How UPF can be selected based on location?

Hello Core Experts:

How UPF can be selected based on location?

The aim is to select closest UPF to the end user to have the best latency in user plane. This is for 5G SA.

Any idea experts?

Are you planing to connect the RAN side to multiples UPF?

If no, you can connect the RAN side only to the new edge UPF.

RAN is already connected to 2 different AMF amd UPF.

I want users get service from the closest UPF by distance.

Dear, I know this is not your “question”, if you allow me: “why you would like to do so?”

CORE side able to map same region AMF with UPF only, which means user can get service from long distance UPF which increase latency.

Distance between 2 UPF is 1000 km.

Have you found the answer?

Don’t have much experience in 5G SA, but if I were to guess you know the location of gNB, have them map to nearest UPF as primary and distant one as secondary.

That’s what I am thinking of, but still looking for the answer.

There could be some solution based on TAC plus DNS.

Maybe that could be accomplished by using the IP network, as a anycast solutions with some cost metrics.

Good idea!

Can you elaborate a bit?

Are you standing up these UPFs in cloud based on user location?

What is the architecture?

Usually, user location is known by the gNB id it attaches with.

Unless you are assigning UPFs in cloud, you can configure UPF based on gNB location.

Cloud base.

If you can share the 3gpp referece or any other source I appreciate :+1:

I may need to understand more details on what you want to achieve, but 3gpp define many architecture targeted for use of MEC you can use one of them for your use case.

So basically it will depends on your UPF, if is supporting UL CL or not, and if is your CU supporting RAN slicing or not, as you may have to create different slice for you packet routed to your main UPF or anchor UPF and MEC or low latency UPF.

Refer to following figure:

1 Like

User mobility pattern recognition, ML/AI helps on this.

1 Like

To get low latency you need to place UPF at the gNB. Otherwise, try to count number of hops and calculate backhaul distance between gNB to all nearest UPFs and then select shorter one in distance and the lowest in number of hops. You may ping from gNB the IP of UPFs and check latency and number of hops.

1 Like

Hi,

Maybe you can solve using slice solution defining some parameters (application, location, device, time, cell congestion, etc…)
For a more dynamic solution allocation, I don’t know how much slicing would work…
Maybe an event-based automation solution will also help (Ansible).

1 Like

In this project that I participated in, having the UPF nearby made a huge difference in latency.

1 Like

For AI ML in 3GPP, following specs can be followed

  • 38.843 - Study on Artificial Intelligence (AI)/Machine Learning (ML) for NR air interface
  • 28.908 - The AI/ML management study
  • 28.809 - Study Management Data Analytics MDA
  • 28.104 - Management Data Analytics MDA
  • 28.105 - Management and orchestration AI/ML Management
  • 37.817 - Study on enhancement for Data Collection for NR and EN-DC
  • 38.423 Xn Application Protocol (XnAP) (Release 18)
  • 38.331 Radio Resource Control (RRC) protocol specification
  • 21.918 - Release 18 Description
  • 21.917 - Release 17 Description
1 Like

Solution: using NRF to detect closest UPF based on CGI cell id of gNodeB.

Oh yeah of course.

I’m guessing some scripting required in NRF to detect the incoming request and spin-up UPF nearest to UE?

Correct.