CRITIQUE OF THE JELINSKI-MORANDA MODEL FOR SOFTWARE RELIABILITY.

The paper discusses some problems associated with an early model for software reliability growth during debugging, first proposed by Jelinski and Moranda. It is suggested that the assumption of all faults contributing equally to the overall failure rate is overly naive and can be improved. Necessary and sufficient conditions are given for ML estimates of the model parameters to be finite. It is shown that there is always a finite probability that N equals infinity . Since other authors have shown from simulated data that ML estimates can be misleading even when finite, it is important that goodness-of-fit tests are performed. Such a test on one set of data shows the model to perform badly; an alternative model due to Littlewood and Verrallis is better