Power Consumption of Google Android`s Video Decoder

Report
School of Computing Science
Simon Fraser University
Power Consumption of Google
Android’s Video Decoder
Mathieu Spénard
Mathieu Spénard
1
Outline
 Goal
 How the experiment was conducted
 Experiment results
 Challenges and Difficulties
 Conclusion
Mathieu Spénard
2
Goal
 Understand how a video decoder consumes power
within a cell phone, when renderin H.264/AVC
encoded videos
 Experiment done on a real phone, not a chip
 Understand how various processes involved in a phone
impacts the overall power consumption of rendering a
video
Mathieu Spénard
3
Tools Used
 Power Tutor [1], which allows measuring power on the
phone using a software with a 5% accuracy for the
HTC G1, HTC G2 and Nexus One
 Ffmpeg [2] to encode videos
 City, Crew and Foreman video sequences [3]
 Linda File Manager [4], a free file management system
that allows for quickly navigating files within an
Android phone
Mathieu Spénard
4
Experiment
 Encoded City, Crew and Foreman using ffmpeg
- 30fps
- CIF
- Bitrates from 100kbps to 3000kpbs
- Noted the mean Y-PSNR
 Transfer the files to a Google Nexus One phone
 Set it to plane mode (no 3G antenna/WiFi)
 Start PowerTutor
 Using Linda Manager, navigate and start an encoded video
 When it is over, stop PowerTutor, save the log and rename it so it
is easy to retrieve
 Repeat for all encoded videos
Mathieu Spénard
5
Results
Mathieu Spénard
6
Results
Mathieu Spénard
7
Results
Mathieu Spénard
8
Results
 Video decoder couldn’t play videos encoded at 2000,
2250 and 2500 kbps, and also from 4000 to 5000 kbps
 Couldn’t stop the Compass process. Consumes very
little power, but could add to the operating system
overhead
Mathieu Spénard
9
Challenges and Difficulties
 Was not parsing the right data at the beginning
 PowerTutor is slow
 PowerTutor has a too coarse granularity
 Not a lot of precision for the OLED screen power
consumption because of an unrooted phone (brightness
only)
Mathieu Spénard
10
Conclusion
 Bitrate and mean Y-PSNR do have an impact on the
power consumption of a H.264/AVC video decoder
 System overhead consumes more power than decoding
 Screen consumes most of power; does not leave a lot of
room for improvement
 A study using a 3D video decoder could give a different
decoder/screen power consumption
Mathieu Spénard
11
References
 [1] “PowerTutor,”
http://ziyang.eecs.umich.edu/projects/powertutor/
 [2] “FFmpeg.” [Online]. Available:
http://www.ffmpeg.org/
 [3] “Video Library and Tools - NSL.” [Online].
Available:
http://nsl.cs.sfu.ca/wiki/index.php/Video_Library_and_
Tools
 [4] “Linda File Manager.” [Online].Available:
http://www.androlib.com/android.application.com/lind
aandny-lindamanager-jpBj.aspx
Mathieu Spénard
12

similar documents