Free tool · RAID calculator
How much space is left – and how long does it run without a net?
Usable capacity is the easy part. This calculator adds how long the array runs unprotected after a disk swap and how likely it is that the rebuild itself fails. Those are the numbers a decision hangs on.
Usable
24 TB
of 32 TB raw · 25 % spent on redundancy
The difference between 24 TB and what the operating system shows is neither loss nor error: manufacturers count 1 TB as one trillion bytes, Windows counts in powers of two. Same amount, two scales – about 9 % appear to be missing, on every disk from every manufacturer.
During the rebuild, 24 TB must be read without a single error. At one unrecoverable read error per 1,000 trillion bits read – the data sheet figure for a 24/7-rated disk – that is 0.19 expected errors. With single parity there is no second checksum to step in: a single read error in the wrong place ends the rebuild.
RAID 5 (single parity)
Long the default, today only defensible with small disks. During a rebuild ALL remaining disks must be read completely – see the risk line below. From about 4 TB per disk, RAID 6 belongs here instead.
And the most important part: None of these levels is a backup. RAID protects against the failure of a device – not against a deleted file, not against ransomware, not against fire, and not against a controller that writes garbage as it dies. In all four cases the whole array is affected immediately and completely.
The underestimated window
Between the disk swap and the finished rebuild lies a whole day without protection.
In the imagination a disk failure is an event: it beeps, the disk is swapped, all is well. In reality it is a period – and a long one. The controller has to fill a complete disk, and for 8 TB that takes around fifteen hours if the array has nothing else to do. If it has something to do, it takes two days.
For that whole time a RAID 5 runs with no protection at all. And it is also the time of the highest load the remaining disks will ever see: they are read from the first sector to the last. Disks bought together, installed together and loaded together do not fail independently of one another.
So the question about RAID 6 is not about the improbable double failure on the same day. It is whether you want to spend that day without a net – and the answer costs one disk.
RAID levels
What each level is for
- RAID 0 (striping)Full capacity, no protection.
- For scratch data you can recreate at any time – video editing caches, render files. If one disk fails, all data is gone, not just part of it. RAID 0 has no place in a server holding business data.
- RAID 1 (mirror)Half the capacity, simplest rebuild.
- The default for the system drive and for small servers. The rebuild is a plain copy from one disk to the other – the least complicated of all levels.
- RAID 5 (single parity)One disk for protection, one may fail.
- Long the default, today only defensible with small disks. During a rebuild ALL remaining disks must be read completely – see the risk line below. From about 4 TB per disk, RAID 6 belongs here instead.
- RAID 6 (double parity)Two disks for protection, two may fail.
- The sensible choice for file storage with large disks. The point is not a second simultaneous failure but a failure DURING a rebuild – and with large disks that window is long.
- RAID 10 (mirrored stripes)Half the capacity, fastest rebuild.
- For databases, session hosts and virtual machines – anything with many small writes. It costs half the capacity and in return gives the shortest unprotected window: the rebuild reads from a single partner disk.
- Single disks (JBOD)Full capacity, each disk on its own.
- Not an array but several separate drives. If one fails, only its data is affected – better than RAID 0, still no protection. Sensible at most as a backup target, never as live storage.
Frequently asked questions
Why does Windows show less capacity than the disk label says?
Because two scales are in use. Manufacturers count in decimal: 1 TB is one trillion bytes. Windows counts in binary: 1 TiB is 1,099,511,627,776 bytes, about 9 % more per unit – but it labels the result 'TB' anyway. A 4 TB disk therefore shows as 3.64 TB. Nothing is missing and nothing is wrong: the same amount, two scales, on every disk from every manufacturer.
Is RAID 5 still acceptable?
With small disks yes, with large ones no longer. The reason is the risk line in the calculator: during a rebuild, every remaining disk must be read completely without error. With eight 16 TB disks that is over 100 terabytes – and at one unrecoverable read error per 100 trillion bits, the data sheet figure for desktop-class disks, the expectation is above one error. With single parity nothing steps in: one read error in the wrong place ends the rebuild. From about 4 TB per disk, RAID 6 belongs there.
How long does a rebuild take?
It depends on the size of ONE disk, not of the array – exactly one disk has to be written in full. At 8 TB and 150 MB/s that is about 15 hours idle, and closer to two days if the array has to serve users at the same time. For that whole period a RAID 5 runs without protection; a second failure means total loss. That window, not a simultaneous double failure, is the real argument for RAID 6.
Does RAID replace a backup?
No, and it is the most important sentence on this page. RAID protects against the failure of a device. It does not protect against a deleted file, ransomware, fire or water, or a controller that writes garbage to every disk as it dies. In all four cases the array is affected immediately and completely – the deletion is mirrored faithfully.
How many disks should an array have?
For RAID 6, six to twelve have proven sensible. Fewer wastes capacity because two disks always go to protection; many more increases the amount that must be read error-free during a rebuild, and with it the risk. More important than the count is a hot spare that sits in the chassis and takes over on its own – then the rebuild starts within minutes, not when someone reads the alert.
Where does the read-error figure come from?
From the manufacturer's data sheet: one unrecoverable read error per 100 trillion bits read for desktop-class disks, per 1,000 trillion for disks rated for 24/7 operation. It is a specification, not a measurement – field data tends to be better, and the calculation counts as pessimistic. We show it anyway because it gets the order of magnitude right and explains why manufacturers sell different disks for servers. The assumption is visible next to the result; switch the disk class and watch the difference.
Read on: Subnet calculator · Colocation in Germany · All English tools · Deutsche Fassung