The world in which IPv6 was a good design (2017)

Original Article ↗ Hacker News Discussion ↗

WiFi, CSMA, and topology

  • Debate over the article’s claim that CSMA/CD is “gone”:
    • Several point out WiFi has always used CSMA/CA, plus RTS/CTS and now OFDMA; collisions are still managed, just differently.
    • Ethernet switched networks no longer need CSMA/CD; WiFi still shares a medium, but access points and PHY tricks make it behave more like a switch.
    • Topology characterization differs: some argue WiFi is logically star over a shared (bus-like) RF medium.

Mobility, sessions, and routing

  • Disagreement on whether changing IPs mid-connection is practical.
  • Critics of the article say it ignores how intermediate routers learn new locations; purely UUID-based session IDs don’t solve L3 routing or DoS issues.
  • Others note QUIC/multipath-style designs decouple L4 identity from L3 and can re-establish sessions after roaming, accepting some packet loss.
  • Mobile IP and Mobile IPv6 are mentioned as real but largely undeployed due to ecosystem and ISP constraints.

IPv6 design, ARP/NDP, and layering

  • Some like the article’s historical overview but object to its negativity about ARP, arguing it enabled simple LAN IP without routers.
  • Others defend IPv6’s Neighbor Discovery as cleaner than ARP (ICMPv6, multicast instead of broadcast), though there’s back-and-forth on whether either truly avoids “layering violations.”
  • A recurring theme: many alternative proposals or “IPv4+” ideas end up functionally equivalent to IPv6, or were actually tried and rejected.

IPv4 + NAT vs IPv6

  • One camp: IPv4 with NAT (and even CGNAT) is “good enough” for most users who just need outbound web access; simple P2P is not seen as essential.
  • Opponents highlight CGNAT pain (e.g., hosting small game servers), loss of easy end-to-end connectivity, and centralization risks.
  • Discussion of IPv6 privacy addressing and fears that per-device global addresses could enable per-user charging; others counter with address rotation mechanisms.

Deployment pain, politics, and dual stack

  • Multiple anecdotes of flaky dual-stack behavior (e.g., apt timeouts) leading admins to disable IPv6.
  • Some blame IAB/IETF “end-to-end purity,” slow adaptation to real-world security/firewall needs, and complex address-selection rules for delaying adoption.
  • Others argue the real issue is path dependence and that any new “better” protocol would require essentially the same painful global upgrade.