Modern File Systems

Beyond the toy:

ext4:

• Linux's most common. Mature, fast, journaled.
• Extents, dir_index, online resize.
• 1 EiB max FS, 16 TiB max file.
• 30+ years of bug fixes.

xfs:

• B-trees for everything. High performance.
• Excellent for large files + many.
• No shrink (only grow).
• Default in RHEL since 7.

btrfs:

• COW, snapshots, RAID 0/1/5/6/10.
• Subvolumes (nested file systems).
• Compression, deduplication.
• Stable for most use; some RAID modes still experimental.

ZFS:

• Originally Solaris (now OpenZFS). Mature.
• Pools (zpool) abstract over devices.
• COW, snapshots, send/receive (incremental backup).
• End-to-end checksums (silent corruption detection).
• ARC (cache) integrated; no double-buffering with kernel.
• ZFS on Linux via OpenZFS project.

APFS (macOS, iOS):

• COW, snapshots, encryption-by-default.
• Optimized for SSDs.
• Fast cloning (clone files = COW reference).

F2FS (Flash-friendly FS):

• Designed for SSD/eMMC characteristics.
• Sequential writes (avoid random in flash).
• Cleaner-friendly: garbage collection of obsolete blocks.

FAT/exFAT:

• Simple. No journaling. Universal compatibility.
• Used in USB drives, SD cards, BIOS-mode boot partitions.

NTFS:

• Windows. Transactional. ACLs.
• Read-only support in Linux is mature; write iffy until ntfs3 (2021).

Network FS:

• NFS (RFC 1813, 5661): standard UNIX network FS.
• SMB/CIFS: Windows network FS.
• 9P: Plan 9; used by some VMs.
• FUSE: user-space FS framework. Implement FS in user space.

Distributed FS:

• CephFS, GlusterFS: scale-out.
• HDFS: Hadoop. Optimized for large block sequential.
• GCS/S3: object stores. Not really FS but increasingly used as one.

Choosing:

• General Linux server: ext4 or xfs.
• Large data + snapshots: btrfs or ZFS.
• macOS: APFS (default).
• Windows: NTFS.
• USB drive: exFAT.
• Embedded SSD: F2FS.