Transcript Presentation kit

1
SMERT: Energy-Efficient Design of
a Multimedia Messaging System
for Mobile Devices
Lin Zhong
Rice University
Bin Wei
A&T Labs-Research
Michael Sinclair
Microsoft Research
Multimedia Messaging
1 Compelling media sources
1
2 Relevant pieces constructed
3 Rich content imposes more
challenges on power usage
3
?
2
news program
Tuesday, May 28, 2006 8:31PM
TeleComm(96 seconds)
text, images, audio, and video
In carefully worded statement, BellSouth, AT&T, and now
Verizon have challenged a report that said the companies gave
tens of millions of consumers' phone records to the National
Security Agency as part of the war on terrorism after 9/11.
3
Energy Cost
 Device power profile

Communications


1600mW, GPRS
System


 Messaging service
370mW, Busy
User interfaces

212mW, Display
Message
Text KFrames Video
Size
140
10K
696K
Auto Dnld (J)
~0.03
~1.2
~80
Manu Dnld (J)
~0.08
~1.7
~114
Consum. cost (J) ~5
~10
~48
Our Goal:
Optimizing multimedia messaging services,
given the constraint of battery lifetime
4
Our Approach
 Reconstruct message content


Different levels
Selective and incremental, instead of all-or-none
 Utilize low-power user interface device


A wrist-worn low-power user interface device
Limited content with control interface
 Reduce user interruptions


Battery-aware message fetching
Automatic downloading without energy consumption on display
5
SMERT: A SMart alERT Messaging System
MediaAlert
messaging
system
Watch
SMS
Notification
Fetching
Media-alerts of different
qualities/formats
Web server
Mobile device
SMERT: A hierarchical multimedia messaging
system for mobile users
6
SMERT SMS vs. MMS/SMS
 140 Bytes
 Short description of the message
 URL to richer content

Different from conventional MMS/SMS



SMS: short message + control information
MMS: more detailed content
The device can choose when to download the content with
which wireless interface


MMS is limited to the use of cellular networks
Our device can choose to use Wi-Fi, which can consume 10 times
less energy than GPRS in data transfer if available
7
Cache-Watch
 Cache-Watch

Caches messages from
phone


Data cache
Low-power secondary
interface to phone

Interface cache
 100 X 132 Dot Matrix LCD
 Three series of touch sensors


1: mode switching
2 and 3: displayed content
manipulation
 Browse/delete/confirm cached
information

Synchronized with the phone
periodically
Cache-watch 1.0
8
Watch with Invisible Technology
 Tech-heavy watches yet to prove wide social acceptance
Invisible text
Fossil Palm Watch
SPOT/MSN-Direct
Fossil
A watch with usual appearance but one line of digital text display
•Short text messages
•Caller IDs
9
New Development at Rice
 2x8 character low-power LCD
 Debugging board
 Both software and hardware
will be OPEN-SOURCE!
10
Energy Efficiency Benefit
 Phone outsources simple-yet-frequent interactive tasks to
Cache-Watch

Display usage reductionenergy reduction
3
# of phone SMS accesses
Equivalent # of 20-second phone SMS access per hour
2.3
2
One reduction in phone
text message accesses
per two hour will justify
Bluetooth connection
every 10 minutes
1.4
1
0.7
0.6
0.5
Benefit increases if lower
power wireless personalarea technologies are used
0
0.5
1
3
5
10
Phone-CacheWatch synchronization interval (minute)
11
SMERT: Fetching and Notification
MediaAlert
Messaging
system
Watch
SMS
Notification
Fetching
Media-alerts of different
qualities/formats
Web server
Mobile device
 Fetching and Notification


Prioritize messages
Battery awareness (Evaluate battery)
 Adaptive synchronization scheme
12
Message Prioritization
 Priority is determined by match quality and quantity


Keywords are grouped in levels of different urgency
Matches are counted across groups; higher level dominates lower ones; more
matches contribute more.
 Delay-Tolerance Score (DTS)
Informative (5-0)
Deferrable (10-0)
Critical (1-0)
l
m
n
i=1
i=1
i=1
∑ C i + E * ∑ I i + F * ∑ Di
DTS =
Critical:
d *(l + m + n)
1/10, 1/20, 1/30, 1/40, …
Informative: 5/10, 9/20, 12/30, 14/40,…
Deferrable: 10/10, 19/20, 27/30, 33/40…
 Priority score is (1-DTS) which is between
1 (most urgent) and 0 (least urgent).
13
Battery Evaluation
 Energy Optimism Score (EOS)
RBC
EOS =
ECR * EWT
RBC: Remaining battery capacity
ECR: Energy consumption rate
EWT: Expected work time
 Battery-aware policies need information from both angles.


Priority score – the application perspective.
Energy optimism score – the device perspective.
14
Battery-Aware Policies
Messages arrive
as SMS
Notification
policy
Phone-watch
synchronization
schedule
Through phone
Send to wrist
Fetching
policy
Bluetooth
Decision sent
back to phone
Fetch
Delete
User
decision
CacheWatch
Default message
notification
Message Gator on phone
15
Battery-Aware Fetching and Notification
 Notification policy depends on priority score and EOS.


Ignore low priority messages when EOS is low
Notify the user through the phone when priority is high
 Automatic or user directed fetching
 Tradeoffs among priority, EOS and message size, as
Fetching Factor (Y) = P * EOS -
10
P:
Y<0, P*EOS <= 0.5: immediate
S
6
priority score
P*EOS >0.5: k-frames
0<Y<0.5:
k-frames
Y>1.5:
clip
EOS: energy-optimism score
S:
the size of the smallest video format
16
Communication Need w.r.t. Message Delay
Total # of phone-watch communications
1000
900
800
Fixed interval
700
Adaptive-MAX=20, STEP=1
600
Adaptive-MAX=40, STEP=1
500
Adaptive-MAX=40, STEP=2
400
300
200
100
0
0
5
10
15
20
Average message delay (minutes)
25
30
35
17
Design Lessons
 Looking beyond the mobile device itself
 Energy efficiency coupled with usability
 Energy efficiency != less energy consumption

Maximize messaging services while achieving the targeted
battery lifetime
18
Conclusion and future work
 SMERT: an energy-efficient multimedia messaging system
 Targets:


Reduce communication requirement
Minimize interruptions to the user
 Methods:


Device hierarchy
Information richness versus battery usage
 Future work


User studies
Multiple wireless networks for higher energy efficiency
(supported by SMERT)
19
Measuring power consumption
 Hardware



0.1 ohm resistor in series inside battery
1 KHz sampling rate
USB measurement device
20
Thermal challenge
Every One Watt increases surface
temperature by about 13 deg C
Simulation
using FloTherm
Phone case temperature will
be 40 deg C higher for a
three-watt SMT5600.
Fuel cell batteries have <50% efficiency: one-watt heat for one-watt electricity
21
Adaptive Communication Scheme
 Dynamically adjust phone-watch communication intervals
 Traces
Message arrival/Phone-watch communication
interval (minutes)
1000
100
10
Alert arrival
Adaptive-MAX=20, STEP=1
Adaptive-MAX=40, STEP=1
Adaptive-MAX=40, STEP=2
1
0
500
1000
1500
2000
2500
3000
3500
0.1
Time (minutes)
22
Related Work
 Internet alerts


Google alerts, Yahoo alerts, etc.
Most of them rely on email
 Balancing application quality and energy conservation


Keeping track of power demand, supply, and usage history
Computation offloaded from mobile devices to wall-powered
computers
 Standard messaging framework

IMS
23