event-triggered tasks Martijn van den Heuvel Reinder J. Bril - - PowerPoint PPT Presentation

Limited preemptive scheduling

• f mixed time-triggered and

event-triggered tasks

Martijn van den Heuvel Reinder J. Bril Xiaodi Zhang Syed Md Jakaria Abdullah Damir Isovic

ETFA 2013

Complexity of industrial applications?

/ Mathematics and computer science PAGE 1 9-9-2013

Efficient resource usage vs complex constraints

Time-triggered task model

/ Mathematics and computer science PAGE 2 9-9-2013

• Job
• Absolute deadline = end of interval

I0 I1 I2 I3 I4 I0 4 6 9 14 20 ...

• Execute jobs in the slots captured in a scheduling table
• Resolve complex task constraints offline

/ Mathematics and computer science PAGE 3 9-9-2013

Offline scheduling

• Offline reservation of time-triggered tasks into intervals
• Each interval is divided into
• spare capacity and
• reserved capacity

P1 P3 P2

Time-triggered tasks

spare capacity reserved capacity length of the interval

1 2 4 5 6 7 8 9 10 11 3

P3

12

P1 P2

I0 I1

Time-triggered task model

/ Mathematics and computer science PAGE 4 9-9-2013

• Job
• Absolute deadline = end of interval

I0 I1 I2 I3 I4 I0 4 6 9 14 20 ...

• Execute jobs according to a scheduling table
• Resolve complex task constraints offline

What to do with the free capacity? Polling is expensive → efficient event handling? How to extend system functionality?

Mixed time- and event-triggered Task model

/ Mathematics and computer science PAGE 5 9-9-2013

Sporadic task ( Si ) A-periodic task ( Ai )

• Use the free time slots for ET jobs:
• Relative deadline = Di
• Absolute deadline = arrival time + Di

Preemptive EDF scheduling: based on absolute deadline

Arrival time di,k job Di

Event-triggered (ET): Time-triggered (TT):

I0 I1 I2 I3 I4 I0 4 6 9 14 20 ...

How (fully preemptive) slotshifting works?

/ Mathematics and computer science PAGE 6 9-9-2013

[Isovic and Fohler 2009]

How (fully preemptive) slotshifting works?

/ Mathematics and computer science PAGE 7 9-9-2013

[Isovic and Fohler 2009]

How (fully preemptive) slotshifting works?

/ Mathematics and computer science PAGE 8 9-9-2013

[Isovic and Fohler 2009]

Slotshifting versus static offline schedules

• Advantages of slotshifting:
• Online reclaiming of spare capacity
• Same worst-case offline analysis as static schedules
• Disadvantages of slotshifting:
• Shifting of slots must allow for arbitrary preemptions
• No support for allocating

subsequent slots

/ Mathematics and computer science PAGE 9 9-9-2013

Non-preemptive shifts

How to overcome these disadvantages?

Slotshifting versus static offline schedules

• Advantages of slotshifting:
• Online reclaiming of spare capacity
• Same worst-case offline analysis as static schedules
• Disadvantages of slotshifting:
• Shifting of slots must allow for arbitrary preemptions
• No support for allocating

subsequent slots

/ Mathematics and computer science PAGE 10 9-9-2013

Restricted shifting of TT tasks! See paper…

Non-preemptive shifts

Problem description

/ Mathematics and computer science PAGE 11 9-9-2013

• What already exists?

Local system Global system

Time-triggered Offline scheduling

New offline analysis+ New online scheduling mechanism

Event-triggered

Standard (SRP +) EDF scheduling

• Slotshifting of preemptive tasks [Isovic and Fohler 2009]
• What is new?
• Extended task analysis for non-preemptive execution

Problem description

/ Mathematics and computer science PAGE 12 9-9-2013

• What already exists?

Local system Global system

Time-triggered Offline scheduling

New offline analysis+ New online scheduling mechanism

Event-triggered

Standard (SRP +) EDF scheduling

• Slotshifting of preemptive tasks [Isovic and Fohler 2009]
• What is new?
• Extended task analysis for non-preemptive execution

Problem: how to disable preemptions?

