Network Slice Location and deployement of model

Q/A
#1

Hello everyone,
1) I have a doubt. I want to know where does Network Slice actually recides in 5G Network. I mean it is logical partitioning of network in other words we can say we virtualize NFs and club them and form a network slice that can provide a service. so where does this resides like in any network functions or where. and how does NSSF knows about it.

  1. I want to develop a federated learning based model for attack detection in network slices. The concept of federated learning is that it has local models and a central model. I am planning to deploy local model on each slices but I am not getting a clue where should i deploy global model.

Kindly help me this.
Thanks

Model Deployement on Free5GC
#2

Hi @virendra,

Here’s my answer to the questions. If something was wrong, it’s welcome to discuss!

  1. You’re right. The network slice is a sort of logical partitioning of network.
    To explain it, let’s start with a UE connecting a slice first. A UE will send registration req to the AMF. The AMF will ask NSSF (in some certain condition) to help UE select the proper slice (or validate requested NSSAI value actually). See more in TS 23.501.
    Who helps a NF to select slice? That’s NRF’s job. It provides NF discovery service to help NF to select proper SBI service producer in a slice (identified by SNSSAI value). If you are interested in this, see more in the spec TS 29.510
    Overall, to distinguish different slices, it’s all about checking SNSSAI value.

  2. I’m definitely no expert in federated learning, but to my best knowledge (learned from Wikipedia :p), the global model acts as a coordinator during the learning process. If so, the global model (or the central server) can be deployed anywhere. Whether it’s cloud or on-prem, just make sure all the worker node is able to reach the global model through the network (not the 5G network in this context).

Hope it helps. :crossed_fingers: