Conclusions
“Lofty Goal:” harden legacy and COTS software components
- For mission-critical systems Without extensive re-engineering to improve robustness
Robustness metric for Operating Systems
- Failure rates look high; true impact depends on operational profile
- Controversy as to whether Abort failures are OK
- Metrics help stimulate funding & desire for improvement
Ballista robustness testing approach
- Scalable, portable, reproducible
- Can include considerable state information (although that’s not obvious)
- Also applied to DoD HLA/RTI simulation backplane
- C++, call-backs, client/server, throws signals for exception handling
- Specifically written for robustness; has lower failure rates than OS code
- Internet-based testing service available