W&M Graduate Research Symposium 2013
DozyAP: Power-Efficient
Wi-Fi Tethering
Speaker
Hao Han
College of William & Mary
3/22/2013
1
Outline




Motivation
Design
Evaluation
Conclusion
2
What is Wi-Fi Tethering

Sharing a cellular data connection via Wi-Fi


An Internet-capable mobile phone acts as a Soft
Access Point (SoftAP)
Also known as “Mobile Hotspot”
soft access point
(a.k.a., mobile hotspot)
3
Wi-Fi Tethering is Desirable


Widely supported on mainstream mobile phones

Built-in feature on iOS (v4.3+), Android (v2.2+) and Windows Phone (v7.5+)

Other third-party tethering tools
Reasons


Cellular data network coverage is ubiquitous
People usually own multiple devices but only one data plan
4
Wi-Fi Tethering is a "Battery
Killer"
Nexus One
(Android)
HTC disabled
HD7
Wi-Fi tethering
(WP)
iPhone 4enabled
Wi-Fi tethering
(iOS)
(idle)
Power
Battery Life
Power
Battery Life
Nexus One
20 mW
259 hours
270 mW
19 hours
HTC HD7
32 mW
150 hours
302 mW
16 hours
iPhone 4
22 mW
247 hours
333 mW
16 hours


Even when idle, battery life is reduced from days to hours
Practical usage will draw battery more quickly
5
Power Saving for Wi-Fi
Tethering is Demanded

Old wisdom: APs are externally powered


Not true for smartphones working in Wi-Fi tethering
DozyAP: Power-Efficient Wi-Fi Tethering



Goals
 reduce power consumption of Wi-Fi tethering
 Retain good user experience
Key idea
 Put a Wi-Fi interface to sleep when network is idle
Challenges
 Avoid packet loss
 Reduce network latency
6
Contributions



First work to study the power efficiency of
Wi-Fi tethering for SoftAPs
Solution without underlying support
First working system on commercial
smartphones
7
Sources of Idle Time


Network traffic is bursty and idle for most of time
Speed discrepancy between cellular and Wi-Fi
Wi-Fi
Cellular
up to 54Mbps for 802.11a/g
up to 600Mbps for 802.11n
up to 2Mbps for 3G
Up to 100Mbps for LTE 4G
Many opportunities SoftAP could and should sleep !
8
Sleep Request-Response
Protocol

Two way hand-shake protocol


Sleep request
Sleep response
Ethernet Header
(type = 0xffff)
The Sleep “Request-Response” Protocol
Type
Sequence
Number
Sleep Duration
0x1: sleep request
0x2: sleep response

SoftAP sleeps only if receiving sleep response
idle > threshold
t0
t1
t2
sleep
t3
any buffered
data
any delayed
data
No
Yes
SoftAP
Client
Yes
No
9
Two-stage Adaptive Sleep
Algorithm

How long to sleep?


Wake up as few times as possible (more energy saving)
Wake up shortly before next packet arrives (less extra delay)
Pkt i
Pkt i+1
sleep
sleep
sleep
Pkt i
Pkt i+1
extra delay
Pkt i
sleep
Pkt i+1
10
Evaluation



Allow Wi-Fi interface sleep for up to 88% of
the total time
Reduce power consumption by up to 33%
Only increase network latency by less than
5.1%
11
Current
Potential to Save More Power
Applications
Kernel & Drivers
Future
Firmware
DozyAP, power saving up to 33%
Proprietary & closed source
Applications
Kernel & Drivers
Firmware
DozyAP, power saving up to 80%
12
Conclusion


Study the characteristics of Wi-Fi tethering
Propose DozyAP with two protocols




Sleep request-response protocol
Two-stage adaptive sleep algorithm
Implement DozyAP on commercial
smartphones
Extensive evaluation
13