IPv6 for ISPs
IPv6 for ISPs: why deploy it, /32 and /48 addressing, dual-stack, 464XLAT and DS-Lite, peering over IPv6, and a practical rollout plan.
Last updated:
IPv6 is the latest version of the Internet Protocol, replacing the exhausted IPv4 address space with 128-bit addresses instead of 32-bit ones. For Internet service providers (ISPs) this means a practically unlimited pool of addresses, the ability to give every customer their own public prefix, and a simpler network because address translation goes away. Deploying IPv6 is no longer a forward-looking option but a baseline requirement for serving modern services and traffic from major content providers.
Why an ISP needs IPv6
The main driver is IPv4 exhaustion. Regional Internet Registries (RIRs) such as RIPE NCC stopped handing out large blocks years ago, and prices on the IPv4 transfer market remain high. To serve a growing customer base from a limited pool, operators deploy CGNAT (Carrier-Grade NAT), sharing one public address among many subscribers. CGNAT carries real costs: it needs capable hardware, complicates the session logging often mandated by law, breaks online gaming, P2P and customer-hosted services, and can cause address reputation problems.
IPv6 removes most of these issues. Every subscriber gets a globally routable prefix, so translation becomes unnecessary and connectivity is fully bidirectional. Traffic to the largest services (Google, Meta, Netflix, CDNs) already flows natively over IPv6 to a large degree. That offloads CGNAT infrastructure and improves connection quality.
Addressing basics
An operator requests an allocation from its RIR, typically a /32 minimum, which corresponds to billions of /64 prefixes. Customer addressing is planned hierarchically, keeping routes aggregated and the layout cleanly split across regions and edge devices. Recommended assignment practices are as follows:
- /32: the typical initial ISP allocation from an RIR and the basis of the whole addressing plan.
- /48: the recommended prefix for a business customer or a site with many subnets.
- /56: a common compromise for residential customers, allowing 256 /64 networks.
- /64: a single subnet; never subdivide it further, as SLAAC depends on it.
Dual-stack and transition mechanisms
The simplest and most reliable approach is dual-stack, running IPv4 and IPv6 in parallel on the same network. The customer receives both addresses, and applications pick a protocol using the Happy Eyeballs algorithm. Dual-stack does not solve the IPv4 shortage, however, so operators combine it with transition mechanisms. DS-Lite tunnels the customer's IPv4 traffic across an IPv6 network to a central AFTR node where translation happens. 464XLAT (NAT64/DNS64 plus a CLAT component on the customer device) lets you run an IPv6-only access network while keeping compatibility with apps that still need IPv4. This approach is especially popular in mobile networks.
Peering, transit and rollout
BGP sessions for IPv6 are established just like IPv4: over Internet exchange points (IXPs) and transit providers, ideally as separate sessions for the IPv6 address family. Announce your /32 prefix as a whole (and any sub-prefixes), maintain correct route6 objects and RPKI ROA entries, and filter your neighbours. In practice, migrate in stages: start with the core and edge routers, then your own services (DNS, mail, web), and finally the access segment and CPE. Monitor both protocols and test paths so that IPv6 actually carries production traffic instead of only existing in the configuration.
Frequently asked questions
Can I drop IPv4 entirely from an ISP network?
Not fully in practice, since some services and devices still require IPv4. You can, however, build an IPv6-only network with 464XLAT or DS-Lite that provides reachability to IPv4 resources through central translation.
What IPv6 prefix should I assign to a residential customer?
A /56 (256 /64 networks) is recommended, or a /48 for more demanding subscribers. Never assign a single /64 if the customer may have multiple segments, and do not split a /64 into smaller subnets.
How does DS-Lite differ from 464XLAT?
DS-Lite tunnels the customer's native IPv4 traffic over IPv6 to an AFTR node where NAT44 happens. 464XLAT uses NAT64/DNS64 plus a CLAT component, so the access network runs IPv6-only and translation to IPv4 occurs on the customer device and in the operator network.
Does IPv6 require separate BGP sessions at an IXP?
Yes, the good practice is to establish separate sessions for the IPv6 address family. Keep correct route6 objects and RPKI ROA entries so your prefixes are accepted and filtered properly.