Homelab Storage: Planning an Array, Choosing Drives and Keeping the Power Down

Homelab storage setup: the order of decisions

A homelab storage server is usually several disks pooled together and running all the time, holding virtual machines, backups and media. The decisions come in an order: the software, the pool layout, the controller, the drives, and then what the whole array draws from the wall. Get the first three right and the drive choice narrows on its own.

This page quotes TrueNAS's documentation for the pool, controller and power settings, and the OpenZFS documentation for the rules beneath them. Unraid, Proxmox, Synology and Windows have their own rules, and nothing here should be taken as a statement about them.

Homelab storage options: mirror or RAIDZ

TrueNAS describes the trade plainly. A mirror vdev's storage capacity is the size of a single disk in the vdev, however many disks it mirrors. RAIDZ2 reserves two disks for parity, and two disks can fail without data loss to the pool. TrueNAS recommends defaulting to a RAIDZ layout generally. Its pool documentation was last modified on 16 October 2026.

OpenZFS states the general rule behind those examples. A RAIDZ group of N disks of size X with P parity disks holds approximately (N minus P) times X and can withstand P devices failing without losing data. The minimum number of devices in a RAIDZ group is one more than the number of parity disks, so three for RAIDZ2, and the manual recommends between 3 and 9 devices to help performance.

Capacity is not the only trade. OpenZFS's tuning guide says that if small random IOPS matter most, mirrored vdevs outperform RAIDZ, because read IOPS on mirrors scale with the number of drives in each mirror while each RAIDZ vdev is limited to the IOPS of its slowest drive. If sequential writes matter most, RAIDZ outperforms mirrors. A pool of virtual machine disks and a media archive therefore ask different questions of the layout.

Worked through for six 20 TB disks, before filesystem overhead (we have not found a document that sizes it): as three two-disk mirrors, usable space is three disks, or 60 TB; as one six-disk RAIDZ2, it is four disks, or 80 TB, and two of the six can fail without data loss. With twelve disks the same arithmetic gives 120 TB as six mirrors against 160 TB as two six-disk RAIDZ2 vdevs. A single twelve-disk RAIDZ2 would hold about ten disks of data, but it sits above the 3 to 9 devices OpenZFS recommends, which is why the example splits it in two. These are our sums from the TrueNAS and OpenZFS statements.

The RAID calculator below works out usable capacity for conventional RAID levels; use it to see how drive count and parity trade against space before you buy anything.

What to avoid in a homelab storage array

TrueNAS's hardware guide, last modified on 11 September 2026, warns against two things directly:

• Drive-managed SMR (DM-SMR) and host-aware SMR (HA-SMR) drives. TrueNAS says they function poorly with TrueNAS, and to expect degraded write performance and longer, less reliable resilver operations. Host-managed SMR (HM-SMR) drives are not supported.
• A hardware RAID card's own RAID. If the card supports HBA mode, also known as passthrough or JBOD mode, TrueNAS says not to use its RAID facility.

The SMR warning covers write performance and resilvers, the rebuild after a disk is replaced, which is when a pool is already a disk down. A seller's title that says CMR is a claim, not a source. A maker's own statement is firmer: Seagate states every IronWolf Pro drive uses CMR.

Homelab storage setup: settings to get right before the pool fills

Some choices cannot be undone. OpenZFS says a vdev's ashift, its alignment shift, is set at vdev creation and is immutable, so it has to suit the drives from the first day.

Others are defaults you can tune dataset by dataset. ZFS datasets use an internal recordsize of 128KB by default, and the property belongs to each dataset, so a dataset of virtual machine disks and a dataset of large media files need not share one value.

A special vdev is an option for a busy pool: a device dedicated to the pool's internal metadata and, optionally, small file blocks. OpenZFS says its redundancy should match or exceed the redundancy of the pool's other normal devices. Small file blocks only go there if you opt in, dataset by dataset, by setting the special_small_blocks property to a nonzero size.

