CSR: SMALL: Robust Algorithms for an Open Source Software Reliability Tool

所有软件成功的关键是它的可靠性。该项目将开发一个开源软件可靠性工具，允许软件工程师自动应用软件可靠性模型，以帮助组织确保他们开发的软件应用程序可以无故障运行。传统算法在数值上不稳定，这意味着如果初始估计不准确，它们可能会失败。算法的失败使得不可能应用软件可靠性模型来做出有用的预测，例如，为了达到期望的可靠性水平，应该测试软件应用程序的额外时间量。这项研究的目标是开发数值稳定的算法，即使初始估计不准确也能成功。针对失效率和非齐次泊松过程(NHPP)软件可靠性模型，提出了期望最大化(EM)和期望条件最大化(ECM)算法。传统的EM算法施加了限制性的假设，将它们的应用限制在最简单的模型上，而ECM算法的潜力还没有被充分挖掘。因此，这一研究挑战将打破现有EM算法的限制，为软件可靠性模型设计高效的ECM算法。这些数值稳定的EM和ECM算法的实现将被合并到开放源码工具中，以确保软件可靠性模型可以成功应用。算法稳定性的增强和该工具的开源性质可能会促进量化软件可靠性模型的广泛使用，使公司和组织能够缩短软件产品的上市或现场开发时间。

The key to the success of all software is its reliability. This project will develop an open source software reliability tool that will allow software engineers to automatically apply software reliability models to help organizations ensure that software applications they develop can operate free of failures. Traditional algorithms are numerically unstable, meaning that they can fail if initial estimates are inaccurate. Failure of an algorithm renders it impossible to apply a software reliability model to make useful predictions such as the amount of additional time a software application should be tested in order to achieve a desired level of reliability. The goal of this research is to develop numerically stable algorithms that will succeed even if the initial estimates are inaccurate. Expectation maximization (EM) and expectation conditional maximization (ECM) algorithms will be developed for failure rate and nonhomogeneous Poisson process (NHPP) software reliability models. Traditional EM algorithms impose restrictive assumptions that limit their application to only the simplest models, while the potential of the ECM algorithm has not been fully explored. Therefore, this research challenge will remove the restrictions of existing EM algorithms and design efficient ECM algorithms for software reliability models. Implementations of these numerically stable EM and ECM algorithms will be incorporated into the open source tool to ensure that software reliability models can be applied successfully. The enhanced stability of the algorithms and the open source nature of the tool may promote widespread use of quantitative software reliability models, enabling companies and organizations to improve time to market or field a software product.