multi-core research topics relevant to automotive and ... · static wcet speed-up of intra-task...

Prof. Dr. Theo Ungerer

Department of Computer Science

University of Augsburg

May 19, 2015

Multi-core Research Topics

Relevant to Automotive and

Avionics Safety-critical

Control Systems

University of Augsburg and my research team Research project parMERASA

Research ideas beyond parMERASA

Contents of talk


University of

Augsburg

60-80 km

Augsburg

About 20 000 students in 7 faculties/departments

Department of Computer Science: about 1500 students


6


Research Team

Multi-/many-core hardware design and

system software for safety-critical systems.

Timing-aware parallelisation techniques.

Application domains: automotive, avionics,

space, automation


Main Research Focus

8

Recent and Current Research Projects

EC parMERASA: Multi-Core Execution of Parallelised Hard Real-Time Applications Supporting Analysability, 2011- 2014

Project coordination Timing predictable parallelisation approach for hard real-time programs Hard real-time System SW and TinyAUTOSAR Memory and NoC structure of timing predictable many-core

Intel Research: Investigation of Intel TSX in context of safety-critical systems

Intel Haswell Hardware Transactional Memory for fault tolerance and fair scheduling of real-time threads

DFG-Research Group OC-TRUST Trust-enhancement of Self-X-Algorithms Trust-Enabling Middleware


Research Team

Multi-Core Execution of parallelised Hard Real-

Time Applications Supporting Analysability

European Community Research Project

Sept. 1, 2011 – Sept. 30, 2014

3.3 Mio EC contribution

Project webpage: http://www.parmerasa.eu

parMERASA 9

Project Partners and IAB

parMERASA 10

Industrial Advisory Board:

Airbus, Toulouse, France

Infineon Technologies UK Ltd, Bristol, UK

Infineon Technologies AG, Munich,

Germany

BMW Group, Munich, Germany

DELPHI, Sweden

Elektrobit Automotive GmbH, Erlangen,

Germany

Daimler AG, Germany

Overview

Motivation and Principal Objective

Principal Developments and Results

Parallelisation Results

Conclusions

Principal Results

11 parMERASA

Hard Real-time Systems

Hard real-time:

a deadline must never be missed

if missed it may cause harm to humans or equipment

Mixed criticality in multi-cores:

combining functionalities with different levels of criticality

within multi-core systems

e.g. sub-systems to be combined that have different

automotive safety integrity levels (ASIL)

12 parMERASA

parMERASA goes one step beyond mixed

criticality demands:

We target future complex control algorithms

by parallelising hard real-time programs to

run on predictable multi-core processors.

Vision of parMERASA

parMERASA 13

parMERASA 14

Uni Augsburg

Honeywell DENSO Bauer

BSC Uni Augsburg TU Dortmund

Uni Augsburg

BSC & TU Dortmund & Uni Augsburg

Uni Toulouse &

Rapita Systems

Avionics RTE Automotive RTE Constr. M. RTE

Avionics Automotive Construction M.

Kernel Library

Predictable Multi-core Hardware

Verification & Profiling Tools

Parallelisation Methods

Project Layout

Overview




Conclusions

Principal Results

15 parMERASA

Parallelisation of four industrial hard real-time applications:

Stereo navigation (Honeywell International s.r.o.)

3D path planning (Honeywell International s.r.o.)

Control algorithm of a Crawler Crane

(BAUER Maschinen GmbH)

Diesel engine management system

(DENSO Automotive Deutschland GmbH)

Principal Results (1)

16 parMERASA

Pattern-based approach to efficiently parallelise industrial

applications for embedded real-time systems.


