Approaches of Wireless TCP Enhancement and A New Proposal Based on Congestion Coherence ∗ Chunlei Liu Lucent Technologies 6100 E Broad St, Columbus, OH 43213 liu.223@osu.edu Raj Jain Department of Computer and Information Science Ohio State University, Columbus, OH 43210 jain@cis.ohio-state.edu Abstract TCP is known to have poor performance over unreliable wireless links where packet losses due to transmission er- rors are misinterpreted as indications of network conges- tion. TCP enhancements proposed in the literature differ in their signaling and data recovery mechanisms, applica- ble network configurations, traffic scenarios and locations where required changes are made. In this paper we catego- rize existing enhancements into several approaches. Moti- vated by these criteria, we propose a new enhancement that requires only local changes, but applies to a broad range of network and traffic configurations. Comparison with exist- ing algorithms show this new enhancement achieves excel- lent performance. 1 Introduction Transmission Control Protocol (TCP), the most widely used reliable transport protocol, was designed mainly for wired networks where transmission errors are rare and the majority of packet losses are caused by congestion. An im- portant assumption of TCP algorithm is that packet losses and the resulting timeout at the source are indications of congestion and the source should reduce its traffic rate on timeout [1]. When applied to wireless networks where transmission errors are frequent, TCP is found to have poor performance if proper enhancement is not used. This is because the as- sumption behind TCP congestion control algorithm that the majority of packet losses are caused by congestion is no longer true. When a wireless loss 1 is treated as a conges- tion loss, the effective TCP transmission rate drops to half. * This research was supported in part by National Science Foundation grants 9980637 and 9809018. 1 We call packet losses caused by wireless transmission errors “wireless losses” and those by network congestion “congestion losses”. If transmission errors happen frequently, the effective trans- mission rate of the wireless link becomes almost zero even though the network is not congested. The rapid development of mobile and wireless networks is a driving force for wireless TCP enhancements. In the past few years, numerous enhancements have been pro- posed. Theses enhancements differ in their signaling and data recovery mechanisms, the applicable network and traf- fic configurations, and locations where needed changes are made. These approaches have big impacts on the feasibil- ity, generality, effort and performance of the enhancements. In this paper we summarize the approaches used by major enhancement proposals in the literature, and use them as a set of criteria to evaluate different enhancements. Motivated by the set of criteria, a new enhancement is proposed and its performance is compared with existing methods. 2 Approaches of Wireless TCP Enhancement Enhancement proposals in the literature differ widely in their mechanism and algorithm. This section is an attempt to categorize them into several approaches. 2.1 End-to-end v.s. Split TCP is an end-to-end protocol — a packet is acknowl- edged only after it is received by its final destination. En- hancements that preserved this semantic are called end-to- end. Some enhancements split the entire path into a wired connection and a wireless connection and run TCP inde- pendently on both connections. When the transmission of a packet is complete in one connection, it is acknowledged to the source and relayed to the next connection. Such en- hancements do not preserve TCP’s end-to-end semantic and are called split enhancements. I-TCP [2] is an early protocol of split enhancements. Its major drawback is that acknowledgments received by the sender do not mean that the packets have been received Proceedings of the 36th Hawaii International Conference on System Sciences (HICSS’03) 0-7695-1874-5/03 $17.00 © 2002 IEEE