Summary

  • WorldLink reports more than one million IPv6-enabled IPoE subscribers and about 70% of its traffic on IPv6; its post does not publish the calculation period or denominator.
  • A DHCPv6 lease can sit on IPoE’s subscriber-session path. CPE, BNG and access-network recovery failures can separate a configured capability from traffic that works.

A green “IPv6 enabled” flag can mean that provisioning is complete. It does not show which protocol carried a customer’s data or whether service will return after an outage. WorldLink’s operating account makes that difference visible.

In an APNIC guest post dated 12 October 2026, WorldLink principal network engineer Kijush Maharjan reports more than one million IPoE broadband subscribers, all IPv6-enabled, with about 70% of the company’s traffic on IPv6. The post says tests began in 2016 and live service in 2019. These are operator claims, not an independent audit. The author gives no observation window, denominator, traffic direction or method for the 70% figure.

The two measures answer different questions. An enabled count records what the provisioning system expects. Traffic share is closer to use, but counts bytes: a small group of heavy users can move it. It cannot show how many customers have working IPv6 or whether their service is reliable.

The account describes IPoE, where DHCP and DHCPv6 lease state helps maintain the subscriber session. IPv4 and IPv6 have separate forwarding planes but share that session. DHCPv6 can assign a CPE address through IA_NA and delegate a home prefix through IA_PD. The current RFC 9915 specifies DHCPv6 addresses and prefixes; standards compliance alone does not establish that CPE and BNG implementations recover compatibly.

Maharjan recounts a fibre cut above an optical line terminal. The BNG did not learn of it; renew and rebind attempts went unanswered. After lease expiry, some CPEs restarted address and prefix exchanges separately. One CPE sent 116 SOLICIT messages in a minute, the post reports. That is a single-device observation, not a network-wide rate.

The post says binding removal during renegotiation could break the dual-stack session while a CPE kept using its old address and prefix. Retries did not restore IPv6; affected traffic returned to IPv4 and added CGNAT pressure. The account also describes multicast and snooping issues that could block DHCP replies. Customers noticed slow services, not a DHCPv6 label. WorldLink says implementation details were generalized, so the source identifies no vendor or incident frequency.

The evidence supports a narrower conclusion than “IPv6 adoption is 70%.” Enablement describes intended configuration; traffic share describes bytes moved under an undisclosed method. Neither alone proves service quality. Together with separate reachability and lease-success measures, they give a more useful view of whether deployment works.