Experiments in software reliability - Life-critical applications

The paper discusses four reliability data gathering experiments which were conducted using a small sample of programs for two problems having ultrareliability requirements, n-version programming for fault detection, and repetitive run modeling for failure and fault rate estimation. The experimental results agree with those of Nagel and Skrivan in that the program error rates suggest an approximate log-linear pattern and the individual faults occurred with significantly different error rates. Additional analysis of the experimental data raises new questions concerning the phenomenon of interacting faults. This phenomenon may provide one explanation for software reliability decay. The fourth experiment underscored the difficulty in distinguishing between observations of deficiencies in the design of the algorithm and observations of software faults for real-time process control software. These experiments are a part of a program of serial experiments being pursued by the System Validation Methods of NASA-Langley Research Center to find a means of credibly performing reliability evaluations of flight control software.