How to share communication resources in networks has been an important question since the development of packet-switched networks. Recently, the explosion of the Internet, in both its population and the number of applications, forces the researchers to revisit this question with respect to the Internet resources, such as link bandwidth and routers.
In this dissertation, we study fairness of congestion control for the Internet. Congestion in the network happens when the input traffic exceeds the capacity of network resources such as links and switches. Congestion control is a set of mechanisms exercised by network elements such as sources of traffic, routers, or receivers to prevent or resolve the state of congestion. In the Internet, congestion control is implemented in the transmission control protocol (TCP).
First, we analyze and compare the two most popular TCP implementations: TCP Reno and TCP Vegas. We compare the performance characteristics of these implementations in two cases: the single connection case, and the multiple connections case. We demonstrate through analysis that TCP Vegas, with its better bandwidth estimation scheme, uses the network resources more efficiently and is fairer than TCP Reno. We also show that TCP Vegas suffers less packet loss in the slow start, and that TCP Vegas is incompatible with TCP Reno. Simulation results are given to confirm our analysis.
Second, we answer the question of whether fair end-to-end congestion control protocols exist. More precisely, we prove by construction the existence of congestion control protocols that converge to a fair equilibrium without the help of internal network nodes or routers. Using an end-to-end protocol, the hosts get implicit feedback from the network such as round-trip delays and throughput but no explicit signals are provided from the network routers.
We formulate a mathematical model, a multiclass fluid model, of data network with window-based control, which explains the dynamics of window-based control. Based on this model, we show that the flow rates of users are a well-defined function of the window sizes and the network topology. We also proved characteristics of the function such as continuity and differentiability. We then generalize the definition of fairness to incorporate the proportional fairness and the max-min fairness . We propose congestion control algorithms which achieves the convergence to the fair equilibrium we proposed. The convergence of protocols is proved using the Lyapunov function technique. Simulation results support our analysis.
Recommendations
Robust and fair Multicast Congestion Control (M2C)
Since 1995 and the Receiver-driven Layered Multicast (RLM) protocol, numerous multicast congestion control protocols have been proposed, such as RLC, FLID-SL, FLID-DL and finally the WEBRC protocol. However these protocols suffer from some limitations ...
Fair multicast congestion control (M2C)
INFOCOM'09: Proceedings of the 28th IEEE international conference on Computer Communications WorkshopsWe propose a new TCP-friendly multicast congestion control for very large groups of receivers. This protocol named M2C is layered and receiver-driven. Like TCP, M2C is composed of a Congestion Avoidance and a Slow Start state. However, M2C can re-...