Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Event-Based Scheduling for Energy-Efficient Quality of Service (eQoS) in Mobile Web Applications

Yuhao Zhu
February 09, 2015

Event-Based Scheduling for Energy-Efficient Quality of Service (eQoS) in Mobile Web Applications

HPCA 2015

Yuhao Zhu

February 09, 2015
Tweet

More Decks by Yuhao Zhu

Other Decks in Education

Transcript

  1. Event-Based Scheduling for Energy-Efficient
    Quality of Service (eQoS) in Mobile Web Applications
    Yuhao Zhu, Matthew Halpern, Vijay Janapa Reddi
    Department of Electrical and Computer Engineering
    The University of Texas at Austin
    HPCA — Feb. 9th, 2015

    View Slide

  2. Responsiveness

    View Slide

  3. Responsiveness
    ▸100 ms latency is the limit for having
    users feel the system is reacting
    responsively [Miller 1968; Card et al. 1991]

    View Slide

  4. Responsiveness
    ▸100 ms latency is the limit for having
    users feel the system is reacting
    responsively [Miller 1968; Card et al. 1991]
    ▸64% of mobile users will not revisit a
    slow Web app [Source: Akamai]

    View Slide

  5. Responsiveness
    ▸100 ms latency is the limit for having
    users feel the system is reacting
    responsively [Miller 1968; Card et al. 1991]
    ▸64% of mobile users will not revisit a
    slow Web app [Source: Akamai]
    ▸Amazon found every 100 ms of
    latency costs them 1% in sales per
    year [Source: Amazon]

    View Slide

  6. Responsiveness Energy-Efficiency

    View Slide

  7. Responsiveness Energy-Efficiency
    Conflicting
    requirements

    View Slide

  8. eQoS: Making a Calculated Trade-off
    3
    Performance degradation
    QoS Experience

    View Slide

  9. eQoS: Making a Calculated Trade-off
    3
    Performance degradation
    QoS Experience
    Imperceptible

    View Slide

  10. eQoS: Making a Calculated Trade-off
    3
    Performance degradation
    QoS Experience
    Imperceptible Tolerable

    View Slide

  11. eQoS: Making a Calculated Trade-off
    3
    Performance degradation
    QoS Experience
    Imperceptible Tolerable Unusable

    View Slide

  12. eQoS: Making a Calculated Trade-off
    3
    Energy Savings
    Performance degradation
    QoS Experience
    Imperceptible Tolerable Unusable

    View Slide

  13. eQoS: Making a Calculated Trade-off
    3
    Energy Savings
    Performance degradation
    QoS Experience
    Imperceptible Tolerable Unusable

    View Slide

  14. eQoS: Making a Calculated Trade-off
    3
    Energy Savings
    Performance degradation
    QoS Experience
    Imperceptible Tolerable Unusable

    View Slide

  15. eQoS: Making a Calculated Trade-off
    3
    Energy Savings
    Performance degradation
    QoS Experience
    Imperceptible Tolerable Unusable

    View Slide

  16. eQoS: Making a Calculated Trade-off
    3
    Energy Savings
    Performance degradation
    QoS Experience
    Imperceptible Tolerable Unusable

    View Slide

  17. eQoS: Making a Calculated Trade-off
    3
    Energy Savings
    Performance degradation
    QoS Experience
    Imperceptible Tolerable Unusable

    View Slide

  18. eQoS: Making a Calculated Trade-off
    3
    Energy Savings
    Performance degradation
    QoS Experience
    Imperceptible Tolerable Unusable

    View Slide

  19. eQoS: Making a Calculated Trade-off
    3
    Energy Savings
    Performance degradation
    QoS Experience
    Imperceptible Tolerable Unusable

    View Slide

  20. Interacting with an App
    4

    View Slide

  21. Interacting with an App
    4
    http://yuhaozhu.com/googlemaps.mp4

    View Slide

  22. Interacting with an App
    4
    http://yuhaozhu.com/googlemaps.mp4

    View Slide

  23. Interacting with an App
    4
    http://yuhaozhu.com/googlemaps.mp4
    bay area

    View Slide

  24. Interacting with an App
    4
    http://yuhaozhu.com/googlemaps.mp4
    bay area
    typing
    Interactions

    View Slide

  25. Interacting with an App
    4
    http://yuhaozhu.com/googlemaps.mp4
    bay area
    typing
    Interactions

    View Slide

  26. Interacting with an App
    4
    http://yuhaozhu.com/googlemaps.mp4
    bay area
    typing
    pressing
    Interactions

