Abstract
Open Source Software (OSS) products are widely used although a general consensus on their quality is far to be reached. Providing results on OSS reliability - as quality indicator – contributes to shed some light on this issue and allows organizations to make informed decisions in adopting OSS products or in releasing their own OSS. In this paper, we use a classical technique of Software Reliability Growth to model failures occurrences across versions. We have collected data from the bug tracking systems of three OSS products, Mozilla Firefox, OpenSuse and OpenOffice.org. Our analysis aims at determining and discussing patterns of failure occurrences in the three OSS products to be used to predict reliability behaviour of future releases. Our findings indicate that in the three cases, failures occurrences follow a predetermined pattern, which shows: a) an initial stage in which the community learns the new version b) after this first period a rapid increase of the failure detection rate until c) very few failures are left and the discovery of a new failure discovery is rare. This is the stage in which the version can be considered reliable.