Performance Analysis of TCP over Wired and Wireless Network

Abstract

Transmission Control Protocol (TCP) currently accounts for about 90% applications and 80% data of network traffic, and plays the dominant role in Internet transmission. In this dissertation, we present three studies of TCP performance over wired and wireless networks. In the first study, we mainly focus on the stability of TCP⁄AQM (Active Queue Management) systems for wired network environment. In particular, we study the local and global stability of TCP-newReno⁄RED under many flows regime. By using a normalized discrete-time model, which is simple, we analyze the global stability in a very efficient manner, and the results show that by properly choosing RED parameters, we can always make the TCP-newReno⁄RED system globally stable.

The second study concerns the TCP performance when the buffer size is scaled up differently from the traditional one at a link with large capacity shared by many flows. Specifically, we consider the buffer size chosen on the order of (Nˆalpha) (0< alpha < 1), where N flows share the link of capacity O(N). We then develop a doubly-stochastic model for a TCP⁄AQM system with many flows by taking into account the packet-level dynamics over fine time scales. We show that, under our scale, the system always performs well in the sense that the link utilization goes to 1 and the loss ratio decreases to zero as the system size $N$ increases. We verify our results using extensive ns-2 simulations.

Finally, we then analyze the impact of TCP flows on the wireless networks from the resource allocation point of view. We propose a TCP-AIMD aware call admission control scheme over wireless networks. Our scheme is based on a two level CAC framework, which takes both the call level and the packet level dynamics into account. The interaction between two levels is characterized through a single metric called the Quality of Service (QoS) guaranteed capacity. The performance of the proposed scheme is then analyzed and extensive simulation results are presented under different scenarios. Our results show that the proposed scheme can improve the system performance both on call and packet levels.