1 evaluation of safety critical software david l. parnas, c acm, june 1990

16
1 Evaluation of Safety Critical Software David L. Parnas, C ACM, June 1990

Upload: melanie-baker

Post on 13-Dec-2015

216 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: 1 Evaluation of Safety Critical Software David L. Parnas, C ACM, June 1990

1

Evaluation of Safety Critical Software

David L. Parnas, C ACM, June 1990

Page 2: 1 Evaluation of Safety Critical Software David L. Parnas, C ACM, June 1990

Software Reliability

Nonetheless, our practical experience is that software appears to exhibit stochastic properties. It is quite useful to associate reliability figures such as MTBF (Mean Time Between Failures) with an operating system or other software product. Some software experts attribute the apparently random behavior to our ignorance. They believe that all software failures would be predictable if we fully understood the software, but our failure to understand our own creations justifies the treatment of software failures as random.

2841f07hamlet19oct30

Page 3: 1 Evaluation of Safety Critical Software David L. Parnas, C ACM, June 1990

MTTF

Mean is an average.

What is the relationship to Expected Value?

3841f07hamlet19oct30

Page 4: 1 Evaluation of Safety Critical Software David L. Parnas, C ACM, June 1990
Page 5: 1 Evaluation of Safety Critical Software David L. Parnas, C ACM, June 1990

5

Table 1

Table I shows that, if our design target

was to have the probability of failure be

less than 1 in 1000, performing between

4500 and 5000 tests (randomly chosen

from the appropriate test case distribution)

without failure would mean that the

probability of an unacceptable product

passing the test was less than 1 in a

hundred.

5841f07hamlet19oct30

Page 6: 1 Evaluation of Safety Critical Software David L. Parnas, C ACM, June 1990

6

Table II

6841f07hamlet19oct30

Page 7: 1 Evaluation of Safety Critical Software David L. Parnas, C ACM, June 1990

7

Practical ultra-reliability forabstract data types

Borislav Nikolik and Dick Hamlet

Softw. Test. Verif. Reliab. 2007; 17:183–203

7841f07hamlet19oct30

Page 8: 1 Evaluation of Safety Critical Software David L. Parnas, C ACM, June 1990

8

Term Redundancy Method (TRM)

8841f07hamlet19oct30

Page 9: 1 Evaluation of Safety Critical Software David L. Parnas, C ACM, June 1990

9

Boolean Stack ADT

9841f07hamlet19oct30

r1: pop(push(s, b))→sr2: top(push(s, b))→bFigure 1. Stack of boolean values TRS.

Page 10: 1 Evaluation of Safety Critical Software David L. Parnas, C ACM, June 1990

10

Rewriting

10841f07hamlet19oct30

pop(push(pop(push(x, y)), b)) = pop(push(x, y))

Page 11: 1 Evaluation of Safety Critical Software David L. Parnas, C ACM, June 1990

11

Post-release testing

11841f07hamlet19oct30

The relatively poor reliability estimate from the pre-release testing phase can now be used to obtain ultra-reliable term evaluations. In the self-checking phase the additional equivalent terms are drawnfrom the test-phase distribution for which the 10−4 bound was obtained. Therefore, if three randomly chosen terms agree, the probability that all of them are failures is less than (10−4)3 = 10−12.

Page 12: 1 Evaluation of Safety Critical Software David L. Parnas, C ACM, June 1990

12

Reliability

12841f07hamlet19oct30

Suppose a constant failure rate θ of π, and n random terms drawn from π, executed on δ without failure. The probability that δ fails on a randomly chosen term from π is θ, and 1 − θ that it will succeed. Given that the n terms are independent, the probability that δ succeeds on all the terms is (1 − θ)n. The confidence bound α on θ is defined as the probability that the failure rate of δ is below θ.The confidence bound is related to the testset size n and the failure rate θ by

α ≤ 1 − (1 − θ)n

Page 13: 1 Evaluation of Safety Critical Software David L. Parnas, C ACM, June 1990

13

Failure rate

13841f07hamlet19oct30

The confidence bound of Equation (1) is used to quantify the probability of failure of a majority values of a self-check. Equation (1) could be used to estimate the confidence bound on the failurerate of δ on a majority of N random terms generated by RBTR. Suppose a successful test (no failures occurred during the test) of δ on n terms is conducted at test time. Half or more of N terms (majority) falsely agreeing at run-time gives a failure rate of at least N/2n. Therefore, substituting N/2n for θ in Equation (1) yields

α ≤ 1 − 1 − N 2nn

Page 14: 1 Evaluation of Safety Critical Software David L. Parnas, C ACM, June 1990
Page 15: 1 Evaluation of Safety Critical Software David L. Parnas, C ACM, June 1990

15

Equation 2

15841f07hamlet19oct30

Equation (1) yieldsα ≤ 1 − 1 − N 2nn (2)

The meaning of the confidence bound is the probability that the failure rate is below N/2n for a repetition of the test. For example, 1 − α = 6.0 × 10−8 with N = 33 and n = 104.

Page 16: 1 Evaluation of Safety Critical Software David L. Parnas, C ACM, June 1990

16

How do we evaluate Hamlet?

16841f07hamlet19oct30