An Empirical Study on Failure Causes in a Commercial Off-the-Shelf Operating System

Empirical studies in software reliability research have predominantly focused on end-user applications. In this paper we present an empirical study on operating system reliability, which considered a sample of 7,007 real operating system failures collected from 566 computers. Our approach considered the clustering of different types of failures, as well as used an algorithm we created to search for and classify the causes of failures. We performed quantitative and qualitative analyses on the failure patterns and causes found, in order to investigate their correlations. The findings showed that failures in operating system services were dominant, especially related to software updates. We also found strong evidences of failure correlation, a property that is often neglected by studies in software reliability. In particular, we found evidences of causality relations between different types of operating system failures.