RAID 5 vs RAID 6: Capacity, Minimum Disks, Performance and a Calculator

RAID 5 vs RAID 6: capacity and failure tolerance

Both levels stripe data across every disk in the group and add parity, so the array can rebuild a lost disk from the others. Synology's DSM help (version 6.2.3) describes both the same way: data and parity information striped across all member drives. The difference is how much parity. QNAP's RAID documentation says RAID 5 loses the capacity of one disk to store parity information, and RAID 6 loses the capacity of two. Dell describes RAID 6 as RAID 5 plus another parity block.

That extra parity buys a second failure. Synology says a RAID 5 pool may be rebuilt using parity information on the other drives if one member drive fails, and a RAID 6 pool even if two member drives fail. Dell says RAID 6, because of its dual parity, can handle two simultaneous drive failures. ZFS has the same split: OpenZFS documents raidz with one, two or three parity disks, meaning the group can sustain one, two or three failures without losing any data.

Note the smallest-drive term in Synology's formulas. In an array of mixed sizes, every drive counts as the smallest one, so a single smaller disk caps the usable capacity of the whole group. Buy matching capacities for a parity array.

RAID 5 vs RAID 6 vs RAID 10

RAID 10 takes a different route to protection. QNAP describes it as every two disks paired using RAID 1, with all pairs then striped together using RAID 0. Synology says RAID 10 provides the performance of RAID 0 in addition to the same data protection level as RAID 1. QNAP credits it with excellent random read and write speeds and high failure protection, but half the total disk capacity is lost, and it recommends the level for databases.

Synology gives the capacity as the capacity of the smallest drive times the number of drives divided by two, and says RAID 10 combines at least four drives and that the number of drives must be even. Its failure tolerance works per pair. RAID 10 survives one failed disk in each mirrored pair, so it can lose more than one drive and keep going, but two failures in the same pair still lose the array. RAID 6 survives any two failures, wherever they fall. RAID 5 survives one.

In short, RAID 10 trades capacity for random-access speed, RAID 6 trades capacity and some write speed for a second failure, and RAID 5 keeps the most capacity with the least margin.

RAID 5 vs RAID 6 minimum disks

• RAID 5: at least three drives (Synology and Dell).
• RAID 6: at least four drives (Synology).
• RAID 10: at least four drives, and always an even number (Synology).
• ZFS raidz: one more device than the number of parity disks (OpenZFS), so two for raidz1 and three for raidz2 at the bare minimum.

A minimum is not a sensible size. The fewer data disks in the group, the larger the share of raw capacity spent on parity: by Synology's formulas, a four-drive RAID 6 gives the same usable capacity as a four-drive RAID 10. IBM's support note for its MegaRAID-based ServeRAID controllers goes further and says RAID 6 with fewer than five drives will suffer a performance penalty, so a RAID 10 array is recommended on those controllers. If you have exactly four bays, that is the choice to weigh.

RAID 5 vs RAID 6 performance

The cost of RAID 6 shows up on writes. Dell's RAID table lists its disadvantage as a write performance decrease due to dual parity calculations: every write has to update two parity blocks instead of one. QNAP, for its part, recommends RAID 6 for critical data protection, business and general storage use.

Rebuilds are the other performance question. Dell's table gives RAID 5 a longer rebuild due to parity recalculation, and during a rebuild a single-parity array has no spare protection left. RAID 10's mirrored pairs are QNAP's pick for random reads and writes, and IBM's pick over RAID 6 on small arrays for its MegaRAID-based ServeRAID controllers.

We have not found a vendor document that gives a numeric RAID 5 vs RAID 6 write penalty, so this page prints none. If write speed decides it, test your own workload on the controller or NAS you will use.

A worked example: six 16 TB drives

Applying the Synology and QNAP capacity rules to six 16 TB drives gives the figures below. This is our arithmetic from the formulas above, before file system overhead, not a figure either maker printed.

• RAID 5: six drives minus one is 80 TB usable; survives one drive failure.
• RAID 6: six drives minus two is 64 TB usable; survives any two drive failures.
• RAID 10: half of 96 TB is 48 TB usable; survives one failure per mirrored pair.

So on six disks, RAID 6 gives up 16 TB against RAID 5 to survive a second failure, and RAID 10 gives up a further 16 TB against RAID 6 for its random-access speed.

Rebuilds and read-error ratings: why drive class matters for RAID 5

When a RAID 5 array loses a drive, rebuilding it means reading every surviving drive in full. Makers rate how often a drive may fail to read a sector, and the rating differs by class. Seagate rates its BarraCuda 3.5-inch 3TB to 8TB drives at 1 nonrecoverable read error per 10^14 bits read. It rates the IronWolf Pro 16TB to 24TB at 1 per 10^15 bits read, maximum, and the Exos X24 at fewer than 1 in 10^15.

The rating can change by capacity within one line. Toshiba's N300 data sheet rates its 12 TB, 14 TB, 16 TB and 18 TB models at an unrecoverable error rate of 1 per 10^14, and its 20 TB, 22 TB and 4 TB to 10 TB models at 1 per 10^15. Check the sheet for the exact capacity and part number you are buying.

Our arithmetic shows why this matters for single parity. In the six-drive example, a RAID 5 rebuild must read five surviving 16 TB drives: 80 TB, or about 6.4 x 10^14 bits. At a rating of 1 per 10^14, that is an expected 6.4 unreadable sectors; at 1 per 10^15, an expected 0.64. These are maker maximum rates turned into an expected value, not a prediction that a rebuild will fail.

TrueNAS reaches similar advice for a different reason: time. Its ZFS primer says the larger the vdev, the longer the resilvering time, and that when replacing a disk in a RAIDZ, another disk could fail before the resilvering process completes. For this reason, it says, RAIDZ1 is not recommended for drives over 1 TiB in size, and it adds that RAIDZ2 offers better data availability and significantly better mean time to data loss than RAIDZ1. Every drive this guide discusses is far larger than 1 TiB.

RAID 5 vs RAID 6 calculator

The calculator starts at RAID 6 with six 16 TB drives, the example above. Change the level to 5 to see the extra capacity single parity gives, or change the drive count and size to match the bays you have.

For more layouts and a drive price block, our English RAID planner covers further levels.

RAID 5 or RAID 6: which to choose

• Large drives, or data you cannot easily replace: RAID 6, or raidz2 on ZFS. TrueNAS advises against RAIDZ1 above 1 TiB because another disk could fail before a resilver completes.
• Exactly four bays: RAID 6 and RAID 10 give the same usable capacity, and IBM recommends RAID 10 over RAID 6 below five drives on its MegaRAID-based ServeRAID controllers.
• Three bays, modest drive sizes and a backup elsewhere: RAID 5 is the only parity level that fits Synology's minimums, at one failure of margin.
• Random-access workloads such as databases: RAID 10, which QNAP recommends for them, at half the raw capacity.
• Any single-parity array: drives rated at 1 per 10^15 rather than 1 per 10^14 lower the expected read errors during a rebuild, by the maker's own ratings.

Whichever level you choose, RAID keeps an array running through a drive failure; it does not protect against deleted files or a failed array, so keep a separate backup.

Who computes the parity is a separate decision, covered on hardware RAID vs software RAID. For drives rated at 1 per 10^15 or better, see Seagate Exos X24 and IronWolf Pro, and for where the array lives, DAS vs NAS.