/ Mathematics and computer science PAGE 13 9-9-2013

Global system Locally Non-preemptive Locally Non-preemptive Local TT task system Local ET task system Time-triggered tasks Event-triggered tasks Globally Non-preemptive Preemption

Currently: fully preemptive EDF of the global system Our contributions: 3 limited-preemptive extensions

Problem: how to disable preemptions?

/ Mathematics and computer science PAGE 14 9-9-2013

Global system Locally Non-preemptive Locally Non-preemptive Local TT task system Local ET task system Time-triggered tasks Event-triggered tasks Globally Non-preemptive Preemption

• 1. Restricted shifting of TT tasks!

See paper…

Global system Locally Non-preemptive Locally Non-preemptive Local TT task system Local ET task system Time-triggered tasks Event-triggered tasks Globally Non-preemptive

Global non-preemptive execution of ET tasks

/ Mathematics and computer science PAGE 15 9-9-2013

New offline analysis + New online scheduling mechanism

Our main contribution:

Problem: how to disable preemptions?

/ Mathematics and computer science PAGE 16 9-9-2013

Global system Locally Non-preemptive Locally Non-preemptive Local TT task system Local ET task system Time-triggered tasks Event-triggered tasks Globally Non-preemptive Preemption

• 2. Intermediate step:

Incorporate EDF+SRP analysis of [Baruah, RTSS 2006]

Offline analysis of ET tasks (Recap)

/ Mathematics and computer science PAGE 17 9-9-2013

[Isovic and Fohler 2009]

Slotshifting: a fully preemptive system (Recap)

/ Mathematics and computer science PAGE 18 9-9-2013

Global system Local TT task system Local ET task system Time-triggered tasks Event-triggered tasks I0 I1 I2 I3 I4 I0 4 6 9 14 20 ...

Admission?

/ Mathematics and computer science PAGE 19 9-9-2013

Offline analysis of ET tasks (Recap)

Schedulability test of sporadic tasks

• Classical demand-based analysis
• Supply based on static offline schedule

Demand during an interval of length L

L

≤

[Baruah 2006]

Demand during an interval of length L

≤

Supply based on offline schedule Supply during an interval of length L

Calculation of supply bound

(t2 − t1)

/ Mathematics and computer science PAGE 20 9-9-2013

L

t1 I0 I1 I2 t2 I3

L

t1 I0 I1 I2 t2 I3

L

t1 I0 I1 I2 t2 I3

− overlapping reserved capacities in d intervals

Calculation of supply bound

/ Mathematics and computer science PAGE 21 9-9-2013

L

t1 I0 I1 I2 t2 I3

L

t1 I0 I1 I2 t2 I3

L

t1 I0 I1 I2 t2 I3

L = (t2 − t1)

Schedulability of ET tasks

/ Mathematics and computer science PAGE 22 9-9-2013

• A sporadic task set Ts is can be scheduled under EDF

together with an offline schedule, iff

where − Demand bound function, dbf [Baruah, RTSS 2006] − New Supply bound function, sbf

Calculation of sbf: (t2 − t1) − overlapping reserved capacities in d intervals

Disable local preemptions of ET tasks

/ Mathematics and computer science PAGE 23 9-9-2013

Preemption Global system Locally Non-preemptive Locally Non-preemptive Local TT task system Local ET task system Time-triggered task Event-triggered task

Intermediate step: Incorporate EDF+SRP analysis of [Baruah, RTSS 2006]

Scheduling rules for local ET system

/ Mathematics and computer science PAGE 24 9-9-2013

Rule

• If an ET task starts executing its critical section

− Case 1: Preemptions from any other ET tasks are disabled − Case 2: Preemptions from TT task with higher priority are allowed

Non-preemptive sections of local ET tasks

/ Mathematics and computer science PAGE 25 9-9-2013

• Local blocking:

S1 S2 I0 I1 rcℓ Blocking Critical section of sporadic task Reserved capacity of interval Sporadic task

Non-preemptive sections of local ET tasks

/ Mathematics and computer science PAGE 26 9-9-2013

• Blocking function [Baruah, RTSS 2006]

If hj denotes length of critical sections B(L)  maximum hj needed by a lower priority task

• Schedulability test

Global system Locally Non-preemptive Locally Non-preemptive Local TT task system Local ET task system Time-triggered tasks Event-triggered tasks Globally Non-preemptive

