Mobile Internet system implementation and measurement

Abstract

It is forecast that mobile data will grow from 2:5EB in 2010 to 18EB in 2014, in which more than 90% will be contributed by laptops and smartphones. As the popularity of mobile Internet increases, both industry and academic communities have been stimulated to investigate performance of existing networking systems for optimizing or creating new networking elements to support ubiquitous Internet access. This thesis investigated the performance of mobile HSPA. Based on the findings, a new HSPA based networking system has been devised for serving robust mobile Internet on the Long Distance Train (LDT). Having provided broadband equivalent speed over cellular data network, recent years have witnessed a rapid expansion of High-Speed Packet Access (HSPA, a 3.5G cellular standard) in the world. In the meantime, sale of mobile devices which are capable of accessing Internet has also proliferated. Fueled by the popularity of mobile Internet and devices, passengers on public transport are more likely to entertain themselves (or for business) via mobile data access. However, it has been a persistent problem to provide ubiquitous wireless connections to all kinds of wireless devices on fast moving vehicles traversing a large expanse of area, such as Long Distance Train. One difficulty is that the performance of HSPA network in mobility at application layer remains largely unknown to research communities. Second, normally only cellular (e.g. UMTS, HSPA and LTE) signals are available during most of the journey, which makes other mobile device without cellular interfaces not capable of networking. So it is imperative for us to provide diversity of wireless accesses based on the cellular connection. Even such techniques are available to provide wireless diversity, the unstable nature of the cellular signals has to be addressed carefully, especially on a fast moving vehicle, to provide certain degrees of QoS. To overcome the first difficulty, an empirical study on the performance of mobile HSPA networks has been conducted in Hong Kong via extensive field tests. The study, from the viewpoint of end users, covers virtually all possible mobile scenarios in urban areas, including subways, trains, off-shore ferries and city buses. Mobility has been confirmed having largely negative impacts on the performance of HSPA networks, as fastchanging wireless environment causes serious service deterioration or even interruption. Meanwhile the field experiment results have shown unexpected new findings and thereby exposed new features of the mobile HSPA networks, which contradict commonly held views. It is surprising to find out that mobility can improve fairness of bandwidth sharing among users and traffic flows. Also the triggering and final results of handoffs in mobile HSPA networks are unpredictable and often inappropriate, thus calling for fast reacting fallover mechanisms. In-depth research has been conducted to furnish detailed analysis and explanations to what have been observed. As revealed in the empirical study, cellular networks often see random blackouts, and the bandwidth of cellular networks is also limited with high mobility. But the study has also demonstrated a number of networking opportunities which can be exploited by a group of mobile HSPA users. As a consequence, to overcome the second difficulty, a prototype system is provided for converting all available cellular wireless networking opportunities into other wireless accessing possibilities. Wi-Fi is chosen as the target wireless technology because of its great popularity, and the scenario is placed on an LDT. As we all know, LDT is so long that many such converting devices, which will be called D-router later, are needed to cover the whole length. The cellular communication capability of all D-routers along the LDT are integrated together, and certain degree of individual D-router failures can be amortized onto the whole D-router chain. Experiments and simulations show that DragonNet significantly increase the Internet connection time by a factor of 4 and aggregate throughput by a factor of 2 on average.