Performance.

① Single-file I/O — YoloFS matches bare ext4.

Single-threaded reads and writes on a 1 GB file with 4 KB requests. YoloFS adds no measurable overhead on the staging area. OverlayFS and BranchFS both pay a cost even on simple pass-through.

② Metadata — faster than OverlayFS, sometimes faster than ext4.

For metadata operations, the target file may live in the base filesystem, a snapshot, or the staging area. YoloFS is faster than OverlayFS across the board. readdir, rename, and unlink on staged files are even faster than ext4 — the in-memory override tree resolves them without a directory-walk syscall. BranchFS is over 20× slower than the others. Adding permission control adds negligible overhead; 4% on stat is the worst case.

③ Snapshot scalability — flat as snapshots grow.

We create a series of snapshots by overwriting a set of ten files and measure whether subsequent operations slow down. YoloFS reads and writes are unaffected — it always keeps a single current override tree, so common operations don't traverse past snapshots. OverlayFS fails past ~50 snapshots (the mount-option length exceeds the kernel limit). BranchFS slows linearly. Commit time is inherently linear in the number of snapshots; YoloFS still commits faster because it reads from a single journal file instead of querying the kernel per snapshot.

④ Realistic workload — as fast as ext4 on a real kernel patch series.

Adapted from an actual Linux kernel patch series: set up a git worktree, build the kernel, then for each patch search/edit/rebuild/ commit, finally commit all changes to base.

• YoloFS — as fast as ext4 across the workflow, with only ~3.5 s of extra time to commit over 100k files.
• OverlayFS — 18% slower: extra time in commands plus remounting whenever a snapshot is taken.
• BranchFS — has a bug and can't run past the initial build; adds over 2 minutes to a 20-second build before failing — a large overhead even compared with LLM response latency.