Hello Nicolas
Let me clarify one issue that I mentioned before. For ingress traffic, network devices generally do not have control over the rate at which packets arrive. Therefore, shaping, which involves delaying and scheduling packets, isn’t feasible for incoming traffic. Instead, policing is used to either drop or mark packets that exceed the bandwidth profile, as buffering large amounts of incoming traffic to shape it would be impractical and could lead to high latency or buffer overflows.
Now concerning your question, typically QoS mechanisms kick in whenever there is congestion. Whenever there is no congestion, traffic will simply be forwarded as soon as it arrives, in the order it arrives. Since there is no congestion, there is no delay. QoS is really only meaningful when there is congestion.
So to answer your question, on ingress, as mentioned before, only policing has any practical application. If you want to have more control over how such traffic is being transmitted, and how it arrives on the interface of a device, it would be preferable to apply any congestion management QoS mechanisms on the outgoing interface of the directly connected device. That’s not always possible, as sometimes this device belongs to the ISP. In such cases, you would have to negotiate with the ISP to ask them to set up the rate at which you want your traffic to arrive.
I hope this has been helpful!
Laz