Social, individual, and greedy optimal controls (rate and window) in wide area high speed packet networks

Rate and window controls are used by the users of packet networks to adapt to the congestion level in the network and to maintain a high performance level in the face of changing network conditions. Increasing network and application speeds imply that the transmission times on links are becoming small in comparison with the (load independent) propagation delays and that the latter need to be modeled explicitly in evaluation of controls. When multiple users share such networks, natural questions arise as to the relative merits of socially optimal controls (rate or window) and individually optimal controls. In addition, the performance impact of one or more greedy users deviating from the prescribed (socially optimal) behavior needs to be understood. In this paper, we formulate open and closed networks of queues models capturing the essence of the issues mentioned above. We use the “power” as the objective function to be maximized and consider two service disciplines at a queueing point: FIFO and 1-limited Cyclic (Head-of-the-Line-Processor Sharing). We analyse these models to provide characterization of the relative performance of the two control approaches (rate and window) under social, individual, and greedy optimization. Special attention is given to the behavior of the solutions as the ratio of the mean propagation delay to the mean service time becomes large