RAID calculator

What a RAID calculator works out

A RAID (redundant array of independent disks) combines several drives into one volume. The SNIA dictionary describes each level as a data placement policy: RAID 0 spreads consecutive blocks over the drives with no redundancy, RAID 1 stores every block on more than one drive, RAID 5 stores parity so that any single drive can fail, RAID 6 stores enough parity to survive any two, and RAID 10 stripes over mirrored pairs. The price of redundancy is capacity. This calculator turns the number and size of your drives into the space you can actually use, what share of the raw capacity that is, how many failures the level survives and how many drives it needs, in decimal terabytes (what the drive label says) and in tebibytes (what many operating systems show).

Capacity, minimum drives and failures by level

Here n is the number of drives. The minimums, formulas and failure counts are from the Synology DSM help article “Choose a RAID Type”; the SNIA dictionary and the 1988 Berkeley paper agree on what each level does. In the paper’s numbering RAID 1 is mirroring, RAID 3 and 4 use one check disk and RAID 5 spreads the check data over all the disks; RAID 0, 6 and 10 are not among the five levels it defines.

Worked example: four 8 TB drives

Four 8 TB drives are 32 TB raw, or 29.1 TiB. RAID 0 uses all of it, 32 TB, with no failure survived. RAID 5 gives (4 − 1) × 8 = 24 TB, which is 75% of the raw space and shows up as 21.83 TiB. RAID 6 gives (4 − 2) × 8 = 16 TB (50%), RAID 10 gives (4 ÷ 2) × 8 = 16 TB, and a four-way RAID 1 mirror keeps just 8 TB but survives 3 failures. Which one is right depends on how much data you can afford to lose and to rebuild; the calculator only does the arithmetic.

Write cost, from the paper that named RAID

Redundancy also costs speed when writing. In the Berkeley paper, a small write to a RAID 4 or RAID 5 array updates parity from the old data, the new data and the old parity, so it takes four accesses: two reads and two writes. A mirrored array writes every block to each copy, as the SNIA definition of RAID 1 says. We do not give a figure for RAID 6 here because we have not found one in a primary source we read; Synology says RAID 6 performs below RAID 5 because of its additional fault tolerance.

Common mistakes

Counting in the wrong unit: a drive sold as 8 TB holds 8 × 10^12 bytes, not 8 TiB, and the same array shows 21.83 TiB of RAID 5 space after the arithmetic. Expecting RAID 10 to survive any two failures: it survives two only if they fall in different mirror pairs. Treating RAID as a backup: a mirror or parity set copies mistakes and ransomware as faithfully as good data. Mixing drive sizes and expecting all the space: three small drives and one big one is a four-small-drive array with spare room on the big drive. Forgetting the overhead the formulas leave out: filesystem metadata, snapshots, hot spares and cache drives take space too.

How do I calculate usable RAID capacity?

Multiply the smallest drive by the number of drives that hold data. RAID 5 gives up one drive’s worth of space to parity, so usable = (n − 1) × size; RAID 6 gives up two, (n − 2) × size; RAID 10 mirrors every drive, (n ÷ 2) × size; RAID 1 keeps one drive’s worth, whatever the number of copies; RAID 0 and JBOD add the drives together. These are the formulas in Synology’s RAID table. With four 8 TB drives, RAID 5 leaves 24 TB, RAID 6 leaves 16 TB and RAID 10 leaves 16 TB.

How many drives do I need for RAID 5, RAID 6 and RAID 10?

Synology’s table gives the minimums: RAID 5 needs 3 drives, RAID 6 needs 4, RAID 10 needs 4 and an even number, RAID 0 needs 2 and RAID 1 needs 2 (more drives make an n-way mirror). The more drives you add to RAID 5 or RAID 6, the smaller the share lost to parity: with 3 drives RAID 5 uses 66.7% of the raw space, with 8 drives 87.5%. Patterson, Gibson and Katz’s 1988 paper made the same point for one check disk per group of 10 data disks (about 91% usable) and 25 (about 96%).

