composition at scale - chess · sensor fus target rec integrated embedded common services...

Institute for Software Integrated SystemsVanderbilt University

Composition at Scale

Janos SztipanovitsJanos Sztipanovits18 August, 2010

Software at ScaleUC BerkeleyBerkeley, CAy,

Outline

Integration Challenge in Software Intensive Systems

Pursuing Compositionality Model-Based System Integration Approach Model-Based Integration Process Software/Systems “Wind Tunnel” Challenges Challenges

Dimensions of System Integration

Components

L Layers

System of Systems System of Systems

Component Integration

Functional: E.g.: Dynamics• Composability and

DigitalController

D/AS/H

PowerAmp.

PlantandSensors

Composability and compositionality are keyconcepts

A/D • Defined for carefully selectedproperties (stability, latency,power,..)Component Integration Platform (e.g. SL/SF)

Software: E.g. Timing• Decomposed into structure, interaction and behavior

p g ( g )

Comp-1 Comp-2 Comp-3 • Challenges: – composition frameworks

providing constructivityComponent Integration Platform (e.g. TTP)

providing constructivityfor essential properties

Layered Systems: Vertical Integration

Command andCommand andControlControl FormationFormation SensorSensor

ProcessingProcessing

Human OrganizationHuman Organization• Inter-layer interactionsCognitive processes

Social interaction C d d t l

Roles Layers Characteristics

System Operation LayerSystem Operation Layer

• Effects propagateacross the layers

• Efficiency and optimization

Command and control

CoordinationData distribution

SW/Component LayerSW/Component Layer

• Efficiency and optimizationdrives toward intractability

• Inter-layer relationship:

Component interactionsComponent behaviorsArchitecture

Resource management

ACCU MUXEI (8-6)

I (2-0) D(3-0) I (8-6)CK

S(3-0)

R(3-0)

I (8-0): Instruction CodeD(3-0): Input DataCK: ClockY(3-0): Output

OS/Network Layer OS/Network Layer - mapping- refinement- synthesis

Resource managementSchedulingSeparation

Timing/performanceF lt t

ALU

MUXOUT

RAMI (5-3)

CKI (8-7)

Y(3-0)

R(3 0)

ALU_OUT(3-0)Rb(3-0)

Ra(3-0)

HW/Systems LayerHW/Systems Layer • Challenges: – modeling, – constraining

i

Fault managementPower management

Heat dissipationCrossover

Materials & Devices Materials & Devices – composingRadiation effects

System of System Integration

• Heterogeneous Future Combat System

Distributed DatabaseInformation LayerI bl

Standards-BasedOpen SoftwareArchitectureCommon Operating

PictureJoint CommonDatabase

Information ManagementInformation ManagementVehicle Applications Mission Applications Business Applications Administration Applications

Human Machine Interface /Machine-Machine Interface

& P

rep

ndin

g

men

t

n n Tr

aini

ng

nt ent

t on oot

Vehicle Applications Mission Applications Business Applications Administration Applications


& P

rep

ndin

g

men

t

n n Tr

aini

ng

nt ent

t on oot

• Open Dynamic Architecture - heterogeneous

t ki

COP

Distributed DatabaseInformation LayerInteroperableexport

HQESO

nnin

g &

Prep

nder

stan

ding

t & E

xecu

tion

ion

ogni

tion

Sust

ainm

ent

Trai

ning

Warfighter Interface

DB Synchronization

InteroperabilityInteroperabilityFIOP

Foundation Infrastructure –(e.g, Network with: COMSEC Crypto Services, Mobility Enhancements, IP Network Appliqué's, )

Operating System

Operating System Abstraction Services

Network InfrastructureServices

SOS Framework ServicesCOTSNDI

SOS Operations ServicesInformation Assurance (IA) Network Mgt (NM) Information Dissemination Mgt (IDM)

