Energy and Power Measurement on Mobile Devices

Tutorial in MobiTools co-located with ISCA 2016

Transcript

1. Energy and Power
   Measurement on
   Mobile Devices
   Yuhao Zhu (@yzhu88)
   The University of Texas at Austin
   MobiTools 2016
   1
   

   View Slide

2. 2
   You can’t fix what you can’t measure
   

   View Slide

3. 3
   8
   6
   4
   2
   0
   Measured Peak Power (W)
   Droid Galaxy S Galaxy Nexus Galaxy S3 Galaxy S4 Galaxy S5 Galaxy S6
   (2009) (2010) (2011) (2012) (2013) (2014) (2015)
   Display Radio CPU T H R OT T L I N G
   

   View Slide

4. Measurement Granularity
   ▸Entire device
   ▹SoC, Radio, Screen
   ▸SoC
   ▹CPU, GPU, Memory, DSP, other accelerators
   ▸Component: CPU/GPU/Memory
   4
   

   View Slide

5. Hardware Platform
   ▸Smartphone (tablets, etc.)
   ▹Mostly close-sources
   ▸Development board
   ▹Designed for hacking
   ▹Mostly open-source (schematic, PCB design, etc.)
   ▹SoC with peripherals (WiFi, USB, etc.)
   ▹Typically does not come with a screen
   5
   

   View Slide

6. Measurement Visibility
   6
   Smartphone Dev board
   Device
   SoC
   Component
   

   View Slide

7. Power Measurement Tools
   ▸NI DAQ
   ▹A range of choices (sampling rate, channels, price)
   ▹An example of a high-end DAQ
   - 2M samples per second
   - 8 channels, 16 bit resolution
   - ~$4400
   ▹ An example of a low-end DAQ
   - 10k samples per second
   - 8 channels, 12 bit resolution
   - $149
   7
   

   View Slide

8. A Typical DAQ Setup
   8
   + -
   Vin+ Vin-
   Vout GND
   Sense resistor
   15mΩ
   SoC
   ARM Cortex A9
   VRM
   Gain
   x50
   Probe
   ▸Do you need an
   amplifier circuit?
   ▸Before or after
   the VRM?
   

   View Slide

9. 9
   Power Measurement Tools
   ▸Monsoon
   ▹5k samples per second
   ▹1 channels, 16 bit resolution
   ▹~$770
   ▹Built-in sense resistor (the reason why people use it!!!)
   

   View Slide

10. 10
    Data Processing Software
    ▸NI LabView
    ▹ Specifically designed for NI DAQ
    ▹ Rich API for manipulating data, well integrated with IDEs
    

    View Slide

11. ▸Monsoon software
    11
    Data Processing Software
    

    View Slide

12. ARM DS-5 Streamline
    12
    

    View Slide

13. ARM DS-5 Streamline
    13
    

    View Slide

14. ▸Correlate over a timeline all your software execution
    activity with power
    ▸Specifications
    ▹No built-in sense resistors
    ▹10k samples per second
    ▹3 channels, 10 bit resolution
    ▸No programming APIs
    14
    ARM DS-5 Streamline
    

    View Slide

15. BattOr: Portable Power Monitor
    ▸ Store power measurements while mobile, stream power
    measurements while stationary.
    ▸ Open source: https://github.com/aschulm/battor
    ▸ Integrated into Android and Chrome trace viewer
    15
    

    View Slide

16. Conclusions
    16
    Smartphone Dev board
    Device
    SoC
    Component
    ▸DAQ
    ▸Monsoon
    ▸BattOr
    ▸Streamline
    

    View Slide