    View Slide

  27. Interacting with an App
    4
    http://yuhaozhu.com/googlemaps.mp4
    bay area
    typing
    pressing
    Interactions

    View Slide

  28. Interacting with an App
    4
    http://yuhaozhu.com/googlemaps.mp4
    bay area
    typing
    pressing
    moving
    Interactions

    View Slide

  29. Interacting with an App
    4
    http://yuhaozhu.com/googlemaps.mp4
    bay area
    typing
    pressing
    moving
    Interactions Events

    View Slide

  30. Interacting with an App
    4
    http://yuhaozhu.com/googlemaps.mp4
    bay area
    typing
    pressing
    moving
    Interactions Events
    keypress
    click
    touchmove

    View Slide

  31. Interacting with an App
    4
    http://yuhaozhu.com/googlemaps.mp4
    bay area Events
    keypress
    click
    touchmove

    View Slide

  32. Interacting with an App
    4
    http://yuhaozhu.com/googlemaps.mp4
    bay area Events
    keypress
    click
    touchmove
    1487 Events

    View Slide

  33. Mobile Web Applications are Event-Driven
    5

    View Slide

  34. Mobile Web Applications are Event-Driven
    5
    Event FIFO Queue

    View Slide

  35. Mobile Web Applications are Event-Driven
    5
    Event FIFO Queue
    Head
    Tail

    View Slide

  36. Mobile Web Applications are Event-Driven
    5
    Touch Screen

    Event
    Event FIFO Queue
    Key Press

    Event


    Head
    Tail Timer

    Event
    Insert to

    Event Queue

    View Slide

  37. Mobile Web Applications are Event-Driven
    5
    Touch Screen

    Event
    Event FIFO Queue
    Key Press

    Event


    Head
    Tail Timer

    Event
    Insert to

    Event Queue
    Event Loop

    View Slide

  38. Mobile Web Applications are Event-Driven
    5
    Touch Screen

    Event
    Event FIFO Queue
    Key Press

    Event


    Head
    Tail Timer

    Event
    Insert to

    Event Queue
    Fetch for

    Execution
    Event Loop

    View Slide

  39. Mobile Web Applications are Event-Driven
    5
    Touch Screen

    Event
    Event FIFO Queue
    Key Press

    Event


    Head
    Tail Timer

    Event
    Insert to

    Event Queue
    Fetch for

    Execution
    Event Loop

    View Slide

  40. Mobile Web Applications are Event-Driven
    5
    Event FIFO Queue
    Key Press

    Event


    Head
    Tail
    Timer

    Event
    Insert to

    Event Queue
    Fetch for

    Execution
    Event Loop

    View Slide

  41. Mobile Web Applications are Event-Driven
    5
    Event FIFO Queue
    Key Press

    Event


    Head
    Tail
    Timer

    Event
    Insert to

    Event Queue
    Fetch for

    Execution
    Event Loop
    Optimize for eQoS at an event-granularity

    View Slide

  42. ▸ Key idea: Execute events with just-enough energy to meet
    user QoS expectation
    ▹ Develop a runtime system that leverages the large scheduling
    space of heterogeneous CPUs
    6
    Executive Summary

    View Slide

  43. ▸ Key idea: Execute events with just-enough energy to meet
    user QoS expectation
    ▹ Develop a runtime system that leverages the large scheduling
    space of heterogeneous CPUs
    6
    Executive Summary
    Event-level
    Characterization

    View Slide

  44. ▸ Key idea: Execute events with just-enough energy to meet
    user QoS expectation
    ▹ Develop a runtime system that leverages the large scheduling
    space of heterogeneous CPUs
    Runtime Mechanics to
    Exploit Latency Slack
    6
    Executive Summary
    Event-level
    Characterization

    View Slide

  45. ▸ Key idea: Execute events with just-enough energy to meet
    user QoS expectation
    ▹ Develop a runtime system that leverages the large scheduling
    space of heterogeneous CPUs
    Runtime Mechanics to
    Exploit Latency Slack
    Heterogeneous
    Resource Utilization
    6
    Executive Summary
    Event-level
    Characterization

