Failure Risk Estimation via Markov Software Usage Models

A software usage models describes the prospective use of a program in its intended environment and allows the generation of random test cases leading to unbiased estimates of the failure risk, i.e., the expected loss by program failure. We concentrate on usage models of Markov type and show that by suitable changes of the probabilities of state transitions during test, the precision of the risk estimate can be optimized. An algorithm for the computation of optimal transition probabilities is presented, and experimental results based on a C++ implementation of this algorithm are reported.