pes: proactive event scheduling for energy-efficient mobile web computing

Slide 1

Slide 1 text

PES: Proactive Event Scheduling for Responsive and Energy-Eﬃcient Mobile Web Computing Yu Feng with Yuhao Zhu Department of Computer Science University of Rochester ISCA 2019

Slide 2

Slide 2 text

Ubiquity of Mobile devices 2

Slide 3

Slide 3 text

Ubiquity of Mobile devices 2

Slide 4

Slide 4 text

Ubiquity of Mobile devices 3

Slide 5

Slide 5 text

Ubiquity of Mobile devices 3 ▸2x times mobile users than desktop users

Slide 6

Slide 6 text

Ubiquity of Mobile devices 3 ▸2x times mobile users than desktop users ▸76% of population are using mobile devices daily

Slide 7

Slide 7 text

Mobile Applications are Event-Driven 4

Slide 8

Slide 8 text

Mobile Applications are Event-Driven 4

Slide 9

Slide 9 text

Mobile Applications are Event-Driven 4 Typing Interactions ISCA 2019

Slide 10

Slide 10 text

Mobile Applications are Event-Driven 4 Scrolling Typing Interactions ISCA 2019

Slide 11

Slide 11 text

Mobile Applications are Event-Driven 4 Scrolling Tapping Typing Interactions ISCA 2019

Slide 12

Slide 12 text

Mobile Applications are Event-Driven 4 Scrolling Tapping scrollstart scrollupdate scrollend tapdown Tapup Keypress Typing Interactions Events ISCA 2019

Slide 13

Slide 13 text

No content

Slide 14

Slide 14 text

No content

Slide 15

Slide 15 text

Event-Driven Execution Model 6 Event

Slide 16

Slide 16 text

Event-Driven Execution Model 6 touchmove onclick timer … Time Event Queue Push Event

Slide 17

Slide 17 text

Event-Driven Execution Model 7 touchmove onclick timer … Time Event Queue Fetch Runtime Scheduler Core

Slide 18

Slide 18 text

Event-Driven Execution Model 7 touchmove onclick timer … Time Event Queue Fetch Runtime Scheduler Core Scheduling knobs: DVFS settings

Slide 19

Slide 19 text

Inefﬁciency of Current Schedulers 8

Slide 20

Slide 20 text

Inefﬁciency of Current Schedulers 8 Time

Slide 21

Slide 21 text

Inefﬁciency of Current Schedulers 8 Time input 1 input 2 QoS Deadline 1 QoS Deadline 2

Slide 22

Slide 22 text

Inefﬁciency of Current Schedulers 8 Time OS Governor input 1 input 2 QoS Deadline 1 QoS Deadline 2

Slide 23

Slide 23 text

Inefﬁciency of Current Schedulers 8 Time OS Governor input 1 input 2 QoS Deadline 1 QoS Deadline 2 E1

Slide 24

Slide 24 text

Inefﬁciency of Current Schedulers 8 Time OS Governor input 1 input 2 QoS Deadline 1 QoS Deadline 2 E1 slack

Slide 25

Slide 25 text

Inefﬁciency of Current Schedulers 8 Time OS Governor input 1 input 2 QoS Deadline 1 QoS Deadline 2 E1 E2 slack

Slide 26

Slide 26 text

Inefﬁciency of Current Schedulers 9 Time input 1 input 2 QoS Deadline 1 QoS Deadline 2 E1 E2 OS Governor

Slide 27

Slide 27 text

Inefﬁciency of Current Schedulers 9 Time input 1 input 2 QoS Deadline 1 QoS Deadline 2 E1 E2 QoS-Aware OS Governor

Slide 28

Slide 28 text

Inefﬁciency of Current Schedulers 9 Time input 1 input 2 QoS Deadline 1 QoS Deadline 2 E1 E2 E1 QoS-Aware OS Governor

Slide 29

Slide 29 text

Inefﬁciency of Current Schedulers 9 Time input 1 input 2 QoS Deadline 1 QoS Deadline 2 E1 E2 E1 E2 QoS-Aware OS Governor

Slide 30

Slide 30 text

Inefﬁciency of Current Schedulers 9 Time input 1 input 2 QoS Deadline 1 QoS Deadline 2 E1 E2 E1 E2 QoS-Aware OS Governor ?

