Control Plane vs User Plane in URLLC: Where Is the Real Latency Bottleneck?

Hello World

While studying URLLC industrial use cases (e.g., 2 ms control loop robotics), I wanted to clarify something regarding 5G architecture.

As we know, AMF and SMF are part of the control plane and are involved during:

  • Registration

  • Authentication

  • PDU Session establishment

  • Slice selection

However, during real-time data transmission, the user plane path becomes:

UE → gNB → UPF → Edge Application (Controller)

So my question is:

In practical deployments targeting sub-5 ms latency, where is the dominant bottleneck typically observed?

  1. RAN scheduling delay (numerology, mini-slot configuration)

  2. Transport delay between gNB and UPF

  3. UPF processing delay

  4. Application processing at MEC

Would appreciate insights from those who have worked on live URLLC or MEC deployments.

RAN scheduling (numerology, mini-slot config) often dominates because ultra-low latency requires very fast, deterministic scheduling at the gNB. Transport delay between gNB and UPF is small if the network is well-optimized and fiber/backhaul is short. UPF processing is minimal with modern hardware. Application processing at MEC can contribute, but MEC placement near the RAN usually keeps this under control.