Preemption Point Selection in Limited Preemptive Scheduling using - - PowerPoint PPT Presentation

Preemption Point Selection in Limited Preemptive Scheduling using Probabilistic Preemption Costs

Filip Marković, Jan Carlson, Radu Dobrin Mälardalen Real-Time Research Centre,

• Dept. of Computer Science and Software Engineering,

Mälardalen University, Sweden

Limited Preemptive Scheduling

Limited Preemptive Scheduling

• An attractive scheduling paradigm instead of fully-preemptive

and non-preemptive scheduling.

• Enables control of preemption related overheads, thus

reducing their impact on schedulability.

Limited Preemptive Scheduling

• An attractive scheduling paradigm instead of fully-preemptive

and non-preemptive scheduling.

• Enables control of preemption related overheads, thus

reducing their impact on schedulability.

• Fixed Preemption Points
• Preemption is allowed only at predefined selected locations

inside the code, called preemption points.

Limited Preemptive Scheduling

• An attractive scheduling paradigm instead of fully-preemptive

and non-preemptive scheduling.

• Enables control of preemption related overheads, thus

reducing their impact on schedulability.

• Fixed Preemption Points
• Preemption is allowed only at predefined selected locations

inside the code, called preemption points. 𝜐" 𝜐#

preemption point

ℎ𝑗𝑕ℎ𝑓𝑠 𝑄 𝑚𝑝𝑥𝑓𝑠 𝑄

Limited Preemptive Scheduling

• An attractive scheduling paradigm instead of fully-preemptive

and non-preemptive scheduling.

• Enables control of preemption related overheads, thus

reducing their impact on schedulability.

• Fixed Preemption Points
• Preemption is allowed only at predefined selected locations

inside the code, called preemption points. 𝜐#

preemption point

𝜐" ℎ𝑗𝑕ℎ𝑓𝑠 𝑄 𝑚𝑝𝑥𝑓𝑠 𝑄

Limited Preemptive Scheduling

• An attractive scheduling paradigm instead of fully-preemptive

and non-preemptive scheduling.

• Enables control of preemption related overheads, thus

reducing their impact on schedulability.

• Fixed Preemption Points
• Preemption is allowed only at predefined selected locations

inside the code, called preemption points. 𝜐#

preemption point

𝜐" ℎ𝑗𝑕ℎ𝑓𝑠 𝑄 𝑚𝑝𝑥𝑓𝑠 𝑄

Limited Preemptive Scheduling

• An attractive scheduling paradigm instead of fully-preemptive

and non-preemptive scheduling.

• Enables control of preemption related overheads, thus

reducing their impact on schedulability.

• Fixed Preemption Points
• Preemption is allowed only at predefined selected locations

inside the code, called preemption points. 𝜐#

preemption point

𝜐" ℎ𝑗𝑕ℎ𝑓𝑠 𝑄 𝑚𝑝𝑥𝑓𝑠 𝑄

Limited Preemptive Scheduling

• An attractive scheduling paradigm instead of fully-preemptive

and non-preemptive scheduling.

• Enables control of preemption related overheads, thus

reducing their impact on schedulability.

• Fixed Preemption Points
• Preemption is allowed only at predefined selected locations

inside the code, called preemption points. 𝜐#

preemption point

𝜐" ℎ𝑗𝑕ℎ𝑓𝑠 𝑄 𝑚𝑝𝑥𝑓𝑠 𝑄

Limited Preemptive Scheduling

• An attractive scheduling paradigm instead of fully-preemptive

and non-preemptive scheduling.

• Enables control of preemption related overheads, thus

reducing their impact on schedulability.

• Fixed Preemption Points
• Preemption is allowed only at predefined selected locations

inside the code, called preemption points. 𝜐#

preemption point

𝜐" ℎ𝑗𝑕ℎ𝑓𝑠 𝑄 𝑚𝑝𝑥𝑓𝑠 𝑄

Motivation

• The existing selection methods account for upper

bounded preemption overheads, thus introducing a potentially high level of pessimism in the results.

Motivation

• The existing selection methods account for upper

bounded preemption overheads, thus introducing a potentially high level of pessimism in the results.

𝜐" 𝜐#

Preemption

• verhead

ℎ𝑗𝑕ℎ𝑓𝑠 𝑄 𝑚𝑝𝑥𝑓𝑠 𝑄

Motivation

• The existing selection methods account for upper

bounded preemption overheads, thus introducing a potentially high level of pessimism in the results.

