Performance analysis of TCP connections sharing a congested Internet link

Many Internet channels are today heavily loaded with traffic, most of which is due to TCP connections supporting bursty applications, such as web browsing. Models that can accurately estimate the performance gains that could be obtained by increasing the link bandwidth are called for to support the Internet design and planning activities. A number of analytical models of the TCP behavior have been presented in the recent literature. Most of them consider the interaction among TCP connections sharing one bottleneck link, but assume that they behave independently of each other, thus neglecting the synchronization induced by periods of high loss probability due to buffer overflows. In this paper, we develop a new model to describe the behavior of a number of TCP connections that share a bottleneck link, considering the effects of the synchronization among connections. The model is based on Markovian assumptions and on two GSPN (generalized stochastic Petri net) descriptions of the system. From the two GSPN descriptions, using a fixed point algorithm, several interesting performance metrics referring to the TCP connections can be derived. A validation of our modeling approach is provided through extensive comparisons of the model's results with the output of the LBL simulator, `ns version 2', that is known to provide quite detailed descriptions of the dynamics of the Internet protocols. Comparisons show that our model succeeds in providing an accurate representation of the behavior of TCP connections under several different settings.