Reducing the Energy Usage of Office Applications

1
Reducing the Energy Usage of Office Applications

• Jason Flinn
• M. Satyanarayanan
• Carnegie Mellon University

Eyal de Lara Dan S. Wallach Willy
Zwaenepoel Rice University
2
Motivation

• Energy is a vital resource in mobile computing
• Previous work shows value of energy-aware
  adaptation
• Reduce fidelity
• Requires modification to application source code
  
• Can this work without source code?
• Middleware-based proxy can help!

3
Validation

• Common office applications can be made
  energy-aware
• Puppeteer uses well-defined data format and API
• Modified PowerPoint to reduce energy usage
• Preview of results
• Energy reductions up to 49 for some activities
• Proxy approach modifies PowerPoint behavior
• Other opportunities for increased
  energy-efficiency

4
Outline

• Motivation
• Background Puppeteer
• Energy benefits of adaptation
• Other opportunities for extending battery
  lifetime
• Future work and conclusions

5
Component-Based Adaptation
Documents often stored on a central file
server Must download and edit on a mobile
client Can save time and energy by editing a
low-fidelity version
Server
Client
6
Puppeteer
Uses well-defined data formats, exported APIs
Server
Client
Data Server
PowerPoint
7
Measuring Application Energy Usage
Digital multimeter samples laptop power
use Applications mark start and end of
events Energy usage is integral of power over time
8
Benchmark Presentations
Benchmark consists of 10 PowerPoint
presentations Size and effect of distillation vary
9
Experiment Loading Documents

• Client IBM 560X laptop with 2 Mb/s wireless
  network
• Server Wall-powered desktop
• Client runs NT, PowerPoint, Puppeteer client
  proxy
• Server runs NT, Apache, Puppeteer server proxy
• Measure energy used to load documents
• From Apache (native mode)
• Using Puppeteer (distilling out multimedia data)

10
Energy Benefit for Loading Documents
Average energy usage reduced 40!
With simple filter, energy usage reduced 49!
11
Experiment Editing Documents

• Load documents on the client
• Measure energy needed to page through document
  for
• Full-quality version
• Distilled version

12
Energy Benefit for Editing Documents
Distillation reduces paging energy 13 Benefit
less on subsequent traversals of document
13
Reducing Computational Fidelity
How much energy is used by background activities?
Not a lot to be gained by disabling most
activities . . . But, the Office Assistant is
quite expensive!
14
Experiment Autosave Frequency

• Documents are saved on the client
• Autosaves over network prohibitively expensive!
• Reducing frequency saves energy
• But, greater possibility of losing data!
• Measure additonal power savings for autosave
  frequency
• One minute (very expensive)
• Five minutes (less expensive)
• No autosave (optimal savings)

15
Effect of Autosave Frequency
Average energy reduction of 11
16
Outline

• Motivation
• Puppeteer component-based adaptation
• Energy benefits of adaptation
• Other opportunities for extending battery
  lifetime
• Future work and conclusions

17
Comparing Puppeteer and Native Mode
Mystery Why does Puppeteer take less time and
energy to load a presentation than native mode?
18
Effect of Network Power Management

• Hypothesis power management slows down transfer
• Network receiver disabled for 100 ms. periods
• Wireless bandwidth 2 Mb/s
• In effect, a high bandwidthdelay product (25
  KB)
• Socket buffer and receive window only 8 KB
• Native mode uses only 1 connection
• Puppeteer uses up to 4 connections
• Verified this hypothesis by measuring
• With 64KB buffer sizes
• Without network power management

19
Network Power Management Results
With 64 KB buffers, native mode uses 26 less
energy Disabling power management saves
additional 18
20
Disk Power Management
Disk power management predicts periods of
inactivity After autosave, waits for additional
activity Disk in high-power state for 30 seconds
21
Transparent Power Management

• Applications and power mgmt. layer dont
  communicate
• Power management balances performance energy
• Tries to minimize impact on applications
• Difficult without knowledge of application
  activity
• Transparent power management can help!
• Applications provide hints about their
  activities
• OS combines hints from all applications
• OS chooses the optimal power mgmt. settings

CPU
Power Mgmt. Settings
Transparent Power Management Layer
Network
Hints
Applications
Disk
22
PowerPoint and Power Management

• Puppeteer could provide PM hints for PowerPoint
• Signals start and end of large transfers
• Power management disabled during transfers
• Uses 18 less energy, 22 less time
• Signals when regular autosave is occurring
• Can spin-down disk immediately after autosave
• Uses 4 less energy

23
Outline

• Motivation
• Puppeteer component-based adaptation
• Energy benefits of adaptation
• Other opportunities for extending battery
  lifetime
• Future work and conclusions

24
Future Work

• System support for energy-aware applications
• Currently open-source implementation (Linux)
• Monitors supply and demand, adjusts fidelity
• Port to closed-source environment (Windows)
• Transparent power management
• Create API for expressing application hints
• Develop algorithms that combine hints
• Investigate other hardware devices (CPU)

25
Conclusions

• No one magic bullet for reducing PowerPoint
  energy use
• But, many opportunities for significant energy
  reduction
• Reducing data fidelity
• Reducing computational fidelity
• Transparent power management
• Puppeteer provides a mechanism for achieving
  reductions
• Proxy approach requires no source-code
  modification
• Takes advantage of exported APIs
• Other potential domains for this work
• Other Microsoft Office applications
• Web browsers