Global non-preemptive execution of ET tasks

/ Mathematics and computer science PAGE 27 9-9-2013

New offline analysis + New online scheduling mechanism

Our main contribution:

Scheduling rules for Global System

/ Mathematics and computer science PAGE 28 9-9-2013

• Rule 1
• Case 1: spare capacity > critical section length

ET task executes globally non-preemptively

• Case 2: spare capacity < critical section length

ET task blocks it-self and other ET tasks

Scheduling rules for Global System

/ Mathematics and computer science PAGE 29 9-9-2013

• Rule 1
• Case 1: spare capacity > critical section length

ET task executes globally non-preemptively

• Case 2: spare capacity < critical section length

ET task blocks it-self and other ET tasks

• Rule 2
• Case 1: another ET task wants to execute during self-blocking

ET task is not allowed to execute

• Case 2: a TT task wants to execute during self-blocking

TT task is allowed to execute

S1 rcℓ Miss I0 I1 I2 I3 Miss Global critical section Reserved capacity of interval Sporadic task Reserved self-blocking slot 1 6 9 16

Global non-preemptive execution of ET tasks

/ Mathematics and computer science PAGE 30 9-9-2013

• Self-blocking mechanism

S1 rcℓ Miss I0 I1 I2 I3 Miss Blocking Global critical section Reserved capacity of interval Sporadic task Reserved self-blocking slot 1 6 9 16 S1 rcℓ I0 I1 I2 I3

Global non-preemptive execution of ET tasks

/ Mathematics and computer science PAGE 31 9-9-2013

• Self-blocking mechanism

Self-blocking mechanism

/ Mathematics and computer science PAGE 32 9-9-2013

• idle bound function ibf(hi):
• Calculate idled away time units in self-blocking.
• ibf(hi) has no relation to any task property;

it reflects the delay of executing a global critical section with length hi

τ1 τ2 rcℓ I0 I1 I2 I3 I4 4 6 9 14 20 Non-preemptive region of sporadic task Reserved capacity of interval Sporadic task Reserved self-blocking slot

Idle bound function calculation

/ Mathematics and computer science PAGE 33 9-9-2013

τ2 rcℓ I0 I1 I2 I3 I4 I0 4 6 9 14 20 Self-blocking Non-preemptive region of sporadic task Reserved capacity of interval Sporadic task Reserved self-blocking slot ...

• Example(cont’d)
• Interval I0 has enough spare capacity larger or equal to critical section
• Assume the critical section starts just too late.

Idle bound function calculation

/ Mathematics and computer science PAGE 34 9-9-2013

τ2 rcℓ I0 I1 I2 I3 I4 I0 4 6 9 14 20 Self-blocking ...

• Example(cont’d)
• Interval I3 has enough spare capacity larger or equal to critical section

Idle bound function calculation

/ Mathematics and computer science PAGE 35 9-9-2013

τ2 rcℓ I0 I1 I2 I3 I4 I0 4 6 9 14 20 Self-blocking ...

• Example(cont’d)
• Interval I4 has enough spare capacity larger or equal to critical section

The maximum idled away time units:

New ET schedulability test

/ Mathematics and computer science PAGE 36 9-9-2013

hi ibf(hi) 2 3 3 6

Self-blocking from lower priorities Add self-blocking to WCET of jobs

Compute inserted idle times for non-preemptive sections:

Problem: how to disable preemptions?

/ Mathematics and computer science PAGE 37 9-9-2013

Global system Locally Non-preemptive Locally Non-preemptive Local TT task system Local ET task system Time-triggered tasks Event-triggered tasks Globally Non-preemptive Preemption

Restricted shifting of TT tasks; Use local SRP [Baruah 2006]; Self-blocking of ET tasks.

Conclusion

/ Mathematics and computer science PAGE 38 9-9-2013

New schedulability analysis for slotshifting: Non-preemptive scheduling of tasks:

• Extended analysis with

− Time-triggered tasks: by construction − Event-triggered tasks: blocking function

• Extended scheduling mechanisms:

− A self-blocking mechanism for event-triggered tasks

Demand during an interval of length L

≤

supply based on offline schedule Supply during an interval of length L