research in modelresearch in model--based product ... · research in modelresearch in model--based...

Research in ModelResearch in Model--Based Product DevelopmentBased Product Developmentat PELAB in the MODPROD Centerat PELAB in the MODPROD Center

Presentation at MODPROD'2011Presentation at MODPROD'2011PELAB PELAB –– Programming Environment LaboratoryProgramming Environment Laboratory

at PELAB in the MODPROD Centerat PELAB in the MODPROD Center

Department of Computer and Information ScienceDepartment of Computer and Information ScienceLinköping University Linköping University

20112011--0202--0808

Peter FritzsonPeter Fritzson

class x {publicint a;float b;int func (int a,int b);Asa asadAsda ad Asd ad ccAac sdscfcc c aAscccv caAscc cac}




Examples of Complex Systems in Examples of Complex Systems in EngineeringEngineeringg gg g• Robotics

A tomoti e• Automotive• Aircraft• Mobile Phone Systems• Mobile Phone Systems• Business Software • Power plantsPower plants• Heavy Vehicles• Process industryy

2 © Peter Fritzson

Vision of IntegratedVision of IntegratedModelModel--Based Product DevelopmentBased Product Developmentpp

FeedbackFeedback

Business Requirements Model-Driven Compilation

SystemSimulation

eedb c

Business Requirements Model-Driven Compilation

SystemSimulation

eedb c

ProcessControl

RequirementsCapture Design

(PIM)& Code Gen

(PSM)

Software &

ProcessControl

RequirementsCapture Design

(PIM)& Code Gen

(PSM)

Software &

Product models

Requirementsmodels

Syst ProductPlatformmodels

Processmodels

Product models

Requirementsmodels

Syst ProductPlatformmodels

Processmodels

Unified Modeling: Meta-modeling& Modelica& UML & OWLUnified Modeling: Meta-modeling& Modelica& UML & OWL

Vision of unified modeling framework for model-driven product development from platform independent models (PIM) to platform specific models (PSM)

3 © Peter Fritzson

to platform specific models (PSM)

Open Source ModelicaOpen Source ModelicaOpen Source Modelica Open Source Modelica ConsortiumConsortium

Open Source EffortOpen Source EffortOpen Source EffortOpen Source Effort

4 © Peter Fritzson

The OpenModelica Open Source Enviroment The OpenModelica Open Source Enviroment www.openmodelica.orgwww.openmodelica.orgwww.openmodelica.orgwww.openmodelica.org• Advanced Interactive Modelica compiler (OMC)

• Supports most of the Modelica Language

• Basic environment for creating models• OMShell – an interactive command handler * ModelicaML UML Profile• OMNotebook – a literate programming notebook * MetaModelica transforms• MDT – an advanced textual environment in Eclipse * OMEdit graphic editor

5 © Peter Fritzson

OSMC OSMC –– International Consortium for Open Source International Consortium for Open Source ModelModel--based Development Tools,based Development Tools, 32 members Dec 201032 members Dec 2010

Industrial members (18)•ABB Corporate Research, Sweden Open source community services

ModelModel based Development Tools, based Development Tools, 32 members Dec 201032 members Dec 2010Founded Dec 4, 2007

•Evonik Energy Services, India• IFP, Paris, France

•Bosch Rexroth AG, Germany •Siemens Turbo Machinery AB, Sweden •CDAC Centre for Advanced Computing, Kerala, India

Open-source community services• Website and Support Forum• Version-controlled source base

Bug database

, ,•InterCAX, Atlanta, USA•MOSTforWATER, Belgium •MathCore Engineering AB, Sweden •Maplesoft, Canada

•CEIT Institute, Spain•Creative Connections, Prague, Czech Republic •Frontway AB, Sweden•Equa Simulation AB, Sweden

• Bug database• Development courses• www.openmodelica.org

•TLK Thermo, Germany •VI-grade, Italy•VTT, Finland •XRG Simulation, Germany

Code Statistics

•Linköping University, Sweden •Hamburg University of Technology/TuTech, Germany FH Bi l f ld Bi l f ld G

University members (14)

•FH Bielefeld, Bielefeld, Germany•Technical University of Braunschweig,, Germany •Technical University of Dortmund, Germany•Université Laval, modelEAU, Canada •Griffith University, Australia •Politecnico di Milano Italy•Politecnico di Milano, Italy •Mälardalen University, Sweden •Technical University Dresden, Germany•Telemark University College, Norway•Ghent University, Belgium•Ecoles des Mines CEP Paris France

