IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 6, NO. 4, AUGUST 1998 485 Comparative Performance Analysis of Versions of TCP in a Local Network with a Lossy Link Anurag Kumar, Senior Member, IEEE Abstract—We use a stochastic model to study the throughput performance of various versions of transport control protocol (TCP) (Tahoe (including its older version that we call OldTahoe), Reno, and NewReno) in the presence of random losses on a wireless link in a local network. We model the cyclic evolution of TCP, each cycle starting at the epoch at which recovery starts from the losses in the previous cycle. TCP throughput is computed as the reward rate in a certain Markov renewal–reward process. Our model allows us to study the performance implications of various protocol features, such as fast retransmit and fast recovery. We show the impact of coarse timeouts. In the local network environment the key issue is to avoid a coarse timeout after a loss occurs. We show the effect of reducing the number of duplicate acknowledgements (ACK’s) for triggering a fast retransmit. A large coarse timeout granularity seriously affects the performance of TCP, and the various protocol versions differ in their ability to avoid a coarse timeout when random loss occurs; we quantify these differences. As observed in simulations by other researchers, we show that, for large packet-loss probabilities, TCP-Reno performs no better, or worse, than TCP-Tahoe. TCP-NewReno is a considerable improvement over TCP-Tahoe, and reducing the fast-retransmit threshold from three to one yields a large gain in throughput; this is similar to one of the modifications in the recent TCP-Vegas proposal. We explain some of these observations in terms of the variation of fast-recovery probabilities with packet- loss probability. Finally, we show that the results of our analysis compare well with a simulation that uses actual TCP code. Index Terms—Congestion control, mobile Internet, models for TCP, TCP performance. I. INTRODUCTION I T IS WELL KNOWN that transport control protocol (TCP) (the transport-layer protocol in the Internet) reacts to all packet losses as if they were caused by congestion, i.e., by stalling for a long timeout period and then dropping its transmission window. Thus, over a cellular wireless channel, in which there can be losses due to link errors and channel handovers, TCP can exhibit very poor performance. Several modifications have been proposed to the TCP loss-recovery and congestion-control mechanism to improve data throughput in the event of random loss; a simulation study of these has been reported in [6]. To study the performance of TCP over wireless channels, several researchers have recently used experimental testbeds to understand TCP behavior and to Manuscript received December 17, 1996; revised April 24, 1998; approved by IEEE/ACM TRANSACTIONS ON NETWORKING Editor A. U. Shankar. The author was with the Wireless Information Networks Laboratory (WIN- LAB), Rutgers University, Piscataway, NJ 08854 USA, on leave from the Department of Electrical and Communication Engineering, Indian Institute of Science, Bangalore 560 094, India (e-mail: anurag@ece.iisc.ernet.in). Publisher Item Identifier S 1063-6692(98)05642-8. propose and evaluate possible solutions; see, for example, [1], [2], [4], and [8]. Our work, reported here, belongs to a line of research that attempts to develop detailed analytical models of the TCP protocol in an effort to predict the performance of the various versions that are being proposed. Such modeling and analysis can also be used as a testbed for evaluating other variations, so that experimentation can be done for the few that are promising. The analysis effort also reveals the reasons for which the various effects are observed in the experimental work. The models we develop are analytical and parametric, so that we can quickly answer “what-if” questions or evaluate the effect of modifying a particular protocol or network parameter. In this paper we develop models and analyses for study- ing the bulk throughput of four versions of TCP, namely, OldTahoe (the original protocol from [7]), Tahoe, Reno, and NewReno [14], [6]. We assume a local network scenario, in which a host on a local area network (LAN) is transmitting bulk data to a mobile host connected to the LAN by a wireless link. Our models incorporate important aspects, such as slow start and congestion avoidance, coarse timers, fast retransmit, and fast recovery. Assuming an uncorrelated random packet- loss model on the wireless link, we obtain the data throughput as a function of the packet-error probability. Our results show how the throughput degrades, for each version, with increasing packet-loss probability. For a given loss probability, our results quantify the performance improvement provided by each ver- sion. We show that our analysis provides accurate quantitative evaluation by comparing its results with a simulation based on actual TCP code. Prior research closest to our work is that of Mishra et al. [12] and Lakshman and Madhow [11]; additional related references are also given in these papers. Mishra et al. only analyze the original protocol ([7]) and provide some supporting simulation and experimental results. They observe a cyclical structure in the TCP transmission process and they identify and analyze the Markov chain of congestion window sizes at loss instants. Both of these elements are key to our analysis as well, but we analyze the newer protocols with the fast-retransmit feature. Lakshman and Madhow consider OldTahoe and Reno, and model and analyze one or more TCP connections whose flow is constrained by a common bottleneck link. The link has a finite buffer, and the bandwidth (i.e., the bottleneck bandwidth) round-trip-delay product for each connection is large. There is, thus, the issue of buffer overflow at the bottleneck link. These authors study buffer sizing to obtain high TCP throughputs and obtain an approximation to the random loss probability so that random loss does not constrain throughput. They do not, 1063–6692/98$10.00  1998 IEEE