Don't Use Iperf3 on Windows

Original Article ↗ Hacker News Discussion ↗

Context and main reactions

  • Many agree the article correctly highlights that popular Windows iperf3 builds are outdated Cygwin ports and can underperform, but several note that for typical LAN / 10 Gbps use they see little or no practical issue.
  • Some argue Microsoft would be better off improving/maintaining a native iperf3 for Windows instead of pushing its own tools (ntttcp, ctsTraffic).

Interoperability and real-world usage

  • A recurring objection: Microsoft’s tools assume Windows on both ends, while iperf is ubiquitous on routers, ISP test endpoints, and assorted Linux/BSD systems.
  • Commenters stress that a network testing tool must be cross‑platform and interoperate; single‑platform or non–wire‑compatible tools are seen as non‑starters.
  • ntttcp packages are limited (Debian/Ubuntu/NetBSD); not available on common router OSes like OpenWRT or commercial network gear, limiting its practicality.

Performance and technical points

  • Some report iperf3 on Windows achieving near‑line‑rate 10 Gbps with a single TCP stream, contradicting the article’s more pessimistic numbers.
  • Others accept that Cygwin/syscall translation adds overhead but question why Microsoft doesn’t contribute a native implementation or wire‑compatible tool.
  • Discussion notes that Microsoft’s newer tools target advanced scenarios (high‑latency/high‑bandwidth, QUIC and offload testing) that iperf doesn’t cover well.
  • There is debate over whether performance gaps could be closed with iperf flags (e.g., parallel streams) versus needing entirely new tooling.

WSL and platform behavior

  • WSL is seen as unsuitable for “pure” benchmarking because it introduces emulation/virtualization layers, even if it may outperform the old Cygwin iperf3.
  • Some mention that enabling WSL2 can itself put Windows into a more virtualized mode, further complicating measurements.

CLI UX, packaging, and app distribution

  • Microsoft’s benchmarking tools are criticized for verbose, unfriendly command lines compared to iperf3 -s style UX.
  • Centralized package managers (winget, choco, Linux repos) are praised for simplicity and safety, but there is pushback against a single Microsoft‑controlled store and against app‑store bureaucracy.
  • One thread notes Linux distro packages can be feature‑reduced (e.g., VLC without non‑free components), so “official repo” does not always mean “best binary.”

iperf2 vs iperf3

  • Clarification that iperf2 and iperf3 are separate, non‑interoperable projects, both actively developed, with different strengths.
  • iperf2 on Windows is reported to perform similarly to Linux in localhost tests, suggesting Cygwin overhead may not always be significant.