In choosing a network evaluation model, we considered two options. The first option is the ‘Omnipotent view’, where one program manages the entire network. The second option is the ‘Distributed view’, where each node is managed by a copy of the program under evaluation.
In the distributed view, the node does not know the overall topography of the network, it only knows about the pipes connected directly to it. Furthermore, the programs in the distributed view can modify flow only on its output pipes.