Improving TCP throughput in 802.11 WLANs with high delay variability.pdfVIP

Improving TCP Throughput in 802.11 WLANs with High Delay Variability Georgios I. Fotiadis and Vasilios A. Siris Foundation for Research and Technology - Hellas (FORTH), Institute of Computer Science P.O. Box 1385, GR 711 10 Heraklion, Crete, Greece {fotiadis,vsiris}@ics.forth.gr Abstract— Reduced throughput of TCP over wireless networks is mainly due to misinterpretation of wireless losses as congestion signals and high delay variability because of fluctuating radio channel quality. High delay variability often leads to spurious TCP timeouts which result in severe throughput degradation. In this paper, we show the negative impact of large delay variations in a mixed wired and 802.11 network scenario with bursty errors, and propose an adaptive algorithm that adds delay to packets in order to prevent spurious timeouts. Our algorithm can adjust to specific network characteristics such as propagation delays, loss rate, and number of mobile nodes, and is applied at the access point of a wireless LAN. I. INTRODUCTION The Transmission Control Protocol (TCP) is the most widely used protocol for reliable data transfer over the In- ternet. TCP was initially designed for wired networks but the growing popularity of mobile applications has established its deployment in wireless environments as well. The main goal of TCP is to probe the available bandwidth of an end-to- end connection and efficiently adapt the sender’s transmission rate so as to prevent network congestion. Packet losses are the most convenient way to detect congestion in wireline networks where transmission errors are very rare and most losses are due to buffer overflow at the intermediate routers. When a loss is detected the sender reduces its transmission rate, so that the overloaded buffers drain. However, network congestion is usually not the reason for packet losses in wireless networks. Errorprone transmissions over the wireless channel and high delay variability are misinterpreted as signals of congestion