Add benchmarks comparing zsync vs rsync vs mutagen

[andrewgazelka]

Goal

Add reproducible benchmarks to README showing zsync's performance advantage. HN loves hard numbers.

Benchmark Scenarios

1. Initial sync (cold)

• Sync a real codebase (~500 files, 50MB) to fresh remote
• Compare: rsync, mutagen, zsync

2. Incremental sync (hot) ⭐ This is where zsync shines

• Change 1 line in a large file
• Change 10 files across codebase
• Add a new file
• Compare transfer size and time

3. Cross-project dedup

• Sync project A, then sync similar project B
• Show that shared chunks aren't re-transferred

Test Setup

• Local: macOS or Linux
• Remote: Linux VM or actual cloud instance (RunPod/Lambda Labs for authenticity)
• Network: Simulate typical cloud latency (~50ms) if local

Metrics to Capture

• Wall clock time
• Bytes transferred (use nethogs or SSH stats)
• CPU usage (optional)

Output Format

| Scenario | rsync | mutagen | zsync |
|----------|-------|---------|-------|
| Initial sync (50MB repo) | 12s | 8s | 6s |
| 1-line change in 10MB file | 10s | 5s | 0.3s |
| 10 scattered file changes | 15s | 8s | 1.2s |

Tools

• hyperfine for timing
• rsync --stats for transfer info
• Custom script to measure zsync transfers

Notes

• Be honest — if rsync wins a scenario, show it
• Use realistic codebases (maybe linux kernel subset, or a real ML project)
• Consider latency sensitivity (zsync's chunking overhead vs rsync's simplicity)