Slide 31

Slide 31 text

Inefﬁciency of Current Schedulers 10 Time input 1 input 2 QoS Deadline 1 QoS Deadline 2 E1 E2 E1 E2 QoS-Aware OS Governor

Slide 32

Slide 32 text

Inefﬁciency of Current Schedulers 10 Time Oracle input 1 input 2 QoS Deadline 1 QoS Deadline 2 E1 E2 E1 E2 QoS-Aware OS Governor

Slide 33

Slide 33 text

Inefﬁciency of Current Schedulers 10 Time Oracle input 1 input 2 QoS Deadline 1 QoS Deadline 2 E1 E2 E1 E2 E1 E2 QoS-Aware OS Governor

Slide 34

Slide 34 text

Inefﬁciency of Current Schedulers 11 Time E1 E2 E3 E4 E5 Schedule across events

Slide 35

Slide 35 text

Inefﬁciency of Current Schedulers 12 Time E1 E2 E3 E4 E5 Current events

Slide 36

Slide 36 text

Inefﬁciency of Current Schedulers 12 Time E1 E2 E3 E4 E5 Current events Future events

Slide 37

Slide 37 text

Proactive Event Scheduler 13

Slide 38

Slide 38 text

Proactive Event Scheduler Proactive Event Scheduler 13

Slide 39

Slide 39 text

Proactive Event Scheduler Proactive Event Scheduler 13 Prediction Web Program Analysis Machine Learning

Slide 40

Slide 40 text

Proactive Event Scheduler Proactive Event Scheduler 13 Prediction Web Program Analysis Machine Learning Scheduling Constraint Optimization

Slide 41

Slide 41 text

Proactive Event Scheduler Proactive Event Scheduler 13 Result reduce energy consumption while improving QoS Prediction Web Program Analysis Machine Learning Scheduling Constraint Optimization

Slide 42

Slide 42 text

Proactive Event Scheduler Proactive Event Scheduler 13 Result reduce energy consumption while improving QoS Prediction Web Program Analysis Machine Learning Scheduling Constraint Optimization

Slide 43

Slide 43 text

Event Sequence Learning 14 Prediction Model

Slide 44

Slide 44 text

Event Sequence Learning 14 Time E1 E2 E3 Event Sequence Prediction Model

Slide 45

Slide 45 text

Event Sequence Learning 14 Time E1 E2 E3 ? Event Sequence Prediction Model

Slide 46

Slide 46 text

Event Sequence Learning 14 Time E1 E2 E3 ? Event Sequence Prediction Model

Slide 47

Slide 47 text

Event Sequence Learning 14 Time E1 E2 E3 ? Event Sequence E4 Prediction Model

Slide 48

Slide 48 text

Event Sequence Learning 14 Time E1 E2 E3 ? Event Sequence E4 Prediction Model

Slide 49

Slide 49 text

Recurrent Prediction 15 Time E1 E2 E3 E4 Event Sequence Prediction Model

Slide 50

Slide 50 text

Recurrent Prediction 15 Time E1 E2 E3 E4 Event Sequence Prediction Model

Slide 51

Slide 51 text

Recurrent Prediction 15 Time E1 E2 E3 E4 Event Sequence E5 Prediction Model

Slide 52

Slide 52 text

Recurrent Prediction 15 Time E1 E2 E3 E4 Event Sequence E5 Prediction Model

Slide 53

Slide 53 text

Recurrent Prediction 15 Time E1 E2 E3 E4 Event Sequence E5 … Prediction Model

Slide 54

Slide 54 text

Prediction Model 16 Prediction Model

Slide 55

Slide 55 text

Prediction Model 17 Prediction Model

Slide 56

Slide 56 text

Prediction Model 17 Prediction Model The distance of click The number of scrolls The number of navigations Features encoding past interactions …

