rsy

A blazing fast parallel copy tool for Unix.

Benchmarks

Criterion-validated benchmarks (Linux x86_64, NVMe SSD, 16 cores). Run yourself: cargo bench

Cold copy - first sync, empty destination

10k files x 4KB (39MB total)

rsy src/ dst/

| Tool | Time (ms) | Effective Throughput (MB/s) | Relative to rsy | Time Saved vs Tool | |------|----------:|----------------------------:|----------------:|-------------------:| | rsy ✓ | 126 | 309.5 | 1.00x | - | | rsync | 483 | 80.7 | rsy is 3.83x faster | rsy saves 357 ms (73.9%) | | cp | 245 | 159.2 | rsy is 1.94x faster | rsy saves 119 ms (48.6%) |

Technical note: Small-file workloads are metadata and syscall heavy; rsy benefits from parallel file traversal and copy scheduling.

500 files x 1MB (500MB total)

rsy src/ dst/

| Tool | Time (ms) | Effective Throughput (MB/s) | Relative to rsy | Time Saved vs Tool | |------|----------:|----------------------------:|----------------:|-------------------:| | rsy ✓ | 35 | 14285.7 | 1.00x | - | | rsync | 272 | 1838.2 | rsy is 7.77x faster | rsy saves 237 ms (87.1%) | | cp | 116 | 4310.3 | rsy is 3.31x faster | rsy saves 81 ms (69.8%) |

Technical note: Parallel I/O and lower per-file overhead produce a large latency win.

100 files x 16MB (1.6GB total)

rsy src/ dst/

| Tool | Time (ms) | Effective Throughput (MB/s) | Relative to rsy | Time Saved vs Tool | |------|----------:|----------------------------:|----------------:|-------------------:| | rsy ✓ | 98 | 16326.5 | 1.00x | - | | rsync | 715 | 2237.8 | rsy is 7.30x faster | rsy saves 617 ms (86.3%) | | cp | 335 | 4776.1 | rsy is 3.42x faster | rsy saves 237 ms (70.7%) |

Technical note: On larger files, rsy keeps multiple workers active and sustains higher end-to-end throughput.

Incremental - re-sync with no changes

10k files x 4KB (39MB total)

rsy src/ dst/   # re-run, nothing changed

| Tool | Time (ms) | Relative to rsy | Time Saved vs Tool | |------|----------:|----------------:|-------------------:| | rsy ✓ | 15 | 1.00x | - | | rsync | 69 | rsy is 4.60x faster | rsy saves 54 ms (78.3%) |

Technical note: With no payload transfer, this reflects metadata scan and change detection cost.

500 files x 1MB (500MB total)

rsy src/ dst/   # re-run, nothing changed

| Tool | Time (ms) | Relative to rsy | Time Saved vs Tool | |------|----------:|----------------:|-------------------:| | rsy ✓ | 2.3 | 1.00x | - | | rsync | 45 | rsy is 19.57x faster | rsy saves 42.7 ms (94.9%) |

Technical note: rsy's parallel check path significantly reduces no-op sync latency.

Delta - re-sync with 10% of data changed

10 files x 10MB (100MB total, 10% mutated)

rsy src/ dst/   # re-run after partial changes

| Tool | Time (ms) | Relative to rsy | Time Saved vs Tool | |------|----------:|----------------:|-------------------:| | rsy ✓ | 1.4 | 1.00x | - | | rsync | 44 | rsy is 31.43x faster | rsy saves 42.6 ms (96.8%) |

Technical note: Rolling checksum plus parallel block scanning minimizes both scan and patch time.

Overall: In this benchmark set, rsy wins every case. Speedup vs rsync ranges from 3.83x to 31.43x.

Why rsy

• Parallel I/O — Rayon distributes file work across all CPU cores
• Rolling-checksum delta — only transfers changed bytes; parallel block scanning
• APFS clonefile — O(1) copies on macOS same-volume syncs
• 1.6 MB binary — zero runtime dependencies

Install

# npm
npm install -g @gundu/rsy

# cargo
cargo install rsy

# from source
cargo build --release

Usage

rsy src/ dst/                     # local sync
rsy src/ user@host:dst/           # push to remote
rsy user@host:src/ dst/           # pull from remote
rsy -a src/ dst/                  # archive (times, perms, owner, group)
rsy --delete src/ dst/            # remove dst files absent from src
rsy -n src/ dst/                  # dry run
rsy --log src/ dst/               # timestamped log, no TUI

Flags

| Flag | Description | |------|-------------| | -a | Archive: preserve times, perms, owner, group | | -v | Verbose | | -n | Dry run | | -u | Skip if destination is newer | | -W | Whole file (skip delta) | | -x | Don't cross filesystem boundaries | | -c | Compare by checksum instead of mtime+size | | -z | Compress (accepted, not yet implemented) | | --delete | Delete destination files absent from source | | --exclude PATTERN | Exclude matching files (repeatable) | | --include PATTERN | Include override for exclude rules | | --max-size SIZE | Skip files larger than SIZE (K/M/G) | | --no-tui | Plain output, no interactive display | | --log | Timestamped per-file log to stderr | | --stats | Print transfer statistics at end | | --jobs N | Worker threads (default: 2× CPU cores) |

Exit Codes

| Code | Meaning | |------|---------| | 0 | Success | | 23 | Partial transfer (some files failed) | | 1 | Fatal error |

License

MIT