Usable Capacity, Fault Tolerance, and What the Array Costs to Leave On
Researched
Seeded with published specifications and cited sources rather than results measured here. Every seeded figure links to where it came from.
80% of the raw capacity is usable, and 8 TB goes to redundancy. Expect no redundancy for about 14.8 h while a replacement rebuilds, and longer if the array stays in use.
This calculator needs JavaScript. The capacity and power formulas are written out below.
Runs in your browser. Nothing you type here is uploaded, logged or sent anywhere. The page is a static file and the arithmetic happens on your machine. Same principle as the rest of the site.
How this is calculated
Capacity is drive arithmetic. Power is the same kilowatt hour calculation as everywhere else on this site, applied to a machine that never turns off.
usable drives = drives − parity drives, or drives ÷ 2 for mirrors
usable TB = usable drives × drive TB
usable TiB = usable TB × 10^12 ÷ 2^40
standing watts = drives × watts per drive + chassis watts
annual cost = standing watts × 24 × 365.25 ÷ 1000 × price per kWh
rebuild hours = drive TB × 10^12 ÷ (MB/s × 10^6) ÷ 3600
| Layout | Usable | Survives |
|---|---|---|
| RAID 0, striping only | all drives | nothing |
| RAID 1, mirror | one drive, whatever the count | all but one drive |
| RAID 5 or RAIDZ1 | drives minus 1 | 1 drive |
| RAID 6 or RAIDZ2 | drives minus 2 | 2 drives |
| RAIDZ3 | drives minus 3 | 3 drives |
| RAID 10, striped mirrors | half the drives | 1 guaranteed, more with luck |
| Constant | Value | Why |
|---|---|---|
| Terabyte, as sold | 1e+12 bytes | Drive makers count in powers of ten |
| Tebibyte, as reported | 1.100e+12 bytes | 2 to the power 40, which is what the OS shows |
| Hours in a year | 8766 | 24 times 365.25, because a NAS does not switch off |
| Drive and chassis watts | Your inputs | From the datasheet or a plug meter. Nothing typical is assumed |
Worked example, the one above: 5 drives minus 1 for parity is 4 usable drives, 32 TB. In tebibytes that is 32 times 10^12 over 2^40, or 29.1 TiB. Power is 5 times 5 W plus 20 W of chassis, which is 45 W standing, 394 kWh a year.
What this can't know
- Usable is not free. ZFS reserves slop space, every filesystem spends something on metadata, and a pool kept close to full fragments and slows down. Plan against roughly eighty percent of the figure above, not all of it.
- The rebuild window is a floor, not an estimate. It assumes the array does nothing else and sustains the speed you typed for the whole rebuild. A parity rebuild reads every remaining drive end to end, and a live array serving files will take considerably longer.
- Fault tolerance counts drives, not events. A power supply that takes out a backplane, a controller failure, or a bad batch of drives from one production run are all single events that can exceed the parity count. Drives bought together tend to fail together.
- Idle watts are not average watts. Drives draw far more spinning up than idling, and a scrub or rebuild moves the whole array to active power for hours. The annual figure here is the standing floor.
- It says nothing about performance. Layout affects read and write speed, IOPS and resilver behaviour in ways this page does not model at all.
- RAID is not a backup. No arrangement of these numbers changes that.
Sources
- The terabyte and tebibyte definitions are IEC 80000-13, the standard that separates the decimal and binary prefixes. iso.org
- Parity and mirror capacity follows from the definition of each layout in the OpenZFS and Linux md documentation. openzfs.github.io
- Drive idle and operating watts come from the datasheet for your specific model, and rebuild throughput from your own array. Nothing typical is assumed here on purpose.
- No figure on this page came off the bench. When runs are published, this tool gains measured idle draws for the machines on it and is re-badged Measured. The method is on the methodology page.
FAQ
Why is my new NAS smaller than the capacity on the box?
Two separate deductions, and people usually only expect one. Parity takes whole drives: 5 drives of 8 TB in RAID 5 or RAIDZ1 leaves 32 TB. Then the units change underneath you, because drive makers count a terabyte as a trillion bytes and operating systems count it as 2 to the power 40. That is another 9 percent gone on paper without a byte being used, so 32 TB shows up as 29.1 TiB. Filesystem metadata and reserved space take a little more again.
Is RAID a backup?
No, and this is the single most expensive misunderstanding in home storage. RAID keeps the array serving files when a drive dies. It does nothing about a deleted file, a ransomware run, a failed power supply taking several drives with it, a filesystem bug, or the building burning down, because every one of those reaches all the drives at once. Three copies, on two kinds of media, one of them off site, and test the restore before you need it.
How long does a rebuild take, and why does it matter?
Longer than the optimistic figure this tool gives. Rebuilding one 8 TB drive at a sustained 150 MB/s is 14.8 hours if nothing else touches the array, and something always does. It matters because single parity tolerates exactly one failure: for the whole rebuild window you have no redundancy left, while every remaining drive is being read end to end. Those drives are the same age and model as the one that just died.
RAIDZ2 or RAIDZ1 for a home array?
The question is really whether you can survive a second failure during the rebuild window, which grows with drive size. Going from single to double parity in this configuration costs one drive of capacity, taking 32 TB down, and buys tolerance of a second simultaneous failure. With drives above about 8 TB and a rebuild measured in days rather than hours, that is the trade most people should take. Change the level in the tool and watch what it costs.
Does spinning down the drives save money?
It saves some, and the arithmetic here shows how much there is to save. The array above draws 45 W standing, which is 394 kWh and $118 a year at $0.3 per kWh. Spin-down cuts the drive portion but not the chassis, adds latency to the next access, and adds start-stop cycles that drive datasheets rate for a finite number. On a NAS that is accessed all day it usually costs more in wear than it saves in electricity.
Where to go next
- The standing draw above is one line in a bigger bill: the power cost calculator.
- An array that must survive a power cut needs a battery sized for it: the UPS sizing calculator.
- Several machines in one room are quieter together than the numbers suggest: the noise combiner.