Impact of Scheduling and Buffer Sizing on TCP Performance over IMT-Advanced Kunimasa Kakuda Graduate School of Computer Science and Systems Engineering Kyushu Institute of Technology, Fukuoka, Japan Email: kakuda@infonet.cse.kyutech.ac.jp Masato Tsuru Network Design Research Center, Kyushu Institute of Technology, Fukuoka, Japan Email: tsuru@ndrc.kyutech.ac.jp Abstract—IMT-Advanced is gaining much attention as the next-generation mobile communication technology. In this study, we evaluate the TCP throughput performance of mobile users competing on an IMT-Advanced shared downlink via a network simulation by using realistic environmental parameters. Our focus is on evaluating the impact of two basic types of time- frequency scheduling in which not only a time-varying wireless condition is used but also a frequency-block-dependent wireless condition of each user is exploited using various buffer sizes at radio base stations and various user moving speeds. We investigated the subtle tension between the scheduling, buffer sizing, and user moving speed for the TCP throughput. Our results imply that more TCP awareness is necessary for sharing the large bandwidth of IMT-Advanced downlinks more efficiently. Keywords-IMT-Advanced, TCP throughput, MaxCIR sched- uler, Proportional Fair scheduler I. I NTRODUCTION Recently, mobile communications have expanded rapidly worldwide. Third-generation mobile communication systems (3G), including wideband code division multiple Access (W- CDMA) and high-speed packet access (HSPA) are now being used. Moreover, cellular-phone-oriented mass contents and individual user-data traffic are increasing in conjunction with each other every year. Therefore, the demand for high-speed mobile communications is currently increasing. Subsequently, the international mobile telecommunication-Advanced (IMT- Advanced) known as 4G, which attains improvement in communication speed, is also gaining much attention [1]. To achieve high-speed wireless access, IMT-Advanced would need a wireless bandwidth wider than the 5 MHz of the wireless bandwidth of W-CDMA. Generally, because the wireless condition of each user, i.e., mobile station (MS), varies in time because of the interference, fading, or the movement of the MS, the shared wireless chan- nel for downlink access is divided into fixed-length time slots at every transmission time interval (TTI). A scheduler at the radio base station (BS) allocates individual time slots to MSs that depend on the wireless condition of each MS and by using feedback information from each MS to each BS. Furthermore, in order to utilize the wider bandwidth efficiently, the entire bandwidth is divided into smaller bandwidths (i.e., frequency blocks or subchannels); and time-frequency scheduling assigns each subchannel to one MS at each time slot, whereas tradi- tional time scheduling assigns the entire bandwidth to one MS for every time slot. A subchannel assignment in a wireless link with time-frequency scheduling is illustrated in Fig. 1, where different colors/patterns denote different users (MSs). The transmission control protocol (TCP) is a dominant transport-layer protocol for end-to-end reliable data transfer, and most of the Internet applications rely on TCP. However, because TCP adaptively controls its sending rate of IP data- grams in response to datagram losses and delay-time varia- tions, its throughput characteristic is highly sensitive to diverse factors, and is difficult to estimate its throughput characteristic analytically when the network condition varies dynamically (e.g, [2]) . On IMT-Advanced environments, the performance of TCP-based applications such as file transfer is expected to be greatly affected by both configurable conditions, such as the type of scheduling mechanism and the buffer sizing of the BS; and environmental conditions, such as the delay time of a core network and the moving speeds of MSs. Therefore, it is indispensable to conduct the evaluation and investigation of the TCP throughput performance of competing mobile users quantitatively. Performance evaluation of IMT-Advanced communications using network simulators have begun to get more attention (e.g., [3]). However, for TCP flow-level performance of com- peting mobile users in IMT-Advanced via realistic network simulations, little is found in the literature. In our previous work [4], the TCP performance in the Evolved UTRA and UTRAN environments by using time scheduling and time- frequency scheduling was evaluated using a network simu- lation by employing realistic packet-error patterns. Building upon that work, here we examine the TCP performance in the new IMT-Advanced environments by using typical time- frequency schedulers (i.e., MaxCIR and Proportional Fair schedulers). Our focus is on evaluating the subtle tension between the scheduling, buffer sizing, and user moving speed for the TCP throughput. The rest of this paper is organized as follows. In Section II, we introduce two typical scheduling algorithms. Section III explains our network simulation settings for large (continuous) file downloads, and the simulation results are investigated in terms of the impact on the base-station buffer size in Section 32 AICT 2011 : The Seventh Advanced International Conference on Telecommunications Copyright (c) IARIA, 2011. ISBN: 978-1-61208-123-6