    View Slide

  46. ▸ Key idea: Execute events with just-enough energy to meet
    user QoS expectation
    ▹ Develop a runtime system that leverages the large scheduling
    space of heterogeneous CPUs
    Runtime Mechanics to
    Exploit Latency Slack
    Heterogeneous
    Resource Utilization
    6
    Executive Summary
    Event-level
    Characterization

    View Slide

  47. Exploit the Slack in Events
    7
    150
    100
    50
    0
    Event Latency (ms)
    Events

    View Slide

  48. Exploit the Slack in Events
    8
    150
    100
    50
    0
    Event Latency (ms)
    Events

    View Slide

  49. Exploit the Slack in Events
    8
    Unusable

    QoS
    150
    100
    50
    0
    Event Latency (ms)
    Events

    View Slide

  50. Exploit the Slack in Events
    8
    keyup
    Unusable

    QoS
    150
    100
    50
    0
    Event Latency (ms)
    Events

    View Slide

  51. Exploit the Slack in Events
    8
    keyup
    Unusable

    QoS
    150
    100
    50
    0
    Event Latency (ms)
    Events

    View Slide

  52. Exploit the Slack in Events
    8
    keyup
    Large Slack
    Unusable

    QoS
    150
    100
    50
    0
    Event Latency (ms)
    Events

    View Slide

  53. Exploit the Slack in Events
    8
    keyup
    Large Slack
    change
    Unusable

    QoS
    150
    100
    50
    0
    Event Latency (ms)
    Events

    View Slide

  54. Exploit the Slack in Events
    8
    keyup
    Large Slack
    change
    Small

    Slack
    Unusable

    QoS
    150
    100
    50
    0
    Event Latency (ms)
    Events

    View Slide

  55. Exploit the Slack in Events
    8
    keyup
    Large Slack
    change
    Small

    Slack
    click
    Unusable

    QoS
    150
    100
    50
    0
    Event Latency (ms)
    Events

    View Slide

  56. Exploit the Slack in Events
    8
    keyup
    Large Slack
    change
    Small

    Slack
    click
    Unusable

    QoS
    ▸ Wide distribution of event latencies. Events exhibit different slacks.
    ▹ How to exploit different slacks for different events?
    150
    100
    50
    0
    Event Latency (ms)
    Events

    View Slide

  57. Event-Based Scheduler (EBS) Overview
    9
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event

    View Slide

  58. Event-Based Scheduler (EBS) Overview
    9
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event
    Thread Scheduling

    View Slide

  59. Event-Based Scheduler (EBS) Overview
    9
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event
    Thread Scheduling

    View Slide

  60. Event-Based Scheduler (EBS) Overview
    9
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event
    Thread-based Scheduler
    Thread Scheduling

    View Slide

  61. Event-Based Scheduler (EBS) Overview
    9
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event
    Thread-based Scheduler
    Thread Scheduling
    Throughput Fairness

    View Slide

  62. Event-Based Scheduler (EBS) Overview
    10
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event

    View Slide

  63. Event-Based Scheduler (EBS) Overview
    10
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event
    Mobile applications

    View Slide

  64. Event-Based Scheduler (EBS) Overview
    10
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event
    Mobile applications
    Event Queue

    View Slide

  65. Event-Based Scheduler (EBS) Overview
    10
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event
    Event-based Scheduler
    Mobile applications
    Event Queue

    View Slide

  66. Event-Based Scheduler (EBS) Overview
    10
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event
    Event-based Scheduler
    Mobile applications
    eQoS
    Event Queue

    View Slide

  67. Event-Based Scheduler (EBS) Overview
    11
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event
    H3 … H2 … H1
    change
    click keyup
    Event Queue

    View Slide

  68. Event-Based Scheduler (EBS) Overview
    11
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event
    H3 … H2 … H1
    change
    click keyup
    Event Queue
    Big/Little

    Hardware

    View Slide

  69. Event-Based Scheduler (EBS) Overview
    11
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event
    H3 … H2 … H1
    change
    click keyup
    Event Queue
    Big/Little

    Hardware
    Event-Based Scheduler

    View Slide

  70. Event-Based Scheduler (EBS) Overview
    11
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event
    H3 … H2 … H1
    change
    click keyup
    Event Queue
    Big/Little

