Scheduling algorithms must be used on the radio side of these technologies in order to map radio sources to incoming/outgoing data packets and satisfy quality of service requirements such as latency and error rate. But are they also used in core network components such as SGW/PGW or UPF for reasons such as delivering latency sensitive data sooner to the radio side? Would scheduling packets in core network side be meaningless since they will be scheduled again in the radio side?


For the most part, the scheduling happens in the eNodeBs (gNBs in 5G). Core network components may use "scheduling", as part of the QoS mechanisms to meet end-to-end latency requirements, etc., but that is different from scheduling over the air that happens in eNodeBs (gNBs) and that takes into account the channel conditions, interference conditions, etc., that are radio related and the core network components have no knowledge about.

Having said that, we cannot say that "scheduling packets in core network side" is "meaningless". As the core network components are still concerned about the network latency. They would do their best based on what they know, which is the QoS parameters like QCI (or QFI in 5G) that they see in the packets, as well as the traffic flows that they have been experiencing, and so on.

The radio components and the core network components each do their part to support goals like end-to-end latency, based on the knowledge that they each have.

| improve this answer | |
  • Thank you for the answer. I'd like to ask one more thing. In 5G core network, UPF doesn't know QCI (5QI) and only knows QFI, which is a value mapped to 5QI in radio. Would a UPF be able to perform any scheduling beneficial to end-to-end latency without knowing the 5QI? – tyb May 5 at 15:04
  • Yes, the UPFs have the QFI, and they can implement their QoS control mechanisms (queuing, etc.) based on that. We can think of it as distributed way to implement end-to-end QoS. The gNBs have certain knowledge related to the air interface, and nodes in the core network have other knowledge. It doesn't need a central entity that "sees" the end-to-end flow of each packet and decides based on that. Besides, it wouldn't be scalable. Feel free to up-vote my answer if it was helpful :-). I'll edit it to include 5G terms as well. – auspicious99 May 5 at 15:25
  • 1
    I understand now, "distributed" part made it clear to me, thank you. I already up-voted but since my reputation is low my up-votes do not change the post score :). – tyb May 5 at 15:45
  • Ah, I see. I just learnt something new today :-). Yes, the distributed processing, I think was inspired by QoS models used in the Internet like DiffServ, MPLS and so on – auspicious99 May 5 at 15:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.