IPv6 Transition Mechanisms: A Comprehensive Performance Analysis of IPv4-as-a-Service and Host-Based Translation Technologies

Abstract

The loss of the IPv4 address space, codified by the loss of the free pool of IANA in February 2011 and then of the free pools of the Regional Internet Registry (RIR), has expedited the adoption of Internet Protocol version 6 (IPv6) worldwide. The transition to IPv6 is still delayed despite its architectural advantages, including a 128-bit address space (about 3.4 x 1038 addresses), Stateless Address Autoconfiguration (SLAAC), built-in IPsec support, and improved routing scalability. This paper gives a technically sound assessment of the trends of adoption of IPv6 and the key transition mechanisms aimed at ensuring interoperability throughout the coexistence phase. We evaluate two broad types of translation mechanisms: IPv4-as-a-Service (IPv4aaS) solutions (464XLAT, Dual Stack Lite (DS-Lite), Mapping of Address and Port with Encapsulation (MAP-E), Mapping of Address and Port with Translation (MAP-T), and Lightweight 4over6) and host-based translation mechanisms. Each mechanism presents distinct advantages and disadvantages: 464XLAT offers strong mobile compatibility but incurs NAT64 state overhead; DS-Lite is broadly deployed but bottlenecks at the stateful AFTR; MAP-E/MAP-T provide stateless, line-rate forwarding at the cost of reduced per-subscriber port availability; lw4o6 offloads Network Address Port Translation (NAPT) to the Customer-Premises Equipment (CPE); D4across6 achieves near-native performance via stateless gateway processing; and host-based translators (BIS, BIA, BIAv2, BIH) enable endpoint translation without infrastructure changes but scale poorly under high flow volumes. We evaluate performance using ns-3 simulations on an ISP-scale topology of 100 nodes. Throughput, end-to-end delay, packet loss ratio, and per-packet protocol overhead are measured under both Transmission Control Protocol (TCP) and User Datagram Protocol (UDP) traffic patterns. Our results show that stateless encapsulation models (MAP-E, D4across6) achieve higher throughput and lower packet loss than stateful NAT64-based systems, whereas host-based translators incur higher Central Processing Unit (CPU) overhead as the number of flows increases. The discussion offers practical recommendations on network operators who develop massive IPv6 migration strategies.