RAID Rebuild URE Calculator
The chance a RAID 5 or RAID 6 rebuild reads every surviving drive without an unrecoverable read error, from the error rate on the drive data sheet. A simple model, with its assumptions stated.
Result
Enter unrecoverable read error rate, from the data sheet to see the result.
https://calcgallery.com/computing/raid-rebuild-ure-calculator/
Results are the bit arithmetic that the named RFCs define, applied to what you enter. They do not check whether an address is reachable, assigned or reserved for a special purpose. Confirm network plans with whoever administers the network. Read the full disclaimer
How to use the RAID Rebuild URE Calculator
- Choose RAID 5 or RAID 6 and enter how many drives the array has, counting the failed ones.
- Enter the capacity of each drive in terabytes as printed on the label.
- Choose the unrecoverable read error rate from the drive's data sheet, or enter the power of ten if it is another figure.
- For RAID 6, say whether one or two drives have failed, and with one failed, enter the chunk size the array uses.
- Read the chance the rebuild completes, and the expected number of read errors along the way.
The formula
p = 1 ÷ 10^k from "1 error per 10^k bits read"bits read = surviving drives × capacity (bytes) × 8chance of no error = (1 − p)^(bits read)RAID 5, or RAID 6 with two drives failed: chance the rebuild completes = chance of no errorRAID 6 with one drive failed: q = chance of two or more errors among the bits of one stripe row chance the rebuild completes = (1 − q)^(stripe rows)A stripe row is one chunk from each drive. Each surviving drive holds capacity ÷ chunk size rows, and every one of them must come back with at most one unreadable chunk.
The assumptions, plainly
This is the URE-only model often used when large RAID 5 arrays are discussed. It assumes read errors are independent, that each bit fails at exactly the data sheet rate, and that nothing else goes wrong. Each of those is a simplification. The data sheet rate is a maximum: Seagate’s sheets label the row “Max”. One IronWolf data sheet lists 1 per 10E15 for its larger models and 1 per 10E14 for its smaller ones, where 10E14 is Seagate’s way of writing 10^14, so the rate differs even within one product line. Use the row for your exact model. The model also ignores a second drive failing during the rebuild, which is a separate risk.
Because of this, real-world rebuild failures are far rarer than this worst-case model suggests. The often repeated conclusion that large RAID 5 arrays can no longer be rebuilt, drawn from this same arithmetic, is disputed for that reason. Read the headline as the chance at the data sheet’s maximum error rate, not as a forecast for your array.
What the model is good for is comparison. The effect of more drives, larger drives, a better error rate or a second parity drive shows up clearly, even when the absolute figure is uncertain.
Reading the numbers
The expected number of errors is the plainest figure: the bits read multiplied by the error rate. Below about 0.1 a RAID 5 rebuild is likely to finish at the stated rate; above 1 it is likely to hit at least one error. The chance of completing turns that into a percentage, and the “1 rebuild in” figure turns a small failure chance into odds that are easier to compare. A chance too close to 0% or 100% to show at four decimal places is given as “less than 0.0001%” or “more than 99.9999%” rather than rounded to a certainty. For RAID 6 with one drive down, the chance of at least one read error is shown as well, since there a single error does not end the rebuild; elsewhere it would be the same as the chance the rebuild fails.
What the diagram shows
One box per drive, with the failed drives struck through and the surviving drives filled, since each of them is read from end to end during the rebuild.
With your numbers
- Bits read = 3 drives × 4 TB × 8 = 96 × 10^12 bits, at 1 error per 10^14
- Expected errors = 0.96
- Completes = (1 − 10^−14)^bits = 38.2893%
Quick reference
A URE-only model at the data sheet’s maximum error rate. Real drives usually do better, and other failures during a rebuild are not counted.
| Example | Chance the rebuild completes, at the data sheet’s maximum error rate |
|---|---|
| RAID 5, 4 × 4 TB, 10^14 | 38.2893% |
| RAID 5, 6 × 18 TB, 10^14 | 0.0747% |
| RAID 5, 6 × 18 TB, 10^15 | 48.6752% |
| RAID 6, one failed, 6 × 18 TB, 10^14 | 99.9999% |
| RAID 6, two failed, 6 × 18 TB, 10^14 | 0.3151% |
| RAID 6, one failed, 8 × 20 TB, 10^15 | more than 99.9999% |
| RAID 5, 3 × 8 TB, 10^15 | 87.9853% |
Common questions about the RAID Rebuild URE Calculator
- What is an unrecoverable read error?
- A sector the drive cannot read back even after its own retries and error correction. Hard drive data sheets state a maximum rate, such as 1 per 10^14 bits read. In a healthy array the controller rebuilds that sector from parity. During a RAID 5 rebuild there is no parity left to spare, so one such error can stop the rebuild or leave a hole in the data.
- Why do the results look so alarming for large RAID 5 arrays?
- Because 10^14 bits is only 12.5 terabytes, and a rebuild of six 18 TB drives reads 90 TB. At the stated rate several errors would be expected. The data sheet figure is a maximum, however, and many drives do much better in practice, so treat the result as the pessimistic end of the range rather than a forecast.
- Why is RAID 6 with one failed drive so much safer?
- With one drive gone, RAID 6 still has one parity block in every stripe, so a single unreadable sector is rebuilt like any other missing block. The rebuild only fails if two errors land in the same stripe at once, which is far less likely. With two drives gone, RAID 6 is in the same position as a degraded RAID 5.
- What does the model leave out?
- Another drive failing outright during the rebuild; errors that cluster rather than falling at random; controller and filesystem behaviour, since some carry on past a bad sector and some do not; and backups, which are what protects the data when any of this goes wrong.