    Hardware
    Event-Based Scheduler
    Event

    View Slide

  71. Event-Based Scheduler (EBS) Overview
    11
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event
    H3 … H2 … H1
    change
    click keyup
    Event Queue
    Big/Little

    Hardware
    Event-Based Scheduler

    Event

    View Slide

  72. Event-Based Scheduler (EBS) Overview
    11
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event
    H3 … H2 … H1
    change
    click keyup
    QoS
    Monitor
    Event Queue
    Big/Little

    Hardware
    Event-Based Scheduler

    Event

    View Slide

  73. Event-Based Scheduler (EBS) Overview
    11
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event
    H3 … H2 … H1
    change
    click keyup
    QoS
    Monitor
    Event Queue
    Big/Little

    Hardware
    Event-Based Scheduler
    Model

    Event

    View Slide

  74. Event-Based Scheduler (EBS) Overview
    11
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event
    H3 … H2 … H1
    change
    click keyup
    QoS
    Monitor
    Model
    Constructor
    Event Queue
    Big/Little

    Hardware
    Event-Based Scheduler
    Model

    Event

    View Slide

  75. Event-Based Scheduler (EBS) Overview
    11
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event
    H3 … H2 … H1
    change
    click keyup
    QoS
    Monitor
    Model
    Constructor
    Event Queue
    Big/Little

    Hardware
    Event-Based Scheduler
    Model

    Event

    View Slide

  76. Event-Based Scheduler (EBS) Overview
    11
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event
    H3 … H2 … H1
    change
    click keyup
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event Queue
    Big/Little

    Hardware
    Event-Based Scheduler
    Model

    Event

    View Slide

  77. Event-Based Scheduler (EBS) Overview
    11
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event
    H3 … H2 … H1
    change
    click keyup
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event Queue
    Big/Little

    Hardware
    Event-Based Scheduler
    Model

    Event

    View Slide

  78. Event-Based Scheduler (EBS) Overview
    11
    ▸ Goal: Find the most energy-efficient architectural configuration
    that meets the QoS target for each event
    H3 … H2 … H1
    change
    click keyup
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event Queue
    Big/Little

    Hardware
    Event-Based Scheduler
    Model
    Recalibrate

    Event

    View Slide

  79. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    Feedback
    How the Model Constructor Works
    12
    ▸ Goal: Estimate the event latency under each config

    View Slide

  80. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    Feedback
    How the Model Constructor Works
    12
    ▸ Goal: Estimate the event latency under each config
    Memory
    Operation
    CPU
    Operation
    Tmemory
    Ndependent
    f
    Event
    Latency
    Xie, et al., Compile-Time Dynamic Voltage
    Scaling Settings: Opportunities and Limits, PLDI’03

    View Slide

  81. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    Feedback
    How the Model Constructor Works
    12
    ▸ Goal: Estimate the event latency under each config
    Memory
    Operation
    CPU
    Operation
    Tmemory
    Ndependent
    f
    Event
    Latency
    Xie, et al., Compile-Time Dynamic Voltage
    Scaling Settings: Opportunities and Limits, PLDI’03
    Event
    Latency
    =

    View Slide

  82. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    Feedback
    How the Model Constructor Works
    12
    ▸ Goal: Estimate the event latency under each config
    Memory
    Operation
    CPU
    Operation
    Tmemory
    Ndependent
    f
    Event
    Latency
    Xie, et al., Compile-Time Dynamic Voltage
    Scaling Settings: Opportunities and Limits, PLDI’03
    Event
    Latency
    = Tmemory
    +

    View Slide

  83. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    Feedback
    How the Model Constructor Works
    12
    ▸ Goal: Estimate the event latency under each config
    Memory
    Operation
    CPU
    Operation
    Tmemory
    Ndependent
    f
    Event
    Latency
    Xie, et al., Compile-Time Dynamic Voltage
    Scaling Settings: Opportunities and Limits, PLDI’03
    Event
    Latency
    = Tmemory
    + Ndependent / f

    View Slide

  84. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    Feedback
    How the Model Constructor Works
    12
    ▸ Goal: Estimate the event latency under each config
    Memory
    Operation
    CPU
    Operation
    Tmemory
    Ndependent
    f
    Event
    Latency
    Xie, et al., Compile-Time Dynamic Voltage
    Scaling Settings: Opportunities and Limits, PLDI’03
    Event
    Latency
    = Tmemory
    + Ndependent / f