6 © Peter Fritzson

•Ecoles des Mines, CEP, Paris, France•University of Ljubljana, Slovenia

New OpenModelica Connection EditorNew OpenModelica Connection EditorOMEditOMEdit

• Supports MSL 3.1E t• Easy to use

• Rather stableI l t d i C Qt lib• Implemented in C++ Qt library

7 © Peter Fritzson

New Interactive New Interactive Simulation with Simulation with OpenModelicaOpenModelica

Examples of Simulation Visualization

Simulation Control

Plot View

RequirementsRequirements

Evaluation View

Liquid Source

MaxLevel

Level h

Level h Domain-Specific

Visualization View

8 © Peter Fritzson

Tank 1 Tank 2

New OpenModelica Optimization SubsystemNew OpenModelica Optimization SubsystemOMOptimOMOptimpp• Parameter

optimizationoptimization• Currently using

genetic optimizationoptimization algorithms in OMOptim 0.9.

9 © Peter Fritzson

New WebNew Web Server Based Teaching/Learning SubsystemServer Based Teaching/Learning SubsystemOMWeb OMWeb

• Runs on ordinary web serverweb server

• Teachers provide modelingmodeling exercises

• Students send inStudents send in solutions

• Running ofgsimulations through server

• Lila EU project

10 © Peter Fritzson

New New DrControlDrControl Interactive Interactive Electronic BookElectronic Book

Examples

QuestionsTheory

ExperimentationModels


ResearchResearchResearchResearch

ModelingModeling--Language DesignLanguage Design

Modeling Support EnvironmentsModeling Support EnvironmentsModeling Support EnvironmentsModeling Support Environments


Important QuestionsImportant Questionspp

• Design of modeling languages for modeling l ( h i l) t (M d li ) i l dicomplex (physical) systems (Modelica) including

precise semantics, extensibility, etc.• How to engineer complex engineering systems of

both hardware and software in a consistent and safe manner?

• Compilation of models for efficient (real-time)• Compilation of models for efficient (real-time) execution on multi-core architectures

• Traceability from requirements to models to implementation


Context of Using of Semantics for Checking Context of Using of Semantics for Checking g gg g


Current Semantics and Current Semantics and MetaModelicaMetaModelica/Modelica 4 /Modelica 4 WorkWorkWorkWork• Defining Core language MKL (Modeling Kernel

Language) and Core MetaModelica for expressing baseLanguage) and Core MetaModelica for expressing base semantics

• Other constructs should be expressible in the Core language

• Defining extensible meta-modeling/meta-programming language primitives (e g MetaModelica)language primitives (e.g. MetaModelica)

• Define compiler and core library in MetaModelica, p y ,migrated to Modelica 4


Integrated HardwareIntegrated Hardware--Software Software ggModelingModeling

ModelicaMLModelicaMLModelicaML ModelicaML UML Profile for ModelicaUML Profile for Modelica

SysMLSysML Modelica IntegrationModelica IntegrationSysMLSysML--Modelica Integration Modelica Integration


Using ECLIPSE as Integration PlatformUsing ECLIPSE as Integration Platformg gg g

UML-ModelicaOpenUP/Basic Capacity Sub-OpenModelicaMDT

UML ModelicaPlug-in

OpenUP/Basic p yProcess Areas

ECLIPSE Process Framework (EPF) Composer Specific components

ECLIPSE Modeling Framework

Graphical Modeling Framework

ECLIPSE Rich Client Platform (RCP) Runtime

ECLIPSE Modeling Framework

Java runtime

OpenModelica runtime

MetaModelica runtime

C/C++ runtime


ModelicaMLModelicaML –– UML Profile for ModelicaUML Profile for Modelica1st Generation (20061st Generation (2006--2008)2008)(( ))

• Extension of SysML subsetExtension of SysML subset• Features:

S pports Modelica constr cts• Supports Modelica constructs• Modelica generic class modeling

M d li t i d fi iti• Modelica syntax in definitions• Equation-based modeling

Si l ti d li• Simulation modeling


ModelicaMLModelicaML UML Profile, 2nd GenerationUML Profile, 2nd GenerationSysMLSysML/UML to Modelica OMG Standardization/UML to Modelica OMG Standardization(with Wladimir Schamai 2009(with Wladimir Schamai 2009 Now)Now)(with Wladimir Schamai, 2009 (with Wladimir Schamai, 2009 –– Now)Now)

• ModelicaML is a UML Profile for SW/HW modeling• Applicable to “pure” UML or to other UML profiles e g SysML• Applicable to pure UML or to other UML profiles, e.g. SysML

