Assumptions
What if
- Pull this week
- On the list
- Clears in
- At risk, next 30 days
- Chance of losing data this year
Fleet
Cleared, week by week
- Risk
- Drives
- Before your change
Pull list
| Rank | Batch | Annual rate | At risk, 30 days | When |
|---|
Under the hood
Each row of the pull list is one batch of drives carried through the same four steps. The example follows whatever is at the top of your list.
Start from the published rate
Every model begins at its lifetime failure rate from Backblaze Drive Stats, with the drive days behind it, so a thin record shows as a wide interval rather than false certainty.
Adjust for age
Failure rates follow a bathtub: a bump in the first months, a floor near year two, then a climb that gets serious after year five. Each batch takes the multiplier for its own age.
Weigh the SMART evidence
In the public fleet, 77 percent of failed drives showed one of five SMART warnings, against 4 percent of healthy ones. A flagged drive is about 18 times likelier to fail; a clean one, a quarter as likely.
Weigh what a failure costs
Every failure starts a rebuild that exposes its whole group. Multiply the chance of failure by the data sitting behind it and the list sorts itself: risk times blast radius.
The wear curve, with your fleet on it
The age multiplier from step two. Each ring is one batch in your fleet, sized by drive count; move Look ahead and watch them climb.
Every model the tool can queue, with its published lifetime failure rate, the 95 percent interval on it, and the record behind it.
| Model | Capacity | Lifetime rate | 95% interval | Drive days | Notes |
|---|
This is the whole model. If you disagree with a constant, you know exactly which one to argue with.
- Annual rate
- lifetime rate × age multiplier × evidence ratio
- Evidence ratio
- 18.26 if flagged, 0.24 if clean normalized to the coverage you set
- Chance of failing in 30 days
- 1 − exp(−annual rate × 30 / 365)
- Rebuild hours
- capacity TB × 106 / rebuild MB/s / 3600
- Exposure per survivor
- 1 − exp(−batch rate × rebuild hours / 8760)
- Chance a group loses data
- P(at least m of the k+m−1 survivors fail in that window)
- Blast radius
- k × capacity the data the group holds
- At risk, 30 days
- drives × chance of failing × blast radius the sort key
- Data actually lost
- the same × chance the group loses data the fleet total
SMART is evidence, not a threshold. Flagged drives carry an 18× likelihood ratio and clean ones 0.24×, renormalized to the coverage you set so the fleet’s total expected failures are preserved.
The age multiplier follows the published bathtub shape, normalized so a mixed-age fleet averages 1.0. Look ahead adds the same years to every batch and reads the curve at the new age: no replacements, no new drives, no change in rates beyond the curve.
Rebuild rate is the effective rate on a live system, not the spec sheet; 150 MB/s is an honest default. Unrecoverable read errors count only where no parity is left.
The list holds drives with a reason. The bands sit at 1.5 percent a year (watch), 5 (schedule) and 15 (pull). The crew works the list top down; weeks after the first assume it also absorbs the drives that newly earn a place each week, which is what the plan chart and the week scrubber show.
A triage heuristic, not a warranty model. Figures are the published lifetime table, rounded; re-pull the current quarter before you spend money, and nothing here replaces watching your own fleet.
Part of Infrabench
Free tools for the people who build data centers, on public data with every constant on the page.
Built by Adrian Mucha
A product designer for the software most designers avoid: data center tools, AI tooling and operator consoles where a wrong click costs real money.
- See the portfolioadrianmucha.us
- Open to new workBased in Los Angeles