Engineering Mobile Proxy Design for Wide-Area Wireless

The vision of a broadly unified and pervasive communication paradigm where people can freely communicate with wide range of devices (phones, PDAs, palmtops etc.) using networks of wired and wireless links is gradually taking shape [3], [11], [12]. Applications that will run over a communication infrastructure composed of heterogeneous links (wired and wireless) will be susceptible to the vagaries of underlying network performance effects. This might be an outcome of the inexperience in the handling of a completely new wireless network technology, or integrating diverse network infrastructure, while still reusing the ‘staple’ (Internet) protocols never attuned for such networks. Manifestation of the network performance effects will result in severe degradation of the application performance, lowering the achievable Quality of Service (QoS) or essentially hindering information access and even impeding data transfer to the end user. In this paper, we focus on how seamlessly we can integrate wide-area wireless networks with the Internet without trading off on performance. The goal is to design and implement a transparent mobile proxy that utilizes novel schemes that yields a better performance and mitigates any existing performance bottlenecks to enable a seamless integration. An integration of wide-area wireless with the Internet is likely to pose many impediments, since very little is known (as of now) on practical performance problems related to networks like GPRS or 3G. While the Internet acts as a testbed for actual measurements for wired networks, similar performance measurements pertaining to wide-area wireless technologies will be available only after their wide-scale deployment. Take for instance, General Packet Radio Service(GPRS), a packet-based wireless extension of GSM, which is being deployed throughout much of the world. GPRS offers an “always on” connectivity to mobile users, with a geographical coverage over a wide-area and data rates comparable to that of conventional fixed-line telephone modems. This will be soon expected to be followed by the deployment of third generation wireless networks offering bandwidths of upto 2Mbps. However, despite the momentum behind GPRS and 3G, little has been done to characterize GPRS/3G links and evaluate protocol performance such as TCP and HTTP on top of it. There are some interesting simulation studies on GPRS [8], [10] in this direction; however, our preliminary measurements indicate actual performance over production GPRS networks to be somewhat different [1], [2], [32]. A vast body of past research have investigated protocol performance issues over wireless. Unfortunately, most of the previous work focus on improving performance over wireless LANs (WLANs), which exhibit link characteristics that are somewhat different from GPRS. For example, we have found GPRS links to have very high RTTs [1] of the order of 10-20 times that of a typical WLAN environment. Apart from the low bandwidth nature of the wide-area GPRS, they also suffer from ack compression as well as link outages (link “stalls”) that could eventually lead to packet loss over the link. Thus, TCP performance over GPRS suffer in many ways [1]: A slow-start phase that takes many seconds (due to high RTTs) for the window to ramp-up and fully utilize the link, Excess queuing over the downlink that can result in gross unfairness to other TCP flows, and a high probability of timeouts during intial connection request, Spurious TCP timeouts due to occasional link ‘stalls’, Slow recovery (many seconds) after timeouts. Following on from this investigation, we focus our current research to delineate what other practical performance problems might surface in a typical wide-area wireless type of environment. As we uncover many of these problems, tackling them would be crucial for better application performance. The aims of this project are as follows: 1) understanding link characteristics of wide-area wireless networks (GPRS as a first step), 2) identifying protocol performance problems experienced by TCP and HTTP, 3) devising mobile proxy based solutions that can overcome or even optimise such bottlenecks and, 4) experimentally demonstrating the overall efficacy of our solution.