Summary

  • Terry Sweetser measured qBittorrent traffic at 44.3% IPv6 before binding the client to an IPv6 interface; his two-day after sample reached 92.6%.
  • The default “Any Interface” option let the client continue making outbound IPv4 connections through NAT even though IPv6 already worked.

Terry Sweetser had done nearly everything an IPv6 enthusiast is told to do. His home gateway had a native /48; local DNS and services used IPv6; qBittorrent had an IPv6 port forward. The application still sent most of its traffic over IPv4. The culprit was not a broken IPv6 route. It was a little menu choice labelled “Any Interface.”

In his APNIC field report, Sweetser describes using NetFlow data to sort high-bandwidth applications by protocol. The 30-day pre-change average for BitTorrent was 44.3% IPv6. Trackers supplied IPv4 peer addresses; the client, allowed to use any interface, initiated connections through the gateway's IPv4 NAT. The IPv6 side accepted inbound traffic, while the IPv4 side was outbound-only because there was no IPv4 port forward. Dual stack had made the application work, then quietly chose the less useful side of it.

Sweetser changed qBittorrent's Network Interface setting to its IPv6 interface. His reported two-day after average was 92.6% IPv6 for the application, with IPv4 traffic falling from a reported 1.08 Mbps before to 0.38 Mbps after. Across the whole network, the report compares a 67.7% IPv6 30-day baseline with 79.2% in a two-day after interval. Those windows are different, and other local changes may affect the whole-network mix. The narrow finding is strong enough: an app that advertises IPv6 support can still choose IPv4 when its default is free to do so.

Sweetser's APRICOT 2026 presentation presents the same interface-binding intervention in rounded terms. Its slide says the application moved from 44% to “ALL IPv6”; the later APNIC report gives the measured two-day figure of 92.6%. The exact figure matters because it stops a dramatic slide from becoming a false promise about sustained zero IPv4 traffic.

The remaining IPv4 traffic had other owners: the report identifies older cameras, web services and ISP-side infrastructure. Sweetser, listed in APNIC's community nominee record, could change his own client's interface binding. He could not change every device or distant service. That is the useful boundary in this experiment. Before buying a grand transition architecture, inspect what the applications already do when nobody tells them which network to use.