Slide 57

Slide 57 text

Prediction Model 18 Features encoding past interactions ln( p 1 − p ) = x β Prediction Model

Slide 58

Slide 58 text

Prediction Model 19 Features encoding past interactions Click ScrollUp ScrollDown ZoomIn ZoomOut … Prediction Model 0.10 0.12 0.58 0.07 0.02

Slide 59

Slide 59 text

Prediction Model 19 Features encoding past interactions Click ScrollUp ScrollDown ZoomIn ZoomOut … Prediction Model ✔

Slide 60

Slide 60 text

Prediction Model 19 Features encoding past interactions Click ScrollUp ScrollDown ZoomIn ZoomOut … Prediction Model All Event Types

Slide 61

Slide 61 text

Prediction Model 19 Features encoding past interactions Click ScrollUp ScrollDown ZoomIn ZoomOut … Prediction Model Filter Events ?

Slide 62

Slide 62 text

20 Program Analysis

Slide 63

Slide 63 text

20 Program Analysis

Slide 64

Slide 64 text

20 Program Analysis

Slide 65

Slide 65 text

20 Program Analysis Viewport

Slide 66

Slide 66 text

…

… Program Analysis DOM Tree Viewport

Slide 67

Slide 67 text

…

… Viewport Program Analysis DOM Tree Viewport

Slide 68

Slide 68 text

…

… Viewport Program Analysis DOM Tree Viewport

Slide 69

Slide 69 text

…

… Viewport Program Analysis DOM Tree Viewport

Slide 70

Slide 70 text

…

… Viewport Program Analysis DOM Tree Viewport

Slide 71

Slide 71 text

Include Program Analysis Features 21 Features encoding past interactions Click ScrollUp ScrollDown ZoomIn ZoomOut … Prediction Model

Slide 72

Slide 72 text

Overview of Predictor 22 Past Event Statistics Current Application State Click ScrollUp ScrollDown ZoomIn ZoomOut … + Prediction Model

Slide 73

Slide 73 text

Overview of Predictor 22 Past Event Statistics Current Application State Click ScrollUp ScrollDown ZoomIn ZoomOut … + Prediction Model Recurrent

Slide 74

Slide 74 text

Overview of Predictor 22

Slide 75

Slide 75 text

Overview of Predictor 22 High Accuracy, low overhead!

Slide 76

Slide 76 text

Proactive Event Scheduler Proactive Event Scheduler 23 Result reduce energy consumption meanwhile improve QoS Prediction Web Program Analysis Machine Learning Scheduling Constraint Optimization

Slide 77

Slide 77 text

Proactive Event Scheduler Proactive Event Scheduler 23 Result reduce energy consumption meanwhile improve QoS Prediction Web Program Analysis Machine Learning Scheduling Constraint Optimization

Slide 78

Slide 78 text

Scheduling events 24 E1 E2 E3 Time From predictor

Slide 79

Slide 79 text

Scheduling events 24 E1 E2 E3 Time From predictor Goal: minimize total energy while meeting deadlines

Slide 80

Slide 80 text

Problem Formulation 25 E1 E2 E3 Time ▸ Scheduling Problem → Constrained Optimization.

Slide 81

Slide 81 text

Objective: Problem Formulation 25 E1 E2 E3 Time ▸ Scheduling Problem → Constrained Optimization. N ∑ i Min. Energy (i)

Slide 82

Slide 82 text

Objective: Problem Formulation 25 E1 E2 E3 Time △Texe 1 △Texe 2 △Texe 3 ▸ Scheduling Problem → Constrained Optimization. N ∑ i △Texe (i) x Min.

Slide 83

Slide 83 text

Objective: Problem Formulation 25 E1 E2 E3 Time △Texe 1 △Texe 2 △Texe 3 ▸ Scheduling Problem → Constrained Optimization. N ∑ i △Texe (i) x Power (i) Min.