    View Slide

  85. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    Feedback
    How the Model Constructor Works
    12
    ▸ Goal: Estimate the event latency under each config
    Event
    Latency
    = Tmemory
    + Ndependent / f
    Event Latency
    Frequency

    View Slide

  86. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    Feedback
    How the Model Constructor Works
    12
    ▸ Goal: Estimate the event latency under each config
    Event
    Latency
    = Tmemory
    + Ndependent / f
    Event Latency
    Frequency

    View Slide

  87. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    How the Detector Works
    13
    ▸ Goal: Find the QoS target for each event
    CamanJS
    Pdf.js
    Crypto
    Zlib
    Paper.js
    Ember.js
    GWT
    Backbone
    jQuery
    sina
    google
    ebay
    Doom
    Rain
    10-4
    104
    106
    100
    Feedback

    View Slide

  88. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    How the Detector Works
    13
    ▸ Goal: Find the QoS target for each event
    CamanJS
    Pdf.js
    Crypto
    Zlib
    Paper.js
    Ember.js
    GWT
    Backbone
    jQuery
    sina
    google
    ebay
    Doom
    Rain
    10-4
    104
    106
    100
    CamanJS
    Pdf.js
    Crypto
    Zlib
    Paper.js
    Ember.js
    GWT
    Backbone
    jQuery
    sina
    google
    ebay
    Doom
    Rain
    10-4
    104
    106
    100
    Event
    Latency (s)
    Event
    Intensity (evt/s)
    Feedback

    View Slide

  89. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    How the Detector Works
    13
    ▸ Goal: Find the QoS target for each event
    CamanJS
    Pdf.js
    Crypto
    Zlib
    Paper.js
    Ember.js
    GWT
    Backbone
    jQuery
    sina
    google
    ebay
    Doom
    Rain
    10-4
    104
    106
    100
    CamanJS
    Pdf.js
    Crypto
    Zlib
    Paper.js
    Ember.js
    GWT
    Backbone
    jQuery
    sina
    google
    ebay
    Doom
    Rain
    10-4
    104
    106
    100
    Event
    Latency (s)
    Event
    Intensity (evt/s)
    Feedback

    View Slide

  90. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    How the Detector Works
    13
    ▸ Goal: Find the QoS target for each event
    CamanJS
    Pdf.js
    Crypto
    Zlib
    Paper.js
    Ember.js
    GWT
    Backbone
    jQuery
    sina
    google
    ebay
    Doom
    Rain
    10-4
    104
    106
    100
    CamanJS
    Pdf.js
    Crypto
    Zlib
    Paper.js
    Ember.js
    GWT
    Backbone
    jQuery
    sina
    google
    ebay
    Doom
    Rain
    10-4
    104
    106
    100
    Event
    Latency (s)
    Event
    Intensity (evt/s)
    Feedback

    View Slide

  91. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    How the Detector Works
    13
    ▸ Goal: Find the QoS target for each event
    CamanJS
    Pdf.js
    Crypto
    Zlib
    Paper.js
    Ember.js
    GWT
    Backbone
    jQuery
    sina
    google
    ebay
    Doom
    Rain
    10-4
    104
    106
    100
    CamanJS
    Pdf.js
    Crypto
    Zlib
    Paper.js
    Ember.js
    GWT
    Backbone
    jQuery
    sina
    google
    ebay
    Doom
    Rain
    10-4
    104
    106
    100
    Event
    Latency (s)
    Event
    Intensity (evt/s)
    NA
    Feedback

    View Slide

  92. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    How the Detector Works
    13
    ▸ Goal: Find the QoS target for each event
    CamanJS
    Pdf.js
    Crypto
    Zlib
    Paper.js
    Ember.js
    GWT
    Backbone
    jQuery
    sina
    google
    ebay
    Doom
    Rain
    10-4
    104
    106
    100
    CamanJS
    Pdf.js
    Crypto
    Zlib
    Paper.js
    Ember.js
    GWT
    Backbone
    jQuery
    sina
    google
    ebay
    Doom
    Rain
    10-4
    104
    106
    100
    Event
    Latency (s)
    Event
    Intensity (evt/s)
    NA
    Feedback
    [1, 10] s
    [50, 100] ms [60, 30] fps