17 parMERASA

WCET analysis and verification tools for multi-cores:

static WCET tool OTAWA (University of Toulouse) and

measurement-based WCET tool RapiTime

all extended for parallel programs

Further tools developed/extended for parallel program

analysis by Rapita Systems Ltd:

RapiTask trace viewer

RapiCheck constraint checker

RapiCover for code coverage

RapiTime dependency analysis tool

Principal Results (3): Tools

18 parMERASA

Hard real-time support in system software

Common Kernel Lib

Tiny automotive, Tiny avionics, BIOS for crawler crane

RTEs


19 parMERASA

Overview




Conclusions

Principal Results

20 parMERASA

Execution times measured by parMERASA simulator

Two types of WCET speed-ups:

based on static WCET bounds reached by OTAWA

based on dynamic WCET estimates based on RapiTime

Results of Parallelisations (1)

21 parMERASA

Observed speed−up =execution time of the sequential program

execution time of the parallelised version

WCET speed−up =WCET estimate of the sequential program

WCET estimate of the parallelised version

3DPP application

speed-ups

with perfect

cache coherence


22 parMERASA

3DPP application

speed-ups

with ODC2

cache coherence


23 23/09/2014 parMERASA

Observed execution time with ODC2 (2.51 million

cycles) is very close to a cache with perfect

coherency protocol (2.42 million cycles for 8 cores)

Compaction machine application


24 23/09/2014 parMERASA

Diesel engine management system

(DENSO Automotive Deutschland GmbH)

1200 runnables

11 time-driven tasks and

1 crank-angle task (interrupt from the camshaft sensor)

(1) Inter-task level: parallel execution of tasks

(2) Intra-task level: parallel execution of runnables

of the same task

(3) Intra-runnable: parallel execution of instruction blocks

or function calls of a runnable.


25 parMERASA

Diesel engine management system:

static WCET speed-up of inter-task parallelization


26 parMERASA


static WCET speed-up of intra-task parallelization


27 parMERASA

Intra- and inter-task parallelism combined:

Longest running task distributed over 2 cores.

Static WCET speedup estimate increased to 5.97 on 8 cores.


measurement-based WCET speed-up of intra-runnable parallel.


28 parMERASA

EC FP-7 parMERASA project (Oct. 1, 2011 - Sept. 30, 2014)

targeted future complex control algorithms by parallelizing

hard real-time programs.

Reasonable WCET speed-ups can be reached with a low

number of cores, e.g. 5.97 on 8 cores for diesel EMS.

Scalability of real-world hard real-time applications that run

successful on single-core is limited.

Results Summary

29 parMERASA

EC FP-7 parMERASA project (Oct. 1, 2011 - Sept. 30, 2014)

targeted future complex control algorithms by parallelizing

hard real-time programs.

Reasonable WCET speed-ups can be reached with a low

number of cores, e.g. 5.97 on 8 cores for diesel EMS.

Scalability of real-world hard real-time applications that run

successful on single-core is limited.

parMERASA project paved the way for future high-performance

embedded systems applications.

More complex control algorithms than today can be applied.

Results Summary

30 parMERASA

Potentially relevant to future AUTOSAR

Investigate combination of Latest-is-best semantic with

Time Synchronous Execution Model

Investigate Real-time scheduling in case of deadline misses

Research Ideas (1)

31

Investigate high-performance control algorithms

(physical simulation, autonomous driving, compound ECU)

Investigate message-passing many-cores in safety-critical

systems (parallelisation, parallel patterns, timing-predictability,

fault tolerance)

Research Ideas (2)

32

Hardware transactional memory in embedded systems

apply techniques developed during the Intel project to

embedded microcontrollers and systems

Potential advantages

non-blocking synchronisation leads to ease of programming

shortened blocking critical code sections

no priority inversion

Hardware managed and fine-grained checkpointing lead

to new lock-step-techniques with fine-grained restart

in case of transient faults

Research Ideas (3)

33

Potentially relevant to current and future Automotive Systems

Combine automotive driver-assistance systems with

medical devices

Possible extension of a planned EC FET proposal on

body-implanted medical devices

- Brain implant for the rapid detection and stopping of

epileptic seizures

- Heart implant (Cardiac-Alarm) for the timely detection of

Ventricular Fibrillation

Research Ideas (4)

34

Thanks for your attention

multi-core research topics relevant to automotive and ... · static wcet speed-up of intra-task...

Documents