Modeling Wireless TCP Connection Arrival Process

Ensuring quality of service (QoS) for high band- width applications over wireless networks is a challenging re- search problem. Accurate modeling of the wireless TCP connec- tion arrival process aids in designing QoS algorithms. In this paper we perform an extensive statistical analysis of collected wireless TCP connection arrival process. Unlike wired TCP connection arrivals which are readily modeled as a heavy-tailed Weibull renewal process, we show that more sophisticated models are needed for wireless. Our proposed model, a multinomial canonical cascade with 3 stages, is able to capture the long- range dependent and multifractal characteristics of wireless TCP connection interarrival times. In addition, the model gives good approximation to the packet loss rate of measured wireless traffic trace. I. INTRODUCTION The transmission control protocol (TCP) is the dominant transport protocol for computer network traffic. The main function of TCP is to ensure end-to-end reliable delivery of packets routed over heteregenous computer networks. Built into TCP is a congestion avoidance mechanism that moderates the transmission of packets so as to prevent the network from being flooded with excessive loads (1). In general, TCP works well for wired networks. However, the unstable wireless channel can significantly reduce the efficiency of TCP in wireless networks unless some redressing actions are taken (2). Wireless communication for data traffic is a rapidly bur- geoning industry as evident by the growth of wireless local area networks (WLANs) and 3G cellular networks. Voice traffic will no longer be the dominant traffic carried over wireless networks due to the increasing popularity of data based applications. Therefore, the classical teletraffic theory of Erlang may become obsolete in light of fractal like charac- teristics of data traffic (3). The goal of this paper is to model the arrival process of TCP connections for HTTP/1.1 (i.e., persistent connections) traffic over wireless channels. HTTP traffic was chosen as it represents the bulk of the traffic in the Internet today. The TCP connection arrival process is important for dimensioning web server proxies, signaling, dynamic routing and evaluation of congestion control algorithms. Unlike previous work (cf. (4) and references therein) on TCP modeling which focused on analyzing the TCP mechanisms for a single host transmitting packets over a wireless link, we adopt a trace driven approach. This approach simultaneously takes into account the different user behaviors, the utilization degree of the network, the effects of TCP protocol, and channel conditions since they are all implicitly captured by the TCP connection arrival process. The derived model will then allow us to simulate realistic workloads for wireless networks since the model is based on actual wireless traces. Previous analysis of TCP mechanisms from a single host perspective (cf. (4) and references therein) provides a better understanding of the TCP protocol but fails to capture the network level impacts of simultaneous transmission by multiple hosts, as is the case in fully deployed wireless networks.