Regarding the Optimality of Speedup Bounds of Mixed-Criticality - - PowerPoint PPT Presentation

regarding the optimality of speedup bounds of mixed
SMART_READER_LITE
LIVE PREVIEW

Regarding the Optimality of Speedup Bounds of Mixed-Criticality - - PowerPoint PPT Presentation

Regarding the Optimality of Speedup Bounds of Mixed-Criticality Schedulability Tests Zhishan Guo Department of Computer Science, Missouri S&T Presented at Dagstuhl Seminar 17131 Mar. 29, 2017 MC & Vestals Interpretation Mixed


slide-1
SLIDE 1

Regarding the Optimality of Speedup Bounds

  • f Mixed-Criticality Schedulability Tests

Zhishan Guo

Department of Computer Science, Missouri S&T Presented at Dagstuhl Seminar 17131

  • Mar. 29, 2017
slide-2
SLIDE 2
  • Mixed Criticality (MC) [Vestal, RTSS’07]

– MC: functionalities of different degree of importance are implemented upon a shared platform. – MC Correctness:

  • All deadlines met under normal mode
  • HI deadlines met under HI mode

Mode switch ci

L

ci

H

t Static Analysis; Pessimistic Measurement Based; Optimistic

MC & Vestal’s Interpretation

slide-3
SLIDE 3
  • Mixed Criticality (MC) [Vestal, RTSS’07]

– MC: functionalities of different degree of importance are implemented upon a shared platform. – MC Correctness:

  • All deadlines met under normal mode
  • HI deadlines met under HI mode

Mode switch ci

L

ci

H

t Static Analysis; Pessimistic Measurement Based; Optimistic

MC & Vestal’s Interpretation

  • E.g.2:
  • 20 cycles
  • 3~9~321 cycles
  • E.g.1: x := a + b takes

Q: Why Mixed Criticality? A: The resources over-provisioned to the critical functionalities can be used to execute the less- critical functionalities instead.

slide-4
SLIDE 4
  • Mixed Criticality (MC) [Vestal, RTSS’07]

– MC: functionalities of different degree of importance are implemented upon a shared platform. – MC Correctness:

  • All deadlines met under normal mode
  • HI deadlines met under HI mode

– Important & Interesting: 300+ papers in 8 years [1] – Hard [2]

Mode switch ci

L

ci

H

t Static Analysis; Pessimistic Measurement Based; Optimistic

MC & Vestal’s Interpretation

Q: Why Mixed Criticality? A: The resources over-provisioned to the critical functionalities can be used to execute the less- critical functionalities instead.

[1] A Burns and R Davis. Mixed-criticality systems, a review. 9th ed, 2017. [2] S Baruah. Mixed criticality scheduling is highly intractable, 2008.

NP-Hard in the strong sense

slide-5
SLIDE 5
  • Mixed Criticality (MC) [Vestal, RTSS’07]

– MC: functionalities of different degree of importance are implemented upon a shared platform. – MC Correctness:

  • All deadlines met under normal mode
  • HI deadlines met under HI mode

– Important, Interesting, & Hard

Mode switch

Speedup Bound

Strategy Work Uniprocessor, Vestal Job 1.618, Best OCBP [Baruah, Li, Soutgie RTAS’10] Uniprocessor, Vestal Task 4/3, Best EDF-VD [Li & Baruah RTSS’10] [Baruah et al. ECRTS’12] Multi processor, Vestal Task 1.618 4/3 MC-Fluid [Lee et al. RTSS’14] 4/3, Best MCF [Our Work, RTSS’15]

MC & Vestal’s Interpretation

A schedulability test has speedup factor s, s≥1, if any task set that is schedulable by any algorithm on platform with processing speed of 1, it will be deemed schedulable by this test upon a processor that is s times as fast.

NP-Hard in the strong sense

slide-6
SLIDE 6
  • Mixed Criticality (MC) [Vestal, RTSS’07]

– MC: functionalities of different degree of importance are implemented upon a shared platform. – MC Correctness:

  • All deadlines met under normal mode
  • HI deadlines met under HI mode

– Important, Interesting, & Hard

Mode switch

Speedup Bound

Strategy Work Uniprocessor, Vestal Job 1.618, Best OCBP [Baruah, Li, Soutgie RTAS’10] Uniprocessor, Vestal Task 4/3, Best EDF-VD [Li & Baruah RTSS’10] [Baruah et al. ECRTS’12] Multi processor, Vestal Task 1.618 4/3 MC-Fluid [Lee et al. RTSS’14] 4/3, Best MCF [Our Work, RTSS’15]

MC & Vestal’s Interpretation

A schedulability test has speedup factor s, s≥1, if any task set that is schedulable by any algorithm on platform with processing speed of 1, it will be deemed schedulable by this test upon a processor that is s times as fast.

