Measurement-Based Worst-Case Execution Time (WCET) Analysis Guillem Bernat and Adam Betts

advertisement

Measurement-Based Worst-Case

Execution Time (WCET) Analysis

Guillem Bernat and Adam Betts

Measurement-Based

Approach the Holy Grail of

WCET Analysis

The real-time constraint

• Computer market dominated by embedded systems

• Majority of embedded systems operate in real-time

• Functional and Temporal correctness required

• Failure has varying degrees of consequences

Why real-time people are useful

• In-flight control system – Death

• Video streaming – Degraded picture quality

• Automotive – Breakdown

• Toaster – Burnt toast

• Mobile phone – Angry girlfriend

Circumventing Catastrophe

How to calculate execution time

Real-Time System broken into number of particular hardware?

• Tasks are scheduled according to their temporal parameters

20 24 37 10

Basics of WCET

• Safe estimates desired

• Static Analysis uses mathematical models of both program and processor

• These data are combined using calculation methods

• Formal proofs ascertain correctness

instruction1 instruction2 instruction3 instruction4 instruction5 instruction6 instruction7 instruction8 instruction9 instruction10

10

1

2

7

18

3

4

10

4

6

WCET = 65

instruction1 instruction2 instruction3 instruction4 instruction5 instruction6 instruction7 instruction8 instruction9 instruction10 instruction11 instruction12 instruction13 instruction14 instruction15 instruction1 instruction2 instruction3

26 instruction4 instruction5 instruction6

34 22 instruction7 instruction8 instruction9 instruction10 instruction11 instruction12 instruction13

56 instruction14 instruction15

10

Where Static Analysis Struggles

• Processor advancements aim to improve average-case performance

• Predictability compromised

• Features disabled or pessimistic assumptions made

• Integration of advanced features into embedded domain gathering speed

instruction1 instruction2 instruction3 instruction4 instruction5

1 2

1

3 4 5

2

1

3 4 5

2 3 4 5

1 2

1

3 4 5

2 3 4 5

It gets worse…

• Interaction between all features yet to be fully explored

• Unrealistic assumptions about processor behaviour

• Intellectual property constrains the scope of static analysis

Measure, you say?

• Scepticism toward measurementbased approach

• Pure end-to-end measurement might not encapsulate the worst-case

• Instrumentation required to collect data

• Probe effect disrupts the temporal behaviour

What we propose

• Combination of static analysis and testing techniques

• Testing measures the real behaviour of the processor

• Static analysis techniques dictate how these data are combined to produce a final estimate

Instr1

Instr4

Instr5

Instr6

Instr7

Instr2

Instr3

Instr8

Instr9

Instr10

Instr11

Instr12

Instr13

Issues in Measurement-Based

Analysis

• Well instrumented programs

• Instrumentation under constraints

• Generating test-data for temporal correctness

• Adequacy criteria for WCET coverage

• Probability that estimates are correct

Conclusion

• Timeliness of many embedded systems relies on WCET analysis

• Static analysis struggles with advanced processor features

• Measurement-Based approach aims to combine static and testing techniques

• Accuracy the key as wider-issue processors become more prevalent

Download