. Understanding latent sector errors and how to protect against them. FAST 2010, section 5.1. (accessed 2026-10-20)
3. Hesela: LET interval union versus event-mean detection delay. (accessed 2026-10-20)

JSON: /data/terms/mean-latent-error-time.json