A measurement-based united congestion control resolution for TCP/IP networks

A united framework for transport layer is proposed to improve its congestion control status quo. Designed to be a two-layer structure, the framework utilizes measurement infrastructure or services in its up layer to understand internal network states, which is then used to derive a common control objective for down layer on amount of extra packets. The down layer, composed of end systems, carries out that objective by window or rate adjustment reacting to lose and delay signals, which is performed in a protocol specific way. To increase system robustness to measurement errors, fuzzy inference technique is employed in objective deriving stage. And the layered design helps to combine two kinds of operations conducted on totally different time scales, i.e. network measurement on minutes and congestion control on several to hundreds of milliseconds. Down layer algorithms for both TCP and UDP end systems are described. Simulations on ns-2 validate that the proposed framework and algorithms can achieve high efficiency and fairness for both protocols. Meantime it exhibits reasonable robustness to errors. And in fulfilling the common objective setting up by up layer, UDP flows behave quite TCP-friendly and smoothly.