1. An Experimental Evaluation on Reliability Features of N-Version Programming Xia Cai, Michael R. Lyu and Mladen A. Vouk ISSRE’2005

2. 2 Outline  Introduction  Motivation  Experimental evaluation Fault analysis Failure probability Fault density Reliability improvement  Discussions  Conclusion and future work

3. 3 Introduction  N-version programming is one of the main techniques for software fault tolerance  It has been adopted in some mission-critical applications  Yet, its effectiveness is still an open question What is reliability enhancement? Is the fault correlation between multiple versions a big issue that affects the final reliability?

4. 4 Introduction (cont’)  Empirical and theoretical investigations have been conducted based on experiments, modeling, and evaluations Avizienis and Chen (1977), Knight and Leveson (1986), Kelly and Avizienis (1983), Avizienis, Lyu and Schuetz (1988), Eckhardt et al (1991), Lyu and He (1993) Eckhardt and Lee (1985), Littlewood and Miller (1989), Popov et al. (2003) Belli and Jedrzejowicz (1990), Littlewood. et al (2001), Teng and Pham (2002)  No conclusive estimation can be made because of the size, population, complexity and comparability of these experiments

5. 5 Research questions  What is the reliability improvement of NVP?  Is fault correlation a big issue that will affect the final reliability?  What kind of empirical data can be comparable with previous investigations?

6. 6 Motivation  To address the reliability and fault correlation issues in NVP  To conduct a comparable experiment with previous empirical studies  To investigate the “variant” and “invariant” features in NVP

7. 7 Experimental background  Some features about the experiment Complexity Large population Well-defined Statistical failure and fault records  Previous empirical studies UCLA Six-Language project NASA 4-University project Knight and Leveson’s experiment Lyu-He study

8. 8 Experimental setup  RSDIMU avionics application  34 program versions  A team of 4 students  Comprehensive testing exercised Acceptance testing: 800 functional test cases and 400 random test cases Operational testing: 100,000 random test cases  Failures and faults collected and studied  Qualitative as well as quantitative comparisons with NASA 4-University project performed

9. 9 Experimental description  Geometry  Data flow diagram

10. 10 Comparisons between the two projects  Qualitative comparisons General features Fault analysis in development phase & operational test  Quantitative comparisons Failure probability Fault density Reliability improvement

11. 11 General features comparison

12. 12 Faults in development phase

13. 13 Distribution of related faults

14. 14 Fault analysis in development phase  Common related faults  Display module (easiest part)  Calculation in wrong frame of reference  Initialization problems  Missing certain scaling computation  Faults in NASA project only  Division by zero  Incorrect conversion factor  wrong coordinate system problem.

15. 15 Fault analysis in development phase (cont’)  Both cause and effect of some related faults remain the same  Related faults occurred in both easy and difficult subdomains  Some common problems, e.g., initialization problem, exist for different programming languages  The most fault-prone module is the easiest part of the application

16. 16 Faults in operational test

17. 17 Faults in operational test (cont’)  These faults are all related to the same module, i.e., sensor failure detection and isolation problem  Fault pair (34.2 & 22.1) : 25 coincidence failures  Fault pair (34.3 & 29.1) : 32 coincidence failures  Yet these two pairs are quite different in nature  Version 34 shows the lowest quality Poor program logic and design organization Hard coding  The overall performance of NVP derived from our data would be better if the data from version 34 are ignored

18. 18 Input/Output domain classification  Normal operations are classified as: Si,j = {i sensors previously failed and j of the remaining sensors fail | i = 0, 1, 2; j = 0, 1 }  Exceptional operations: S others

19. 19 Failures in operational test  States S 0,0, S 1,0 and S 2,0 are more reliable than states S 0,1, S 1,1, S 2,1  Exceptional state reveals most of the failures  The failure probability in S 0,1 is the highest  The programs inherit high reliability on average

20. 20 Coincident failures  Two or more versions fail at the same test case, whether the outputs identical or not  The percentage of coincident failures versus total failures is low: Version 22: 25/618=4% Version 29: 32/2760=1.2% Version 32: (25+32)/1351=4.2%

21. 21 Fault density  Six faults identified in 4 out of 34 versions  The size of these versions varies from 1455 to 4512 source lines of code  Average fault density: one fault per 10,000 lines  It is close to industry-standard for high quality software systems

22. 22 Failure bounds for 2-version system  Lower and upper bounds for coincident failure probability under Popov et al model  DP1: normal test cases without sensor failures dominates all the testing cases  DP3: the test cases evenly distributed in all subdomains  DP2: between DP1 & DP3 Version pair DP1DP2DP3 Lower bound Upper bound Lower bound Upper bound Lower bound Upper bound (22,34)0.0000070.0001300.0003420.0067210.0003530.008396 (29,34)0.0000000.0000010.0000090.0001310.0000470.000654 Average in our project 1.25*10 -8 2.34*10 -7 6.26*10 -7 0.0000127.13*10 -7 0.000016 Average in NASA project 2.32*10 -7 0.0000070.0000230.0001030.0000720.000276

23. 23 Quantitative comparison in operational test  NASA 4-university project: 7 out of 20 versions passed the operational testing  Coincident failures were found among 2 to 8 versions  5 out of 7 faults were not observed in our project

24. 24 Observations  The different on fault number and fault density is not significant  In NASA project: The number of failures and coincident failures in NASA project is much higher Although there is coincident failures in 2- to 8-version combinations, the reliability improvement for 3-version system still achieves 80~330 times better  In our project: Average failure rate is 50 times better The reliability improvement for 3-version system is 30~60 times better

25. 25 Invariants  Reliable program versions with low failure probability  Similar number of faults and fault density  Distinguishable reliability improvement for NVP, with 10 2 to 10 4 times enhancement  Related faults observed in both difficult and easy parts of the application

26. 26 Variants  Compared with NASA project, our project: Some faults not observed Less failures less coincident failures Only 2-version coincident failures (other than 2- to 8- version failures) The overall reliability improvement is an order of magnitude larger

27. 27 Discussions  The improvement of our project may attributed to stable specification better programming training experience in NVP experiment cleaner development protocol different programming languages & platforms

28. 28 Discussions (cont’)  The hard-to-detected faults are only hit by some rare input domains  New testing strategy is needed to detect such faults: Code coverage? Domain analysis?

29. 29 Conclusion  An empirical investigation is performed to evaluate reliability features by a comprehensive comparisons on two NVP projects  NVP can provides distinguishable improvement for final reliability according to our empirical study  Small number of coincident failures provides a supportive evidence for NVP  Possible attributes that may affect the reliability improvement are discussed

30. 30 Future Work  Apply more intensive testing on both Pascal and C programs  Conduct cross-comparison on these program versions developed by different programming languages  Investigate the reliability enhancement of NVP based on the combined set of program versions

31. Thank you ! Q & A