Flow and congestion control of interconnected networks

This thesis deals with flow and congestion control in interconnected packet switched networks providing virtual circuit services. Flow control is defined as a mechanism to regulate traffic flowing from source to destination so that the source does not send data at a rate greater than the receiver can process it. On the other hand, congestion control is defined as a more global mechanism accomplished by internal network nodes and gateways so as to prevent network and gateway congestion. Two different control schemes are presented based on how the source network restricts the internet traffic. The first control scheme is end-to-end flow control in which the source network controls its internet traffic based on acknowledgements returned from the end point of the destination network. The second control scheme is network-by-network flow control in which each network including the source network controls the internet traffic based on acknowledgements returned from the end point of its own virtual circuit. For the purpose of analyzing the performance of these two schemes, we develop queueing models: a loss system assuming that packets generated while a virtual circuit has been blocked by flow control are lost, a lossless system in which such packets are assumed to wait in a user input buffer of infinite capacity without any loss, and a loss-wait system in which newly generated packets are assumed to be lost when a user input buffer of finite capacity is full. In contrast to the loss system, the lossless and the loss-wait system reflect the important effect of admission delay experienced in real flow control. Because flow control functions in queueing networks cause population size constraints, there exist no product form solutions. Therefore, these system models are approximately analyzed by means of the Norton equivalent technique (decomposition). Simulation results are presented to verify the approximation. As a performance criterion, we employ the concept of power which is defined as throughput over delay. In each system model, we attempt to determine the optimal flow control parameters to maximize power.