Carrier transition: DS-Lite, MAP-E, MAP-T and CGNAT
Compare DS-Lite, MAP-E and MAP-T as IPv4-as-a-service mechanisms over an IPv6 access network. Apply the CGNAT requirements (RFC 6888) and shared address space (RFC 6598). Select a mechanism for a given access network.
Take the quiz · pass mark 70% · 9 questions
Explanation
Access providers run an IPv6-only access network and deliver IPv4 as a service. The three main mechanisms differ in where the IPv4 NAT happens and whether IPv4 is tunnelled or translated.
DS-Lite (RFC 6333)
Dual-Stack Lite tunnels the subscriber's IPv4 inside IPv6. The CPE (the B4) encapsulates IPv4-in-IPv6 and sends it across the IPv6 access network to the AFTR in the core, which decapsulates and performs a centralised NAT44 to public IPv4. All NAT state lives in the carrier's AFTR. It is simple for the CPE but concentrates state and is stateful in the core. Because IPv4 is encapsulated in IPv6, raise the access MTU (the encapsulation adds a 40-byte IPv6 header) so the softwire does not fragment.
MAP-E and MAP-T (RFC 7597, RFC 7599)
MAP (Mapping of Address and Port) pushes the NAT to the CPE and makes the core stateless, which scales far better. A central set of mapping rules deterministically ties each CPE to a shared public IPv4 address plus a port range (a Port-Set, identified by a PSID), so many subscribers share one IPv4 by splitting the port space. The core just forwards according to the rules — no per-subscriber state.
- MAP-E (RFC 7597) encapsulates IPv4 in IPv6 (like DS-Lite, but stateless in the core thanks to the rules).
- MAP-T (RFC 7599) translates IPv4 to IPv6 (double stateless translation, SIIT-based) instead of encapsulating, so the core sees native IPv6 packets — easier to inspect and ECMP, but translation caveats apply.
Trade-off: DS-Lite is simplest but stateful and centralised; MAP-E/MAP-T are stateless and scale, at the cost of provisioning and distributing mapping rules and giving each subscriber only a slice of ports.
CGNAT requirements and shared address space (RFC 6888, RFC 6598)
Whenever many subscribers share public IPv4 (DS-Lite AFTR, MAP port-sets, or plain CGN), the carrier-grade NAT must meet RFC 6888 (BCP 127): support required protocols, provide enough ports, log address/port mappings for abuse traceability, and behave predictably (e.g. endpoint-independent mapping). The private IPv4 between CPE and CGN uses the RFC 6598 Shared Address Space, 100.64.0.0/10, which exists precisely so operators do not collide with subscribers' own private address space.
ictNetworks scenario
ictNetworks' fibre access runs IPv6-only and offers IPv4aaS by MAP-T where it wants a stateless, inspectable core, and DS-Lite in a legacy region where centralised AFTR NAT is acceptable. Its CGN/AFTR meets RFC 6888 (per-subscriber port quotas and mapping logs for law-enforcement traceability), and it uses 100.64.0.0/10 (RFC 6598) between CPE and core so it never clashes with customer private addressing. It raises the access MTU to absorb encapsulation.
Drawing diagram…
Worked examples
Example 1: DS-Lite encapsulation overhead
DS-Lite puts IPv4 inside an IPv6 softwire, adding a 40-byte IPv6 header. The operator should raise the B4-to-AFTR path MTU by at least 40 bytes (RFC 6333 suggests a 1540-byte softwire MTU) so the tunnel does not fragment.
Example 2: Shared Address Space size
The RFC 6598 Shared Address Space 100.64.0.0/10 contains 2^(32-10) = 4194304 addresses — a large private pool between CPE and CGN that will not collide with customer private (10/8, 172.16/12, 192.168/16) space.
Example 3: Ports per subscriber with a sharing ratio
With a sharing ratio of 1:64 (64 subscribers per public IPv4), and ignoring the excluded well-known range, each subscriber's port-set is 65536 / 64 = 1024 TCP/UDP ports. This is why MAP provisioning must size the sharing ratio against application needs.
Lab
Goal: Build DS-Lite between a CPE and an AFTR and trace a flow through it.
You need: a containerlab topology (Linux ip6tnl and nftables acting as B4 and AFTR) with an IPv6-only access link, a subscriber IPv4 host, and a public IPv4 destination; ping, curl, tcpdump and the conntrack/nft tools. (Encapsulation and NAT syntax vary by kernel/version — check the docs.)
- Configure the B4 on the CPE to encapsulate subscriber IPv4 in IPv6 toward the AFTR across the IPv6-only link.
- Configure the AFTR to decapsulate and NAT44 to a public IPv4, using 100.64.0.0/10 (RFC 6598) on the inside.
- Raise the softwire MTU to absorb the 40-byte overhead and confirm large flows do not fragment/black-hole.
- From the subscriber, reach the public IPv4 destination and trace the flow: capture the IPv6 softwire, the decapsulation, and the NAT translation; find the entry in the AFTR's state/log.
- Note what RFC 6888 would additionally require in production (port quotas, mapping logs) and contrast with a stateless MAP design.
What to submit: the B4/AFTR configuration, the MTU setting, the end-to-end capture trace, the AFTR state/log entry, and the DS-Lite-vs-MAP trade-off note.
Summary
- Carriers run IPv6-only access and deliver IPv4aaS; DS-Lite (RFC 6333) tunnels IPv4-in-IPv6 to a stateful central AFTR.
- MAP-E (RFC 7597) encapsulates and MAP-T (RFC 7599) translates, both stateless in the core using rule-based address+port-set sharing.
- CGN/AFTR must meet RFC 6888 (ports, logging, predictable behaviour); the CPE-to-CGN side uses RFC 6598 Shared Address Space 100.64.0.0/10.
- DS-Lite is simplest but centralised and stateful; MAP scales statelessly at the cost of provisioning and per-subscriber port slices.
- Next lesson (L6.07): making applications IPv6-ready.
Quiz imported · 9 questions · pass mark 70%. Lab questions are not auto-marked.