𝜐" 𝜐#

Preemption

• verhead

ℎ𝑗𝑕ℎ𝑓𝑠 𝑄 𝑚𝑝𝑥𝑓𝑠 𝑄

Motivation

• The existing selection methods account for upper

bounded preemption overheads, thus introducing a potentially high level of pessimism in the results.

𝜐" 𝜐#

preemption

• verhead

Preemption

• verhead

ℎ𝑗𝑕ℎ𝑓𝑠 𝑄 𝑚𝑝𝑥𝑓𝑠 𝑄

Motivation

• The existing selection methods account for upper

bounded preemption overheads, thus introducing a potentially high level of pessimism in the results.

𝜐" 𝜐#

deadline miss

Preemption

• verhead

preemption

• verhead

ℎ𝑗𝑕ℎ𝑓𝑠 𝑄 𝑚𝑝𝑥𝑓𝑠 𝑄

Motivation

• The existing selection methods account for upper

bounded preemption overheads, thus introducing a potentially high level of pessimism in the results.

• Can we reduce the pessimism by considering

probabilistic information about overheads?

𝜐" 𝜐#

deadline miss

Preemption

• verhead

preemption

• verhead

ℎ𝑗𝑕ℎ𝑓𝑠 𝑄 𝑚𝑝𝑥𝑓𝑠 𝑄

Contributions

• We propose a probabilistic distribution model of
• verheads and preemption point selection method which

provides controllable probabilistic relaxations.

Contributions

• We propose a probabilistic distribution model of
• verheads and preemption point selection method which

provides controllable probabilistic relaxations.

𝜐#

Contributions

• We propose a probabilistic distribution model of
• verheads and preemption point selection method which

provides controllable probabilistic relaxations.

𝜐#

Preemption

• verhead

Contributions

• We propose a probabilistic distribution model of
• verheads and preemption point selection method which

provides controllable probabilistic relaxations.

𝜐#

preemption overhead upper bound

Contributions

• We propose a probabilistic distribution model of
• verheads and preemption point selection method which

provides controllable probabilistic relaxations.

𝜐#

preemption overhead upper bound empirical samples

• f preemption overheads

Contributions

• We propose a probabilistic distribution model of
• verheads and preemption point selection method which

provides controllable probabilistic relaxations.

𝜐#

preemption overhead upper bound empirical samples

• f preemption overheads

probability density function probability

Preemption Point Selection Algorithm

Preemption Point Selection Algorithm

• Input
• Task set with potential preemption points
• Associated probabilistic overhead distributions

• Input
• Task set with potential preemption points
• Associated probabilistic overhead distributions
• Output
• Selected preemption points

Preemption Point Selection Algorithm

• Input
• Task set with potential preemption points
• Associated probabilistic overhead distributions
• Output
• Selected preemption points
• Algorithm
• Gradually decreases probabilistic factor for preemption
• verheads in order to find preemption point selection

Preemption Point Selection Algorithm

• Input
• Task set with potential

preemption points

• Associated probabilistic
• verhead distributions.
• Output
• Selected preemption points
• Algorithm
• Gradually decreases

probabilistic factor for preemption overheads in order to find preemption point selection.

27

probability of a deadline miss (part of the future work) iteration focused overhead 1 2 3 1 1 1 implies selection

• f different points

Preemption Point Selection Algorithm

• Input
• Task set with potential

preemption points

• Associated probabilistic
• verhead distributions.
• Output
• Selected preemption points
• Algorithm
• Gradually decreases

probabilistic factor for preemption overheads in order to find preemption point selection.

28

probability of a deadline miss (part of the future work) iteration focused overhead 1 2 3 1 1 1 implies selection

• f different points

Preemption Point Selection Algorithm

• Input
• Task set with potential

preemption points

• Associated probabilistic
• verhead distributions.
• Output
• Selected preemption points
• Algorithm
• Gradually decreases

probabilistic factor for preemption overheads in order to find preemption point selection.

29

probability of a deadline miss (part of the future work) iteration focused overhead 1 2 3 1 1 1 implies selection

• f different points

Preemption Point Selection Algorithm

Preliminary results

• Goal of the experiment: To investigate to what extent the

relaxation of the considered overheads facilitates finding solutions to the preemption point selection problem.

0.8 0.82 0.84 0.86 0.88 0.9 0.92 0.94 0.96 0.98 1

Utilisation

20 40 60 80 100

Task sets for which a selection is found (%)

Upper bounds Quantile selection

Summary and Future work

• Contributions
• Probabilistic overhead model
• Preemption point selection based on probabilistic overhead

distributions

• Future work
• Probabilistic schedulability analysis techniques for tasks with fixed

preemption points and associated probabilistic overheads

• Novel preemption point selection strategies to maximize

schedulability