PODS revisited-a study of software failure behaviour

Abstract

A description is given of an empirical study of the failure characteristics of software defects detected in the programs developed in the Project on Diverse Software (PODS). The results are interpreted in the context of a state machine model of software failure. The results of the empirical study case doubts on the general validity of the assumption of constant software failure probability and the assumption of constant software failure probability and the assumption that all defects have similar failure rates. In addition, an analysis of failure dependency lends support to the use of diversity as a means of minimizing the impact of design-level faults. Here, nonidentical faults exhibited coincident failure characteristics approximately in accord with the independence assumption, and some of the observed positive and negative correlation effects could be explained by failure masking effects, which can be removed by suitable design.<>