Modeling the effect of short-term rate variations on TCP-friendly congestion control behavior

Transmission Control Protocol (TCP) is the dominant transport protocol in today\´s Internet. To maintain stability of the Internet, flows other than TCP must be "friendly" to TCP flows, or share network bandwidth fairly with TCP traffic. Usually a flow is claimed to be TCP-friendly when its throughput is theoretically the same as the throughput of a TCP flow when they experience the same congestion signals. However, when flows compete for bandwidth, they may not have the same perception of congestion. Therefore, measured bandwidth shares of flows are not necessarily equal, even when all flows are theoretically designed to be TCP-friendly. To study the effect on bandwidth sharing of interactions among a set of competing TCP-friendly flows, we built a hybrid state-space-based model of TCP using differential equations\´ and event-driven switches. We modified the TCP model, using TCP\´s additive-increase multiplicative-decrease (AIMD) congestion avoidance algorithm with different increase and decrease parameters, to create theoretically TCP-friendly protocols with various short-term transmission rates. We prove that TCP-friendly flows result in a stable attractor if the backing off of flow transmission rates is synchronized. Experiments using our model and using ns simulator with unsynchronized backing off show unfairness among competing flows with different short-term behaviors