The Emerging Dominance of IPoE

For two decades, Point-to-Point Protocol over Ethernet (PPPoE) was the default way an ISP connected a subscriber's CPE to the network. That era is ending. Across fiber, cable, and fixed-wireless access, IP over Ethernet (IPoE) has moved from an "alternative" subscriber-management model to the emerging default — and the shift is accelerating for reasons that are architectural, economic, and operational rather than merely fashionable.

A protocol built for a network that no longer exists

PPPoE was designed for the DSL and ATM era. When copper-fed DSLAMs aggregated subscribers over ATM virtual circuits, operators needed a way to carry point-to-point session semantics — authentication, per-user state, and accounting — across a shared medium that had no native concept of a subscriber. PPPoE answered that need by wrapping every subscriber's traffic in a PPP session, discovered and established through the familiar PADI/PADO/PADR/PADS handshake, then authenticated with PAP or CHAP and addressed through IPCP.

That machinery made sense when the access network was a dumb ATM cloud. It makes far less sense today. Modern access — GPON and XGS-PON, active Ethernet, DOCSIS 3.1 and DOCSIS 4.0, and fixed wireless — is Ethernet-native from the CPE to the edge. The subscriber is already identifiable at Layer 2 through the access node, the port, and VLAN tags. Re-imposing a point-to-point session protocol on top of an already-Ethernet network is solving a problem the network no longer has, and paying for the privilege.

The overhead tax

The most tangible cost of PPPoE is the encapsulation itself. The PPPoE header consumes eight bytes, dropping the usable MTU from 1500 to 1492. That seemingly small number produces a long tail of operational pain: TCP MSS clamping on the BNG, path-MTU black holes, fragmentation, and the perennial support tickets that trace back to a handful of applications that don't handle a reduced MTU gracefully. IPoE carries no such tax. The subscriber gets a clean 1500-byte MTU, native to the Ethernet fabric, with no session wrapper to account for.

Simpler subscriber management, not less control

The objection operators historically raised was that PPPoE gave them control — authentication, per-session accounting, a clean tear-down when a subscriber went away. IPoE delivers the same control through mechanisms that are better matched to modern networks. Address assignment moves to DHCPv4, SLAAC, and DHCPv6. Subscriber identification rides on DHCP Option 82 (and its DHCPv6 equivalents), populated by the access node with the exact line, port, and VLAN context the operator needs. AAA still runs through RADIUS, so authentication, authorization, and accounting are fully preserved — the BNG simply keys them to an IP session and the Layer-2 circuit rather than to a PPP session.

The result is that operators lose none of the policy control they actually use — rate limiting, tiered plans, lawful accounting, dynamic policy from RADIUS CoA — while shedding the session-establishment plumbing that generated most of the complexity.

Aligned with IPv6, multicast, and IPTV

Three forces of modern broadband all favor IPoE. First, IPv6 dual-stack deployment is materially simpler over IPoE, where SLAAC and DHCPv6 prefix delegation are the native tools, than over PPPoE's IPv6CP negotiation. Second, IPoE is a far better substrate for multicast: IPTV and other multicast video services map cleanly onto IGMP/MLD snooping across an Ethernet access network, whereas PPPoE's point-to-point unicast model forces awkward workarounds and replicated streams. Third, the entire Broadband Forum architecture — TR-101, TR-178, and the CUPS-oriented models that follow — treats Ethernet-based, DHCP-driven subscriber management as the reference design. The standards momentum is unambiguous.

Scale and the economics of disaggregation

Every PPPoE subscriber is a session — a stateful construct the BNG must establish, keep alive with periodic LCP echoes, and tear down. At the scale of a modern edge, that per-session state is a real constraint on subscriber density and on failover behavior. IPoE's lighter, IP-session model scales more gracefully, and it fits the direction the whole BNG market is moving: away from monolithic, chassis-based routers and toward disaggregated, software-based platforms running on commodity x86.

This is where the economic argument becomes decisive. A disaggregated softwire BNG on standard servers strips the premium hardware markup out of the edge, and its control-plane simplicity translates directly into lower deployment, operational, and support cost. IPoE amplifies that advantage. Fewer moving parts in the subscriber-management path means fewer failure modes, simpler zero-touch provisioning, cheaper CPE that no longer needs a PPPoE client configured, and a support organization that spends less time chasing MTU and session-flap tickets. For an ISP or WISP weighing cost per subscriber, the combination of commodity hardware and a leaner protocol stack compounds.

Why WISPs never looked back

Fixed-wireless operators are the clearest tell. Most WISPs never carried the DSL-era PPPoE baggage; their networks were Ethernet-and-IP from the first radio. For them IPoE isn't a migration — it's simply how the network was always meant to work. As fixed wireless has grown into a serious broadband access category, it has quietly demonstrated that a large, profitable subscriber base can be run entirely without PPPoE, and run more cheaply for it.

The trajectory

None of this means PPPoE vanishes overnight; large installed bases persist, and some operators will run it for years out of inertia. But the direction of travel is set. New builds — fiber greenfields, DOCSIS upgrades, fixed-wireless expansion — are being provisioned IPoE-first, because the access network is Ethernet-native, the standards point that way, IPv6 and multicast demand it, and the economics reward it. IPoE's dominance is emerging not because it is new, but because it is finally the model that matches the network operators actually run today. A purpose-built, IPoE-only platform isn't a narrowing of scope — it's a bet on where the market has already decided to go.