[en] In this brave new world of smartphone-dependent society, dependability is a strong requirement and needs to be addressed properly. Assessing the dependability of these mobile systems is still an open issue, and companies should have the tools to improve their devices and beat the competition against other vendors. The main objective of this dissertation is to provide the methods to assess the dependability of mobile OS, fundamental for further improvements. Mobile OS are threatened mainly by traditional residual faults (when errors spread across components as failures), aging-related faults (when errors accumulate over time), and misuses by users and applications. This thesis faces these three aspects. First, it presents a qualitative method to define the fault model of a mobile OS, and an exhaustive fault model for Android. I designed and developed AndroFIT, a novel fault injection tool for Android smartphones, and performed an extensive fault injection campaign on three Android devices from different vendors to analyze the impact of component failure on the mobile OS. Second, it presents an experimental methodology to analyze the software aging phenomenon in mobile OS. I performed a software aging analysis campaign on Android devices to identify the impacting factors on performance degradation and resource consumption. Third, it presents the design and implementation of a novel fuzzing tool, namely Chizpurfle, able to automatically test Android vendor customizations by leveraging code coverage information at run-time.
