Official URL: http://www.hamilton.ie/publications/baruch_even_th...

The performance of the Internet is determined not only by the network and hardware technologies that underlie it, but also by the software protocols that govern its use. In particular, the TCP transport protocol is responsible for carrying the great majority of traffic in the current internet, including web traffic, email, file transfers, music and video downloads. TCP provides two main functions. First, it provides functionality to detect and retransmit packets lost during a transfer thereby providing a reliable transport service to higher layer applications. Second, it enforces congestion control. That is, it seeks to match the rate at which packets are injected into the network to the available network capacity. A particular aim here is to avoid so-called congestion collapse, prevalent in the late 1980s prior to the inclusion of congestion control functionality in TCP. Over the last decade or so, the link speeds within networks have increased by several orders of magnitude. While the TCP congestion control algorithm has proved remarkably successful, it is now recognised that its performance is poor on paths with high bandwidth-delay product, e.g. see [13, 8, 14, 26, 12] and references therein. With the increasing prevalence of high speed links, this issue is becoming of widespread concern. This is reflected, for example, in the fact that the Linux operating system now employs an experimental algorithm called BIC-TCP[26] while Microsoft are actively studying new algorithms such as Compound-TCP[25]. While a number of proposals have been made to modify the TCP congestion control algorithm, all of these are still experimental and pending evaluation as they change the congestion control in new and significant ways and their effects on the network are not well understood. In fact, the basic properties of networks employing these algorithms may be very different to networks of standard TCP flows. The aim of this thesis is to address, in part, this basic observation.