Every 5G data packet a subscriber sends or receives passes through the UPF. It forwards traffic, enforces QoS, and connects the network to external data networks. Place it centrally and latency suffers. Place it at the edge and latency-sensitive applications work. The UPF is where that decision is made.
In 5G Standalone, the data plane is a separate network function: the UPF. It forwards every packet, applies QoS, and provides the connection to external data networks. Where it sits determines the round-trip latency a subscriber experiences.
A centralised data plane works for general internet traffic. It does not work for the applications operators want 5G to enable: industrial automation, AR, low-latency edge compute. Those need the UPF distributed to the edge, close to the radio. A data-plane function that cannot be placed and scaled independently of the control plane forces a trade-off between architecture and deployment cost.
gNB sends traffic — UPF anchors the data path
The UPF is the anchor point for the data path in a 5G SA network. It receives traffic from the gNB over N3, forwards it to external data networks over N6, and applies the forwarding, QoS, and charging rules programmed by the SMF over N4 using PFCP. Where a session spans multiple UPFs, traffic passes between them over N9. The UPF enforces per-session QoS, generates usage reports for charging, and supports uplink classification for local breakout.
Distributed P-Gateway: place the UPF at the edge
The Ouroboros UPF is the data-plane half of the gateway that runs as GGSN for 2G/3G and PDN-GW for LTE. The distributed P-Gateway architecture already in that platform is what places the UPF close to the subscriber. Operators deploy UPF instances at the edge for low-latency traffic and centrally for general data, all controlled by one SMF.
QoS, charging, and local breakout — enforced in the data plane
IPv4 and IPv6 are supported natively. Local breakout routes selected traffic to a local data network without backhauling it to a central site.
Edge compute, national data, private 5G, network slicing — the data plane that moves with the traffic.
Edge compute and low-latency applications
The UPF deploys at the network edge, close to the gNB, so traffic for latency-sensitive applications breaks out locally instead of routing to a central site. Round-trip latency drops without changing the control plane.
National 5G data access
Centrally deployed UPF instances carry general subscriber data traffic to the internet and external networks, with per-session QoS applied from SMF policy.
Private 5G with local breakout
A campus, port, or industrial private 5G network runs the UPF on-site so operational traffic stays local. Production data does not leave the facility, and latency is controlled locally.
Network slicing
Dedicated UPF instances carry the traffic for specific slices, isolating an enterprise or IoT slice's data path from consumer traffic.
Quote placeholder — client input needed.
Operator CTO
CTO · Major MNO
Distributed edge deployment
UPF instances place the data plane close to the gNB, with the SMF controlling each over N4. Latency-sensitive traffic breaks out at the edge while general traffic routes centrally. One control plane manages both.
Data-plane continuity from PGW to UPF
The same platform that runs GGSN and PDN-GW provides the UPF data plane in 5G SA. The move to 5G extends the existing gateway rather than introducing a separate data core.
N4 control by the SMF
Forwarding, QoS, and charging rules are programmed by the SMF over PFCP. Data-plane capacity scales independently. Add UPF nodes without reconfiguring the control plane.
Local breakout
Selected traffic routes to a local data network without backhaul to a central site. Private and edge deployments keep operational traffic local, cutting latency and backhaul load.
Per-session QoS enforcement
QoS rules apply at the packet level per PDU session. Slice and plan commitments are honoured in the data path, not just in policy.
Native IPv4 and IPv6
Dual-stack forwarding per session. Operators running IPv6 or mixed addressing require no separate data-plane configuration.
The UPF only earns its place if it can sit where the traffic needs it. The Ouroboros UPF inherits the distributed P-Gateway architecture already proven in the platform's LTE data gateway, so edge placement is a deployment choice rather than a re-architecture. The control plane and the data plane scale separately, which is the entire point of the 5G split.
It is the data-plane half of the gateway that runs as GGSN and PDN-GW, so operators reaching 5G SA extend a known platform. Carrier-grade, in production since 2004, deployable on COTS, virtualised, or cloud.
OPEX rental is available, so edge UPF capacity grows with the traffic it carries.
Where are your edge sites, and what latency are you targeting? We'll map a UPF deployment, central and edge, controlled from one SMF.