    View Slide

  93. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    Feedback
    How the QoS Monitor Works
    14
    ▸ Goal: Predict the for the next event

    View Slide

  94. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    Feedback
    How the QoS Monitor Works
    14
    ▸ Goal: Predict the for the next event
    Predicted Latency
    Frequency
    Big Core
    Little Core

    View Slide

  95. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    Feedback
    How the QoS Monitor Works
    14
    ▸ Goal: Predict the for the next event
    QoS Target
    Predicted Latency
    Frequency
    Big Core
    Little Core

    View Slide

  96. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    Feedback
    How the QoS Monitor Works
    14
    ▸ Goal: Predict the for the next event
    QoS Target
    Predicted Latency
    Frequency
    Big Core
    Little Core

    View Slide

  97. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    Feedback
    How the QoS Monitor Works
    14
    ▸ Goal: Predict the for the next event
    QoS Target
    Predicted Latency
    Frequency
    Big Core
    Little Core
    ▸ Scheduling overhead (~120 us):
    ▹ Core migration
    ▹ Frequency switching

    View Slide

  98. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    Feedback
    How the QoS Monitor Works
    15
    ▸ Goal: Predict the for the next event
    QoS Target
    Predicted Latency
    Frequency
    Big Core
    Little Core

    View Slide

  99. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    Feedback
    How the QoS Monitor Works
    15
    ▸ Goal: Predict the for the next event
    QoS Target
    Predicted Latency
    Frequency
    Big Core
    Little Core

    View Slide

  100. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    Feedback
    How the QoS Monitor Works
    15
    ▸ Goal: Predict the for the next event
    QoS Target
    Predicted Latency
    Frequency
    Big Core
    Little Core
    ▸ Fine-tune the execution upon over-
    prediction or under-prediction

    View Slide

  101. Feedback
    Detector
    QoS
    Monitor
    Model
    Constructor
    Event

    Info
    Model
    Recalibrate

    Feedback
    How the QoS Monitor Works
    15
    ▸ Goal: Predict the for the next event
    QoS Target
    Predicted Latency
    Frequency
    Big Core
    Little Core
    ▸ Fine-tune the execution upon over-
    prediction or under-prediction
    ▸ Recalibrate if it mispredicts too often

    View Slide

  102. Evaluation Methodology
    ▸ Implemented inside Google Chromium Web browser
    ▸ Representative hardware platform
    ▹ Exynos 5410 SoC (A15 + A7)
    16

    View Slide

  103. Evaluation Methodology
    ▸ Baseline Mechanisms
    ▹ Highest performance (Perf) — Standard to guarantee responsiveness
    ▹ Minimal energy (Energy) — Minimize energy consumption
    ▹ Interactive governor (Interactive) — Android default
    ▹ On-demand governor (Ondemand)
    17

    View Slide

  104. Evaluation Methodology
    ▸ Baseline Mechanisms
    ▹ Highest performance (Perf) — Standard to guarantee responsiveness
    ▹ Minimal energy (Energy) — Minimize energy consumption
    ▹ Interactive governor (Interactive) — Android default
    ▹ On-demand governor (Ondemand)
    17

    View Slide

  105. Evaluation Methodology
    ▸ Baseline Mechanisms
    ▹ Highest performance (Perf) — Standard to guarantee responsiveness
    ▹ Minimal energy (Energy) — Minimize energy consumption
    ▹ Interactive governor (Interactive) — Android default
    ▹ On-demand governor (Ondemand)
    17
    PH
    Imperceptible
    0
    PI
    PU
    PL
    QoSU
    QoSI
    Tolerable Unusable
    -ESU
    ESI
    EST
    Quality-of-Service (QoS)
    Energy Savings (ES)
    Performance Degradation
    ▸ Scheduling Scenarios

    View Slide

  106. Evaluation Methodology
    ▸ Baseline Mechanisms
    ▹ Highest performance (Perf) — Standard to guarantee responsiveness
    ▹ Minimal energy (Energy) — Minimize energy consumption
    ▹ Interactive governor (Interactive) — Android default
    ▹ On-demand governor (Ondemand)
    17
    PH
    Imperceptible
    0
    PI
    PU
    PL
    QoSU
    QoSI
    Tolerable Unusable
    -ESU
    ESI
    EST
    Quality-of-Service (QoS)
    Energy Savings (ES)
    Performance Degradation
    ▸ Scheduling Scenarios
    ▹ Scheduling for imperceptibility

