Q uantitative E valuation of E mbedded S ystems

Quantitative Evaluation of Embedded Systems• Mutual introductions• The context of the course:

Model Based / Driven Design• Organisation of the course

Introducing the lecturers

Anne Remke (UT)

Pieter Cuijpers (TU/e)

Marielle Stoelinga (UT)

Marco Zuniga (TUD)

Why a tele-lecture ?

• Link between education and research• 3TU cooperation :

Specialization in research vs Broad engineering education• Efficiency

Why a class-room ?

• More time for questions & (tele)-communication• Rewind button• Better insight in your progress• More convenient homework

flipped

Last years evaluation (warning)

• Bad tele-connections• Three (too) different topics• Too many notational conventions• Too abstract for hands-on

embedded systems enthousiasts• Too much mandatory homework

Who are you?

BSc Electrical Eng BSc Computer Science Other

TU/e

TUD

UT

Who are you?

Logic & Set-theory

Petri-nets

Finite Autom.

Linear algebra

Prob.th. Model checking

TU/e

TUD

UT

Model-based Designtiming energy

memorycost

worst-case

best-caseaver

age-

case

throughput

latency

time-outs

package loss

up-timechance of failure

deadline miss

overflow

bandwidth

battery drain

robustness

measurements

THE COST OF FIXING SOFTWARE BUGS (BOEHM)

Specification

Design

Implementation

Deployment & Maintenance

The Engineering Design Cycle

Specification

Design

Implementation

Deployment & Maintenance

Model Based Design

Model Checking

Specification

Design

Implementation

Deployment & Maintenance

Model Driven Design

State space explorationProgramming paradigms

Code Generation

Next Generation Computing Trends: Complex Highly networked Failures = fact of life

Quality = Quantity Deadlines Power usage Fault tolerance Performance

Needed: Systematic Quant.

Analysis at Design-time Multi-disc. approach QEES!

Model CheckingDynamic Behavior

Properties

Continuous

TimedDiscrete

State based

CTL*

ParameterizedDataProbabilistic

Event based

Qualitative(Logical) Quantitative

(Numerical)

LTL modal µ-calculus

pCTLtCTL

linear

convex

monotone

Automata

Petri-nets

Differentialequations

Max-plus algebra

Contents of the course

• 3 Typical quantitative formalisms: Dataflow, Timed Automata, Markov Chains

• 1 Quantitative analysis method for Dataflow• 3 Model-checking methods for TA and MC• 3 Tools: SDF3, UPPAAL, PRISM• 1 Case study

Case: Cyber Physical Systems

Computation

Physical World

Sensing ActingCyber

PhysicalControl

Communication network

Case: Cyber Physical Systems

Comp.Inner control

Physical World

Sensor 1Temperature

Actor 1Valve

Actor 2Motor xyzComp.

Emergency detection

Comp.Image processing

Sensor 2Pressure

Sensor 4Microphone

Actor 3Motor rot.

Sensor 3Camera

Determine an appropriate communication schedule that guarantees given latency and

throughput constraints for this control network and predict the

associated network load.

General planning of QEES

• Dataflow - Timed Automata - Probabilistic Automata• Tele-lectures & flipped classroom• Watch videos at home…

…make exercises in class• Some additional material in class• One mandatory assignment (pass/fail)

(One case-study document – to be updated 3 times)

• One exam

Program for DataflowDate Weblecture - at home Additional – in class Exercises – in class

11-11-’13 1 – intro dataflow(This one we’ll watch in class)

Intro to QEESCounters and daters

Simulate a given graph and set up matrix equations for it.

15-11-’13 2 – throughput3 – periodic schedule(watch these at home, in the train, wherever, but not in class!)

Eigenvalues and linear programs

Determine the MCM and periodic schedule for a graph.

18-11-’13 4 – latency 15 – latency 2

Monotonicity Determine the latency of a graph.

22-11-’13 6 – buffering7 – latency 3

TDMA + intro assignment, multi-rate, intro to SDF3

Determine minimum buffersizes of a graph.

Deadline: 9-DEC-2013 : As a first step in the case study, you will model a small cyber-physical control network in SDF3, in which TDMA communication using wirelessHART is used. You will analyse worst-case latency and throughput that is achieved, and add buffers to determine the network load.

Program for Timed AutomataDate Weblecture - at home Additional – in class Exercises – in class

25-11-’13 Intro Timed Automata Intro Timed Automata Modeling and analysis of a small resource scheduling problem.

29-11-’13 ES-Day in Delft : GUEST LECTURE Arjen Mooij : Model Based Design

2-12-’13 UPPAAL under the hood UPPAAL under the hood Practice semantics and composition

6-12-’13 UPPAAL under the hood Practice R.A.

9-12-’13 Paper – “Scheduling of data paths in printers”

Discussion paperIntro part 2 of the assignment

Wrap up of exercisesQ&A

Deadline: 6-JAN-2013 : As a second step in the case study, you will use UPPAAL to make an optimal TDMA schedule for the cyber-physical system and see how latency and throughput are improved. These results are then fed back into the dataflow model.

Program for Markov ChainsDate Weblecture - at home Additional – in class Exercises – in class

13-12-’13 CTL (This one we’ll watch in class) Markov Chains CTL

16-12-’13 CTL model checking Discr. Timed Markov Chains CTL model checking

20-12-’13 PCTL model checking DTMC & PCTL model checking

6-1-’14 Paper: “Wireless HART” Perf. eval. of Wireless HARTCont. Timed Markov Chains

10-1-’14 CSL model checking CSL model checking

13-1-’14 Q&A

17-1-’14 Q&ADeadline: 17-JAN-2013 : As a final step in the case study, you will analyze the effect of message loss on the optimal TDMA schedule you found previously. Furthermore, you will discuss in a concluding chapter the different roles of the three formalisms studied in this course in the engineering design-cycle.