Application Program Interfaces –Common Services

COTSNDI

Mis

sion

Pla

nnin

g &

Situ

atio

n U

nder

sta

Bat

tle C

omm

and

Inte

grat

ed S

usta

inm

Targ

et R

ecog

nitio

nS

enso

r Fu

sion

Em

bedd

ed M

issi

on

Nav

igat

ion

Con

trol

s

Pro

puls

ion

Hyd

raul

ic

Ele

ctric

alFu

el S

ys

Hea

lth M

anag

emen

Eng

inee

ring

Pro

cure

men

tFa

cilit

ies

Logi

stic

sP

erso

nnel

Tran

spor

tatio

n

Dis

posa

l

Sys

tem

Man

agem

eR

emot

e Se

rver

Mgt

Sof

twar

e D

istr

ibut

i

Use

r Man

agem

ent

Sof

twar

e U

pgra

deS

oftw

are

Inst

all

Rem

ote

Trou

bles

ho


Operating System







COTSNDI

Mis

sion

Pla

nnin

g &

Situ

atio

n U

nder

sta

Bat

tle C

omm

and

Inte

grat

ed S

usta

inm

Targ

et R

ecog

nitio

nS

enso

r Fu

sion

Em

bedd

ed M

issi

on

Nav

igat

ion

Con

trol

s

Pro

puls

ion

Hyd

raul

ic

Ele

ctric

alFu

el S

ys

Hea

lth M

anag

emen

Eng

inee

ring

Pro

cure

men

tFa

cilit

ies

Logi

stic

sP

erso

nnel

Tran

spor

tatio

n

Dis

posa

l

Sys

tem

Man

agem

eR

emot

e Se

rver

Mgt

Sof

twar

e D

istr

ibut

i

Use

r Man

agem

ent

Sof

twar

e U

pgra

deS

oftw

are

Inst

all

Rem

ote

Trou

bles

ho

networking - heterogeneouscomponents

UE/HQESO

EPLRSSINCGARSVHF

Link 4ALink 11Link 16WIN T

Mis

sion

Pla

Situ

atio

n U

n

Battl

e M

gmt

Sens

or F

us

Targ

et R

eco

Inte

grat

ed S

Embe

dded

T

Common Services

Information Management

Computing and Networking

HQ

BattleCommand

Reachback

HHQ XX

InteroperabilityInteroperability

• Very high level concurrency with complex interactions

WIN-T Hierarchical Ad-Hoc Network

DataImagesVoice

UGS

WNW WNW

stubnetJTRS

• Challenges: – understanding and– predicting behaviorL COP L COP L COP L COP

Video Vetronics

Common VehicleSubsystems

EO/IR EO/IRSAR/MTI

PlatformPlatformNetworked CommandNetworked Command

Outline



Integration Inside AbstractionLayers: Composition

Plant DynamicsModels

Controller Models

Dynamics: • Properties: stability, safety, performance• Abstractions: continuous time, functions,

signals, flows,…Ph i l d i

1( ) ( ( ), ..., ( ))p jB t B t B t

s g a s, o s,Physical design

SoftwareArchitecture

Models

Software Component

CodeS ft d i

Software : • Properties: deadlock, invariants,

security,…• Abstractions: logical-time concurrency

1( ) ( ( ), ..., ( ))c kB i B i B i

Software design Abstractions: logical time, concurrency, atomicity, ideal communication,..

System Architecture

Models

ResourceManagement

Models

Systems : • Properties: timing, power, security, fault

toleranceAb t ti di t ti d l

1( ) ( ( ), ..., ( ))j p i k iB t B t B t

Sztipanovits: 8

System/Platform Design • Abstractions: discrete-time, delays, resources, scheduling,

Integration Across Abstraction Layers: Much Unsolved Problems

Plant DynamicsModels

Controller Models

Ph i l d i

Controller dynamics is developedwithout considering implementation uncertainties (e.g. word length, clock accuracy ) optimizing performancePhysical design accuracy ) optimizing performance.

Assumption: Effects of digital implementationcan be neglectedX

SoftwareArchitecture

Models

Software Component

Code

Software architecture models are developed without explicitly consideringsystems platform characteristics, eventhough key behavioral properties

Software design depend on it.