Slide 84

Slide 84 text

Objective: Constraints: Problem Formulation 25 E1 E2 E3 Time △Texe 1 △Texe 2 △Texe 3 ▸ Scheduling Problem → Constrained Optimization. N ∑ i △Texe (i) x Power (i) Min.

Slide 85

Slide 85 text

Objective: Constraints: Problem Formulation 25 E1 E2 E3 Time △Texe 1 △Texe 2 △Texe 3 ▸ Scheduling Problem → Constrained Optimization. Order: ≤ Tend (i) Tstart (i+1) N ∑ i △Texe (i) x Power (i) Min.

Slide 86

Slide 86 text

Objective: Constraints: Problem Formulation 25 E1 E2 E3 Time Tstart 1 Tstart 2 Tstart 3 TQoS 1 TQoS 2 TQoS 3 △Texe 1 △Texe 2 △Texe 3 ▸ Scheduling Problem → Constrained Optimization. Order: ≤ Tend (i) Tstart (i+1) Deadline: ≤ Tstart (i) △Texe (i) TQoS (i) + N ∑ i △Texe (i) x Power (i) Min.

Slide 87

Slide 87 text

Slide 88

Slide 88 text

Objective: Constraints: Problem Formulation 26 ▸ Scheduling Problem → Constrained Optimization. N ∑ i △Texe (i) x Power (i) Min. Order: ≤ Tend (i) Tstart (i+1) Deadline: ≤ Tstart (i) △Texe (i) TQoS (i) +

Slide 89

Slide 89 text

Each Event: Objective: Constraints: Problem Formulation 26 ▸ Scheduling Problem → Constrained Optimization. N ∑ i △Texe (i) x Power (i) Min. △Texe (i) = Order: ≤ Tend (i) Tstart (i+1) Deadline: ≤ Tstart (i) △Texe (i) TQoS (i) +

Slide 90

Slide 90 text

Each Event: Objective: Constraints: Problem Formulation 26 ▸ Scheduling Problem → Constrained Optimization. N ∑ i △Texe (i) x Power (i) Min. △Texe (i) = Tmemory + Order: ≤ Tend (i) Tstart (i+1) Deadline: ≤ Tstart (i) △Texe (i) TQoS (i) +

Slide 91

Slide 91 text

Each Event: Objective: Constraints: Problem Formulation 26 ▸ Scheduling Problem → Constrained Optimization. N ∑ i △Texe (i) x Power (i) Min. Tcpu △Texe (i) = Tmemory + Order: ≤ Tend (i) Tstart (i+1) Deadline: ≤ Tstart (i) △Texe (i) TQoS (i) +

Slide 92

Slide 92 text

Each Event: Objective: Constraints: Problem Formulation 26 ▸ Scheduling Problem → Constrained Optimization. N ∑ i △Texe (i) x Power (i) Min. △Texe (i) = Tmemory + Ncycles / f (i) Order: ≤ Tend (i) Tstart (i+1) Deadline: ≤ Tstart (i) △Texe (i) TQoS (i) +

Slide 93

Slide 93 text

Each Event: Objective: Constraints: Problem Formulation 26 ▸ Scheduling Problem → Constrained Optimization. N ∑ i △Texe (i) x Power (i) Min. △Texe (i) = Tmemory + Constants Ncycles / f (i) Order: ≤ Tend (i) Tstart (i+1) Deadline: ≤ Tstart (i) △Texe (i) TQoS (i) +

Slide 94

Slide 94 text

Each Event: Objective: Constraints: Problem Formulation 26 ▸ Scheduling Problem → Constrained Optimization. N ∑ i △Texe (i) x Power (i) Min. △Texe (i) = Tmemory + Constants Oﬄine proﬁle Ncycles / f (i) Order: ≤ Tend (i) Tstart (i+1) Deadline: ≤ Tstart (i) △Texe (i) TQoS (i) +

Slide 95

Slide 95 text

Each Event: Objective: Constraints: Problem Formulation 26 ▸ Scheduling Problem → Constrained Optimization. N ∑ i △Texe (i) x Min. △Texe (i) = Tmemory + Constants Oﬄine proﬁle Ncycles / f (i) Order: ≤ Tend (i) Tstart (i+1) Deadline: ≤ Tstart (i) △Texe (i) TQoS (i) + Pmap (i)

