Performance study of backpressured packet networks using a soft fluid flow model

In a backpressured network, single-hop backpressure control messages are sent from each congested node to the immediate neighbors, in order to stop the flow of incoming traffic. A backpressured network can provide loss-free communication among its nodes. This is a significant advantage that greatly simplifies the design of end-to-end protocols. However, we show in this paper that, along with the above advantage, backpressured networks are susceptible to three important problems: lack of fairness, unnecessary spread of backpressure signals, and the possibility of deadlock. Another contribution of this paper is the development of a soft fluid flow analytical model for the study of backpressured networks. Performance study of a backpressured network is complicated due to the inherent on-off behavior of the transmission links resulting from the backpressure messages. In our soft fluid flow model, we alleviate this problem by replacing the abrupt transitions encountered in the actual network with smooth and gradual changes. Using numerical studies and simulation, we have verified that the main features and performance of a backpressured network are almost invariant to the above substitution