UPS for a NAS: Protect Your Data, Shut Down Cleanly, Come Back Automatically

What actually goes wrong when a NAS loses power

A NAS rarely loses its whole volume from one power cut. The damage is usually quieter: a file that was being written is truncated, a database inside a container is left inconsistent, or a parity array has to run a long consistency check after reboot. The risk compounds with every unplanned cut, and some of it is invisible until you need the data.

Write caches at three levels

Data headed for disk sits in volatile memory at several points: the operating system's page cache (often several seconds of writes), any SSD read-write cache you have configured, and each hard drive's own onboard DRAM cache. A filesystem that issues flush commands correctly will survive losing the in-flight data, but the application that thought it had saved a file may not. Consumer SSDs used as NAS write caches typically lack power-loss protection capacitors, which is one reason several NAS makers warn about read-write SSD caching without a UPS.

The RAID write hole

In RAID 5 and RAID 6 (including hybrid schemes built on them, such as Synology SHR), updating a stripe means writing data blocks and a parity block. Those writes are not atomic. If power fails between them, parity no longer matches data, and the array cannot tell which is wrong. Nothing looks broken until a drive fails and the rebuild reconstructs garbage from the stale parity. Some implementations mitigate this (a journal or partial parity log), and copy-on-write designs like ZFS RAIDZ avoid it by never overwriting live blocks, but a clean shutdown sidesteps the problem entirely.

Filesystem and metadata integrity

Journaled filesystems (ext4, XFS) and copy-on-write filesystems (Btrfs, ZFS) are designed to remount consistently after a crash. "Consistent" means the metadata is intact, not that every recent write survived. After an unclean shutdown many NAS operating systems also schedule a full parity scrub or array resync, which can run for many hours on large drives, during which performance drops and a second failure is more dangerous.

How much power a NAS draws

NAS power draw is dominated by the hard drives. A 3.5-inch drive typically idles at 3 to 6 W and reads or writes at 5 to 9 W; the NAS board, fans and CPU add 10 to 30 W for compact units and more for x86 boxes with a discrete network card. Measure your own unit if you can (see how to measure power draw), or start from the device power draw database.

Spin-up surge

A 3.5-inch drive pulls roughly 1.5 to 2.5 A from its 12 V rail for a few seconds while the platters accelerate, so 20 to 30 W per drive. Most NAS units stagger spin-up so the peak never stacks fully. This matters for the UPS only at boot, when you are usually back on mains, and when drives wake from spin-down during an outage. It is well inside the surge tolerance of any UPS sized for the steady load, so you do not need to oversize for it.

Waveform

Small NAS units with external power bricks are tolerant of simulated sine output. Larger rackmount NAS and DIY servers with active-PFC power supplies can occasionally shut off or buzz on a stepped waveform at the moment of transfer. If your NAS has an internal ATX or server supply, a pure sine UPS removes that risk; see pure vs simulated sine wave.

Worked example: a 4-bay NAS with its network

A home has a 4-bay NAS (measured 38 W during normal use), a router (12 W) and a small switch (8 W). The router and switch must be on the UPS so other devices can keep reaching the NAS and so network shutdown signals get through. Total: 58 W.

Candidate: a typical 850 VA / 510 W line-interactive unit with one 12 V 9 Ah battery (about 108 Wh nominal).

• New battery: 108 Wh x 0.80 x 60 / 58 W = about 89 minutes. (We use 0.80 rather than 0.90 because small inverters are less efficient at light load.)
• Battery at three years, assuming about 70% of original capacity: 108 x 0.70 x 0.80 x 60 / 58 = about 63 minutes.
• Load as a share of the watt rating: 58 / 510 = 11%. Plenty of headroom for drive wake-ups.

The NAS needs perhaps 3 minutes to shut down. So the sensible plan is to ride through short blips, then shut down after 5 minutes on battery, leaving well over half the battery in reserve. That reserve is what makes the setup robust when the battery ages and the runtime estimate becomes optimistic. The UPS runtime calculator models battery aging and nonlinear discharge more carefully than this formula.

