Software System Reliability Analysis

Reliability is a crucial characteristic in safety-critical systems, yet it can only be measured late in the software development process when changes are difficult and costly. We propose a framework for estimating the risk of software that can be applied early in the software development life-cycle. This framework will use 21st century source code metrics--derived from advanced static analysis and symbolic simulation--coupled with machine learning techniques to identify software modules that are likely to contain faults. It also uses static analysis techniques to estimate the severity of a fault in a module. This framework will enable engineers to predict and resolve costly reliability problems long before the software is subject to system tests or deployment, allowing risks to be mitigated or eliminated early in the software development life-cycle.