Estimate RAID rebuild duration or the throughput needed for a target window from explicit reconstruction work and rate assumptions. A separate illustrative URE exposure model uses aggregate read volume.
RAID Rebuild Time Formula
T = (C * 1,000,000 * U) / (S * F * 3600)
- T is the rebuild time in hours
- C is the failed drive capacity in TB (decimal, 1 TB = 1,000,000 MB)
- U is the fraction of one replacement member that is reconstructed: 1 for full-member work, or an explicit partial-work percentage divided by 100. Only use partial work when the actual repair skips the excluded blocks.
- S is the entered sustained rate in MB/s, on the same replacement-member reconstruction basis as the work amount. Raw sequential read speed or aggregate surviving-disk traffic is not automatically this rate.
- F is the user-assumed usable fraction of the entered rate. A measured effective rebuild rate uses F = 1, avoiding a second overhead deduction.
- RAID level and load affect realized throughput, but the calculator applies no universal RAID or workload factor. Lower F values are explicit scenario assumptions, not vendor performance specifications.
The duration modes convert one replacement member’s reconstruction work into bytes and divide by effective throughput. Full and partial modes differ only in their explicit work fraction. Partial repair and used-block skipping depend on the actual implementation and configuration; a half-empty pool does not universally rebuild in half the time. The target-rate modes rearrange the same equation to find the entered rate needed for the supplied window:
S = (C * 1,000,000 * U) / (T * F * 3600)
The separate URE modes illustrate a Poisson independent-error scenario for the actual aggregate bytes read. This is a mathematical exposure model, not an observed rebuild-failure or data-loss probability:
P = 1 - exp(-B / E)
- P is the probability of at least one error in the stated hypothetical model.
- B is the aggregate number of bits read across the actual repair path, not the replacement member’s reconstructed bytes alone.
- One byte is 8 bits. Member-read mode multiplies member capacity by the explicit equivalent members read and read fraction. Nested RAID layouts require the affected group; total drives minus one is not a universal read count.
- E is the assumed bits read per error for the scenario. A datasheet limit does not establish a measured constant independent error probability, and hardware classes do not share one universal rating.
Illustrative Rebuild Scenarios and Priority Settings
The following arithmetic scenarios use a full decimal-capacity member and an entered rate of 160 MB/s, with explicit usable fractions of 58.5%, 45.5% and 35.75%. They preserve the prior numerical examples without treating them as universal RAID-level benchmarks. Changing the usable fraction changes the scenario; actual repair work and measured throughput must come from the system.
| Failed Drive Size | F = 58.5% | F = 45.5% | F = 35.75% |
|---|---|---|---|
| 4 TB | 11.9 h | 15.3 h | 19.4 h |
| 8 TB | 23.7 h | 30.5 h | 38.9 h |
| 12 TB | 35.6 h | 45.8 h | 58.3 h |
| 16 TB | 47.5 h | 61.1 h | 77.7 h |
| 20 TB | 59.4 h | 76.3 h | 97.1 h |
| 24 TB | 71.2 h | 91.6 h | 116.6 h |
Rebuild priority or throttle controls can affect resource allocation, but they are not direct percentages of achieved device throughput. Controller capabilities, repair algorithm, media condition and application workload also matter. Review the documentation for the installed version and measure progress before extrapolating a duration.
| Platform | Control to review | Interpretation |
|---|---|---|
| Linux mdadm | RAID synchronization speed limits | Inspect current per-system and per-array limits; targets do not guarantee sustained throughput. |
| Synology DSM | RAID resync speed and supported Fast Repair options | Availability and skipped-block behavior depend on the supported configuration. |
| QNAP QTS | RAID resync priority | Review the installed version’s host-I/O versus resync priorities; no fixed speedup is assumed. |
| TrueNAS / ZFS | Resilver priority scheduling where supported | A higher-priority window is not a full-speed guarantee; work scope depends on the repair path. |
| Dell PERC / Broadcom MegaRAID | Documented rebuild-rate settings | Defaults and behavior depend on controller/version. A rate setting is not a linear bandwidth percentage. |
RAID Rebuild Time Example Problems
Example 1: An 8 TB reconstruction scenario. Suppose the entered rate is 160 MB/s and you explicitly assume that 45.5% is usable for replacement-member reconstruction. The effective rate is 72.8 MB/s. Rebuilding 8,000,000 MB takes about 109,890 seconds, or 30.5 hours, at approximately 262 GB per hour. RAID level alone does not establish either input rate.
Example 2: A 16 TB reconstruction in 24 hours. Required effective throughput is 16,000,000 / (24 * 3600) = 185.2 MB/s. If the explicitly assumed usable fraction is 46.75%, the required entered rate is approximately 396.1 MB/s. These are planning requirements; the calculation does not determine whether a particular HDD, SSD, controller or workload can sustain them.
RAID Rebuild Time Frequently Asked Questions
Why does my controller estimate a different time? The model assumes constant reconstruction throughput. Production I/O, rate settings, retries, thermal limits and repair-path changes can alter progress. The estimate is not a guaranteed completion time or a mathematical best-case bound. Compare the actual work basis and sustained rate before changing operational settings.
Does the URE illustration predict RAID safety? No. For an illustrative repair reading 56 TB, an assumed independent error rate of one per 10^15 bits gives P = 1 – exp(-0.448), about 36.1%; one per 10^14 gives about 98.9%. A drive specification does not establish those exact observed probabilities. An encountered URE and its recoverability depend on the error pattern, remaining redundancy, repair algorithm and controller handling. The model gives no universal drive-size safety threshold or RAID recommendation. Maintain and verify backups independently of this estimate.
Does a hot spare make the rebuild faster? A compatible available hot spare can reduce the delay before reconstruction starts. The reconstruction rate still depends on the actual spare, controller, topology and workload. Total degraded time can include both waiting time and reconstruction time; do not assume that removing the wait guarantees a particular rebuild rate.
