Author here. Correct link: https://github.com/VincentDu2021/uring_sync Key findings:

  - 4.2x faster than cp for 100K × 100KB files (local NVMe)
  - 58% faster than rsync using kTLS for network transfer
  - Single worker + deep queue beats multi-threading

Happy to answer questions about io_uring quirks or benchmarking methodology.