ICCPS 2012 1 Dionisio de Niz, 1 Lutz Wrage, 2 Nathaniel Storer, 2 - - PowerPoint PPT Presentation

iccps 2012
SMART_READER_LITE
LIVE PREVIEW

ICCPS 2012 1 Dionisio de Niz, 1 Lutz Wrage, 2 Nathaniel Storer, 2 - - PowerPoint PPT Presentation

On Resource Overbooking in an Unmanned Aerial Vehicle ICCPS 2012 1 Dionisio de Niz, 1 Lutz Wrage, 2 Nathaniel Storer, 2 Anthony Rowe, and 2 Raj Rajkumar 1 Software Engineering Institute 2 Electrical & Computer Engineering Carnegie Mellon


slide-1
SLIDE 1

On Resource Overbooking in an Unmanned Aerial Vehicle

1Dionisio de Niz, 1Lutz Wrage, 2Nathaniel Storer, 2Anthony Rowe, and 2Raj Rajkumar 1Software Engineering Institute 2Electrical & Computer Engineering

Carnegie Mellon University

ICCPS 2012

slide-2
SLIDE 2

2

Surveillance Coverage Collision Avoidance Takeoff-to-Surveillance Distance Route Planning Flight Control Amount of Intelligence

Safety-Critical Mission-Critical

(Prevent Damage) (Provide Value)

true

Motivation

slide-3
SLIDE 3

3

Optimization of Mission-Critical Value in Mixed-Criticality Systems

Mission-Critical Reserve

Safety-Critical

  • Flight Control
  • Collision Avoidance
  • Route Planning

Mission-Critical

  • Surveillance Coverage
  • Takeoff-to-Surveillance Distance
  • Amount of Intelligence Information

Conservative approach over-allocates resources to safety-critical tasks

Safety-Critical Reserve (Based on Worst-Case)

slide-4
SLIDE 4

4

Underutilized Most of the Time

Mission-Critical Reserve

Safety-Critical

  • Flight Control
  • Collision Avoidance
  • Route Planning

Mission-Critical

  • Surveillance Coverage
  • Takeoff-to-Surveillance Distance
  • Amount of Intelligence Information

Safety-Critical Reserve

Optimization of Mission-Critical Value in Mixed-Criticality Systems

Conservative approach underutilizes resources

slide-5
SLIDE 5

5

Used for mission-critical most of the time

Mission-Critical Reserve

Safety-Critical

  • Flight Control
  • Collision Avoidance
  • Route Planning

Mission-Critical

  • Surveillance Coverage
  • Takeoff-to-Surveillance Distance
  • Amount of Intelligence Information

Reclaim unused resources & use them to optimize utility of mission-critical tasks while preserving timing guarantees of safety-critical tasks

Safety-Critical Reserve

Overbooking

Reclaiming Unused Resources: Overbooking

slide-6
SLIDE 6

6

t1 t2

2

½

2.5 2 4 8 Task Period Criticality WCET NCET t1 Surveillance Cov. 4 Mission 2 2 t2 Collision Avoid. 8 Safety 5 2.5

Overloading in Mixed-Criticality Systems

2

slide-7
SLIDE 7

7

t1 t2

2

½

2.5 2 1 1 4 8 Task Period Criticality WCET NCET t1 Surveillance Cov. 4 Mission 2 2 t2 Collision Avoid. 8 Safety 5 2.5

Zero-Slack Rate Monotonic

Zero-Slack Instant

slide-8
SLIDE 8

8

t1 t2

2

½

2.5 2 1 1 4 8 Task Period Criticality WCET NCET t1 Surveillance Cov. 4 Mission 2 2 t2 Collision Avoid. 8 Safety 5 2.5

