Supporting Development of Highly Dependable Software Through Continuous, Automated, In-process, and Individualized Software Measurement Validation

从本质上讲，高度可靠的软件是可预测的。例如，人们可以自信地预测软件在哪些环境下可以工作，在哪些环境下会失败。基于经验的创建可预测软件的方法基于两个假设：(1)历史数据可用于开发和校准生成经验预测的模型，以及(2)软件的内部属性（即具体属性，如规模、工作量和缺陷）和软件的外部属性（即抽象属性，如质量或故障时间）之间存在关系。软件度量验证是确定可用的内部属性和相应的有用的外部属性之间的预测关系的过程。本研究的总体目标是设计、实现和验证支持以下方面的软件度量：(1)易出错模块的识别，使质量保证资源的分配更加高效和有效；(2)通过增量的和特定于开发人员的通知和分析的增量软件开发方法。在喷气推进实验室任务数据系统项目中使用这些方法和措施的经验评估将推进软件测量验证的理论和实践。该研究还将产生一个用于软件测量验证的开源软件系统。

Highly dependable software is, by nature, predictable. For example, one can predict with confidence the circumstances under which the software will work and under which it will fail. Empirically-based approaches to creating predictable software are based on two assumptions: (1) historical data can be used to develop and calibrate models that generate empirical predictions, and (2) there exists relationships between internal attributes of the software (i.e. concrete attributes such as size, effort, and defects) and external attributes of the software (i.e. abstract attributes such as quality or time to failure.). Software measurement validation is the process of determining a predictive relationship between available internal attributes and correspondingly useful external attributes.The general objective of this research is to design, implement, and validate software measures that support: (1) identification of fault-prone modules, enabling more efficient and effective allocation of quality assurance resources, and (2) an incremental software development method through incremental and developer-specific notifications and analyses. Empirical assessment of these methods and measures during use on the Mission data System project at Jet Propulsion Laboratory will advance both the theory and practice of software measurement validation. The research will also produce an open source software system for software measurement validation.