Core-Stateless Fair Queueing: Achieving Approximately Fair Bandwidth Allocations in High Speed Networks

Under the assumption that maintaining of state information to achieve fair queueing implementation is overly costly, this paper proposes a solution to push the relatively more expensive operations to the edge routers and keeping core routers simple and extremely low latency. This is achieved by using packet labeling at the edge routers and provide hints to core routers to influence their dropping policy. Furthermore, the paper gives theoretical bounds that the combined effect of core and edge router fair queueing does not allow a user to exceed its fair share.

This architecture assumes that core routers and edge routers are configured under one entity. In reality, this is not true. Tier 1 ISPs frequently peer with each other, and packets often enter and leave these administrative domains. If all peering points are considered edge routers, would that add enough complexity to the system to render it impractical? If not, what is preventing ISP a to fake its label to gain priority on ISP b's network?
A large scale deployment of such configuration or simulation results with multiple administrative domain would help to clarify this issue and would be interesting to investigate.