How the other layouts compare using the same number and size of disks. Your current choice is highlighted.
| Layout | Geometry | Usable | Efficiency | Survives losing |
|---|
These are estimates. Real pools also spend space on metadata, and gain it back with compression. Efficiency is shown for the record size selected above; small records on wide RAIDZ or dRAID vdevs store less efficiently.
RAIDZ is the classic ZFS parity layout: each vdev is one group of disks with one, two, or three disks' worth of parity, and a failed disk is rebuilt onto a separate hot spare by walking the whole block tree.
dRAID (distributed RAID) spreads fixed-width parity groups and spare capacity across every disk in the vdev. When a disk fails, all the surviving disks rebuild into the distributed spare space at once, sequentially, so redundancy is restored in a fraction of the time. The trade-off is a fixed stripe width, which wastes space on small blocks.
dRAID pays off on large pools, typically 20 or more disks per vdev, built from high-capacity drives where a traditional resilver would take days. For smaller pools, RAIDZ2 is simpler, stores small blocks more efficiently, and is the usual recommendation.
If a dRAID pool will hold many small files or small-block zvols, add a mirrored special vdev for them so they do not each consume a full stripe.
Four things add up: parity (and distributed spares on dRAID); padding, because RAIDZ rounds every allocation up to a multiple of parity + 1 sectors and dRAID pads to a full stripe; the slop reserve, about 1/32 of the pool held back so ZFS can always complete a delete; and units. Drives are sold in decimal terabytes, but ZFS reports binary tebibytes, so a 16 TB disk shows as roughly 14.55 TiB before any overhead.
Parity is calculated per record, not per disk. A large record spreads across the full width of the vdev and reaches the ideal data-to-parity ratio. A small record, such as a 16K zvol block on a wide RAIDZ2, still needs its full parity plus padding, so efficiency can fall well below the headline figure. Try the 16K setting on a wide vdev above to see the effect.
ZFS is copy-on-write, so it needs free space to write new data before releasing the old. Performance and fragmentation degrade as a pool fills; planning to stay under about 80% is the common rule of thumb, which is why the calculator shows that figure alongside the absolute usable capacity.
OpenZFS recommends between 3 and 9 disks per RAIDZ group. Wider vdevs store slightly more but resilver more slowly and deliver the random IOPS of roughly a single disk each. For more performance, use several narrower vdevs: a pool stripes across all of its vdevs.