软件可靠性混沌模型

摘要

在分析软件失效机理后认为：有些软件失效行为具有混沌性，所以可以用混沌方法来处理其软件可靠性推断问题.但在应用混沌方法前先要进行系统辨识，确定为混沌系统后，才能应用嵌入空间技术从软件失效时间序列重建系统相空间和吸引子，进而用吸引子所揭示的混沌属性来估计软件可靠性.文中在三个标准数据集的基础上对此进行了实证分析，结果表明其中两个数据集源于混沌机制，他们的吸引子具有低维的小数极限维数，而且预测与实际可靠性吻合较好.值得指出的是文中所提混沌方法突破了软件可靠性一贯使用随机分析的局限.%Computers affected almost every aspect of human lives. As thedependency on computer systems of human beings grows, so does the need for the technology of reliability of computer systems. In contrast to computer hardware, software is far more complicated. Thus the key is to improve the reliability of software if the overall reliability of a system is to be improved. Although scientists, in the past few decades, proposed lots of reliability models for software, which greatly enhanced the reliability and productivity of software products, these models are far from satisfactory. To build models of high accuracy and to improve the existing models is therefore of practical significance. Conventional theory of software reliability assumes that the failure processes of software are completely random, whereas authors of this paper, on the basis of careful investigation on physical mechanics of software failures,suggest that some dynamics of software failures are of chaotic features. Thus the reliability issue of these systems can be addressed with chaotic approaches. But before applying chaotic methodology to estimate the reliability of the software under consideration, the first thing to do is system identification that uses certain standards to distinguish chaotic dynamics from stochastic ones. In cases of chaos, the technology of embedding space is employed to reconstruct, from a time series of failures, the phase space and the attractor which reveals the chaotic properties that are further used to assess the reliability of the software product of interest. Based on three data sets including two of Musa's, the empirical study indicates that two of the data sets arise from chaotic dynamics rather than from stochastic ones, as the reconstructed attractors have low and limiting fractional dimensions. Interestingly, the predictions using chaotic methods in such cases fit quite well with actual reliabilities. This phenomenon is worth particular notice because it goes beyond the conventional limitation of stochastic assumptions of reliability analysis. To sum up, this paper proposed a chaotic model for software reliability. The validity, feasibility and applicability of the model are verified theoretically and experimentally.