QoS flow — The finest grain at which the network promises anything
Sessions, flows and slices
Where it sits
What it is
What it is. The finest granularity of QoS differentiation in a PDU session — TS 23.501 §5.7.1.1 says exactly that. Traffic with the same QFI within a session "receives the same traffic forwarding treatment".
Two kinds, and the split runs through everything:
| GBR | Non-GBR | |
|---|---|---|
| Promise | a guaranteed flow bit rate | best effort within its priority |
| Extra parameters | GFBR and MFBR, uplink and downlink | may carry a Reflective QoS Attribute |
| Admission | the RAN must be able to reserve it | no reservation |
| Typical | voice, real-time video | everything else |
What characterises a flow, per §5.7.1.1 — three things, going to three different places:
a QoS profile, given by the SMF to the RAN via the AMF over N2 · one or more QoS rules, given by the SMF to the device over N1 — or derived by the device through reflective QoS · one or more packet detection rules, given by the SMF to the UPF over N4.
One description of a flow, three audiences. The RAN is told how to schedule it, the device how to mark it, the UPF how to recognise it. All three come from the SMF, which is why the SMF is the session's single owner.
Every session has a default flow, permanently. §5.7.1.1: the flow associated with the default QoS rule "is required to be established for a PDU Session and remains established throughout the lifetime of the PDU Session", and "should be a Non-GBR QoS Flow". So connectivity never depends on a reservation succeeding.
Reflective QoS is the clever economy. Rather than signalling an uplink rule, the network lets the device infer it from how the downlink was marked — see SDAP, which implements the mapping half. The rule arrives without a message being sent for it.
Why flows and not bearers. A bearer in LTE was an end-to-end construct: creating one meant signalling through the core and the radio together. A flow is a marking inside an existing session, so differentiating traffic costs a rule, not a setup procedure. That is what makes it reasonable to have many of them, and why 5G needed SDAP at the bottom of the stack to map them onto radio bearers.
Read on
This concept is read as part of one argument in ref-session, alongside the rest of its group.
Before this concept, the hierarchy says to learn the following — the full chain, in order: