Outline of the presentation

A measurement study of vehicular internet access using in situ Wi-Fi networks

Vladimir Bychkovsky, Bret Hull, Allen Miu, Hari Balakrishnan, and Samuel Madden

MIT Computer Science and Artificial Intelligence Laboratory

By – Anup Jaya Prakash

Outline of the presentation Introduction Questions Answers – The Measurement Study The Experiment Connectivity Results Packet Loss and Data Transfers Discussions Conclusion

Introduction 14.3 million homes in the US have Wi-Fi

access points. Most of the links are often idle. Can be used by others by providing

controlled access. What would be the performance?

Questions What is the distribution of the duration of

connectivity per AP? What is the distribution of the duration of

disconnectivity How long does it take for a client to scan, associate,

and obtain an IP address? What is the distribution of the coverage region of an

AP? What is the distribution of packet loss and data

transfer rates? What is the effect of a car’s speed on these metrics?

Answers – The Measurement Study A set of in situ open AP’s

deployed in and around Boston.

9 Distinct cars fitted with embedded computers.

They try to associate with an AP and if successful, try to obtain an IP address.

Next they do an end-to-end ping to a well known IP.

If this is successful, they start periodic local pings to next hop IP router and initiates a transfer to the internet site.

Answers… The Measurement study is based on uploads. Two

reasons are provided for this New Applications treat vehicles as data source in wireless

sensor networks ( CarTel Project ) Download performance will be at least as good as the

upload performance

The Results are Divided into 2 categories Connectivity Data transfer performance

The Experiment CarTel Embedded Computer – Has a 802.11b

Wi-Fi card, GPS unit, 128 MB RAM, 1 GB Flash memory, running Linux 2.4.31.

Scanping – Application used for the experimentation purpose

GPS used to find position and speed of the car. Computer draws power from the car. Boots up

when ignition is turned

The Experiment - Processes Scan – Scan for AP’s in the Area Association – Try to Associate with one

of the AP’s Address Configuration – Acquire an IP Single end-to-end ping – Try for an end-

to-end ping Connectivity and uploads – Measure

bandwidth and connectivity

Scan Discovery of the AP’s in an area For each discovered AP, the ESSID,

BSSID, Frequency, Signal strength and privacy bits (if any) are logged.

Done till at least one AP is discovered, then proceeds to next step

Association Scanping issues a command to Wi-Fi

interface to associate with the AP. For multiple APs, highest signal strength

is taken into account The result of association along with start

time and duration of operation is logged Jumps to first step if failed else runs

tcpdump to monitor furthur networking activity.

Address Configuration Uses dhcpcd to obtain an IP address. Checks local cache for information on

current AP. If exists then uses the information. Otherwise tries to obtain an IP address If it fails, the client ties out after 5

secs and proceeds to step 1.

Single End-to-End Ping Once it obtains an IP, it starts a

end-to-end ping every 200ms until the first successful ping or until 2 sec elapse.

This is to estimate end-to-end connectivity duration

Connectivity and Uploads 2 processes in parallel

AP Pings – Ping the first hop router every 100ms and log the time and the result of the ping

TCP uploads – Establish a TCP connection to central sever and deliver data.

Timeline of the Operations

Data Summary

Connectivity Results The Connectivity results are organized

into 4 parts namely Wi-Fi Association and IP Address Acquisition

– first 3 graphs Connectivity Duration – 4th, 5th and 6th

graphs Periods without connectivity – 7th graph AP coverage – 8th, 9th and 10th graphs

Connectivity Results…

CDF for distribution of time for various phases of activities after a successful association


Distribution of scan and association times


Distribution of time for different types of IP acquisition


CDF of Association Durations


CDF of Average speeds for Associations


Plot of Connection duration Vs Speed


CDF of time between connectivity events between 4 types of events


CDF of no. of AP’s discovered in successful scan


Fraction of associations to any give AP


CDF of Connection Coverage

Packet Loss and Data Transfers

These Results are divided into two categories Wi-Fi packet loss rates TCP throughput

Wi-Fi Packet Loss Rates

CDF of fraction of AP pings that succeeded per connection

Wi-Fi Packet Loss Rates…

Plot of Wi-Fi packet delivery rate vs the car speed

TCP Throughput

CDF of duration between association and first TCP data ACK

TCP Throughput…

CDF of per connection end-to-end throughput

TCP Throughput…

CDF of per connection bytes received at server

Discussion The idea of the paper holds much

promise What are the other issues in

implementation of such networks Lets see them one by one

Open Wi-Fi Networks Incentives to users and ISPs Tiered security level with different

levels of access controll and data rates at different tiers

Connectivity and Transport Continuous connectivity to mobile

users If not continuous then some

modeling change would be needed

Fast and Friendly Connection Establishment Fast connection establishment and fair

utilization of bandwidth necessary Three possible optimizations

Connection initiation timing – optimize transport protocol to address this issue

Fairness – Rate limiting at AP’s or use cooperative TCP protocols

Conclusion With just 3.2% of total AP’s participating in the

system, the clients remain connected for 24 sec on an average.

The mean period of disconnectivity was 260 seconds which would reduce if the participation increased

Several commercial, legal and policy issues need to be addressed in order to make this vision a reality

Questions ?

Documents

Outline of the presentation