U3 Storage and Data
Storage and NAS: where your data should live
Almost every homelab starts with the same question: where do I put my stuff? The answer shapes everything else, so it is worth planning storage for the next four years instead of the next four weeks.
At a glance
- Keep the NAS a storage appliance on its own hardware. Run applications on other machines that mount its shares.
- Filesystems with checksums need direct access to each disk. Use a host bus adapter in pass-through mode, not a hardware RAID card.
- Buy conventional magnetic recording (CMR) drives for any redundant array. Shingled drives can turn a routine rebuild into a disaster.
- Plan capacity for three to four years, and plan to keep the pool below about 80 percent full.
- Redundancy keeps you running when a drive fails. It is not a backup.
Why storage gets its own box
It is tempting to put the shares, the hypervisor, the media server and every container on one machine. The trouble starts on the first day you need to update or repair it. Storage that is shared with everything else gets rebooted with everything else, and a problem in one application can take your data offline.
A dedicated NAS can be modest. It needs enough bays, reliable disk controllers, adequate memory and good network connectivity. It does not need a fast CPU unless it transcodes video. Applications then run on hypervisor hosts or container hosts that mount the NAS over the network. When an application breaks, the data does not care.
Your options
| Approach | Strengths | Trade-offs |
|---|---|---|
| Turnkey NAS appliance | Easy setup, low power, quiet, polished apps for photo backup and PC backup, simple expansion one drive at a time on many models | Proprietary parts of the software, limited memory and expansion, you depend on the vendor's update and support policy |
| DIY build with a ZFS-based NAS distribution | Strong data integrity, snapshots, replication, no licensing or activation, any hardware you choose | Requires planning of drive groups up front, more to learn, you are the support desk |
| NAS operating system that pools mixed drives | Mix drive sizes freely, add one drive at a time, friendly application ecosystem | Usually slower than striped arrays, different integrity features, often licensed per drive count |
| General Linux server with shares | Maximum control, excellent for learning how sharing really works | Everything is configured and maintained by hand |
There is no wrong answer here, only a question of values. If your priority is ease, low power and a household that can use the photo app without you, a turnkey unit is a legitimate choice. If your priority is control, openness and learning, build your own. Check the vendor's published support lifecycle and security track record before buying any appliance.
Hardware that matters
Disk controllers
A hardware RAID card presents one large virtual disk to the operating system and hides the individual drives. ZFS and similar filesystems need to see every disk directly so they can checksum, scrub and repair data. Use the motherboard's SATA ports or a host bus adapter (HBA) flashed to pass-through mode, often called IT mode. Many used servers can be bought with the controller already in that mode.
Boot drives
The NAS operating system belongs on its own small device, separate from the data drives. A mirrored pair of small, inexpensive SSDs is ideal; 32 to 64 GB is plenty, and fast NVMe is wasted here. Avoid USB flash drives for boot, because they wear out. Most NAS distributions keep their whole configuration in a single exportable file. Save it after every change, and a dead boot drive becomes a 20 minute reinstall.
Memory and CPU
File serving needs little CPU. Memory is used as a read cache, so more is better, but the old rules about a fixed amount of RAM per terabyte are not real requirements. A small pool runs well on 8 GB; 16 to 32 GB is comfortable for most homes. ECC memory is worth having on a machine that holds important data, but it is not a prerequisite.
Network
Gigabit is fine for household file sharing and streaming. If the NAS stores virtual machine disks, receives large backups or serves video editing, a 10 GbE link to the machines that use it pays off quickly. See Home Network Fundamentals.
Power and noise
A careful DIY build with four to six drives can idle around 40 to 60 W. A used storage server with many bays may idle at well over 100 W before drives are counted. Spinning drives add roughly 4 to 8 W each. Check fan noise and idle draw before you buy, and see Racks, Power and Cooling.
Choosing drives
- Insist on CMR. Shingled magnetic recording (SMR) drives write in overlapping bands and slow to a crawl under sustained writes. In a redundant array that can make a rebuild take days or fail entirely. Check the recording technology of the exact model before buying.
- Buy drives meant for always-on arrays. NAS and enterprise models handle vibration, heat and constant operation better than desktop drives.
- Mix batches. Drives from the same production batch can share a defect and fail close together. Buy from more than one seller or spread purchases over time.
- Burn in every drive. Run a full surface test and a long self-test before adding a drive to a pool. Failures cluster early in a drive's life.
- Keep a spare. A matching spare on the shelf turns a failure into a 20 minute job instead of a week of anxiety.
- Watch temperatures. Heat shortens drive life. Keep drives below about 45 degrees Celsius.
Drive and memory prices rose sharply during 2025 and 2026. Recertified and used enterprise drives are a reasonable way to stretch a budget if you test them thoroughly, buy from sellers who offer a warranty, and treat redundancy and backups as non-negotiable.
Myth
"Drive health reports will warn me before a failure." Built-in drive health data is useful and you should monitor it, but many drives fail with clean reports, and some report problems for years without failing. Redundancy, scrubs and backups protect you; health reports only sometimes warn you.
Sharing protocols
| Protocol | Best for | Notes |
|---|---|---|
| SMB | Desktops and laptops of every operating system | Supports per-user permissions; snapshots can appear to users as previous versions of files |
| NFS | Linux servers, containers and hypervisor storage | Simple and fast; restrict exports to specific hosts or networks |
| iSCSI | Block storage for a specific host or application | Looks like a local disk; one mistake can corrupt the whole volume |
For virtual machine storage in a homelab, NFS is usually the better choice. Each VM disk is an ordinary, thinly provisioned file you can snapshot, copy and recover individually, and it is far more forgiving than iSCSI when something goes wrong. Carry NFS and iSCSI traffic between the NAS and your hypervisor hosts on a dedicated storage network that does not pass through the firewall. To share one NAS with many containers, follow Shared Storage for Containers.
Layout and permissions
Design the structure of your storage on paper before you copy data onto it. Moving data between separate filesystems or datasets later is a full copy, not a quick rename.
- Create separate datasets or shares for different kinds of data, such as documents, photos, media, backups and VM storage. Each can have its own snapshot schedule, permissions and settings.
- Use separate credentials for administration and everyday file access. A compromised laptop should not hold the keys to delete snapshots.
- Bind the admin interface only to the management network, and file services only to the networks that need them.
Applications on the NAS
Most NAS platforms offer plugins and container apps. They are convenient, but the versions they offer often lag behind the projects themselves, and every app adds load and attack surface to the box that holds your data. Our approach: keep the NAS for storage, snapshots, replication and a file-sync service that needs local access to the data. Run everything else on hypervisor or container hosts.
Capacity planning
Estimate what you store today, add your yearly growth and plan for three to four years. Then remember three things:
- Redundancy costs capacity. Mirrors use half the raw space; double parity across six drives uses a third.
- Drive labels use decimal terabytes while operating systems report binary tebibytes, so a 10 TB drive shows as about 9.1 TiB.
- Copy-on-write filesystems slow down and fragment as they fill. Plan to stay below about 80 percent.
Our Storage Capacity Calculator works through these numbers for any drive count and layout.
Distributed storage
Distributed storage systems spread data across several servers and heal themselves when a disk or a whole node fails. They are impressive and worth learning if your career points that way. They also want at least three, and preferably five or more, nodes, fast dedicated networking and real operational attention. Their strength is availability and scale, not raw speed. For most homes, one well-built NAS plus good backups is simpler, faster and safer.