• Standardized Mapping UML/SysML to Modelica• Defines transformation/mapping for executable modelsDefines transformation/mapping for executable models• Being standardized by OMG

• ModelicaML• Defines graphical concrete syntax (graphical notation for diagram) for

representing Modelica constructs integrated with UML• Includes graphical formalisms (e.g. State Machines, Activities, g p ( g , ,

Requirements) • Which do not exist in Modelica language • Which are translated into executable Modelica code• Which are translated into executable Modelica code

• Is defined towards generation of executable Modelica code• Current implementation based on the Papyrus UML tool +

O M d li


OpenModelica

ModelicaML: Graphical NotationModelicaML: Graphical Notation

Structure Requirements

B h iBehavior


Example: Representation of System StructureExample: Representation of System Structure

Interconnections

Inheritance

Components


Example: Representation of System BehaviorExample: Representation of System Behavior

State Machine of the

Controller

State Machine of

the Tank

Conditional Algorithm

(Activity Diagram)


ExampleExample: : Representation of System RequirementsRepresentation of System Requirements

Textual Requirement Formalized RequirementTextual Requirement Formalized Requirement


ExampleExample: : Simulation and Requirements Simulation and Requirements EvaluationEvaluation

Req. 001 is instantiated 2 times (there are 2 tanks in

the system)

tank-height is 0.6m

Req. 001 for the tank2 is violated

Req. 001 for the tank1 is not violated


Parallel Execution Parallel Execution Compilation to MultiCore Compilation to MultiCore


Towards HighTowards High--Level Parallel Level Parallel Modeling and SimulationModeling and Simulation

Single-processor Performance Scaling

14.0

16.0

Parallelismgg• Simulations are

time-consuming10.0

12.0

14.0

edu

p

Throughput incr. 55%/year

Device speed

Parallelism

Assumed increase17%/year possibletime consuming

• Moore’s ”Law”: (since 1965) • #devices per chip area doubles

every 18 months 4.0

6.0

8.0

Log

2 S

pee

Limit: Clock rate

Pipelining

Device speed

every 18 months• CPU clock rate also doubled every

18 months – until 2003,then: heat and power issues, limited ILP, ... superscalar technology has reached

0.0

2.0

1984198619881990199219941996199820002002200420062008201020122014201620182020

Limit: RISC ILP

Pipelining

RISC/CISC CPI2006gy

its limits,only (thread-level) parallelism can increasethroughput substantially

• The consequence:

65 nm45 nm 32nm 22nm90 nmSource: Doug Burger, UT Austin 2005

• The consequence: Chip multiprocessors (+ clusters)• Multi-core, PIM, ... (for general-purpose computing)

Need parallel programming/modeling/parallelization• Need parallel programming/modeling/parallelization• Automatic parallelization• Explicit parallel modeling and parallel programming


Towards EasyTowards Easy--toto--UseUseModeling & Simulation using Parallel ComputersModeling & Simulation using Parallel Computers

Modeling using drag’n’drop

Parallel E i

Visualization

TranslationExecution

Parallel SimulationSimulation

Code


Integrating Parallelism and Mathematical ModelsIntegrating Parallelism and Mathematical ModelsThree ApproachesThree Approachespppp

• Automatic Parallelization of Mathematical Models (ModPar)Automatic Parallelization of Mathematical Models (ModPar)• Parallelism over the numeric solver method.• Parallelism over time. • Parallelism over the model equation systemParallelism over the model equation system

• ... with fine-grained task scheduling

• Coarse-Grained Explicit Parallelization Using ComponentsCoarse-Grained Explicit Parallelization Using Components• The programmer partitions the application into computational components

using strongly-typed communication interfaces. • Co-Simulation, Transmission-Line Modeling (TLM), g ( )

• Explicit Parallel Programming • Providing general easy-to-use explicit parallel programming constructsProviding general, easy to use explicit parallel programming constructs

within the algorithmic part of the modeling language. • NestStepModelica


Modelica Simulations Modelica Simulations –– Parallelization ApproachParallelization Approach

• Simulation = solution of (hybrid) DAEs from modelsSimulation solution of (hybrid) DAEs from models

00

),,(),,,(

tYXhtYXXg

• In each step of numerical solver:• Calculate in g (and Y in h)

0),,( tYXh

X g ( )

• Parallelization approach: perform the calculation of in parallel

X

X

p• Called parallelization over the system.

