New Linux benchmark tests corruption, snapshots, rebuilds on 26 layouts
Open-source Linux benchmark tests btrfs, ZFS, bcachefs, ext4 and XFS across 26 configs for corruption, snapshots, rebuilds and ENOSPC handling.
A new open-source project called modern-fs-benchmark evaluates btrfs, ZFS, bcachefs, ext4 and XFS as full multi-device storage systems rather than single-disk fio targets. Its 26 configurations span md/LVM, dm-integrity, native and LUKS encryption, parity layouts, and XFS on a ZFS zvol, testing silent corruption recovery, snapshot aging, rebuild/scrub behavior, and near-full/hard-ENOSPC handling. Results show checksumming filesystems can detect and repair a deliberately corrupted redundant copy, while some classic md/LVM stacks silently returned bad data without surfacing an error.
The hosted CI dashboard runs on loop devices over GitHub VMs, so it's meant for correctness and behavioral comparisons rather than absolute performance. A separate real-hardware dashboard, run on a dual-NVMe Hetzner server donated by bcachefs creator Kent Overstreet, completed three full runs before an NVMe controller failure during a fourth run corrupted part of the RAID0 boot setup; that incomplete run was excluded as a hardware fault, not a filesystem result.
The author is now seeking access to mixed-media hardware (HDD/SSD/NVMe) to test tiering scenarios like ZFS special vdevs, bcachefs foreground/background targets, and dm-cache modes — tests that can't be represented honestly on loop devices sharing one cloud disk. All code, methodology and raw datasets are published under Apache-2.0 and CC BY 4.0.