Slide 96

Slide 96 text

Slide 97

Slide 97 text

Each Event: Objective: Constraints: Problem Formulation 27 ▸ Scheduling Problem → Constrained Optimization. N ∑ i △Texe (i) x Pmap (i) Min. Order: ≤ Tend (i) Tstart (i+1) Deadline: ≤ Tstart (i) △Texe (i) TQoS (i) + { Ncycles / f (j) } △Texe (i) =Tmemory +∑ * ⍺ (i, j) ⍺ (i, j) in {0,1}

Slide 98

Slide 98 text

Slide 99

Slide 99 text

Slide 100

Slide 100 text

Slide 101

Slide 101 text

Slide 102

Slide 102 text

Slide 103

Slide 103 text

Putting Things Together 28 Web Application Hardware Architecture

Slide 104

Slide 104 text

PES Putting Things Together 28 Web Application Hardware Architecture

Slide 105

Slide 105 text

PES Putting Things Together 28 Web Application Hardware Architecture Predictor Events

Slide 106

Slide 106 text

PES Putting Things Together 28 Web Application Hardware Architecture Predictor Events Scheduler Predictions

Slide 107

Slide 107 text

PES Putting Things Together 28 Web Application Hardware Architecture Predictor Events Speculative Schedules Scheduler Predictions

Slide 108

Slide 108 text

PES Putting Things Together 28 Web Application Hardware Architecture Predictor Controller Events Speculative Schedules Uncommitted Results Scheduler Predictions

Slide 109

Slide 109 text

PES Putting Things Together 28 Web Application Hardware Architecture Predictor Controller Events Speculative Schedules Uncommitted Results Scheduler Predictions

Slide 110

Slide 110 text

PES Putting Things Together 28 Web Application Hardware Architecture Predictor Controller Events Speculative Schedules Commit Uncommitted Results Scheduler Predictions

Slide 111

Slide 111 text

PES Putting Things Together 28 Web Application Hardware Architecture Predictor Controller Events Speculative Schedules Commit Uncommitted Results Recover Scheduler Predictions

Slide 112

Slide 112 text

Proactive Event Scheduler Proactive Event Scheduler 29 Result reduce energy consumption meanwhile improve QoS Prediction Web Program Analysis Machine Learning Scheduling Constraint Optimization

Slide 113

Slide 113 text

Proactive Event Scheduler Proactive Event Scheduler 29 Result reduce energy consumption meanwhile improve QoS Prediction Web Program Analysis Machine Learning Scheduling Constraint Optimization

Slide 114

Slide 114 text

Experimental Setup 30

Slide 115

Slide 115 text

Experimental Setup 30 Implemented our framework in Chromium on top of Android system

Slide 116

Slide 116 text

ODORID XU+E development board, which contains an Exynos 5410 SoC Experimental Setup 30 Implemented our framework in Chromium on top of Android system

Slide 117

Slide 117 text

ODORID XU+E development board, which contains an Exynos 5410 SoC Experimental Setup 30 Implemented our framework in Chromium on top of Android system UI-level record and replay for reproducibility. [ISPASS’15]

Slide 118

Slide 118 text

Evaluation ▸Baseline Mechanisms ▹Interactive governor (Interactive) — Android default ▹EBS: a state-of-the-art QoS-aware scheduler ▹Oracle: optimal scheduler 31 31 Time Oracle input 1 input 2 QoS Deadline 1 QoS Deadline 2 E1 E2 E1 E2 E1 E2 QoS-Aware OS Governor

Slide 119

Slide 119 text

Evaluation ▸Baseline Mechanisms ▹Interactive governor (Interactive) — Android default ▹EBS: a state-of-the-art QoS-aware scheduler ▹Oracle: optimal scheduler 31 ▸Metrics ▹Energy Consumption ▹QoS Violation 31

Slide 120

Slide 120 text

Slide 121