• Drawback: Numeric solver might become bottle-neckDrawback: Numeric solver might become bottle neck


Example Example –– Task Graphs and Parallelized Task Graphs and Parallelized ApplicationApplicationpppp

125 0

2.5

3Speedup

32

21

4 5 60 0 0

1.5

2

41

52

62

7 80 010 0.5

1

71

81

1 2 4 8 16# Proc

Clustered Task Graph Thermofluid Pipe Application


Modified Approach Modified Approach ––InliningInlining and Pipeliningand Pipelininggg p gp g

• Try to keep communication as close as iblpossible

• Only communicate in one direction inside a• Only communicate in one direction inside a time step.

• Solver Inlining – distribute the solver across all the processorsthe processors

• Some parallelism across the method – parallel p pevaluation of Runge-Kutta step


Task Merging Task Merging vsvs Approach with Approach with Pipelining/Pipelining/InliningInlining

S

Use a graph iti

p gp g gg

• Try to keepEquation

Equatio

Equatio

SSS

rewriting system to merge tasks into larger

Try to keep communication as close as possiblen tasks

on tasks

on tasks

S

into larger tasks, based on latency and bandwidth

• Only communicate in

di ti

S

SS

Sbandwidth.

Some tasks are duplicated

one direction inside a time step.

S

Eq EE SS

are duplicated to avoid communication within a step

• Solver Inlining –distribute the solver across allquation ta

Equation

Equation

SSwithin a step solver across all

the processors


asks

tasks

tasks

SS

Measurements PipelinedMeasurements Pipelined(100000 steps Flexible Shaft Model, Fall 2008)(100000 steps Flexible Shaft Model, Fall 2008)

Task-merging, MPI, SGI Altix Pipelined, Pthreads, SGI, Intel Xeon

Relative speedup

4

Relative speedup

7

3

3,5

5

6 SGI

1,5

2

2,5

Spee

dup

3

4

Spee

dup Intel

0,5

1

1

2

00 5 10 15

Number of processors

SGI Altix 3700 Bx2

00 2 4 6 8 10

Number of processors

Intel Xeon SGI Altix 3700 Bx2


SGI Altix 3700 Bx2

332/14/2011 PELAB - Linköping University

Speedup Measurements on NVIDIA (Speedup Measurements on NVIDIA (novnov 2009)2009)Modelica Model, Generated Code, Function of Problem SizeModelica Model, Generated Code, Function of Problem Size, ,, ,

• New PlatformNVIDIA Fermi 20502 Teraflop peak

• In Use from January 2011


New Scalable OpenModelica Parallel Code New Scalable OpenModelica Parallel Code GeneratorGenerator

• Work in Progress• Both task parallelism and data parallelism• Handling non-expanded arrays efficiently• Planned use of TLM-partitioning for more parallelism

• Generating OpenCL code for platform independence• Template based code generator


The The OpenModelicaOpenModelica CompilerCompilerImplemented mainly in MetaModelica and C/C++; 114 packagesImplemented mainly in MetaModelica and C/C ; 114 packages

Mod Connect

Types.Mod SCode.Mod Exp.Componentre f DAE

Main BackendDumpDAE

Flat ModelicaPrefix.Pre fix

Lookup

(Env, name)SCode.Class Derive VarTransform

PrefixBackendDAEEXT

Exp.IdentPrefix.Prefix

Types.Mod SCode.Mod p p

ClassLoader

SCodeUtilParser BackendInst

Absyn SCodeDAE: Equations

Algorithms

( , )

DAE F i

DAE, substlist

DAEExp.Exp Exp.Exp

.mo

ClassLoader

SCode.Exp(Exp.Exp,

Type s.Type) TemplateCodeGen

ClassInf

DAE: FunctionsSimCode

ClassInf.Event ClassInf.State

Values.Value

Exp.Exp

Ceval

Static

Exp.Exp

(Env, name)

Absyn

Data Type Modules:

SCode Utility

Modules:

C code

Patternm

Ceval

Types

Algorithm

DAE

Expression

Dump Debug

ModUtilBuiltin

Util MetaUtil

DFA

36 © Peter Fritzson 36

go t p ess o

SystemPrint RTOpts

u t

Summary of MODPROD Research in PELABSummary of MODPROD Research in PELAByy

• Modeling language design (semantics, type t t d li t ibilit )systems, meta-modeling, extensibility)

• Modelica-SysML integration• Requirements traceability• Compilation to multi-core platformsCompilation to multi core platforms• Compilation and performance measurements for

real time simulationreal-time simulation

Contact: [email protected]


research in modelresearch in model--based product ... · research in modelresearch in model--based...

Documents