We have developed a novel technique for estimating soft-ware reliability in a Markov software usage model, taking
account of pre-information on failure probabilities. Specif-ically the procedure uses Importance Sampling technique
based on Ali-Silvey distance. The reliability estimator with
maximum precision is obtained from the test results. To show
the method's performance, we have applied it to estimate
the software reliability of Gutjahr's Markov usage model,
the good performance of the method has been demonstrated
experimentally.
However, the statistical testing models are modeled by
Markov usage models, and the number of nodes is not
too large to avoid state space explosion. This indicates
two obvious directions for future work: extension to non-Markov state models, and developing more efficient methods
for handling large state spaces such as parallel technique.
Furthermore, it may be possible to provide more solid
mathematical foundations, such as a proof of the convergence
of the parameter vector sequence.