Hard Real-time Guarantee of Automotive Applications during

Hard Real-time Guarantee of Automotive Applications during Mode Changes

P. Dziurzanski, A.K. Singh, L.S. Indrusiak

University of York

B. Saballus

Robert Bosch GmbH

Overview

Automotive technology: an overview

Modes in Engine Control Units (ECU)

DemoCar use case

Design flow

Conclusion

2

DreamCloud Kickoff 09/2013

Powertrain

systems Safety systems Comfort systems

Automotive technology: an overview

3

Source: Robert Bosch GmbH, Future mobility: automated, connected, electric, Belgian Insurance Conference, 2014

DreamCloud Kickoff 09/2013

Engine control system

800 – 1600 SW components

1000 – 1500 runnables (basic executable entities) in about 10 tasks

3000 – 6000 sub functions

7000 – 15000 messages

hard real-time and safety requirements

4

Source: Robert Bosch GmbH

Current vs future trends in cars

different HW units for different functionality

more than 100 ECUs in a premium car

static distribution

bus-based

5

2015 Future

multiple applications per HW unit

dynamic distribution

Network on Chips

Limited number of operating modes

Ignition key-On: engine off state

Cranking: engine starting state

Idle: a throttle valve is not opened

Part load: a throttle valve is partially opened

Wide open throttle: a throttle valve is wide open

6

Source: J.S. Park et al., Mode-Dynamic Task Allocation and Scheduling for an Engine Management Real-Time System Using a Multicore Microcontroller, SAE Int. J. Passeng. Cars – Electron. Electr. Syst. 7(1):133-140, 2014.

DemoCar- Minimal functionality gasoline Engine Control Unit

7

18 runnables

6 tasks

61 labels (data elements that can be read or written by runnables)

Problem formulation

8

π0,1 π1,1 π2,1

π0,0 π1,0 π2,0

Goals: • no deadline violation • minimal #used resources • minimal energy dissipation

Finite State Machine describing mode changes in DemoCar use case

9

The current state is stored in a certain variable.

Runnables execution time in different modes

10

Runnable Min op. Max op.

CylNumObserver 134 345

InstructionsDeviation 3728 5921

Runnable Min op. Max op.

CylNumObserver 245 543

InstructionsDeviation 728 921

Steps of the proposed dynamic resource allocation method

11

Design-time

optimisation

Taskset Platform

Run-time Platform

Resource Manager

Current Mode

Allocation

Result

Allocations

Modes

Mode detecting / clustering

12

Reasons: • Runnables have similar

runtime and resource consumption in neighbouring modes.

• Transition is required to be done immediately.

PowerDown PowerUp Cluster_1

Spanning tree construction

13

Static mapping for initial mode

14

M. Norazizi

Using energy dissipation as a criterion, huge

penalties for deadline violations

Genes in chromosomes

15

... ...

Runnable mapping (n genes) Core P-mode (x∙y genes)

τ1

τ2

τ3

τn

... π1,2 π2,2 πx,1

π1,1 π2,1 πx,0

π1,y π2,y πx,y

...

...

...

...

... ...

...

π0,1 π1,1

π0,0 π1,0

Optimization result

16

zzz

zzz zzz

18r

61l

P0

Static mapping for non-initial modes minimizing amount of migrated data

Multicriteria optimization: migration cost from the previous

mode energy dissipation

Huge penalties for deadline violations

17

Pareto frontier - Cluster_1

18

Data to be migrated [bytes] Energ

y u

sed in m

ode C

luste

r_1 [

μJ]

No deadline misses.

9650

9700

9750

9800

9850

9900

9950

10000

10050

10100

200000 300000 400000 500000 600000 700000

Idle hardware in states - 2x2 NoC

• 3 idle cores • 16 idle links



3093.01 µJ per hyperperiod



19

Deadlines for runnables

20

APedVoterSWCRunnableEntity

ThrottleCtlrRunnableEntity

ThrottleChangeSWCRunnableEntity

TransFuelMassSWCRunnableEntity

TotalFuelMassSWCRunnableEntity

BaseFuelMassRunnableEntity

IgnitionSWCRunnableEntity

ThrottleActuatorRunnableEntity

0 1 2 3 4 5 6 7 8 9 10

Task

_10

ms

IgnitionSWCSyncRunnableEntity

InjectionSWCSyncRunnable

0 2.5 5

Act

uat

orT

ask

runnable execution

data written to labels









0 1 2 3 4 5 6 7 8 9 10

Task

_10

ms



0 2.5 5

Act

uat

orT

ask

Schedulability analysis for taskset in any mode

21

schedulability analysis for all jobs

that can be executed in this time slot

schedulability analysis for all packets that

can be transmitted in this time slot









0 1 2 3 4 5 6 7 8 9 10

Task

_10

ms



0 2.5 5

Act

uat

orT

ask

0 10Per

iod

ic s

erve

r

Schedulability analysis for taskset during mode changes

22

schedulability analysis for all packets that can be transmitted in this time slot together with migrations

(done with periodic servers)

Network bandwidth selection

Router latency [ns]

Link latency [ns]

#Hyperperiods (100ms)

100 200 1

100 400 1

200 500 1

400 800 2

500 1000 2

23

Router latency [ns]

Link latency [ns]

#Hyperperiods (100ms)

100 200 1

100 400 2

200 500 2

400 800 3

500 1000 3

Data to be migrated [bytes]

Energ

y u

sed in m

ode C

luste

r_1 [

μJ]

9650

9700

9750

9800

9850

9900

9950

10000

10050

10100

200000 400000 600000 800000

How many hyperperiods are required to migrate all the necessary data using periodic server?

Conclusion

Static mapping of modes and dynamic migration at runtime on mode switch has been proposed to decrease the number of cores in ECUs.

The technique benefits from the modal nature of the ECU.

GA has been used to determine the allocation for all the ECU's modes, minimizing the energy dissipation and migration.

NoC bandwidth for full schedulability has been determined.

24

Thank you!

25

Estimation of dissipated energy

26

π0

π1

t0,0 t0,1

t1,0

E0=E0,init+fd0(t0,0)+fd0(t0,1)+fs0(ts)

E1=E1,init+fd1(t1,0)+fs1(ts)

fd0(t)=const0∙t

fd1(t)=const1∙t

Dynamic energy: Static energy:

ts

fs0(t)=const2∙t

fs1(t)=const3∙t

Documents

Hard Real-time Guarantee of Automotive Applications during