Assumption: Effects of platform properties can be neglectedX

System Architecture

Models

ResourceManagement

Models

System-level architecture defines implementation platform configuration. Scheduling, network uncertainties, etc. are introduce time variant delays that may

X

Sztipanovits: 9

System/Platform Design introduce time variant delays that may require re-verification of key properties on all levels.

Challenge to Compositionality: Heterogeneity

• Consequence of the lack of composability across system layers– intractable interactions– unpredictable system level behavior– full-system verification does not scale

• Active research: simplification strategies– Decoupling: Use design concepts that

decouple systems layers for selectedproperties p p

– Cross-layer Abstractions: Develop methods that can handle effects of cross-layer i t ti

Sztipanovits: 10

interactions

Decoupling Example 1: Robust Implementation of R-T Systems

Abdellatif, Combaz, Sifakis [2010]:Model Based Implementation ofModel-Based Implementation of Real-Time Applications

Abstract Model • : Based on Logical Execution Time (LET)

Based on Timed Automata Actions are atomic and timeless

M

implementationtime safety:

They can be executed after release time andbefore the due time

• : : real-time system Models the behaviorM

Real-Time Model

• : : real-time system. Models the behaviorof the software on a platform. Actions are assigned with WCET

M

Sztipanovits: 11

time robustness

Decoupling Example 1: Robust Implementation of R-T Systems

Abdellatif, Combaz, Sifakis [2010]:Model Based Implementation of

However, essential system properties such as stability safetyModel-Based Implementation of

Real-Time Applicationsproperties such as stability, safety, performance are expressed interms of physical behavior

Abstract Model • : Based on Logical Execution Time (LET)

Based on Timed Automata Actions are atomic and timeless

M

implementationtime safety:

They can be executed after release time andbefore the due time

• : : real-time system Models the behaviorM

Real-Time Model

• : : real-time system. Models the behaviorof the software on a platform. Actions

are assigned with WCET

M

Sztipanovits: 12

time robustness

Decoupling Example 2: Passive Dynamics

Physical models

Goals:

• Compositional verification ofpessential dynamic properties

− stability− safetyimplementation

Abstract Model

• Passivity guarantees stabilityindependently from implementation induced

t i ti

implementation

uncertainties− time varying delays− network uncertainties

(packet drops delays)

time safety:

Real-time Model

(packet drops, delays)• Decreased verification complexity

Kottenstette N J Hall X Koutsoukos P J

Sztipanovits: 13

time robustness

Kottenstette, N., J. Hall, X. Koutsoukos, P. J. Antsaklis, and J. Sztipanovits, "Digital Control of Multiple Discrete Passive Plants Over Networks", Int. J. of Systems, Control and Communications 2010

Illustration of Passive Dynamics

Experimental Setup Joint Angle and Reference

• Two CrustCrawler robotic arms Time delay (Robot 2 and PJ)arms– 4 DOF with AX-12 smart

servos at each joint• Novint haptic paddle

Sztipanovits: 14

Novint haptic paddle• Five networked Windows

PCs with Matlab/Simulink

Open Research Problems

• Extend theory for decoupling

• Develop theory of compositionality among system layers• Develop theory of compositionality among system layers(vertical composition)

E t d iti lit f lti l ti• Extend compositionality for multiple properties, e.g. stability, safety and invariants

• Exploit compositionality in software synthesis

Sztipanovits: 15

Outline

• Integration Challenge in Software Intensive Systems

• Pursuing Compositionality• Model-Based System Integration

– Approach– Model-Based Integration Process– Software/Systems “Wind Tunnel”

Challenges– Challenges

Sztipanovits: 16

System of System Integration

Future Combat System

Distributed DatabaseInformation LayerInteroperable

Standards-BasedOpen SoftwareArchitectureCommon Operating

PictureJoint CommonDatabase

Information ManagementInformation ManagementVehicle Applications Mission Applications Business Applications Administration Applications


lann

ing

& P

rep

Und

erst

andi

ngm

man

d

d Su

stai

nmen

t co

gniti

on

usio

n

d M

issi

on T

rain

ing

n n nage

men

t

ng ent

atio

n

anag

emen

ter

ver M

gtD

istr

ibut

ion

agem

ent

Upg

rade

nsta

ll

roub

lesh

oot

Vehicle Applications Mission Applications Business Applications Administration Applications