Homelab storage array drives: bay ratings decide the shortlist

For an array, the bay rating on a NAS drive's sheet rules drives in or out before price does. Toshiba rates the N300 20TB for up to 8 bays and Seagate the IronWolf Pro ST20000NE000 for up to 24, while the IronWolf Pro ST20000NT001 is rated for unlimited bays and WD sells the Red Pro for RAID-optimised NAS systems with no bay limit. An eight-bay box filled with N300s sits exactly at its limit; a twelve-bay chassis rules it out.

Enterprise drives are positioned for this kind of work too: Seagate describes the Exos X20 as for high-capacity density RAID storage. The two IronWolf Pro 20TB drives share a name, so the part number is what tells you which bay rating you are buying.

Workload ratings, warranties, noise and idle power drive by drive, NAS-class against enterprise at 20 TB, are compared on home server hard drive; this page keeps to what changes when there are many drives.

What a homelab storage array draws, in kWh

An array multiplies idle power by the disk count. At the makers' idle figures, six Seagate Exos X20 drives at 5.4 W draw 32.4 W, about 284 kWh a year, and six IronWolf Pro 20TB at 5.5 W draw 33 W, about 289 kWh a year. Twelve disks double either figure. These are our sums, for the disks alone, with each sheet measuring idle under its own conditions; the board, processor, controller and fans come on top, so meter the whole server at the wall for a real figure.

To turn kWh into money, multiply by your own unit rate. Ofgem's cap average for 1 October to 31 December 2026 is 26.32 pence per kWh by Direct Debit, a reference point rather than your tariff.

Spinning disks down: what TrueNAS offers and what it saves

TrueNAS SCALE sets spin-down per disk. Its HDD Standby setting chooses the minutes of inactivity before a drive enters standby: Always On, or 5, 10, 20, 30, 60, 120, 240, 300 or 330 minutes.

It also exposes Advanced Power Management levels. TrueNAS describes Level 1 as minimum power usage with standby, which saves the most power, with the lowest draw and the least heat and noise, and responds the slowest.

The IronWolf Pro 20TB sheet shows what is at stake: 5.5 W average idle against 1.0 W typical in standby, a difference of 4.5 W per drive. For six drives asleep twelve hours a day, that is 6 times 4.5 W times 12 hours times 365 days, about 118 kWh a year by our arithmetic. Whether a pool really sleeps that long depends on what touches it, and we have not found a TrueNAS figure for that.

The trade is response time. A pool serving virtual machines through the day gets little sleep and pays the wake-up delay each time, so Always On may suit it better; a backup or archive pool that sits idle for hours is where standby earns its keep.

Used enterprise SATA drives for a homelab array on Amazon.co.uk

The table below answers one homelab question: what a terabyte of used or recertified enterprise SATA storage costs today. It ranks those listings on Amazon.co.uk by lowest pounds per TB, inc VAT. Lowest £/TB in it: £20.49/TB. The table shows 8 listings right now.

Two checks before a used drive joins a pool. The seller's title is a claim, so look up the part number. And given TrueNAS's warning about drive-managed SMR, take the recording technology from a maker document for that exact part number, not from the title; where the table shows one, that is where it comes from. Each row is marked used, and each has a Buy on Amazon button.

Choosing a homelab storage solution: a checklist

• Choose the software first, and find its documentation on SMR and controllers; TrueNAS's is quoted above.
• Choose a layout: TrueNAS defaults to RAIDZ, OpenZFS recommends 3 to 9 devices per RAIDZ group, and mirrors favour small random IOPS.
• Set ashift correctly when you create each vdev, because it cannot be changed later.
• Run any RAID card in HBA mode where it has one, on TrueNAS's advice.
• Avoid DM-SMR and HA-SMR drives; HM-SMR is not supported.
• Check the bay rating against your disk count: up to 8 for the N300 20TB, up to 24 for the IronWolf Pro ST20000NE000.
• Multiply idle watts by the number of disks and by 8.76 to get kWh a year, and decide whether the pool can sleep.

If used enterprise SAS disks are on your list too, the interface question is the subject of SAS vs SATA.