Beyond Deployments and Testbeds: Experiences
with Public Usage on Vehicular WiFi Hotspots
Joshua Hare, Lance Hartung, and Suman Banerjee
DDept. of Computer Sciences, University of Wisconsin-Madison
Jan. 31, 2013
WiRover System
WiRover has been running on public transit buses for 18 months
providing a WiFi hotspot for passengers.
9,331 hours operation
17,567 unique devices connection
337.53GB download, 48.19 upload
Goal: Share experiences with deploying a vehicular internet access
service on public transit buses
Joshua Hare, Lance Hartung, and SumanBeyond
Banerjee
Deployments and Testbeds: Experiences with Public UsageJan.
on Vehicular
31, 2013 WiFi
2 / Hotspots
17
WiRover System
Figure: An overview of WiRover system
Joshua Hare, Lance Hartung, and SumanBeyond
Banerjee
Deployments and Testbeds: Experiences with Public UsageJan.
on Vehicular
31, 2013 WiFi
3 / Hotspots
17
Design Issues
Encapsulation and simple flow scheduling
Flow migration during link failures
Traffic optimizations
Remote management of a vehicular Internet service
Joshua Hare, Lance Hartung, and SumanBeyond
Banerjee
Deployments and Testbeds: Experiences with Public UsageJan.
on Vehicular
31, 2013 WiFi
4 / Hotspots
17
System Implementation
Figure: An overview of hardware which is installed on 2 Metro buses and 3 Coach
buses
Joshua Hare, Lance Hartung, and SumanBeyond
Banerjee
Deployments and Testbeds: Experiences with Public UsageJan.
on Vehicular
31, 2013 WiFi
5 / Hotspots
17
Usage Characteristics
Repeat usage comprise a more significant fraction of our sessions
Figure: The growth of devices observed
over time
Figure: The growth of sessions observed
over time
Joshua Hare, Lance Hartung, and SumanBeyond
Banerjee
Deployments and Testbeds: Experiences with Public UsageJan.
on Vehicular
31, 2013 WiFi
6 / Hotspots
17
Usage Characteristics
Usage of different devices on different buses
Figure: Comparison of average usage characteristics for various device types
Joshua Hare, Lance Hartung, and SumanBeyond
Banerjee
Deployments and Testbeds: Experiences with Public UsageJan.
on Vehicular
31, 2013 WiFi
7 / Hotspots
17
Usage Characteristics
Study on uplink and downlink traffic
Figure: Average volume of uplink and downlink traffic over a user’s session time.
Joshua Hare, Lance Hartung, and SumanBeyond
Banerjee
Deployments and Testbeds: Experiences with Public UsageJan.
on Vehicular
31, 2013 WiFi
8 / Hotspots
17
Usage Characteristics
Smartphone usage
Figure: Distribution of various device type observed each month
Joshua Hare, Lance Hartung, and SumanBeyond
Banerjee
Deployments and Testbeds: Experiences with Public UsageJan.
on Vehicular
31, 2013 WiFi
9 / Hotspots
17
Usage Characteristics
Traffic destinations
Figure: The top 15 web destinations sorted by the total number of received bytes
Joshua Hare, Lance Hartung, and SumanBeyond
Banerjee
Deployments and Testbeds: Experiences with Public Usage
Jan.
on 31,
Vehicular
2013 WiFi
10 /Hotspots
17
Data Reduction Techniques
Compression
Figure: The top 10 HTTP content types sorted by the number of bytes that would
have been saved if the content had been compressed using gzip
Joshua Hare, Lance Hartung, and SumanBeyond
Banerjee
Deployments and Testbeds: Experiences with Public Usage
Jan.
on 31,
Vehicular
2013 WiFi
11 /Hotspots
17
Data Reduction Techniques
Packet-level redundancy elimination
Figure: Savings in a 10-hour trace using Packet Caches between the controller and the
gateway
Joshua Hare, Lance Hartung, and SumanBeyond
Banerjee
Deployments and Testbeds: Experiences with Public Usage
Jan.
on 31,
Vehicular
2013 WiFi
12 /Hotspots
17
Data Reduction Techniques
Caching
Figure: The top 5 HTTP content types sorted by the total number of bytes served
locally by the web cache
Joshua Hare, Lance Hartung, and SumanBeyond
Banerjee
Deployments and Testbeds: Experiences with Public Usage
Jan.
on 31,
Vehicular
2013 WiFi
13 /Hotspots
17
Failure Analysis
Figure: The number of failures by type
Figure: The number of individual cellular connectivity failures by provider
Joshua Hare, Lance Hartung, and SumanBeyond
Banerjee
Deployments and Testbeds: Experiences with Public Usage
Jan.
on 31,
Vehicular
2013 WiFi
14 /Hotspots
17
Open Research Challenges
The ability to track and monitor the health of mobile devices
is challenging
Traffic optimizations are needed to help reduce the network
load on cellular back-haul links
The ability to predict link failures could greatly improve the
reliability of the system
Joshua Hare, Lance Hartung, and SumanBeyond
Banerjee
Deployments and Testbeds: Experiences with Public Usage
Jan.
on 31,
Vehicular
2013 WiFi
15 /Hotspots
17
Contributions
Experiences in designing and implementing a simple yet
sustainable vehicular Internet service
Characteristics of user behavior and usage
Some potential optimizations to improving user experience
and their potential usefulness
Exploration of the various types of failures experienced by our
system
Joshua Hare, Lance Hartung, and SumanBeyond
Banerjee
Deployments and Testbeds: Experiences with Public Usage
Jan.
on 31,
Vehicular
2013 WiFi
16 /Hotspots
17
Thank You!
Questions?
Joshua Hare, Lance Hartung, and SumanBeyond
Banerjee
Deployments and Testbeds: Experiences with Public Usage
Jan.
on 31,
Vehicular
2013 WiFi
17 /Hotspots
17