A New Protocol to improve TCP Performance in Network Mobility

Abstract

Network mobility (NEMO) is a mobility management solution that allows various types of moving networks, e.g. network of sensors deployed in a vehicle, to be permanently connected to the Internet. An onboard mobile router (MR) connects the moving network to the wired infrastructure by means of high-speed cellular or any other wide area mobile data services. One application of NEMO attracting commercial interest is the deployment of wireless local area networks inside public transport vehicles, e.g., trains and buses, to provide Internet access to passengers. However, unlike the traditional terminal mobility, where the mobile hosts (MHs) connect to the cellular base station directly, passengers using the NEMO solution encounter an additional wireless link (MR-MH) before their MHs get connected to the wired infrastructure. In this paper, we analyze and quantify the impact of the additional wireless link on the performance of the widely used TCP protocol. Our analysis reveals that TCP performance schemes designed for conventional terminal mobility are not as effective in network mobility. We propose on-board TCP (obTCP) to effectively address the wireless link related issues in network mobility. We compare its performance against a classical scheme, called snoop, known for its effectiveness in terminal mobility. Using analytical means we demonstrate that the performance gain of obTCP over snoop increases linearly with delays and exponentially with the loss probabilities in the wireless links. These analytical observations are validated through extensive ns-2 simulations. We then extend these analyses to obtain throughput models of snoop and obTCP in NEMO. Our simulations further demonstrate that obTCP can coexist with snoop in the same infrastructure (e.g., cellular base stations) without causing serious unfairness to each other.