    View Slide

  107. Evaluation Methodology
    ▸ Baseline Mechanisms
    ▹ Highest performance (Perf) — Standard to guarantee responsiveness
    ▹ Minimal energy (Energy) — Minimize energy consumption
    ▹ Interactive governor (Interactive) — Android default
    ▹ On-demand governor (Ondemand)
    17
    PH
    Imperceptible
    0
    PI
    PU
    PL
    QoSU
    QoSI
    Tolerable Unusable
    -ESU
    ESI
    EST
    Quality-of-Service (QoS)
    Energy Savings (ES)
    Performance Degradation
    ▸ Scheduling Scenarios
    ▹ Scheduling for imperceptibility
    ▹ Scheduling for tolerability

    View Slide

  108. Evaluation Methodology
    ▸ Baseline Mechanisms
    ▹ Highest performance (Perf) — Standard to guarantee responsiveness
    ▹ Minimal energy (Energy) — Minimize energy consumption
    ▹ Interactive governor (Interactive) — Android default
    ▹ On-demand governor (Ondemand)
    17
    PH
    Imperceptible
    0
    PI
    PU
    PL
    QoSU
    QoSI
    Tolerable Unusable
    -ESU
    ESI
    EST
    Quality-of-Service (QoS)
    Energy Savings (ES)
    Performance Degradation
    ▸ Scheduling Scenarios
    ▹ Scheduling for imperceptibility
    ▹ Scheduling for tolerability

    View Slide

  109. Evaluation Results
    18
    QoS Violations (%)
    0.0
    1.5
    3.0
    4.5
    6.0
    emberjs
    gwt
    jquery
    backbone
    paperjs
    sina
    google
    ebay
    EBS Perf Interactive Ondemand Energy

    View Slide

  110. 19
    QoS Violations (%)
    0.0
    1.5
    3.0
    4.5
    6.0
    emberjs
    gwt
    jquery
    backbone
    paperjs
    sina
    google
    ebay
    EBS Perf Interactive Energy
    Evaluation Results
    No QoS Violations

    View Slide

  111. 20
    QoS Violations (%)
    0.0
    1.5
    3.0
    4.5
    6.0
    emberjs
    gwt
    jquery
    backbone
    paperjs
    sina
    google
    ebay
    EBS Perf Interactive Energy
    Evaluation Results
    No QoS Violations

    View Slide

  112. 21
    QoS Violations (%)
    0.0
    1.5
    3.0
    4.5
    6.0
    emberjs
    gwt
    jquery
    backbone
    paperjs
    sina
    google
    ebay
    EBS Perf Interactive Energy
    9.4
    17.8
    58.1
    6.9
    Evaluation Results

    View Slide

  113. 22
    QoS Violations (%)
    0.0
    1.5
    3.0
    4.5
    6.0
    emberjs
    gwt
    jquery
    backbone
    paperjs
    sina
    google
    ebay
    EBS Perf Interactive Energy
    9.4
    17.8
    58.1
    6.9
    Evaluation Results

    View Slide

  114. 23
    QoS Violations (%)
    0.0
    1.5
    3.0
    4.5
    6.0
    emberjs
    gwt
    jquery
    backbone
    paperjs
    sina
    google
    ebay
    EBS Perf Interactive Energy
    9.4
    17.8
    58.1
    6.9
    Evaluation Results
    Energy (J)
    0.0
    1.0
    2.0
    3.0
    4.0
    emberjs
    gwt
    jquery
    backbone
    paperjs
    sina
    google
    ebay

    View Slide

  115. 24
    Energy (J)
    0.0
    1.0
    2.0
    3.0
    4.0
    emberjs
    gwt
    jquery
    backbone
    paperjs
    sina
    google
    ebay
    8.2
    7.7
    Evaluation Results
    QoS Violations (%)
    0.0
    1.5
    3.0
    4.5
    6.0
    emberjs
    gwt
    jquery
    backbone
    paperjs
    sina
    google
    ebay
    EBS Perf Interactive Energy
    9.4
    17.8
    58.1
    6.9