lann

ing

& P

rep

Und

erst

andi

ngm

man

d

d Su

stai

nmen

t co

gniti

on

usio

n

d M

issi

on T

rain

ing

n n nage

men

t

ng ent

atio

n

anag

emen

ter

ver M

gtD

istr

ibut

ion

agem

ent

Upg

rade

nsta

ll

roub

lesh

oot

• Heterogeneous

COP

Interoperableexport

HQESO

lann

ing

& Pr

ep

Und

erst

andi

ng

mt &

Exe

cutio

n

usio

n

cogn

ition

d Su

stai

nmen

t

d Tr

aini

ng

Warfighter Interface

DB Synchronization

InteroperabilityInteroperabilityFIOP


Operating System






COTSNDI

Mis

sion

PS

ituat

ion

UB

attle

Com

Inte

grat

edTa

rget

Rec

Sen

sor

Fu

Em

bedd

ed

Nav

igat

ion

Con

trol

s

Pro

puls

ion

Hyd

raul

icE

lect

rical

Fuel

Sys

Hea

lth M

a

Eng

inee

rinP

rocu

rem

eFa

cilit

ies

Logi

stic

sP

erso

nnel

Tran

spor

ta

Dis

posa

l

Sys

tem

Ma

Rem

ote

SeS

oftw

are

D

Use

r Man

aS

oftw

are

US

oftw

are

I

Rem

ote

Tr


Operating System







COTSNDI

Mis

sion

PS

ituat

ion

UB

attle

Com

Inte

grat

edTa

rget

Rec

Sen

sor

Fu

Em

bedd

ed

Nav

igat

ion

Con

trol

s

Pro

puls

ion

Hyd

raul

icE

lect

rical

Fuel

Sys

Hea

lth M

a

Eng

inee

rinP

rocu

rem

eFa

cilit

ies

Logi

stic

sP

erso

nnel

Tran

spor

ta

Dis

posa

l

Sys

tem

Ma

Rem

ote

SeS

oftw

are

D

Use

r Man

aS

oftw

are

US

oftw

are

I

Rem

ote

Tr

• Open Dynamic Architecture - heterogeneousnetworking

UE/HQESO

EPLRSSINCGARSVHF

Link 4ALink 11Link 16WIN T

Mis

sion

Pl

Situ

atio

n U

Battl

e M

gm

Sens

or F

u

Targ

et R

e

Inte

grat

ed

Embe

dded

Common Services

Information Management

Computing and Networking

HQ

BattleCommand

Reachback

HHQ XX

networking - heterogeneouscomponents

WIN-T Hierarchical Ad-Hoc Network

DataImagesVoiceVideo V t i

UGS

WNW WNW

stubnetJTRS

• Very high level concurrency with complex interactions

L COP L COP L COP L COP

Video Vetronics

Common VehicleSubsystems

EO/IR EO/IRSAR/MTI

Networked CommandNetworked Command

• Challenges: – understanding and– predicting behavior

Sztipanovits: 17

How to achieve predictability with limited/partial composability?

PlatformPlatformNetworked CommandNetworked Command

Real-Life SoS Development

• All integration categories are present (component, layer, SoS)SoS)

• Systems are evolving along “spiral-outs”• New technical challenges are emerging and potentialNew technical challenges are emerging and potential

solutions need to be rapidly explored• All layers of the system are subject to modifications,

there are no well defined synchronization points in the development process

• Integration is inherently incremental; deployedIntegration is inherently incremental; deployed systems need to be integrated with components on different level of maturity: prototypical and with i l t d t / t

Sztipanovits: 18

simulated systems/components.

How Is It Solved Today?

