Choose inputs that match the rebuild
Use the capacity printed for one member and keep the decimal unit that matches it. The throughput input is the effective aggregate source-read rate used by this model, not the interface speed on a drive box. A controller's displayed rebuild rate may measure replacement-disk progress instead, so don't substitute it without checking what it counts. For the read-error input, use the nonrecoverable read-error specification from a drive datasheet such as the Seagate IronWolf manual.
Full member is the default read scope for a generic rebuild. Choose a smaller scope only when your implementation really reads a smaller populated portion. The page cannot infer that behavior from the RAID level alone.
How to choose a realistic rebuild rate
Use observations from a comparable rebuild under the workload you expect. Normal file-copy speed, link speed, and a drive's maximum sequential rating can describe different bottlenecks. If you don't have a measurement, compare several plausible rates and keep the result as a range of scenarios.
For a two-drive RAID 1 with one failed 2 TB member, the model reads 2 TB from the surviving member. At 100 MB/s, that takes 20,000 seconds, about 5 hours 33 minutes. A two-hour replacement delay makes the total degraded window about 7 hours 33 minutes. These are decimal units and constant-rate assumptions, not an observed rebuild.
Why layouts read different amounts
The model estimates the source members needed to reconstruct one failed member. It keeps that topology visible instead of hiding it inside a risk label.
| Layout | Modeled source read | Extra read-error tolerance in the model |
|---|---|---|
| RAID 1 and RAID 10 | 1 source member | 0 |
| RAID 5 | N - 1 source members | 0 |
| RAID 6 | N - 2 minimum | 1 |
RAID 6 uses a minimum reconstruction-read model of N - 2 full-member equivalents. A real implementation may read more. The Intel RAID guide explains why parity and mirror rebuilds do not have the same read pattern.
Read the result as exposure, not a verdict
The total degraded window adds replacement delay to copy time. Source media read is the work used for the time estimate, while modeled read-error exposure is the chance of at least one specified read error in that work.
The percentage treats the selected datasheet read-error rate as an independent per-bit probability. A specification is not a measured probability for your individual drive, and real errors need not occur independently. Read-error tolerance is a separate parity margin. Neither value is data-loss probability, rebuild-success probability, or a guarantee about a controller.
What this model leaves out
This page models one failed member and one rebuild. It does not model a second failure arriving during the window, AFR, MTTDL, fleet correlation, filesystem behavior, controller scheduling, mixed drive sizes, or a drive recommendation.
Verify the live rebuild next
Compare the entered throughput and read scope with the controller or filesystem status while the array is degraded. The Intel degraded-array guidance is a useful starting point. This calculation runs only in this browser. Inputs are not stored or put in the URL. For adjacent planning, use the RAID capacity calculator or the ZFS capacity calculator.