    View Slide

  116. 25
    Energy (J)
    0.0
    1.0
    2.0
    3.0
    4.0
    emberjs
    gwt
    jquery
    backbone
    paperjs
    sina
    google
    ebay
    8.2
    7.7
    Evaluation Results
    QoS Violations (%)
    0.0
    1.5
    3.0
    4.5
    6.0
    emberjs
    gwt
    jquery
    backbone
    paperjs
    sina
    google
    ebay
    EBS Perf Interactive Energy
    9.4
    17.8
    58.1
    6.9

    View Slide

  117. 26
    Energy (J)
    0.0
    1.0
    2.0
    3.0
    4.0
    emberjs
    gwt
    jquery
    backbone
    paperjs
    sina
    google
    ebay
    8.2
    7.7
    Evaluation Results
    QoS Violations (%)
    0.0
    1.5
    3.0
    4.5
    6.0
    emberjs
    gwt
    jquery
    backbone
    paperjs
    sina
    google
    ebay
    EBS Perf Interactive Energy
    9.4
    17.8
    58.1
    6.9

    View Slide

  118. 26
    Energy (J)
    0.0
    1.0
    2.0
    3.0
    4.0
    emberjs
    gwt
    jquery
    backbone
    paperjs
    sina
    google
    ebay
    8.2
    7.7
    Evaluation Results
    QoS Violations (%)
    0.0
    1.5
    3.0
    4.5
    6.0
    emberjs
    gwt
    jquery
    backbone
    paperjs
    sina
    google
    ebay
    EBS Perf Interactive Energy
    9.4
    17.8
    58.1
    6.9
    37.9% - 41.2% energy savings, 0.1% more QoS violations

    View Slide

  119. What Have We Learnt So Far
    27
    ▸ User interactivity and human perceptibility
    are important because it has economic
    ramifications and impacts user device choice,
    as well as end-user satisfaction

    View Slide

  120. What Have We Learnt So Far
    27
    PH
    Imperceptible
    0
    PI
    PU
    PL
    QoSU
    QoSI
    Tolerable Unusable
    -ESU
    ESI
    EST
    Quality-of-Service (QoS)
    Energy Savings (ES)
    Performance Degradation
    ▸ Need a systematic way of understanding and
    QoS-energy trade-offs. To that end we
    advocate eQoS to strike a balance between
    user satisfaction and energy reduction
    ▸ User interactivity and human perceptibility
    are important because it has economic
    ramifications and impacts user device choice,
    as well as end-user satisfaction

    View Slide

  121. What Have We Learnt So Far
    27
    PH
    Imperceptible
    0
    PI
    PU
    PL
    QoSU
    QoSI
    Tolerable Unusable
    -ESU
    ESI
    EST
    Quality-of-Service (QoS)
    Energy Savings (ES)
    Performance Degradation
    ▸ Need a systematic way of understanding and
    QoS-energy trade-offs. To that end we
    advocate eQoS to strike a balance between
    user satisfaction and energy reduction
    ▸ Develop the event-based scheduling (EBS)
    mechanism that exploits the fundamental
    event-driven execution pattern of mobile
    applications to achieve better eQoS
    ▸ User interactivity and human perceptibility
    are important because it has economic
    ramifications and impacts user device choice,
    as well as end-user satisfaction
    Heterogeneous
    Hardware
    Event
    Queue
    H3 … H2 … H1 Dispatch

    PI, PU
    Detector
    QoS
    Monitor
    Model
    Constructor
    Recalibrate
    Event-Based Scheduler
    Event
    Info
    Models
    onkeyup=“H1 () {…}”
    keyup
    onchange=“H2 () {…}”
    change
    Event execution
    feedback
    onclick=“H3 () {…}”
    click

    View Slide

  122. Mobile is One Example
    28
    Web Mobile

    View Slide

  123. Mobile is One Example
    28
    Web Mobile Internet-of-Things
    Sensor networks Cloud

    View Slide

  124. Mobile is One Example
    28
    Web Mobile Internet-of-Things
    Sensor networks Cloud
    Event-based processing is a fundamental
    computation pattern.

    View Slide

  125. Mobile is One Example
    28
    Web Mobile Internet-of-Things
    Sensor networks Cloud
    Event-based processing is a fundamental
    computation pattern.

    View Slide

  126. ▸ Another talk @ Best of CAL session
    ▸ Wednesday after the keynote
    ▸ How computer architects can improve Web technologies

    View Slide

  127. Thank you

    View Slide