A Markovian Model for the Stationary Behavior of TCP

This paper presents a discrete-time Markov chain model for the Reno version of TCP, the transmission control protocol for reliable transport on the Internet. The purpose is the evaluation of stationary TCP flows behavior using performance measures such as the mean throughput. The model is based on previous works which are generalized by taking into account the slow start phases that appear after each time-out recovery. We consider the three different phases of the protocol : time-out, slow start and congestion avoidance and we obtain analytical expressions for the mean number of segments sent during each of these phases and their mean duration. Our model also allows us to evaluate the proportion of losses due to the timer expiry and the duplicate acknowledgments