Some Corrections and new Results from a Scheme to Improve TCP Over Wireless Networks Lucas D. P. Mendes #1 , Jos´ e M. C. Brito #2 # National Institute of Telecommunications P.O. Box 05 - 37540-000 Santa Rita do Sapuca´ ı - MG - Brazil 1 palhao@mtel.inatel.br 2 brito@inatel.br Abstract—Many models for TCP Throughput Prediction were proposed in the literature. However, the first ones only considered the most common networks at that time: wired networks. Thus, these models could not predict the behavior of a network that included wireless links. Also, different Link Layer protocols had to be considered, resulting in different modifications of the former throughput prediction models. During our research on TCP Models, we have found an interesting modification including ARQ, which is one of our objects of study. Also, this modification proposed a new frame retransmission scheme, which could reduce the TCP segment loss probability in a wireless link. However, when we tried to use this modified model, some errors appeared. For instance, the equation to calculate the average frame retransmissions number did not consider all the error possibilities. Thus, we decided to make some corrections in order to use this model in future works. Based on the corrected model, new comparisons were made and new conclusions, different from the original ones, were achieved. I. I NTRODUCTION TCP Throughput Prediction Models have been used lately to simulate a network behavior before it is actually implemented. Many models were created before wireless technologies came to be part of networks. For instance, Lakshman and Madhow [1] created a model capable of predicting TCP Reno and Tahoe throughputs. Later, Pinto and Brito [2] extended the TCP Model to include a wireless node in the network, applying Forward Error Correction (FEC), and stop-and-wait (SW) and go-back-N (GBN) ARQ protocols to the wireless link, in order to mitigate the channel losses harmful effects on TCP. Furthermore, we extended Pinto’s work by including a Cross- Layer Design to maximize throughput when using SW-ARQ [3] and GBN-ARQ [4] at the Data Link Layer. Another example of TCP Model is Padhye’s Model [5]. Compared to Lakshman’s, it is a simpler model. Furthermore, Liu et al. [6] extended Padhye’s work to include an ARQ protocol at the Data Link Layer. We tried to use this modified model to compare results with the modified version of Laksh- man’s Model. If we had achieved results close to simulations, we would have tried to apply our Cross-Layer Designs to Liu’s extension, since it is simpler and computationally more efficient than Pinto’s extension. However, it was not possible to validade Liu’s results. Moreover, analysing his equations, we have found errors that we propose to correct in this work. The remainder of this work is organized as follows: next section shows Liu’s system used to derive his equations; in section III we explain Liu’s original proposed and conventional schemes, and also our corrected schemes; section IV shows an overview of Padhye’s Model, explaining how Liu’s modifica- tions are applied; the results achieved with our corrections are showed in section V; and, finally, in section VI we present our conclusions and proposals for future works. II. SYSTEM DESCRIPTION The system used by Liu et al. [6] to propose their scheme to improve TCP throughput is depicted in Fig. 1. A TCP connection will be started between the fixed and the mobile hosts through a base station. A TCP packet takes D seconds to be transmitted and propagate from the fixed host to the base station, and it has a probability L of being lost on the wired link. Then, the base station breaks the packet in N frames and stores them in the Data Link Layer buffer. Each frame is transmitted through the wireless link to the mobile host with delay RT T L and loss probability P E . Fig. 1. System Model. There is a maximum number of frame retransmissions, M r , which has different meanings for de proposed and conventional schemes, explained in the next section. If M r retransmissions are reached and the packet could not be fully transmitted to the mobile host, all the frames corresponding to that packet are 978-1-4244-3584-5/09/$25.00 2009 IEEE ISWCS 2009 244