Analytic models of TCP performance

[摘要] ENGLISH ABSTRACT: The majority of tra c on the Internet uses the Transmission Control Protocol (TCP) as atransport layer protocol for the end-to-end control of information transfer. Measurement,simulation and analytical models are the techniques and tools that can be used to understandand investigate the Internet and its performance. Measurements can only be used to exploreexisting network scenario or otherwise become costly and inexible with the growth andcomplexity of the Internet. Simulation models do not scale with the growth of networkcapacities and the number of users. Computationally e cient analytical models are thereforeimportant tools for investigating, designing, dimensioning and planning IP (Internet Protocol)networks.Existing analytical models of TCP performance are either too simple to capture the internaldynamics of TCP or are too complex to be used to analyze realistic network topologies withseveral bottleneck links. The literature shows that thexed point algorithm (FPA) is a veryuseful way of solving analytical models of Internet performance. This thesis presents fast andaccurate analytical models of TCP performance with the FPA used to solve them.Apart from what is observed in experimental literature, no comprehensive proof of the convergenceand uniqueness of the FPA is given. In this thesis we show how the FPA of analyticalmodels of reliable Internet protocols such as TCP converges to a uniquexed point. Thethesis speci es the conditions necessary in order to use the FPA for solving analytical modelsof reliable Internet protocols. We also develop a general implementation algorithm of theFPA of analytical models of TCP performance for realistic and arbitrary network topologiesinvolving heterogenous TCP connections crossing many bottleneck links.The models presented in this thesis give Internet performance metrics, assuming that onlybasic network parameters such as the network topology, the number of TCP connections, linkcapacity, distance between network nodes and router bu er sizes are known. To obtain theperformance metrics, TCP and network sub{models are used. A closed network of :=G=1queues is used to develop each TCP sub-model where each queue represents a state of a TCPconnection. An M=M=1=K queue is used for each network sub{model which represents theoutput interface of an IP router with a bu er capacity of K ������������1 packets. The two sub-modelsare iteratively solved. We also give closed form expressions for important TCP performance values and distributions.We show how the geometric, bounded geometric and truncated geometric distributions canbe used to model reliable protocols such as TCP. We give models of the congestion windowcwnd size distribution by conditioning on the slow start threshold ssthresh distribution andvice-versa. We also present models of the probabilities of TCP timeout and triple duplicateACK receptions.Numerical results based on comparisons against ns2 simulations show that our models aremore accurate, simpler and computationally more e cient than another well known TCPmodel. Our models can therefore be used to rapidly analyze network topologies with severalbottlenecks and obtain detailed performance metrics.

[发布机构] Stellenbosch University