Hello Mirko
In the context of QoS LLQ, the bandwidth values you see in the policy-map configurations relate to the actual speed or rate at which packets are sent out of the queue, not the size of the queue in terms of packet storage.
When you configure a policy map and specify a bandwidth for a particular class within the policy map, you are essentially telling the router to reserve that amount of bandwidth for that class under congestion conditions. It’s a way to guarantee that certain traffic (like voice or critical data) will have a certain amount of bandwidth available even when the network is busy. Remember, QoS is all about reserving and guarenteeing bandwidth for particluar packets.
LLQ is specifically designed to provide strict priority queuing for delay-sensitive traffic like voice or video. The traffic that falls into the LLQ is given priority over all other traffic, up to the amount of bandwidth specified by the priority command. This means that the LLQ will be served before other queues, and up to the configured rate, before other classes are able to send traffic.
So the bandwidth command in a policy-map class specifies the rate at which packets are forwarded out of the queue. It is not a storage size limit for the queue itself. The size of the queue, which determines how many packets it can store, is configured separately, often with the queue-limit command. The default queue-limit is typically 64 packets on most platforms, but can be adjusted. An excellent Cisco resource on understanding the queue limits and output drops can be found here:
I hope this has been helpful!
Laz