NP-Hard in the strong sense

4/3 1

“Optimal” Strategy

slide-7
SLIDE 7
  • Mixed Criticality (MC) [Vestal, RTSS’07]

– MC: functionalities of different degree of importance are implemented upon a shared platform. – MC Correctness:

  • All deadlines met under normal mode
  • HI deadlines met under HI mode

– Important, Interesting, & Hard

Mode switch

Speedup Bound

Strategy Work Uniprocessor, Vestal Job 1.618, Best OCBP [Baruah, Li, Stougie RTAS’10] Uniprocessor, Vestal Task 4/3, Best EDF-VD [Li & Baruah RTSS’10] [Baruah et al. ECRTS’12] Multi processor, Vestal Task 1.618 4/3 MC-Fluid [Lee et al. RTSS’14] 4/3, Best MCF [Our Work, RTSS’15]

MC & Vestal’s Interpretation

A schedulability test has speedup factor s, s≥1, if any task set that is schedulable by any algorithm on platform with processing speed of 1, it will be deemed schedulable by this test upon a processor that is s times as fast.

NP-Hard in the strong sense

1 3/4

“Optimal” Strategy

slide-8
SLIDE 8
  • Mixed Criticality (MC) [Vestal, RTSS’07]

– MC: functionalities of different degree of importance are implemented upon a shared platform. – MC Correctness:

  • All deadlines met under normal mode
  • HI deadlines met under HI mode

– Important, Interesting, & Hard

Speedup Bound

Strategy Work Uniprocessor, Vestal Job 1.618, Best OCBP [Baruah, Li, Soutgie RTAS’10] Uniprocessor, Vestal Task 4/3, Best EDF-VD [Li & Baruah RTSS’10] [Baruah et al. ECRTS’12] Multi processor, Vestal Task 1.618 4/3 Best MC-Fluid [Lee et al. RTSS’14] 4/3, Best MCF [Our Work, RTSS’15]

MC & Vestal’s Interpretation

A schedulability test has speedup factor s, s≥1, if any task set that is schedulable by any algorithm on platform with processing speed of 1, it will be deemed schedulable by this test upon a processor that is s times as fast.

DOMINATE

slide-9
SLIDE 9
  • Mixed Criticality (MC) [Vestal, RTSS’07]

– MC: functionalities of different degree of importance are implemented upon a shared platform. – MC Correctness:

  • All deadlines met under normal mode
  • HI deadlines met under HI mode

– Important, Interesting, & Hard

Speedup Bound

Strategy Work Uniprocessor, Vestal Job 1.618, Best OCBP [Baruah, Li, Soutgie RTAS’10] Uniprocessor, Vestal Task 4/3, Best EDF-VD [Li & Baruah RTSS’10] [Baruah et al. ECRTS’12] Multi processor, Vestal Task 1.618 4/3 Best MC-Fluid [Lee et al. RTSS’14] 4/3, Best MCF [Our Work, RTSS’15]

MC & Vestal’s Interpretation

A schedulability test has speedup factor s, s≥1, if any task set that is schedulable by any algorithm on platform with processing speed of 1, it will be deemed schedulable by this test upon a processor that is s times as fast.

Demand-Based MC [Easwaran, RTSS’13] [Ekberg & Yi, RTS 50(1)] DOMINATE

slide-10
SLIDE 10
  • Mixed Criticality (MC) [Vestal, RTSS’07]

– MC: functionalities of different degree of importance are implemented upon a shared platform. – MC Correctness:

  • All deadlines met under normal mode
  • HI deadlines met under HI mode

– Important, Interesting, & Hard

Speedup Bound

Strategy Work Uniprocessor, Vestal Job 1.618, Best OCBP [Baruah, Li, Soutgie RTAS’10] Uniprocessor, Vestal Task 4/3, Best EDF-VD [Li & Baruah RTSS’10] [Baruah et al. ECRTS’12] Multi processor, Vestal Task 1.618 4/3 Best MC-Fluid [Lee et al. RTSS’14] 4/3, Best MCF [Our Work, RTSS’15]

MC & Vestal’s Interpretation

  • Best (Optimal) speedup factor ≠ Optimal scheduling strategy/analysis

Lazy Priority Adjustment [Gu et al., RTCSA’13] LE-EDF [Guo & Baruah, RTNS’15] DOMINATE

slide-11
SLIDE 11
  • Mixed Criticality (MC) [Vestal, RTSS’07]

– MC: functionalities of different degree of importance are implemented upon a shared platform. – MC Correctness:

  • All deadlines met under normal mode
  • HI deadlines met under HI mode

– Important, Interesting, & Hard

Mode switch

Speedup Bound

Strategy Work Uniprocessor, Vestal Job 1.618, Best OCBP [Baruah, Li, Soutgie RTAS’10] Uniprocessor, Vestal Task 4/3, Best EDF-VD [Li & Baruah RTSS’10] [Baruah et al. ECRTS’12] Multi processor, Vestal Task 1.618 4/3 MC-Fluid [Lee et al. RTSS’14] 4/3, Best MCF [Our Work, RTSS’15]

MC & Vestal’s Interpretation

A schedulability test has speedup factor s, s≥1, if any task set that is schedulable by any algorithm on platform with processing speed of 1, it will be deemed schedulable by this test upon a processor that is s times as fast.