Can I mix drives of different sizes?

Yes, but in RAID 1, 5, 6 and 10 every drive counts as the smallest one. Four drives of 4, 4, 8 and 8 TB hold 24 TB raw, but RAID 5 gives (4 − 1) × 4 = 12 TB, an efficiency of 50%; the extra 4 TB on each large drive is simply unused. Synology’s table lists the sum of all drive sizes for RAID 0 and JBOD, and Synology Hybrid RAID (SHR) is its feature for combining different sizes with less waste. Check how your own controller or NAS handles mixed sizes before buying.

How many drive failures can RAID 10 survive?

At least one, and at best half the drives. RAID 10 stripes data over mirrored pairs; the SNIA dictionary says it survives “a single storage device failure in each RAID 1 set.” If the two drives that fail are in different pairs, the array lives on; if both drives of one pair fail, the data is gone. Synology puts it the same way: multiple failures are survivable as long as they are not within the same sub-group. So with 4 drives RAID 10 is guaranteed to survive 1 failure and can survive up to 2, while RAID 6 survives any 2.

Why does my array show less space than the drives say?

Drive makers count in decimal: Seagate’s data sheets state that one terabyte equals one trillion bytes, and warn that the operating system may report a different, lower figure. Many systems divide by 1,024 per step instead (a tebibyte, TiB, is 2^40 bytes, per the NIST/IEC binary prefixes), so 24 TB of usable RAID 5 space shows up as 21.83 TiB. Some of each drive is also used for formatting and other functions, which Seagate notes as well. The calculator shows both figures; the data storage converter converts any size between the two systems.

Is RAID a backup?

No. RAID keeps the array running when drives fail; that is what SNIA’s definitions describe, such as surviving a single device failure for RAID 1 and RAID 5 and two for RAID 6. A mirror writes every block to more than one device, so a deleted or overwritten file is deleted or overwritten on every copy, and nothing in the array protects against a controller fault, theft, fire or a bad update. Keep a separate backup on another device or in another place.

How long does a rebuild take, and how risky is it?

Nobody can give you a universal number, but there are two firm limits. First, the replacement drive has to be written in full: even at the 190 MB/s maximum sustained rate that Seagate lists for its BarraCuda 3.5-inch drives (quoted at the outer diameter, so a best case), an 8 TB drive needs at least 11.7 hours and a 20 TB drive 29.2 hours, and Synology notes that read and write performance is severely impacted while a RAID 5 array is degraded. Second, a RAID 5 or RAID 6 rebuild reads every surviving drive: 24 TB, or 1.92 × 10¹⁴ bits, for four 8 TB drives. Seagate rates BarraCuda drives at no more than 1 nonrecoverable read error per 10^14 bits and IronWolf Pro drives at no more than 1 per 10^15 bits, so that rebuild reads 1.92 times the BarraCuda interval and 0.19 times the IronWolf Pro one. These ratings are maximums printed on data sheets, not measured probabilities, and the calculator does not turn them into one; they are worth knowing because a RAID 5 array has no redundancy left while it rebuilds, whereas RAID 6 can still lose one more drive, and neither replaces a tested backup.

Sources: D. Patterson, G. Gibson and R. Katz, “A Case for Redundant Arrays of Inexpensive Disks (RAID),” SIGMOD 1988 (RAID levels 1 to 5, parity update for small writes, usable percentages for one check disk per group); SNIA Dictionary entries for RAID 0, RAID 1, RAID 5, RAID 6 and RAID 10; Synology DSM Knowledge Center, “Choose a RAID Type” (minimum drives, tolerable failures, capacity formulas); Seagate data sheets for IronWolf Pro (DS1914.21-2206US) and BarraCuda 3.5-inch HDD (DS17.2-2603US) for the read-error ratings, sustained data rate and the definition of a terabyte; NIST, “Prefixes for binary multiples.” Related: data storage converter for TB, TiB and bytes, download time calculator for how long a copy or restore takes, uptime calculator for the availability of redundant components, and subnet calculator for the network around the storage.