Btrfs vs. ZFS vs. bcachefs Under Real Workloads Synthetic Benchmarks Ignore

A hands-on comparison of three modern Linux copy-on-write filesystems — Btrfs, ZFS, and the newer bcachefs — that argues the standard benchmark suites tell an incomplete story. The author’s framing is that classic tests (sequential and random fio runs, single-metric throughput numbers) are easy to reproduce but poorly represent how these filesystems behave under the messier, mixed I/O patterns of production and desktop use.

The stated goal is to measure the three under those overlooked workloads rather than the idealized ones vendors and reviewers typically publish, putting features like snapshots, compression, and checksumming under conditions closer to real usage. bcachefs is the notable inclusion here: as a recent mainline addition it has far less benchmarking history than the mature Btrfs and ZFS, so head-to-head data on it is still scarce.

Specific methodology, per-workload measurements, and the author’s conclusions were not available in the material provided, so they are not reproduced here to avoid misstating the results.