RAID Rebuild Time & URE Risk Calculator
Estimate how long a degraded RAID group or ZFS vdev may take to rebuild, then see the read volume and URE exposure behind the estimate. The result separates the chance of encountering a read error from the chance of losing a multi-parity array.
Estimated rebuild time30.3 hours (1.26 days)
- Estimated completion time
- Sunday, November 1, 2026 at 9:33 AM
- Data reconstructed
- 12 TB (10.91 TiB)
- Surviving-drive data read
- 60 TB (54.57 TiB)
- Expected URE count
- 0.48
- Chance of encountering at least one URE
- 38.12%
- Clean-read probability
- 61.88%
- Risk interpretation
- This is the chance of encountering a URE, not the chance of losing the array: double or triple parity can recover some read errors when sufficient redundancy remains.
Read-error probabilities use the selected published per-bit URE rating and an independent-error model.
Allocated-block mode is a planning approximation for ZFS; actual resilver behavior depends on pool structure and implementation.
How to use this calculator
- Choose the RAID layout for the affected RAID group or ZFS vdev.
- Enter its drive count and the decimal nameplate capacity of one replacement drive.
- Use full-device mode for ordinary block RAID, or allocated-block mode for a ZFS planning estimate.
- Enter an observed effective rebuild speed when possible, then select the URE rating from the drive datasheet.
How the rebuild and URE estimates work
The time estimate starts with the amount that must be reconstructed. For a full-device rebuild, that is the replacement drive’s decimal capacity. For an allocated-block ZFS estimate, it is that capacity multiplied by the allocated percentage. A 12 TB drive therefore represents 12 × 10^12 bytes, not 12 TiB.
C = D × 10^12 × f and T = C ÷ (S × 10^6)
- C is reconstructed bytes, D is capacity in decimal TB, and f is 1 for full-device mode or allocated percentage ÷ 100.
- T is rebuild seconds and S is the entered effective throughput in MB/s.
- For RAID 1 and RAID 10, the estimate reads one surviving mirror member:
R = C. For parity RAID and RAIDZ, it usesR = (N − 1) × C.
That read-volume assumption is deliberately simple. RAID 10 concerns the degraded mirror pair, not every drive in the array. Parity layouts are modeled as reading the surviving members of the affected group. Real controllers, stripe geometry, metadata, retries, and concurrent workloads can make actual I/O different.
What the URE probability means
The calculator converts surviving-drive reads to bits and applies the selected published maximum nonrecoverable read error rate r. The expected count is λ = 8 × R × r. Assuming independent, constant per-bit errors, the clean-read probability is P(0) = exp(−λ), and the chance of at least one encounter is 1 − exp(−λ).
This is read-error exposure, not an array-loss forecast. In RAID 5 or RAIDZ1, a read error while the only parity is already needed can make the affected data unrecoverable. In RAID 6, RAIDZ2, and RAIDZ3, one or more read errors can remain recoverable when enough parity remains. The calculator does not try to turn the total URE probability into a double- or triple-parity failure probability because that depends on where errors fall and how the implementation repairs them.
What changes the estimate most
Capacity and effective throughput set the clock. Sequential media speed is not the same as rebuild speed: foreground NAS activity, controller limits, disk position, heat, retries, and fragmentation can all reduce it. More drives in a parity group increase the modeled read exposure even though they may improve throughput in a particular implementation.
What this calculator leaves out
The model excludes a second whole-drive failure, controller or cabling faults, power loss, firmware defects, latent corruption, and operator error. Published URE ratings are upper-bound specifications, not measured probabilities for an individual drive, and real errors can be correlated. A scrub, SMART review, and verified independent backup remain useful preparation, but none can guarantee a successful rebuild.
Worked example
Consider a six-drive RAID 6 group with 12 TB drives, a full-device rebuild at 110 MB/s, and a 1 per 10^15 bit URE rating. The replacement drive contributes 12,000,000,000,000 bytes, so the time estimate is 12,000,000,000,000 ÷ (110 × 10^6) = 109,091 seconds, or 30.30 hours (1.26 days).
The model reads five surviving drives’ worth of reconstructed data: 60 TB, or 480,000,000,000,000 bits. Multiplying by 10^−15 gives an expected URE count of 0.48. The independent-error model gives a clean-read probability of exp(−0.48), about 61.88%, and a URE encounter probability of about 38.12%. Because this is RAID 6, that 38.12% is not a 38.12% chance of losing the array.
Common questions
How long does rebuilding an 8 TB, 12 TB, or 20 TB RAID take?
Divide the reconstructed decimal bytes by the effective rebuild speed in bytes per second. At 110 MB/s, a full 8 TB, 12 TB, and 20 TB device takes about 20.2, 30.3, and 50.5 hours before accounting for slowdowns. Observed end-to-end throughput is more useful than a drive’s maximum sequential transfer rate.
Why does foreground NAS activity increase rebuild time?
A rebuild competes with normal reads, writes, parity work, and filesystem metadata operations for disks and controller resources. Random or latency-sensitive workload can prevent the sustained sequential access that produces a fast rebuild. Disk position and thermal throttling can also change the effective rate during the job.
What is a URE rating and where is it listed?
A nonrecoverable read error rating is a drive datasheet reliability specification, commonly stated as a maximum number of errors per bits read. Look for a row such as “nonrecoverable read errors per bits read, max” in the manufacturer’s specifications. It is not a promise that a particular number of bits will fail on your drive.
Does one URE always destroy a RAID 5 rebuild?
Not necessarily for every location and implementation, but RAID 5 has only one parity result available while one drive is missing. A URE on data needed to reconstruct an affected stripe can leave that data unrecoverable. Treat a URE during a degraded single-parity rebuild as serious and restore from tested backups if required.
Why is URE encounter probability not the same as RAID 6 failure probability?
RAID 6 and RAIDZ2 have two parity results, and RAIDZ3 has three, so they can repair some read errors while a drive is absent. Whether errors cause loss depends on their stripe placement, remaining redundancy, and implementation behavior. A total read-error exposure model cannot determine that arrangement-specific probability.
Does ZFS resilver only allocated data?
ZFS resilvering can avoid copying unused space, unlike a traditional whole-device rebuild, but the amount and pattern of work depend on pool structure and the OpenZFS implementation. Allocated-block mode here is therefore a planning approximation, not a promise of a particular resilver algorithm or duration.