Research Abstracts - 2006

A Measurement Study of the Vehicular Internet Access Using Unplanned 802.11 Networks

Vladimir Bychkovsky, Bret Hull, Allen Miu, Eugene Shih, Hari Balakrishnan & Samuel Madden

Project Overview

The penetration and proliferation of 802.11 Wi-Fi in homes and offices around the world has been phenomenal---Jupiter Research estimates the number of home-deployed Wi-Fi access points (APs) in the United States alone to be 14.3 million (65% of online households), with the number of deployed APs rising every year. Many home users deploy Wi-Fi in their homes and connect to the Internet over broadband cable modem, DSL, or even fiber. Because these Wi-Fi networks and upstream broadband access links are often idle, they offer the potential for using these home networks for Internet access to other users. We are interested in understanding what sort of performance one might expect from these unplanned, grassroots community networks, if they were to "open up" and provide Internet service to mobile users.

Before describing the technical questions and contributions of this project, we must note that there are several important issues concerning policy, business decisions, and law that must be resolved before this vision of an ``open Wi-Fi'' Internet access network can become real. We believe, however, that with the increasing deployment of open urban Wi-Fi networks in many cities around the world [1] and commercial activity (for example [2]) in this space, such networks are likely to become real in the next few years.

This effort is part of the CarTel mobile sensor computing project.

Project Objectives

The key technical question addressed in this project is: What is the performance of open Wi-Fi networks for mobile users, particularly for users in automobiles, as they move in urban and suburban areas where APs are currently deployed in great abundance?

More specifically, we answer the following questions:

1. For the mobility patterns and speeds observed by cars, what is the distribution of connection time per AP? What is the distribution of ``disconnection'' time (between connections to different APs)?
2. What is the distribution of throughput (and number of bytes) that could be achieved by single TCP connections uploading and downloading data in such a network?
3. What are the frame and packet loss rates observed, particularly as cars enter and leave the region ``covered'' by APs?
4. What is the effect of a car's speed on these metrics?

We answer these questions by running a carefully controlled measurement study of open APs as they are deployed currently in and around Boston and Seattle. In future, we hope to extend this experiment to other locations.

Initial Progress

So far in this study, six cars driving over a six-month period under ``typical'' driving conditions in and around these cities attempt to connect via existing deployed open APs and, with experiments carefully constructed to not ``steal'' bandwidth or disrupt the performance for owners of these APs, estimate the performance characteristics of these wireless networks. Though there has been other work [3] on characterizing Wi-Fi connectivity for moving vehicles, they have been limited in scope, being restricted to a single access point. We believe that ours is the first study to characterize connectivity ``in the wild,'' with cars traveling on roads with actual traffic at the speed at which traffic naturally flows, and at various times of day.

Our driving patterns discovered over 13,000 access points, about 1,300 of which were open; 500 of these permitted end-to-end access. The distribution of connection time to an AP had a median of 6 seconds with some simple caching optimizations to avoid time-consuming DHCP lookups. Our calculations suggest that the median throughput is likely to be bottlenecked by the broadband bandwidth limits in people's homes (a median of about 20 KBytes/s, with the 90th percentile about 35 KBytes/s). The median duration between observing open access points in our experiments is about 70 seconds (\ie, this is the median disconnection time), and which could increase by a factor of 10 or 20 if more access points were to participate in this "grassroots" network in the future. We also found that the probability of a successful connection was uniform at all the speeds we measured (primarily between 10 and 80 km/h).

Our study was motivated largely by curiosity, to see what the characteristics of such grassroots networks set up by home users might be. Our initial results are promising and we are continuing this study. We are also interested in enabling such networks. Of course, there are several commercial, legal, and policy issues to be ironed out for this vision to become reality, but our technical conclusion is that such grassroots Wi-Fi networks are both viable for a variety of vehicular applications, particularly ones that can tolerate some intermittent connectivity (mobile sensor networks with cars, numerous messaging applications, Web access using caching). We are optimistic that such alternate networks will appear in the near future supporting both peripatetic clients and users in cars driving more continually between access points.

Funding

This project is funded by Quanta Corporation.

References:

[1] B. Tedeschi. Big Wi-Fi Project for Philadelphia. New York Times, September 2004.

[2] FON home page. http://en.fon.com

[3] J. Ott and Dirk Kutscher. Drive-thru Internet: IEEE 802.11b for Automobile Users. In Proc. IEEE INFO- COM, Hong Kong, March 2004.