Private 5G Slice with Centralized UPF
A manufacturer deploys a 5G SA private slice across two plants 400 km apart. The 5G core is centralized: AMF/SMF in a regional data center, a single UPF in the same facility, and N6 to the enterprise application in Plant A. The slice SLA is 15 ms round-trip for control traffic and 99.99% availability; the application is latency-sensitive but not hard real-time. At peak, 1,200 AGVs and sensors attach. Field measurements show 38–45 ms RTT at Plant A and 60 ms at Plant B, plus session drops when devices move between gNBs because the UPF is far from the RAN. The enterprise refuses to re-IP devices and requires session continuity. You must redesign the user plane without duplicating the whole core.
1. Explain why the central UPF violates the latency SLA and what part of the core you would relocate.
Read the worked answer
The latency is dominated by user-plane hairpinning. Traffic from Plant A goes gNB → transport → regional UPF → N6 → enterprise application, and Plant B traffic travels even farther. The control plane can remain centralized because AMF/SMF signaling is not in the per-packet path, but the UPF and N6 breakout are. At 400 km, round-trip transport alone adds several milliseconds, and the measured 38–60 ms indicates queueing, backhaul, and security inspection as well.
Relocate the user plane to the plant edge: place a local UPF or a local PSA (PDU session anchor) near each plant, with local N6 to the enterprise application. Keep AMF/SMF centralized for policy and mobility. This removes the hairpin for local traffic and gives a realistic 15 ms budget. Do not move the whole core; the goal is to move the packet path, not duplicate control-plane functions.
2. Compare a full UPF at each plant with an ULCL and central anchor UPF. Which meets IP preservation and session continuity?
Read the worked answer
A full UPF at each plant is simple for local breakout but weak for continuity: when a device moves between plants or gNBs anchored to different UPFs, its IP address and session context must be re-established unless complex inter-UPF mobility is built. That conflicts with the no-re-IP requirement.
An ULCL with a central anchor UPF is better. The anchor UPF keeps the UE IP address stable, while an edge ULCL/branching point sends local traffic to a local PSA and local N6. Traffic that must reach the central network still follows the anchor. For mobility, use SSC mode 3 where supported so a new PSA can be established before the old one is released. This gives local latency for plant traffic and IP preservation for device sessions. The trade-off is more complex N9 and policy design, but it avoids duplicating the full core and meets the continuity requirement.
3. What redundancy and monitoring would you require before go-live?
Read the worked answer
Require an edge UPF/PSA pair per plant with redundant N3 and N6 links, and redundant N9 paths to the anchor. Use BFD or equivalent fast failure detection so a local user-plane failure does not silently black-hole traffic. The central anchor should also be paired, with clear failure behavior: local breakout may continue if policy allows, but new sessions must not be stranded.
Monitor per-slice latency at N3, N6, and N9; packet loss; session establishment success; UPF CPU and throughput; and path asymmetry. Set alerts against the 15 ms SLA and 99.99% availability, with per-plant dashboards. Test failover during a live maintenance window, including a gNB handover while the local PSA is unavailable. Acceptance should prove both local breakout and anchor continuity, not just ping latency at low load.