ZFS Mirror vs RAID 1

A ZFS mirror is a mirror vdev inside a ZFS pool. Two (or more) disks hold identical copies, but ZFS sits on top as both the volume manager and the filesystem, so it knows what every block is supposed to contain. The Proxmox and TrueNAS installers label this option RAID1 or mirror, which is why people search for "RAID 1 ZFS" in the first place. Under the hood it is a copy-on-write mirror, not classic block RAID.

Traditional RAID 1 mirrors two disks below the filesystem. On Linux that is usually mdadm software RAID: the kernel presents a single /dev/md0 device and you put ext4 or XFS on top. It can also be a hardware RAID controller that mirrors in firmware and shows the OS one disk. Either way, the mirror layer copies blocks around without understanding the data inside them.

This is the whole reason to pick ZFS. Every block in a ZFS pool carries a checksum. On read, ZFS verifies the block against that checksum, and if it fails, ZFS pulls the good copy from the other mirror member, returns correct data, and rewrites the bad block. A scheduled scrub walks the entire pool doing the same check proactively. Silent corruption, the kind that comes from bit rot, a flaky SATA cable, or a drive returning wrong bytes without flagging an error, gets caught and fixed.

Traditional RAID 1 has no such layer. mdadm mirrors blocks but stores no checksum, so when the two copies disagree it cannot tell which one is correct. A `check` scrub only counts mismatches; a `repair` blindly copies one side over the other and can overwrite the good copy with the bad one. Hardware RAID 1 is the same story with even less visibility. The data is duplicated, but its correctness is never verified, and for long-lived NAS data that is the gap that matters.

The most repeated objection to ZFS is RAM. The "1 GB of RAM per TB of storage" rule came from old FreeNAS guidance and was always conservative, aimed at large arrays running deduplication. For a two-disk mirror it is overstated. ZFS runs fine on a home NAS with 8 GB; the ARC simply borrows whatever RAM is free as a read cache and hands it back under pressure. Only turn on dedup if you have studied it and have the memory for it, because that is the feature that actually drives the heavy RAM math.

On performance, both mirror types read from both drives, so sequential reads scale similarly and a single disk already saturates a gigabit link. ZFS adds ARC caching, optional compression (lz4 is nearly free and often a net speedup), and copy-on-write, which sidesteps the RAID 1 write-hole risk on a sudden power loss. mdadm carries slightly lower CPU overhead on very weak hardware. For a homelab, the integrity and snapshot features outweigh that small CPU difference.

ZFS pools are portable. Pull both disks, plug them into any machine running OpenZFS (TrueNAS, Proxmox, Ubuntu, FreeBSD), run `zpool import`, and the pool comes up with its data, snapshots, and settings intact. There is no controller to match and no metadata to rebuild by hand.

mdadm is similarly portable across Linux: `mdadm --assemble --scan` reassembles the array on any modern kernel. Hardware RAID 1 is the weak link. The on-disk layout is the controller's own format, so recovering the array can require the same card model or a compatible one. When that card dies three years later, sourcing a match becomes a real problem. This is exactly why ZFS people tell you to flash the HBA to IT mode and give ZFS the raw disks rather than building a controller mirror underneath it.

The other half of the "RAID 1 ZFS" question is mirror versus RAIDZ1, the ZFS take on RAID 5. On two disks there is no contest, since a mirror is the only redundant option. On three or more disks a mirror still resilvers faster, expands more easily (add another two-disk mirror vdev whenever you want), and gives better random IOPS, at the cost of holding only 50 percent of raw capacity.

RAIDZ1 spreads single parity across three or more disks for more usable space, but resilvers are slower and stress every disk, and the single-parity rebuild risk on large drives is the same warning that applies to RAID 5. For a two- to four-disk homelab pool that holds data you would miss, mirrors keep things simple and fast; reach for RAIDZ2 when capacity matters more than easy expansion. Run your disk count through the RAID and ZFS calculator to see usable space for each layout, and if you are still picking a filesystem see Btrfs vs ZFS.