Getting the shutdown signal to the NAS

A UPS without a signal path just delays the crash. There are two ways to deliver the signal.

Most NAS platforms (Synology DSM, QNAP QTS, TrueNAS, Unraid) use Network UPS Tools under the hood, which is why a Synology can act as a UPS server for a TrueNAS box and vice versa. Platform walk-throughs: Synology, TrueNAS, Unraid. QNAP's settings live under its external device or UPS section and follow the same logic.

If the NAS does not detect the UPS over USB, check the cable (many UPS units use a special USB-A to RJ45 or USB-B cable) and see UPS not detected.

Shutdown timing: build a budget

Choose the trigger deliberately. NAS settings typically offer either "after X minutes on battery" or "when the battery reaches low battery". The time-based trigger is more predictable, because a UPS's low-battery estimate is based on a model of a healthy battery and is often too optimistic on an old one.

1. Measure the shutdown duration. Trigger a normal shutdown during typical activity and time it until the power draw falls to standby.
2. Add client shutdown time. If PCs or a hypervisor use the NAS as storage, they must finish first (see below).
3. Set the trigger. A common choice is 2 to 5 minutes on battery. Most outages that end on their own end within a minute or two, so waiting longer mostly spends battery.
4. Check the remaining margin. Trigger time plus shutdown time should use no more than about half of your aged-battery runtime estimate.
5. Test it. Pull the UPS plug from the wall during a quiet period and watch the whole sequence. Repeat after any software update.

Multiple clients that depend on the NAS

If a hypervisor runs VMs from NFS or iSCSI storage on the NAS, the VMs and host must shut down before the NAS, or the VMs lose their disks mid-shutdown. In NUT terms, the NAS that owns the USB link is the primary and the others are secondaries. When the primary decides to shut down it raises a forced-shutdown flag, and by default it waits for secondaries to disconnect (up to a configurable time) before shutting itself down. Give clients a shorter trigger than the NAS, or rely on that wait, but test that the order is right. The homelab server guide covers multi-host ordering in detail.

Safe mode, power-off and auto restart

Platforms differ in what "shutdown" means. Synology DSM, for example, can enter a Safe Mode on UPS signal: services stop and volumes unmount, but the unit may not fully power off. Other platforms perform a full OS shutdown. Behavior also differs between software versions, so read the help page for your exact version.

Then enable restart. On a NAS this is usually a setting such as "restart automatically when power supply issue is fixed" or "restore to previous power state". On a DIY TrueNAS or Unraid build it is the BIOS/UEFI option often called Restore on AC Power Loss set to Power On or Last State.

Choosing the UPS

Prefer a model whose USB HID interface is listed as compatible with your NAS (or in the NUT hardware list). Line-interactive is the right topology for nearly all home NAS use; online units help mainly where mains voltage is chronically unstable.

Common mistakes

• No data cable. The NAS sits on battery until the UPS dies, then crashes anyway.
• Network gear left off the UPS. NUT clients lose contact with the server the moment power fails and cannot be told to shut down. Some clients treat a lost link as "assume on battery", others do nothing.
• Shutting down at "low battery" on a five-year-old battery. The battery collapses before the shutdown completes. See when to replace a UPS battery.
• Plugging the NAS into a surge-only outlet. Many UPS units have both kinds; only the battery-backed ones carry the load in an outage.
• Never testing. A shutdown chain that has never run is a guess.

When a UPS is not the whole answer

A UPS protects against power events, not against the far more common causes of data loss: drive failure, ransomware, accidental deletion and fire. It is not a backup. If the NAS holds the only copy of anything, a second copy elsewhere is worth more than any UPS. And if your goal is keeping a media server streaming through a four-hour outage, a UPS is the wrong tool for the duration: a portable power station or a generator with a UPS in front of it is the better fit.