• Systems are integrated when all components are delivered

– Acquisition pushes in this direction

• Integration means: “Make it working somehow” • System Integration Labs do not offer support for

spiral developmentTh i h t d l ith i l t• There is no approach to deal with incomplete specifications and components

System Integration is the highest risk most

Sztipanovits: 19

System Integration is the highest risk, most expensive, least predictable step in SoS development

Emerging Solution: Model-Based Integration

Apply Models Earlypp y y Apply Models Often Use Every OpportunityUse Every Opportunity Requirements/Architecture Integration Architecture/Design Integration Design Assessment/Verification Prototyping/Scaling Implementation Scaling

Testing Testing

Tool Chain for Architecture Exploration in FCS

ADeVS, IONS RELEX

Excel

Component

SystemModelSegment

IDD

GReATTransform

SystemIntegration

SystemComponentAdapters

C++______________________________

IDL______________________________

SystemIntegrationTest Harness

C++______________________________

IDL________________________

GReATTransform

XML______

RoseCAT file

RuntimeGlue(Deployed)

______

IntegrationLaboratoryTools

________________________

Risk Mitigation: SurrogateModeling and Synthesis

GME System Models

GME Component ModelsDeployment Instance Topology Networks

GME Component Models

Interfaces,Business

Interfaces

Code GeneratorCode Generator

BusinessLogic

InputInterfaces

OutputInterfaces

BusinessLogic(Generated)

InputInterfaces

OutputInterfaces

AcquiredBusinessLogic

“Real”BC

“Real”BC

BC Surrogate Component

(Generated)

System Of Systems Common Operating Environment

BC Surrogate Component

Logic ComponentComponent

Outline

• Integration Challenge in Software Intensive Systems

• Pursuing Compositionality• Model-Based System Integration

– Approach– Model-Based Integration Process– Software/Systems “Wind Tunnel”

Challenges– Challenges

Sztipanovits: 23

Heterogeneous Simulation Integration

Organization/Coordination Controller/Vehicle Dynamics Processing (Tracking) 3-D Environment (Sensors)

CPN Devs Delta3DSL/SFCPNAdaptive Human

Organization

MixedInitiative

Controller

Context Dep.Command

Interpretation

AdaptiveResourceAllocation

Coordination Decision Support

HCI AbstractCommands

PlatformCommands

AssignedPlatform

Commands

CPN Devs Delta3DSL/SF

How can we integrate the models?How can we integrate the simulated heterogeneous system components?H i t t th i l ti i ?

CPN

Data Distribution Network

Support

PlatformStatus

COPElements

COPElements

COPElements

Model-Integrated System and Software Laboratory Environment: C2 Windtunnel

How can we integrate the simulation engines?

GME GMESimulation Interaction Simulation Architecture

OMNETNetwork Architecture

Model-based Integration Architecture

“Virtual”

Model Integration LayerComponents

ExperimentSpecification

& ConfigurationController

ModelsNetworkModels

Org.Models

FusionModels

Models

Env.Models

Si l ti I t ti Pl tf (HLA)

Instrumentation Layer

DEVSFederate.

OmNet++Federate

CPNFederate.

OGREFederate

SimulinkFederate

Instrumentation Layer

Run-time

Simulation Data Distribution/Communication Middleware

Simulation Integration Platform (HLA)

Distributed Simulation PlatformDistributed Simulation Platform

https://wiki.isis.vanderbilt.edu/OpenC2WT

Experiments: Impact of Cyber Attacks

Network attack: A sub-network with hundreds of zombie nodes attacks a critical

router on the main network. Flood attack on udp, tcp or ping p, p p g

Full Zombie subnet

network

Outline



Open Research Problems: Modeling at Scale

Model Versioning Granularity Semantic conflict detection with DSL extensibility

C di ti d l i i Coordinating model versioning with evolving modeling languages, tools, and tool

integration frameworksintegration frameworks with other development artifacts

Collaborative distributed modelingCollaborative distributed modeling Conflict visualization and collaborative resolution Merge consistency guarantees

Summary

• System Integration is a grand challengef i i l l tof engineering large-scale systems

• Composition/compositionality have practical and theoretical limits in complexpractical and theoretical limits in complex heterogeneous systems

• Model-based methods provide promising• Model-based methods provide promising solutions for hard problems

Sztipanovits: 29

composition at scale - chess · sensor fus target rec integrated embedded common services...

Documents