SOFTWARE-RELIABILITY MODEL - A SIMPLE ESTIMATOR FOR ERROR INTENSITY

This paper introduces an estimator for the error intensity of software packages after a period of testing, in which bugs are detected and removed concurrently. The estimator is the number of bugs detected in latter half of the testing period divided by the period's duration. It is simple in structure, consistent, asymptotically unbiased and insensitive to the number of bugs in the package and the distribution of bug's size. Its mean square error (or equivalently, sum of variance and square bias) is asymptotically better than or as good as another well-known estimator built for a case in which all bugs are removed at end of the testing period. Its application in stopping rule for software testing is illustrated and compared with another stopping rule known in the literature and found it is slightly better in performance in spite of its simple structure.