The prolonged coexistence of IPv4 and IPv6 motivates exploring whether IPv4 can be extended beyond its 32-bit address limitation. IPswen encodes variable-length addresses within standard IPv4 options, allowing legacy routers to forward packets based on the 32-bit base address without modification. However, whether such option-bearing packets can traverse the real Internet remains empirically unvalidated. This paper presents the first large-scale measurement of unknown IPv4 option reachability across approximately 75,000 ASes from four geographically distributed vantage points. We find that options ≥ 8 bytes achieve 38–64% reachability across core Internet paths, with bounded RTT overhead under 32%. Motivated by this partial reachability, we design a pragmatic deployment model centered on domain-controlled WAN overlays: native forwarding where options pass, with automatic GRE-based fallback where they do not. We implement this model in a multi-continent testbed featuring Automatic Tunnel Switching, a BGP Virtual Direct Connection scheme for route distribution, and an extended Linux kernel FIB supporting variable-length prefixes. Our evaluation demonstrates acceptable control-plane convergence overhead and data-plane scalability up to 10M routes, establishing a concrete platform for studying incremental IPv4 evolution.

The exhaustion of IPv4 addresses has forced ISPs to deploy Carrier-Grade NAT (CGNAT), a stateful middlebox that breaks end-to-end connectivity for billions of users. IPv6 could restore addressability but requires an incompatible protocol stack that, after 25 years, remains only partially deployed. We present ISP4x, a single-stack architecture that provides a scalable proof-of-concept for improving end-to-end connectivity within IPv4. ISP4x places a 16-bit extension in the standardized IPv4 Options field by leveraging the IPswen addressing framework to expand the effective address space by a factor of 65, 536 per base IP, and introduces Stateless NATswen, an O(1), protocol-agnostic address translation requiring zero per-flow state. We implement an XDP-accelerated edge gateway and evaluate it in a 500-node emulated ISP network. Under 90% load, ISP4x sustains 86% of link capacity while CGNAT degrades to 46%, with p99 jitter of 1.8 ms versus 45 ms. Notably, ISP4x requires zero per-flow memory, removing the state-exhaustion vulnerability inherent in CGNAT. These preliminary results suggest that ISP4x offers a pragmatic, incrementally deployable path toward restoring peer-to-peer connectivity within and across participating ISP domains without abandoning IPv4 infrastructure.

Since the advent of IPv6 in the late 1990s, the Internet has included two parallel protocol stacks: TCP/IPv4 and TCP/IPv6. Although the “legacy” IPv4 part of the Internet is generally recognized as being “ossified” due to the decade-long depletion of the small IPv4 address space, it is still the dominant part of the public Internet, and will coexist with IPv6 at any time in the foreseeable future. In this paper we discuss a unique opportunity for the continued evolution of the IPv4 Internet based on IPswen and its coexistence strategy with IPv6. We here emphasize two crucial aspects of this approach, namely, (1) making use of IPswen's special technical advantages for the long-term evolution of the Internet, and (2) avoiding conventional “IPv6 groupthink” to enable healthy coexistence and independent development of the two IP technology bases. We argue that it is necessary as well as responsible for the Internet community to maintain and extend the TCP/IPv4 architecture using IPswen (or the like extensions) exactly like what we have successfully done using previous evolutionary technological innovations including CIDR, NAT44 and VPN, while deploying IPv6 in more limited application domains or highly controlled environments for better coexistence and interworking with the public Internet.