High-confidence Software for Cyber Physical Systems

Drexel University Philadephia, PA

Vanderbilt University Nashville, Tennessee

Aniruddha Gokhale*, Sherif Abdelwahed

{a.gokhale,s.abdelwahed}@vanderbilt.edu

www.dre.vanderbilt.edu/~gokhalewww.isis.vanderbilt.edu/~sherif

*Proposed research ideas are based partly on prior work done for the DARPA PCES and ARMS programs.

Nagarajan Kandasamykandasamy@cbis.ece.drexel.

eduwww.ece.drexel.edu/

~kandasamy

2

• Network-centric, dynamic, large-scale “systems of systems”

• Service-oriented architecture of distributed collaborating services

• Stringent simultaneous QoS demands, e.g., “never die,” time-critical, secure.

• Highly diverse, complex, integrated & autonomous application domains

• On demand computing needs

Traits of Cyber Physical Systems

Key Requirements for High Confidence Software

• Trustworthiness - delivering multiple, simultaneous QoS

• Autonomicity – self healing, self configuring, self optimizing

• Analyzability – amenable to validation and verification

3

Step 1. Algorithms for Distributed Control & Diagnosis

• System management tasks are posed as control/optimization problems and solved under dynamic and uncertain operating conditions

• Online parameter tuning and model-learning techniques can be integrated within the control framework to improve the quality of partially specified system models as well as adapt to changes in the system model itself over time

• Diagnosis algorithms will detect, isolate, and estimate the state of corrupted hardware and software components using concepts from continuous and discrete-event diagnosis, and consistency-based causality analysis.

Enterprise computing system

PerformanceOptimizer

System model (M)

Learning structure

Environment Inputs (i)

Estimators

Estimated inputs

System response (r)

System state (x)

Control decisions (d)

Control inputs

State feedback

rdi

ix

r’System Model (M’)

Faultdetection/recovery

Recovery/reconfiguration actions

Model-based control

Model-based diagnosis

Enterprise computing system

PerformanceOptimizer

System model (M)

Learning structure

Environment Inputs (i)

Estimators

Estimated inputs

System response (r)

System state (x)

Control decisions (d)

Control inputs

State feedback

rdi

ix

r’System Model (M’)

Faultdetection/recovery

Recovery/reconfiguration actions

Model-based control

Model-based diagnosis

Focus is on developing algorithms to realize incorruptible and self-healing CPSs via a combination of control and diagnostics

4

Step 2. MDE Tool Chainwww.dre.vanderbilt.edu/cosmicwww.dre.vanderbilt.edu/CIAO

Modeling toolsModeling tools

Model Model InformationInformation

Domain, Deployment, SRG, FOU,

Connection QoS, Security

injectionReplica Placement,

Bandwidth allocation, Security

model GeneratorsGenerators

Augmented Augmented Deployment Deployment

PlanPlan

Middleware Bus

Container

…

SecurityReplication TransactionPersistence

Container

… …• Capture trustworthiness dimensions

(e.g.,RT, FT and Security) via DSMLs• Generative programming approach that

uses QoS specs, control algorithms and middleware features to synthesize CPS artifacts

Focus is on resolving accidental complexities and automating system configuration, deployment, adaptation and conducting analyses.

5

Step 3. Trustworthy Middleware Framework• Decouple system adaptation policy from system application code & allow

them to be changed independently from each other• Decouple system deployment framework & middleware from core system

infrastructure to allow CPSs to be dynamically reconfigurable

System ObserversSystem ObserversSystem Condition Observers

System Deployment AgentsSystem D&C Actors

& Middleware

Adaptation Planner

ControlAlgorithmControlAlgorithm

AdaptationPlan

SystemConditions

Running Systems

Control and diagnosticsSelf healing

Self configuring & optimizing

Reflective capabilities

Focus is on realizing a scalable, trustworthy runtime environment.

6

Step 4. System Execution Modeling Tools

“What if” analysis

Validate design conformance

Validate design rules

Focus is on continuous QoS integration and validation via design-time analysis and automated empirical testing/validation

www.dre.vanderbilt.edu/cosmicwww.dre.vanderbilt.edu/CUTS