Asymptotic analysis and congestion control of TCP networks

Abstract

Congestion control mechanisms in the Internet. represent one of the largest deployed feedback systems. However, many deployed control mechanisms recently appeared is beyond the reach of analytical modeling and feedback control theory. In the congestion control mechanisms, two types of studies are of fundamental interest. First, it is important to characterize the equilibrium point that can be obtained from a given congestion control protocol from the point of fairness, efficiency in use of resources and dependence on network parameters. Second, it is importance to concern the dynamics of congestion control protocols to avoid an occurrence of the network congestion and to dissolve the congestion if the congestion occurrence cannot be avoided. In additions, keeping the fairness among the active connections is an essential feature for the network. However, theoretical research on this topic has dealt mostly with simple scenarios (e.g., single-bottleneck). Transmission Control Protocol (TCP) is the most widely used transport protocol for reliable data transfer in the Internet. TCP uses window-based flow control that a destination sends acknowledgement message for packets that are correctly received. However, there are many problems with the current TCP congestion control schemes. The current congestion control mechanisms use packet loss or timeout to discover congestion in the network. Usually, losses occur if the buffer at a network has reached its capacity and incoming packet is dropped. Hence, in TCP, there are no means to discover congestion before a packet is lost. This blind rate adaptation mechanism causes a global synchronization of the flows. Consequently, it leads to unfairness, low utilization, and a greater oscillation in the queueing delay. Thus the first problem in TCP is to avoid an occurrence of the network congestion, and to dissolve the congestion if the congestion occurrence cannot be avoided. In addition, TCP should throttle the window size when...