Accurate analysis of TCP on channels with memory and finite round-trip delay

In this paper, we present an accurate analytical model for transport control protocol (TCP) over correlated channels (e.g., as induced by multipath fading) taking into account a finite round-trip delay. In particular, we develop models and analysis for studying four versions of TCP, namely, Old Tahoe, Tahoe, Reno, and New Reno. We focus on a single wireless TCP connection by modeling the correlated packet loss/error process as a Discrete Time first-order Markov chain. Our model explicitly incorporates important aspects such as slow start, congestion avoidance, fast retransmit and fast recovery. The main findings of this study are that: 1) an increasing round-trip time may significantly affect the throughput performance of TCP, especially when an independent channel is considered; 2) New Reno performs better than Reno and Tahoe when the channel is uncorrelated, whereas Tahoe's recovery strategy is the most efficient when the channel correlation is high; and 3) the maximum window size does not play a determinant role in increasing throughput performance in both correlated and independent channels. While some of these conclusions confirm what other authors have observed in simulation studies, our analytical approach sheds some new light on TCP's behavior.