Overbooking

Zero-Slack Rate Monotonic

slide-9
SLIDE 9

9

t1 t2

2

½

2.5 Task Period Criticality WCET NCET Utility t1 Surveillance Cov. 4 Mission 2 2 {2,2.5} t2 Collision Avoid. 8 Safety 5 2.5 t3 Amount of Intelligence 4 Mission 2 2 {2,2.5}

t3

4 8

Reclaimed Resources

Reclaiming Resources in Mixed-Criticality Systems

slide-10
SLIDE 10

10

t1 t2

2

½

2.5 1 1 Task Period Criticality WCET NCET Utility Levels t1 Surveillance Cov. 4 Mission 2 2 {2,2.5} t2 Collision Avoid. 8 Safety 5 2.5 t3 Amount of Intelligence 4 Mission 2 2 {2,2.5}

t3

1 1

1 2 2 2.5 2 2.5 1 2 Resource Utility Utility

Using Reclaimed Resources to Maximized Utility

4 8 1 1 Total Utility = 2.5

Utility Diminishes: Utility ≠ Criticality

slide-11
SLIDE 11

11

t1 t2

2

½

2.5 1 1 Task Period Criticality WCET NCET Utility Levels t1 Surveillance Cov. 4 Mission 2 2 {2,2.5} t2 Collision Avoid. 8 Safety 5 2.5 t3 Amount of Intelligence 4 Mission 2 2 {2,2.5}

t3

1 1

1 2 2 2.5 2 2.5 1 2 Resource Utility Utility

Total Utility = 4

ZS-QRAM: More mission-critical utility from same resources

Using Reclaimed Resources to Maximized Utility

4 8

slide-12
SLIDE 12

12

ZSQRAM: Period Degradation

2 4 6 8 10 100%

slide-13
SLIDE 13

13

ZSQRAM: Period Degradation

2 4 6 8 10 100%

slide-14
SLIDE 14

14

Two Enforcement Points: Before & After Overload

Period degradation Suspend lower utility

slide-15
SLIDE 15

15

Two Enforcement Points: Before & After Overload

Period degradation Suspend lower utility

slide-16
SLIDE 16

16

Metric: Utility Degradation Resilience (UDR)

32 + 10 = 42

Overload Vector Deadline Misses utility

16 + 20 = 36

Overload Vector Deadline Misses Utility Meet deadline Meet deadline?

Scheduler favors high utility Scheduler does not favor high utility

slide-17
SLIDE 17

17

Drone RK (www.drone-rk.org)

slide-18
SLIDE 18

18

AR Drone Hardware

slide-19
SLIDE 19

19

Experiment Setup

slide-20
SLIDE 20

20

Software Structure

Low-Level Flight Control Actuation Task Sensor Data Task Aux Sensor Data Video Navigation Task Object Detection Task Video Streaming Task httpd crond ntpd

Safety Critical Mission Critical Non-Real-Time

slide-21
SLIDE 21

21

UAV Taskset Parameters

Task Util1 Util2 C Co T1 T2 ZS Actuation 1 1 30 30 Sensor Data 0.1 0.1 65 65 Aux Sensor Data 0.5 0.5 50 50 Navigation 11 11 50 49 Object Detection 7 15 40 100 87 Video Streaming 2 4 10 10 120 40 40

All figures in milliseconds

slide-22
SLIDE 22

22

Utility Functions

Nominal Overloaded

slide-23
SLIDE 23

23

Demo

slide-24
SLIDE 24

24

Object Detection vs. FPS with RMS

slide-25
SLIDE 25

25

Object Detections vs. FPS with ZS-QRAM

slide-26
SLIDE 26

26

Concluding Remarks

CPS requires new scheduling mechanisms that can cope with uncertainty in the environment (variable execution time) Criticality-based overbooking protects safety-critical tasks allowing them to steal cycles from mission-critical ones

  • But fails to encode diminishing returns within mission-critical tasks

ZS-QRAM optimizes mission-critical value

  • Encoded as concave utility functions
  • Overbooking within mission-critical tasks

Developed a metric that captures the capacity of the scheduler to retain utility in overloads Demonstrated in a surveillance mission on our drone-rk platform