Filesystems

The candidates

★ ext4 — boring default; works everywhere; reasonable speed; no fancy features.
★ ZFS (OpenZFS) — copy-on-write; snapshots; checksums; send/receive replication; pools across many disks. The default for "I care about my data."
btrfs — copy-on-write; snapshots; subvolumes; native to recent Linux. Stable for single-disk and RAID-1; RAID-5/6 still has issues.
XFS — fast for big files; the default on Red Hat / CentOS Stream / Rocky / Alma.
f2fs — flash-optimized; common on Android.
bcachefs — newer copy-on-write FS; in mainline Linux 6.7+; promising but not yet ZFS-stable.
NTFS / exFAT / APFS — non-Linux; you'll touch them when interfacing with other OSes.

Cloud / VM root disks: ext4 or XFS. Boring is right.
Bulk storage / NAS: ZFS. Snapshots + integrity checking are worth it.
Single-disk laptop with snapshots: btrfs (or ZFS if you've installed the modules).
Database hosts: XFS for many DBs (Postgres / MySQL); some prefer ext4. Don't run on btrfs.
Container hosts: ext4 or XFS for the Docker dir; CoW issues plus btrfs is a pain combo.
NVMe / mobile / phones: f2fs.

Pools and datasets — pool is the storage; datasets are mountable subdivisions.
Compression on by default — zstd is free space.
Snapshots — zfs snapshot pool/data@daily-2026-05-09. Cheap and instant.
zfs send | zfs receive — replicate datasets between hosts; encrypted backup over SSH.
ARC sizing — ZFS uses memory aggressively. Limit zfs_arc_max on hosts that run other things.
Avoid mixing disk sizes in vdevs.
RAID-Z2 minimum for important data; mirrors for performance.
Scrubs — schedule monthly; zpool scrub checks integrity.

Subvolumes are like cheap separate filesystems.
Snapshots are instant.
btrfs send | btrfs receive — same idea as ZFS.
Avoid btrfs RAID 5/6 in production. Use RAID-1 / RAID-10 or use mdadm + ext4/XFS underneath.
Scrub regularly — btrfs scrub start /mnt/data.

ZFS / btrfs snapshots protect against fat-finger / ransomware on the same pool.
Only off-host replication (send/receive to a different machine) protects against the host dying.
See Backup.

★ ZFS for any system whose data matters. The integrity story is unique.
Compression on. zstd: free perf + space.
Schedule scrubs. Monthly is fine for HDDs; weekly for SSDs.
Atomic snapshots before risky changes (zfs snapshot + apt upgrade).
Don't run DB-heavy workloads on btrfs. Postgres + btrfs has well-documented gotchas; XFS/ext4 is preferred.