How does Diffserv work?
A packet may be marked with a codepoint “anywhere” in the network (but marking probably occurs at domain boundaries)
All packets with the same codepoint get the same behavior (providing aggregation and scalability)
Per-flow state stays at network edges
Marking can be based on microflow identification, the packet ingress link, the measured temporal characteristics of a microflow or aggregate, etc. (Diffserv-capable equipment will include traffic conditioners you can configure.)