NP-Hard in the strong sense

4/3 1

“Optimal” clairvoyant Strategy

NECESSARY CONDITIONS SUFFICIENT CONDITIONS

slide-12
SLIDE 12
  • Mixed Criticality (MC) [Vestal, RTSS’07]

– MC: functionalities of different degree of importance are implemented upon a shared platform. – MC Correctness:

  • All deadlines met under normal mode
  • HI deadlines met under HI mode

– Important, Interesting, & Hard

Speedup Bound

Strategy Work Uniprocessor, Vestal Job 1.618, Best OCBP [Baruah, Li, Soutgie RTAS’10] Uniprocessor, Vestal Task 4/3, Best EDF-VD [Li & Baruah RTSS’10] [Baruah et al. ECRTS’12] Multi processor, Vestal Task 1.618 4/3 Best MC-Fluid [Lee et al. RTSS’14] 4/3, Best MCF [Our Work, RTSS’15]

MC & Vestal’s Interpretation

  • Best (Optimal) speedup factor ≠ Optimal scheduling strategy/analysis
  • Speedup against clairvoyant ≠ Speedup over online optimal

Lazy Priority Adjustment [Gu et al., RTCSA’13] LE-EDF [Guo & Baruah, RTNS’15]

slide-13
SLIDE 13

Thank you!

Zhishan Guo guozh@mst.edu

  • Mar. 29, 2017
slide-14
SLIDE 14

Acknowledgement

  • Prof. Sanjoy Baruah, UNC Chapel Hill (Ph.D. advisor)
  • Prof. Nathan Fisher, Wayne State U
  • Prof. Luca Santinelli, ONERA France
  • Prof. Mohammed Al Faruque, UC Irvine
  • Dr. Zhenkai Zhang, Vanderbilt U
  • Prof. Abusayeed Saifullah, Wayne State U
  • Prof. Sajal Das, MST
  • Prof. Haoyi Xiong, MST
  • Ms. Sai Sruti, MST
  • Mr. Ashik Bhuiyan, MST
slide-15
SLIDE 15

Backup Slides

slide-16
SLIDE 16

Motivation

  • I think there is a world market for maybe five computers.
  • Thomas Watson,

Chairman of IBM, 1943

  • There is no reason anyone would want a computer in

their home.

  • Ken Olson,

President of Digital Equipment Corporation, 1977

slide-17
SLIDE 17

About MCF’s 4/3 speedup bound

slide-18
SLIDE 18
  • Fluid Scheduling Model

– Provide ideal schedule for multiprocessor – Each task τi executes at a given rate θi

About MCF

slide-19
SLIDE 19
  • Fluid Scheduling Model

– Provide ideal schedule for multiprocessor – Each task τi executes at a given rate θi

  • MCF Scheduling Model

– A task τi is executed with

· LO-rate θi

L under normal mode

· HI-rate θi

H after mode switch

Fig from [WMC 2015]

About MCF

Mode switch ci

L

ci

H

t Static Analysis; Pessimistic Measurement Based; Optimistic

slide-20
SLIDE 20
  • Fluid Scheduling Model

– Provide ideal schedule for multiprocessor – Each task τi executes at a given rate θi

  • MCF Scheduling Model

– A task τi is executed with

· LO-rate θi

L under normal mode

· HI-rate θi

H after mode switch

– Proof Idea

Fig from [WMC 2015]

About MCF

Any task set that is schedulable on a c-speed platform: Will be schedulable by MCF

  • n a unit-speed platform :

SUFFICIENT NECESSARY

c c c

slide-21
SLIDE 21
  • Fluid Scheduling Model

– Provide ideal schedule for multiprocessor – Each task τi executes at a given rate θi

  • MCF Scheduling Model

– A task τi is executed with

· LO-rate θi

L under normal mode

· HI-rate θi

H after mode switch

– Proof Idea

Fig from [WMC 2015]

About MCF

Any task set that is schedulable on a ¾-speed platform: Will be schedulable by MCF

  • n a unit-speed platform :

SUFFICIENT NECESSARY

slide-22
SLIDE 22

Missouri S&T Michigan Tech Colorado Mines SD Mines New Mexico Mining & Tech Illinois Tech Cal Tech Florida Tech Georgia Tech Clarkson New Jersey Tech Stevens Tech WPI Rensselaer U of A - Huntsville MIT

slide-23
SLIDE 23

1,125 International Students in Fall 2016